Ictd 2009 Proceedings

3rd International Conference on Information and Communication Technologies and Development

PROCEEDINGS April 17-19, 2009 Carnegie Mellon University in Qatar Education City, Doha, Qatar

Host Organization:

Organizing Partner:

Platinum Sponsors:

Gold Sponsors:

Bronze Sponsors:

Media Partner:

Technical Sponsors:

Table of Contents

Introduction

1

A Global Empirical Evaluation of New Communication Technology Use and Democratic Tendency Victoria Stodden and Patrick Meier

3

A Review of the Research on Mobile Use by Micro and Small Enterprises (MSEs) Jonathan Donner and Marcela Escobari

17

An Evaluation of the Use of ICT within Primary Education in Malawi David Hollow and Paola Masperi

27

Claim Mobile: Engaging Conflicting Stakeholder Requirements in Healthcare in Uganda Melissa R. Ho, Emmanuel K. Owusu and Paul Aoki

35

Computer Games in the Developing World: The Value of NonInstrumental Engagement with ICTs, or Taking Play Seriously Beth E. Kolko and Cynthis Putnam

46

Content Creation and Dissemination by-and-for Users in Rural Areas Sheetal K. Agarwal, Arun Kumar, Amit Anil Nanavati, Nitendra Rajput

56

E for Express: “Seeing” the Indian State through ICTD Renee Kuriyan and Isha Ray

66

Evaluating the Accuracy of Data collection on Mobile Phones: A Study of Forms, SMS, and Voice Somani Patnaik, Emma Brunskill and William Thies

74

FOLKSOMAPS – Towards Community Driven Intelligent Maps for Developing Regions Arun Kumar, Dipanjan Chakraborty, Himanshu Chauhan

85

HIV Health Information Access using Spoken Dialogue Systems: Touchtone vs. Speech Aditi Sharma Grover, Madelaine Plauché, Etienne Barnard, Christiaan Kuun

95

ICT4What? – Using the Choice Framework to Operationalize the Capability Approach to Development Dorothea Kleine

108

ICTD for Healthcare in Ghana: Two Parallel Case Studies Rowena Luk, Matei Zaharia, Melissa Ho, Brian Levine and Paul M. Aoki

118

Improving Child Literacy in Africa: Experiments with an Automated Reading Tutor G. Ayorkor Mills-Tettey, Jack Mostow, M. Bernadine Dias, Tracy Morrison Sweet, Sarah M. Belousov, M. Frederick Dias, Haijun Gong

129

Improving Literacy in Rural India: Cellphone Games in an After-School Program Matthew Kim, Anuj Kumar, Shirley Jain, Akhil Mathur, and John Canny

139

Kelsa+: Digital Literacy for Low-Income Office Workers Aishwarya Lakshmi Ratan, Sambit Satpathy, Lilian Zia, Kentaro Toyama, Sean Blagsvedt, Udai Singh Pawar, Thanuja Subramaniam

150

Mapping the Dynamics of Social Enterprises and ICTD in Cambodia Kelly Hutchinson and Alemayehu Molla

163

Political Incentives and Policy Outcomes: Who Benefits from TechnologyEnabled Service Centers? Jennifer Bussell

173

Results from a Study of Impact of E-government Projects in India Subhash C. Bhatnagar and Nupur Singh

183

The Contribution of User-Based Subsidies to the Impact and Sustainability of Telecenters – the eCenter Project in Kyrgyzstan Michael L. Best, Dhanaraj Thakur and Beth E. Kolko

192

Poster Papers

201

A Speech Enabled Indian Language Text-to-Braille Transliteration System Tirthankar Dasgupta and Anupam Basu

201

Analyzing Statistical Relationships between Global Indicators through Visualization Prabath Gunawardane, Erin Middleton, Suresh Lodha, Ben Crow and James Davis

212

ATMosphere: A System for ATM Microdeposit Services in Rural Contexts Michael Paik and Lakshminarayanan Subramanian

222

Building a Transportation Information System Using Only GPS and Basic SMS Infrastructure Ruth E. Anderson, Anthony Poon, Caitlin Lustig, Waylon Brunette, Gaetano Borriello, Beth E. Kolko

233

Challenges in Health Information Systems Integration: Zanzibar Experience Edwin Nyella

243

Cross Technology Comparison for Information Services in Rural Bangladesh Faheem Hussain and Rahul Tongia

252

Decentralization, Clientelism and Popular Participation - Is There a Role for ICTs to Improve Local Governance? Björn-Sören Gigler

267

Design and Deployment of a Blood Safety Monitoring Tool S. Thomas, A. Asuntogun, J. Pitman, B. Mulenga, and S. Vempala

280

Dimensions of IT Literacy in an Arab Region Study in Barkha(Oman) Sherif M. Aziz

288

Emergency Communication and System Design: The Case of Indian Ocean Tsunami R. Chen, J. Coles, J. Lee, and F. R. Rao

300

Empowering Muslim Youth through Computer Education, Access, Use: A Gender Analysis Farida Khan and Rehanan Ghadially

310

eServices Provisioning in a Community Development Context through a JADE MAS Platform Mamello Thinyane, Alfredo Terzoli, Peter Clayton

320

Extending the Technology-Community-Management Model to Disaster Recovery in Asia Arul Chib and A.L.E. Komathi

328

Featherweight Multimedia for Information Dissemination Gerry Chu, Sambit Satpathy, Kentaro Toyama, Rikin Gandhi, Ravin Balakrishnan, S. Raghu Menon

337

ICT Governance in Higher Education: Case Study of the Rise and Fall of Open Source in a gulf University Sofiane M. Sahyraoui

348

ICTD State of the Union: Where Have We Reached and Where Are We Headed? Rabin Patra, Joyojeet Pal, Sergiu Nedevschi

357

Information Communication Techology and Sustainable Communities in Africa: The Case of the Niger Delta Region of Nigeria (Feb. 2009) Uduak A. Okon

367

Integrating Health Information Systems in Sierra Leone Johan Sæbø, Edem Kwame Kossi, Romain Tohouri Golly-Kobrissa, Ola Titlestad, Jørn Braa

379

Mobile Telephony Access and Usage in Africa A. Chabossou, C. Stork, M. stork, Z. Zahonogo

392

Numeric Paper Forms for NGOs Gusharan Singh, Leah Findlater, Kentaro Toyama, Scott Helmer, Rikin Gandhi, Ravin Balakrishnan

406

Rajnikant’s Laptop: Computers and Development in Popular Indian Cinema Joyojeet Pal

417

Regulatory Independence and Wireless Market Development: A Comparative Analysis of Two African Nations Annemijn F. van Gorp and Carleen F. Maitland

427

Social Enterprises: A Vocational Entrepreneurship Framework for Street Youth Paul Javid, Kentaro Toyama, Manna Biswas

437

Speech vs. Touch-Tone: Telephony Interfaces for Information Access by Low Literate Users Jahanzeb Sherwani, Sooraj Palijo, Sarwat Mirza, Tanveer Ahmed, Nosheen Ali, Roni Rosenfeld

447

The Case for SmartTrack Michael Paik, Ashlesh Sharma, Arthur Meacham, Giulio Quarta, Philip Trahanas, Brian Levine, Mary Ann Hopkins, Barbara Rapchak, Lakshminarayanan Subramanian

458

Uses of Mobile Phones in Post-Conflict Liberia Michael L. Best, Edem Wornyo, Thomas N. Smyth and John Etherton

468

Demos

478

An Automated Braille Writing Tutor with Multilingual Exercises and Educational Games M. Bernadine Dias, M. Freddie Dias, Sarah Belousov, Mohammed Kaleemur Rahman, Saurabh Sanghvi, Imran Fanaswala, Wael Ghazzawi, Ameer Abdulsalam, Noura El-Moughny, and S. Raghu Menon

478

Boosting European Market Access to Malian Mango Growers Saskia Harmen

479

Creating a Mobile-Phone Based Geographic Surveillance System for Asian Flu Yibo Lin and Claire Heffernan

480

Design of a Blood Flow System A. Osuntogun, S. Thomas, J. Pitman, S. Basavaraju, B. Mulenga and S. Vempala

481

DISHA: DISease and Health Awareness for Children on Multiple Input Devices Mohit Jain, Aakar Gupta, Navkar Samdaria, Praveen Shekhar and Joyojeet Pal

482

Freedom Fone: Dial-up Information Service Bev Clark and Brenda Burrell

483

FrontlineSMS and Ushahidi – A Demo Ken Banks and Erik Hersman

484

Global Youth Connectivity (GYC) – Using ICT for Peaceful Recovery and Long-term Change Anne Bertrand

485

Implementing E-Government Accessible to Illiterate Citizens D. Kettani and A. El Mahdi

486

Improving Data Quality with Dynamic Forms Kuang Chen, Harr Chen, Neil Conway, Heather Dolan, Joseph M. Hellerstein, and Tapan S. Parikh

487

IWB4D – Interactive Whiteboards for Development John Traxler and Lee Griffiths

488

Livestock, Learning and Diagnostics: New Directions in Veterinary Tele-medicine Jun Yu and Claire Heffernan

489

Metamouse: Multiple Mice for Legacy Applications Kurtis Heimerl, Divya Ramachandran, Joyojeet Pal, Eric Brewer, and Tapan Parikh

490

Mobile Phone Job Services: Linking Developing-country Youth with Employers, via SMS Amber Houssain, Mohammad Kilany, and Jacob Korenblum

491

MultiMath: Numeric Keypads for Math Learning on Shared Personal Computers Sunil Garg, Charlotte Robinson, Clint Tseng, Heather Underwood, Richard Anderson, Joyojeet Pal

492

A New Generation of Open Source Data Collection Tools Yaw Anokwa, Carl Hartung, Adam Lerer, Brian DeRenzi, Gaetano Borriello

493

RuralScope: An Information System for Tracking Rural Disbursements Sai Gopal Thota, Rabin Ratra, Murali Medisetty, Sivananda Reddy, Vivek Mungala, Joyojeet Pal

494

T-Cube Web Interface in support of Real-Time Bio-surveillance Program Artur Bubrawski, Maheshkumar Sabhnani, Michael Knight, Michael Baysek, Daniel Neill, Saswati Ray, Anna Michalska and Nuwan Waidyanatha

495

Web Search over Low Bandwidth Jay Chen, Lakshminarayanan Subramanian, and Jinyang Li

496

Author Index

497

ICTD 2009 – Organizing Committee Honorary Chairs Dr. Hessa Sultan Al-Jaber Secretary General ictQATAR

Dr. Kentaro Toyama Assistant Managing Director Microsoft Research India (MSRI)

Conference Chair Dr. M. Bernardine Dias Carnegie Mellon University

Program Committee Chairs Dr. Richard Heeks University of Manchester

Dr. Rahul Tongia Carnegie Mellon University

Advisory Board Dr. Vallampadugai S. Arunachalam Center for Study of Science, Technology, and Policy (CSTEP), Bangalore

Dr. Francois Bar University of Southern California

Dr. Michael Best Georgia Institute of Technology

Dr. Kenneth Keniston Massachusetts Institute of Technology

Dr. Balaji Parthasarathy International Institute of Information Technology (IIIT), Bangalore

Dr. Krithi Ramamritham Indian Institute of Technology (IIT), Bombay

Dr. Raj Reddy Carnegie Mellon University

Dr. AnnaLee Saxenian University of California, Berkeley

Dr. Kentaro Toyama Microsoft Research India

Dr. Ernest Wilson University of Southern California

Senior Program Committee Members Dr. Francois Bar University of Southern California

Dr. Michael Best Georgia Institute of Technology

Dr. Eric Brewer University of California, Berkeley

Dr. Chris Coward University of Washington

Dr. Robert Davidson City University of Hong Kong

Dr. Hernan Galperin Universidad de San Andrés

Dr. Shirin Madon London School of Economics

Dr. Alemayehu Molla Royal Melbourne Institute of Technology

Dr. Balaji Parthasarathy International Institute of Information Technology (IIIT)-Bangalore

Dr. Krithi Ramamritham Indian Institute of Technology (IIT)-Bombay

Dr. Kentaro Toyama Microsoft Research India

Dr. Tim Unwin Royal Holloway University of London

Panels and Workshops Chairs Dr. Joseph Mertz Carnegie Mellon University

Dr. Joyojeet Pal University of Washington

Publications Chairs Dr. Yonina Cooper Carnegie Mellon University in Qatar

Dr. Thrishantha Nanayakkara Harvard University and University of Moratuwa, Sri Lanka

Poster Chairs Dr. Faheem Hussain Carnegie Mellon University in Qatar

Dr. Tapan Parikh University of California, Berkeley

Demo Chairs Dr. Khaled Harras Carnegie Mellon University in Qatar

Bill Thies Massachusetts Institute of Technology

Regional Chairs

Africa Dr. Gary Marsden University of Capetown, South Africa

Dr. Tim Waema University of Nairobi, Kenya

Middle East Dr. Adnan Abu Dayya Qatar University

Dr. Fouad Mrad American University of Beirut

East Asia Dr. Jack Linchuan Qiu The Chinese University of Hong Kong

Rinalia Abdul Rahim Global Knowledge Partnership, Malaysia

South Asia Anita Gurumurthy IT for Change, India

Dr. Umar Saif Lahore University of Management Sciences, Pakistan

South and Central America Dr. Nicolau Reinhard Universidade de São Paulo, Brazil

Dr. Osvaldo Rodriguez La Planta University, Argentina

North America Dr. John Bennett University of Colorado at Boulder

Claudia Morrell Multinational Development of Women in Technology

Europe John Traxler University of Wolverhampton, UK

Oceania Christina Higa University of Hawaii

Dr. Esther Batiri Williams The University of the South Pacific

Technical Program Committee Members Dr. Jessica Aalami University of California, Berkeley

Dr. Salam Abdallah Abu Dhabi University

Dr. Reuben Abraham Indian School of Business, Hyderabad

Dr. Erwin Alampay University of Philippines

Dr. Richard Anderson University of Washington

Dr. Peng HwaAng Nanyang Technological University

Akhtar Badshah Microsoft

Dr. V. Balaji ICRISAT

Dr. Anupam Basu Indian Institute of Technology (IIT)-Kharagpur, India

Dr. John Bennett University of Colorado, Boulder

Dr. Subhash Bhatnagar Indian Institute of Management, Ahmedabad

Robert Bichler University of Salzburg

Dr. John Canny University of California, Berkeley

Dr. Royal Colle Cornell University

Dr. Rahul De Dr. Andy Dearden Indian Institute of Management, Bangalore Sheffield Hallam University Dr. Uday Desai Indian Institute of Technology (IIT)-Bombay

Dr. Antonio Diaz University of Auckland

Dr. Jonathan Donner Microsoft Research

Dr. Ayman Elnaggar Sultan Qaboos University

Dr. Kevin Fall Intel/Berkeley

Dr. Ping Gao University of Manchester

Dr. Alison Gillwald University of Witwatersrand

Dr. Gillian Green University of Bolton

Pat Hall Kathmandu University

Dr. Saskia Harmsen International Institute for Communication and Development (IICD)

Dr. Claire Heffernan University of Reading

Dr. Bill Hefley Carnegie Mellon University

Dr. Heather Hudson University of San Francisco

Dr. Faheem Hussain Carnegie Mellon University in Qatar

Mahad Ibrahim University of California, Berkeley

Dr. Ashok Jhunjhunwala Indian Institute of Technology (IIT)-Madras

Dr. Muhammadou M. O. Kah American University of Nigeria

Matt Kam University of California, Berkeley

Dr. Sherif Kamel American University of Cairo

Dr. Atreyi Kankanhalli National University of Singapore

Dr. Srinivasan Keshav University of Waterloo

Dr. G. R. Kiran London School of Economics

Dr. Dorothea Kleine University of London

Dr. Jim Koch Santa Clara University

Dr. Beth Kolko University of Washington

Richa Kumar Massachusetts Institute of Technology

Dr. Gillian Marcelle WITS University

Dr. Victor Mbarika Southern University and A&M College

Dr. Wagner Meira UFMG – Federal University at Minas Gerais

Dr. Michel Menou City University of London

Dr. Harekrishna Misra Institute of Rural Management, Anand

Dr. Amit Mitra Cranfield University

Dr. Beda Mutagahywa University of Dar es Salaam

Dr. Shrikant Naidu Motorola Labs, India

Dr. Amit Nanavati IBM

Dr. Solomon Negash Kennesaw State University

Dr. Joyojeet Pal University of Washington

Dr. Tapan Parikh University of California, Berkeley

Dr. Fay Payton North Carolina State University

Francisco Proenza Food and Agriculture Organization of the United Nations (FAO)

Dr. Ranjini Raghavendra Dr. Nicolau Reinhard Indian Institute of Management, Bangalore University of Sao Paolo Marijn Rijken TNO (Netherlands Organisation for Applied Scientific Research)

Dr. Osvaldo Rodriguez La Planta University

Dr. Roni Rosenfeld Carnegie Mellon University

Dr. Sofiane Sahraoui American University of Sharjah

Dr. Tony Salvador Intel

Dr. Maung Sein University of Agder

Dr. Afzal Sher Swedish Program on ICT in Developing Regions (SPIDER)

Jahanzeb Sherwani Carnegie Mellon University

Dr. Nirvikar Singh University of California, Santa Cruz

Dr. Hettie Soriyan Obafemi Awolowo University

Dr. Christoph Stork Wits University

Dr. Eswaran Subrahmanian CMU/NIST/CSTEP

Dr. Lakshmi(narayanan) Subramanian Bill Thies Courant Institute of Mathematical Sciences Massachusetts Institute of Technology Dr. Cathy Urquhart University of Auckland

Dr. Carinade Villiers University of Pretoria

Dr. Timothy Waema University of Nairobi

Dr. Ernest Wilson University of Southern California

Dr. Adel El Zaim International Development Research Centre (IDRC) Canada

Local Organizing Committee Chairs Murry Evans Carnegie Mellon University in Qatar

Elaine Farah ictQatar

Erin Stewart Carnegie Mellon University in Qatar

Publicity Coordinators Noha Al Afifi Carnegie Mellon University in Qatar

Andrea Zrimsek Carnegie Mellon University in Qatar

Logistics Coordinators Renee Barcelona Carnegie Mellon University in Qatar

Sarah Belousov Carnegie Mellon University

Kara Nesimiuk Carnegie Mellon University in Qatar

Ermine Teves Carnegie Mellon University

Website Coordinators M. Freddie Dias Carnegie Mellon University

Daniel Freeman Carnegie Mellon University

Local Organizing Committee Members Ray Corcoran, Bob Gaus, Jim Gartner, Shamila Khader, Aaron Lyvers Carnegie Mellon University in Qatar

1

Introduction to the Proceedings of ICTD2009 Welcome to the 3rd International Conference on Information and Communication Technologies and Development (ICTD2009). It is with great pleasure that we present the ICTD2009 proceedings, which include all of the full papers presented at the conference in Doha, Qatar, held on 17-19 April 2009. ICTD is the premier series of scholarly conferences on the use of ICTs for development, spanning technical and social science domains. The call for papers attracted a record 250 submissions. All papers were put through a double-blind peerreview process. The Program Co-Chairs assigned papers to our Senior Program Committee members who oversaw a review process involving three Program Committee reviewers per paper: one with deep expertise about the subject matter; another with broad background in the area; and one drawn from an altogether different discipline. Our continuing hope is that this encourages a convergence of vocabulary and ideas within the ICTD field, while maintaining the integrity of different disciplines. The Senior Program Committee members then meta-reviewed the papers. Authors were allowed a brief rebuttal to reviewer comments before final acceptance decisions were made and revisions were finalized. Ultimately, 19 papers were selected for oral presentation, and another 27 papers were chosen as full papers for poster presentation; an acceptance rate of just over 18%. These papers represent some of the best work being done in ICTD today. They focus on a wide variety of development goals, and involve a broad and innovative range of digital technologies. They draw from all continents of the global South, and focus on all stages of the ICTD lifecycle: from readiness through design and adoption to use and impact. They also tell us about all levels, from the individual through communities and projects to ICTD programmes and policies. We hope that you will find them an insightful, provocative, and informative contribution to our fast-growing field of research and practice. We wish to thank our keynote speakers William H. Gates, Chairman of Microsoft Corporation and Co-Chairman of the Bill and Melinda Gates Foundation, and Dr. Carlos A. Primo Braga, Director of Economic Policy and Debt in the Poverty Reduction and Economic Management Network (PREM) at The World Bank for their insightful presentations. All the other conference presenters also deserve our gratitude for the variety of content and insights they added to the conference program. We also need to thank a number of people without whom the program could not have been put together. First, we wish to thank our Honorary Chairs, Dr. Hessa Sultan AlJaber, Secretary General of The Supreme Council of Information and Communication Technology (ictQatar), and Dr. Kentaro Toyama, Assistant Managing Director of Microsoft Research India (MSRI) for their tremendous support in making this conference a success. We are of course deeply indebted to our Program Committee: those who did the hard work of reviewing and our senior PC members who managed the review process so effectively; all together it is they who ultimately steer the course of ICTD. We thank the Regional Chairs for their assistance in promoting the conference, and our Advisory Board which provided guidance and moral support. We also thank the Publication Chairs, Yonina Cooper and Thrishantha Nanayakkara, who

2

made this proceedings possible; and Faheem Hussain and Tapan Parikh, who carried the load in organisation of the poster presentations. The conference program for ICTD2009 went well beyond papers, and was significantly enhanced by demonstrations, organised by Bill Thies and Khaled Harras, and by a series of panels and workshops, organised by Joe Mertz and Joyojeet Pal. We are very grateful for their input and hard work. This conference would not have been possible without the tireless efforts of a number of organizers and volunteers, notably our local organizing committee chairs Elaine Farah, Murry Evans, and Erin Stewart, and the many others on the local organizing committee, especially Dean Charles Thorpe, Sarah Belousov, Ermine Teves, Renee Barcelona, Kara Nesimiuk, Andy Zrimsek, Noha Al Afifi, Shams Hassan, Aaron Lyvers, Shamila Khader, Ray Corcoran, Bob Gaus, Freddie Dias, Daniel Freeman, and Jim Gartner. If there are others we did not name explicitly, it is our lapse. The success of the conference is in part due to our many sponsors and partners. We are extremely grateful to our organizing partner, The Supreme Council of Information and Communication Technology (ictQatar), to our media partner Al Jazeera Children’s Channel (JCC), to our technical sponsors IEEE and ACM, and to our financial sponsors, the Qatar National Research Fund (Platinum Sponsor); Canada’s International Development Research Centre (Platinum Sponsor); Qatar Telecom (Platinum Sponsor); ExxonMobil (Gold Sponsor); Microsoft Corporation (Gold Sponsor); IBM (Bronze Sponsor); the Computer Science program at Carnegie Mellon University in Qatar (Bronze Sponsor); and other sponsors who asked not to be publicly acknowledged. Finally, we are indebted to Carnegie Mellon University in Qatar for hosting the conference at their campus in Education City, and to the numerous employees of Carnegie Mellon University (in both the Pittsburgh and Doha campuses), especially the TechBridgeWorld team, who truly made this conference a success. Thanks to everyone who contributed to ICTD2009 in so many ways, and thank you for participating! We are confident you will find the following papers, abstracts, and information not only interesting and useful, but the seeds for further research, innovation, and developmental impact. Richard Heeks, University of Manchester Rahul Tongia, Carnegie Mellon University Program Chairs

M. Bernardine Dias, Carnegie Mellon University Conference Chair

3

A Global Empirical Evaluation of New Communication Technology Use and Democratic Tendency Victoria Stodden

Patrick Meier

Berkman Center for Internet and Society Harvard Law School Cambridge, MA 02138 [email protected]

Berkman Center for Internet and Society Harvard Law School Cambridge, MA 02138 [email protected]

Abstract—Is the dramatic increase in Internet use associated with a commensurate rise in democracy? Few previous studies have drawn on multiple perception-based measures of governance to assess the Internet’s effects on the process of democratization. This paper uses perception-based time series data on “Voice & Accountability,” “Political Stability,” and “Rule of Law” to provide insights into democratic tendency. The results of regression analysis suggest that the level of “Voice & Accountability” in a country increases with Internet use, while the level of “Political Stability” decreases with increasing Internet use. Additionally, Internet use was found to increase significantly for countries with increasing levels of “Voice & Accountability.” In contrast, “Rule of Law” was not significantly affected by a country’s level of Internet use. Increasing cell phone use did not seem to affect either “Voice & Accountability,” “Political Stability” or “Rule of Law.” In turn, cell phone use was not affected by any of these three measures of democratic tendency. When limiting our analysis to autocratic regimes, we noted a significant negative effect of Internet and cell phone use on “Political Stability” and found that the “Rule of Law” and “Political Stability” metrics drove ICT adoption. Index terms—cell phone, democracy, fixed effects model, ICT, internet

I. I NTRODUCTION Does the globalization of the Internet have a democratizing effect? The question has already been posed by numerous studies but these have largely taken the form of qualitative case studies and/or large theoretical analyses. In terms of a rigorous, quantitative establishment of the democratization effects of the Internet, however, the jury is still out [1]. At the heart of this debate, moreover, lies a more fundamental question about the essence of democracy. In fact, “unless we are clear about what democracy means to us, and what kind of democracy we envision, technology is as likely to stunt as to enhance the civic polity” [2]. The purpose of this paper is to contribute more rigorous data-driven analysis to the literature on Internet and democracy since “there is no doubt that rigorous and datadriven analysis of this relationship will benefit scholars and policymakers alike” [1]. Previous research on the topic of Internet and democracy can be characterized as lacking a serious perusal of the

democracy and regime transitions literature. To be sure, “the trouble with the zealots of technology as an instrument of democratic liberation is not that they misconceive technology but that they fail to understand democracy” [2]. ‘In other words, “it turns out there is no simple general answer to the question: Is the technology democratizing?’ until we have made clear what sort of democracy we intend.” We address this question first before proceeding with a more detailed literature review. Barber’s notion of “strong democracy” comprises the careful and prudent judgment of citizens who participate in deliberative, self-governing communities. Schmitter and Karl write that, “modern political democracy is a system of governance in which rulers are held accountable for their actions in the public realm by citizens, acting indirectly through the competition and cooperation of their elected representatives”[3]. The two authors emphasize that citizens are the most distinctive element in democracies. “All regimes have rulers and a public realm, but only to the extent that they are democratic do they have citizens” [3]. In contemporary studies of democracy and particularly in pluralist theory, “a vibrant civil society is usually regarded as an essential for good governance and effective democratic consolidation” [4]. In other words, regular elections are not sufficient. As Zakaria noted, illiberal democracies have free elections but citizens remain cut off from real power due to the lack of civil liberties [5]. Huber et al. write that the most basic feature of democracy is power sharing [6]. They identify three clusters of power as primarily relevant for the chances of democracy: (1) the balance of power in civil society; (2) the balance of power between state and society; and (3) the transnational balance of power that shape the first two and constrain political decisionmaking. By remaining diverse and independent of the state, political participation by civil society acts as a channel of public voice and accountability, and a way of challenging and checking the unbridled power of authoritarian regimes [4]. The structure of state-society relations is equally relevant for democracy. As Huber et al. note, “the power of the state needs to be counterbalanced by the organizational strength of

4

the civil society to make democracy possible; the state must not be so strong and autonomous from all social forces as to overpower civil society and rule without accountability.” Clearly then a governing body that fails to follow the “rule of law,” should not be considered democratic [3]. These elements of democracy are not sufficient conditions for a stable democracy, but they are necessary and indispensable to the persistence of democratic governance. The italicized terms above represent the fundamentals behind the sort of democracy we intend: active citizen participation, good governance, accountability, power sharing, balance of power and rule of law. The few quantitative studies that do exist on Internet and democracy tend to aggregate these fundamentals of democracy into a single index. Doing so means these lose important information on how these individual components of democracy may be affected by the growing prevalence of global Internet access. Furthermore, past quantitative and qualitative studies tend to focus primarily on the impact of the Internet on established democracies. They also focus on the 1990s almost exclusively, a serious limitation that remains surprisingly understated in the literature. Equally problematic in the current literature is the interchangeable use of the terms “Internet” and “information revolution.” The terms are purposefully not differentiated on the basis that the predominant feature of the information society is the spread of the Internet. While this is true of Western democracies, it is certainly not true for the majority of developing, nondemocratic countries, where cell phones are the most widely spread communication technology after radios [7]. Indeed, the irony is that “those who might most benefit from the net’s democratic and informational potential are least likely to have access to it, the tools to gain access, or the educational background to take advantage of the tools” [2]. This paper seeks to redress each of these shortcomings. First, since the boundaries of the term “democracy,” and how it is measured, is subject to lively debate, we use multiple perception-based measures of governance for our dependent variables. Governance indicators provide a better set of proxies for the sort of democracy we intend as identified above. We therefore draw on the following three World Bank indicators: (1) Voice and Accountability (VA) measuring perceptions of the extent to which a country’s citizens are able to participate in selecting their government, as well as freedom of expression, freedom of association, and a free media; (2) Rule of Law (RL) measuring perceptions of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence; and (3). Political Stability and Absence of Violence (PS) measuring perceptions of the likelihood that the government will be destabilized or overthrown by unconstitutional or violent means, including politically motivated violence and terrorism. These metrics are drawn from the World Bank Governance Indicators Research Database (see http://info.worldbank.org/governance/wgi/). Second, we draw on data from 2000 through 2006, a time

when Internet access and cell phone use is significantly more prevalent and globalized than in the 1990s. If a statistically significant relationship between Internet and democracy does exist, then it is more likely to manifest itself now and not in the 1990s. Third, we draw on both Internet and cell phone data per 100 inhabitants per 181 countries to assess the impact of the information revolution on democratization. We use regression analysis to determine whether Internet or cell phone use has had a correlative effect on measures of democratic tendency. We also model whether the collection of democratic measures has had a correlative effect on Internet or cell phone use. The paper is structured as follows: the first section reviews the current debate and literature on Internet and democracy. The second section explicates the datasets used in this study while the third section formalizes the statistical models employed in the regression analysis. Section four reviews the results and provides an interpretation of the findings. The fifth and final section concludes the study. II. L ITERATURE R EVIEW The Internet and democracy literature comprises two distinct schools of thought each comprising a host of qualitative research and some quantitative inquiry. In this section we review in some detail the qualitative and quantitative literatures that have contributed to both schools of thought over the past ten to fifteen years. In so doing, we compare and contrast the main arguments along with the respective findings. As noted in the introduction, one common shortcoming of the Internet and democracy literature is the tendency to oversimplify our understanding of democracy. The purpose of this literature review is thus to redress this gap evident in previous studies. The first school of thought is often referred to as the more populist school of thought. This strand of the literature subscribes to the viewpoint that the Internet has democratic relevance and impact [8], [9], [10], [11]. According to these authors, the Internet will decentralize access to communication and information while increasing citizen access [12]. Best and Wade write that “the Internet’s collective characteristics (e.g., low cost, multidirectional capability, etc.), helps make this possible.” We first review the qualitative literature that comprises this school of thought followed by quantitative studies. Dahl previously observed that telecommunications technologies have a key role in making possible the advanced democratic country, where policy is firmly anchored in the judgment of the “demos” [13]. In his list of the procedural minimal conditions that must be present for modern political democracy to exist, Dahl thus argues that citizens should have the right to seek out alternative sources of information. Rheingold has dubbed the Internet as “the great equalizer” because it can “equalize the balance of power between citizens and power barons” [14]. The idea here “is that the Internet will serve as a mass audience, and will politicize them in the process” [12]. Anderson et al. draw on sociological research to show that electronic networks lead to the “break-down of status-based social structures” and “increased participation in

5

discussion, decision-making, and task processes by those who typically are political or economically disadvantaged” [15]. Etzioni sees in the Internet the possibility of an advancement of the state of public affairs through “teledemocracy” [16]. Other scholars claim that the Internet will “enable a Jeffersonian revolution” [17]. Grossman argues that a “third epoch of democracy is arriving by the hand of technology,” and claims that a “new, hybrid electronic republic’ is now forming to displace the creation of Montesquieu, Locke, Madison and their contemporaries” [18], [12]. The practice of “electronic democracy,” according to Browning, will differ substantially from previous renderings of democracy [19]. Perhaps the most provocative claim associated with this more populist school of thought is the one made by Negroponte, who argues that the nation-state will evaporate as a consequence of the information revolution [20]. Snider suggests that citizens need only be potentially informed in order to hold government accountable [21]. Barber opines that by favoring decentralization, the multiplication of choice, and consumer sovereignty, new technologies such as the internet or cell phone have already, albeit inadvertently, benefited democratic political culture [2]. According to Barber, “democracy is a form of government that depends on information and communication. It is obvious then that new technologies of information and communications can be nurturing to democracy. They can challenge passivity, they can enhance information equality, they can overcome sectarianism and prejudice, and they can facilitate participation in deliberative political processes.” Hill and Hughes argue that those who subscribe to the populist school have reasons to be optimistic: “If the mere fact that political discourse against repressive governments is taking place is a good in itself, then the utopians have reason to celebrate [22]. Perhaps the Internet will bring about a wider democratic revolution in the world” [22]. Bimber is more cautious, arguing that while the Internet is accelerating the process of issue group formation and action (in America), the structure of political power has not been revolutionized or qualitatively transformed into a new epoch of democracy [12]. According to Bimber, other scholars believe that the Internet has a “transformative potential” because it facilitates a kind of “one-to-one interaction among citizens and between citizens and government.” Along these lines, some scholars such as Corrado and Firestone write that the Internet has the potential to promote “unmediated” communication and thereby decrease citizens’ reliance on officials and organizations [23]. In sum, what distinguishes the populist enthusiasm for the Internet is the “idea that elites and political intermediaries will grow less important” [12]. In the more contemporary, qualitative literature, Steele and Stein, argue that the Internet amplifies trends in international relations [24]. Rosenau and Johnson address the impact of the Internet at both the individual and international level [25]. At the individual level, the authors argue that the Internet can be used as a tool to organize collectively to effect social and political change around the world (see http://www.DigiActive.org,

for example). At the international level, the authors make the bold claim that the Internet has “contributed to the rise of a more multicentric world structure in which nation-states have seen their preeminence lessen and non-governmental actors take the stage” [1]. The salient point here is that groups and individuals can far more efficiently form coalitions of consequence with a range of powerful collectives. As Best and Wade rightly note, there are obvious democratic elements to this, including the need for “nation-states to provide democratic rights to their citizens so as to build legitimacy on the global stage” [1]. We now turn to the quantitative studies that comprise the first school of thought. One of the earlier statistical studies on this side of the literature was carried out by Kedzie, who provides an account of how information communication technologies contributed to the “third wave” of democracy [26]. Prior to the fall of the Soviet Union and the proliferation of new democracies in Eurasia, the mainstream theory of democratization held that democracy followed economic growth and development [27]. To be sure, one of the few robust findings in the literature is that democracy is more likely in more developed countries [28], [29], [30]. Longdregan and Poole have also shown that the most significant predictor of transitions to authoritarianism is poverty [31], [32]. “In short, after 20 years of observation and analysis during the third wave of academic interest in democratization, we can be reasonably certain that a positive relationship between development and democracy exists, though we do not know why” [33]. Kedzie, however, was more interested in testing another potential causal mechanism, the “dictator’s dilemma” hypothesis, which suggests that the globalization of markets places pressure on authoritarian regimes to keep their countries’ communication borders open. He reasoned that the ensuing massive flow of information would not only allow for “the efficient passage of commercial information, but also for more ’democratic’ information” [1]. As Bimber observes, the most important predictions about the Internet’s impact on politics amount to “causal claims regarding the effect of information flow on political participation and the organization of interests” [12]. Other scholars have made related arguments. Webster, for example, writes that the Internet has helped to facilitate a new form of capitalism called “information capitalism” in which global labor markets require highly flexible workers who continuously adapt and learn [34]. Regimes that impose restrictions on information capitalism forgo the financial returns possible by tapping into the information economy [35], [36], [37]. In his study, Kedzie employs regression analysis to compare how much of the variation in democracy is explained by both traditional predictors of democracy and the strength of Internet diffusion by drawing on data from 144 countries [26]. For his set of control variables, Kedzie included more traditional predictors of democracy including economic development, education, human development and health. He also included indicators of pre-Internet information communication technologies (ICTs). His results suggest that the Internet is indeed

6

a strong predictor of democracy, more so than traditional determinants of democracy. In a follow up study, Richards assessed the relationship between the Internet and physical integrity [38]. His findings support Kedzie’s. However, the latter study faces an important limitation since Kedzie’s (rather simple) longitudinal analysis draws on data from 1993. At this point during the early 1990s, the Internet was hardly globalized. Best and Wade recognize this important short coming in Kedzie’s study and therefore explore the global effect of the Internet on democracy over a ten year period, 1992−2002 [1]. They aggregate political and civil rights data from Freedom House to formulate a democracy index, which serves as their dependent variable. The number of Internet users per 1, 000 represents their independent variable while the following measures are used as control variables: GDP per capita, education and literacy rates, life expectancy, urbanization, prevalence of non-Internet ICTs. Their analysis shows that a statistically powerful correlation exists between Internet diffusion and level of democratization. “The more salient observation to make, however, is that while economic prevalence and literacy maintain relatively constant correlations with democracy, the correlation for Internet prevalence gradually strengthens, almost to the same level as economic prevalence” [1]. The authors suggest that this dynamic reflects the growing significant relationship between Internet prevalence and democracy: “perhaps this is an indication that the Internet has come of age as a correlate of democracy” [1]. Indeed, they posit that this growth in correlation strength might “be expected given the positive network externalities, the network effect’ that is a salient property of the Internet” [1]. However, the coefficients from the regression analysis reveal that Internet usage is only able to predict a minimal amount of the variation in democracy: “to generate one point of democracy, an extra 500 Internet users per 1,000 citizens is needed, or an extra $5, 882 of GDP per capita is needed” [1]. The scale of democracy runs from 2 to 14. In terms of democracy’s traditional determinants, GDP was a weak predictor while literacy turned out to have no significance whatsoever. The other control variables used were either insignificant or internally correlated. While Best and Wade’s important contribution to the literature on Internet and democracy is one of the few contemporary quantitative studies carried out thus far, their approach does face a number of important limitations [1]. For one, their democracy index needs to be unpacked and “its constituent components, such as freedom of the press, openness of the electoral process,” for example, tested against traditional determinants of democracy to determine whether one component provides more explanatory power than others. Another limitation of their data is the fact that it extends only to 2002. This should be updated today due to the rapid pace of ICT diffusion over the past several years. In addition, several scholars have criticized the Freedom House data with regards to conceptualization, measurement and aggregation issues (Munck and Verkuilen, 2002; Rydland et al., 2008). Furthermore, as discussed subsequently, there is little to no

variation in the Freedom House data, which makes meaningful statistical analysis more difficult. In contrast to the populist literature, the second school of thought disputes the majority of claims that exist vis-`a-vis the relationship between Internet and democracy. The counterarguments are based on both qualitative and quantitative research. In terms of qualitative research, several scholars argue that the Internet is merely an extension of the ruling class and centralized control [39], [40], [41], [34], [42]. According to Neuman, even if the increase of ICTs had led to an increase in the motivation to communicate - which he argues has not happened - then ICTs would have become centralized by government turning them into social control mechanisms [43]. Scholars who subscribe to this school of thought maintain that mass media information technologies discourage collective behavior, ”unless the rise in couch potatoes can be considered a social movement” [44], [45], [46]. In contrast to Snider’s argument about the mere potential of citizens being informed acting as a source of accountability, if power is measured by the potential for “monopoly and control over information and communication, it is evident that the new technology can become a dangerous facilitator of tyranny” [21], [2]. Indeed, while the Internet may enable citizens to subvert political hierarchy, Barber notes that with increased participation comes the peril of political and economic surveillance. The populist school of thought is often blind to “how easily liberating technologies become tools of repression” [2]. Bimber rejects the supposition that the Internet will have significant effects on public life, point out that “both theory and empirical evidence cast grave doubt on the communication-action connection at the core of the populist theory” [12]. Lippmann argued that the capacity of ICTs to recreate politics is constrained by human nature, ie. cognitive processing, and not by the technical properties of the media themselves [47]. The Internet, then, is no different than other ICTs even if the new medium differs from previous technologies in a fundamental way, namely allowing social bonding to occur asynchronously. In sum, the Internet is “hardly producing the first dramatic expansion in communication: telephone, radio, and television also expanded communication profoundly.” There seems no compelling reason to believe that the communication capacity of the Net will have such a dramatically different effect than have other advances in pointto-point and broadcast communication” [12]. Moreover, Page argues, new ICTs may very well overcome spatial distance but his far from sufficient for establishing vibrant forms of political communication and deliberation [48]. Furthermore, “if democracy is to be understood as deliberative and participatory activity on the part of responsible citizens, it will have to resist the innovative forms of demagoguery that accompany innovative technology and that are too often overlooked by enthusiasts [2]. Aristotle wrote that the basis of a democratic state is liberty. Barber adds that a “free society is free only to the degree that its citizens are informed and that communication among them is open and informed [2]. However, recent research and empirical work

7

confirms that governments increasingly have the upper hand in controlling and regulating the impact of the information revolution [49], [50], [51], [52], [53], [54], [55], [56], [57], [58]. As Goldsmith observes, “if governments can raise the cost of Net transactions, they can regulate the transactions” [59]. Beilock and Dimitrova found that countries with lower Freedom House scores for civil liberties had significantly lower Internet usage (even when controlling for economic development) [60]. De Mesquita and George Downs also argue that government elites (e.g., in Singapore) have learned to “stifle the bottom-up democratic potential of the Internet and still promote economic growth, contrary to Kedzie’s dictator’s dilemma argument” [61]. As Bimber notes, the “central theoretical problem for the populist claim is the absence of a clear link between increases in information and increases in popular political action” [12] To this end, McLuhan’s old dictum may be wrong: “the medium is not the whole message. Content matters, and there is simply no overwhelming reason to believe that a new medium will necessarily enhance the political quality of communicative content” [12]. In short, “technology need not inevitably corrupt democracy, but its potential for benign dominion cannot be ignored” [2]. In terms of quantitative analysis, Scheufele and Nisbet’s 2002 statistical study suggests that the Internet does not increase democracy. “Through linear regression, they find that mass media broadcasting (e.g., television, newspapers) plays a far more effective role than the Internet in promoting democratic citizenship” [1], [62]. Given that an established body of quantitative research on this topic has yet to materialize, Scheufele and Nisbet do caution against generalizing the results of their study, which focused exclusively on the United States. Other scholars interested in this line of research have questioned the supposed direction of causation drummed up by the populist school of thought. Using multiple measures of regime type, Milner’s statistical analysis demonstrates that, ceteris paribus, democracies permit much greater online access, both in terms of Internet users per capita and Internet hosts per capita [63]. To this end, the information revolution may merely be reinforcing pre-existing dynamics. Milner’s study uses data from 1991 − 2001 to measure the influence of regime type of adoption of the internet. This study attempts to address a slightly different question—whether there is a relationship between measures of democracy and ICT penetration—but we build on her work by extending the range of years to 2006 (although we begin measurement in 2000). We adopt a fixed effect model and control for the time component directly in the model. Beilock and Dimitrova develop a model to explain global country differences in Internet use using income, measures of freedom, region dummies, and development indices [60]. Their data is a cross section from 2001 and does not take democratic variables directly into account. Best and Wade ask the question closest to that addressed in this paper [1]. They ask whether Internet penetration has an effect on the level of democracy in a country. Their study

is global in scope and uses time series data from 1992 to 2002. We seek to build on their work by using data from 2000 to 2006 and using the World Bank Governance metrics as our measures of democratic tendency. Best and Wade combined the Freedom House metrics of political rights and civil liberties as their measure of democratic tendency. As described subsequently we feel the Freedom House data are not well suited to a regression study such as this one. A. Our Approach to Measuring Democratic Tendency Using the World Bank Governance Indicators Dahl characterizes a government with power vested in a plurality as follows [13]: 1) Control over governmental decisions about policy is constitutionally vested in elected officials. 2) Elected officials are chosen and peacefully removed in relatively frequent, fair and free elections in which coercion is quite limited. 3) Practically all adults have the right to vote in these elections. 4) Most adults also have the right to run for the public offices for which candidates run in these elections. 5) Citizens have an effectively enforced right to freedom of expression, particularly political expression, including criticism of the officials, the conduct of the government, the prevailing political, economic, and social system, and the dominant ideology. 6) They also have access to alternative sources of information that are not monopolized by the government or any other single group. 7) Finally, they have an effectively enforced right to form and join autonomous associations, including political associations, such as political parties and interest groups, that attempt to influence the government by competing in elections and by other peaceful means. The first four points largely describe procedural aspects of a democracy, whereas the last three points delineate the communication aspect necessary for a well-functioning democratic regime. In fact, Diamond goes further and notes that “[s]ome insist on a fairly robust (though still procedural) definition of democracy, like Dahl’s ‘polyarchy.’ By this conception, democracy requires not only free, fair, and competitive elections, but also the freedoms that make them truly meaningful (such as freedom of organization and freedom of expression), alternative sources of information, and institutions to ensure that government policies depend on the votes and preferences of citizens” [64]. Expanding on Dahl, Diamond has developed his own list of characteristics of a democracy [65]: 1) Control of the state and its key decisions and allocations lies, in fact as well as in constitutional theory, with elected officials (and not democratically unaccountable actors or foreign powers); in particular, the military is subordinate to the authority of elected civilian officials. 2) Executive power is constrained, constitutionally and in fact, by the autonomous power of other government in-

8

3)

4)

5)

6)

7)

8) 9)

10)

stitutions (such as an independent judiciary, parliament, and other mechanisms of horizontal accountability). Not only are electoral outcomes uncertain, with a significant opposition vote and the presumption of party alternation in government, but no group that adheres to constitutional principles is denied the right to form a party and contest elections (even if electoral thresholds and other rules exclude small parties from winning representation in parliament). Cultural, ethnic, religious, and other minority groups (as well as historically disadvantaged majorities) are not prohibited (legally or in practice) from expressing their interests in the political process or from speaking their language or practicing their culture. Beyond parties and elections, citizens have multiple, ongoing channels for expression and representation of their interests and values, including diverse, independent associations and movements, which they have the freedom to form and join. There are alternative sources of information (including independent media) to which citizens have (politically) unfettered access. Individuals also have substantial freedom of belief, opinion, discussion, speech, publication, assembly, demonstration, and petition. Citizens are politically equal under the law (even though they are invariably unequal in their political resources). Individual and group liberties are effectively protected by an independent, nondiscriminatory judiciary, whose decisions are enforced and respected by other centers of power. The rule of law protects citizens from unjustified detention, exile, terror, torture, and undue interference in their personal lives not only by the state but also by organized nonstate or anti-state forces.

Like Dahl, Diamond includes procedural aspects of a democratic regime (points one through three) and he enshrines what he considers essential communication requirements in points four through seven. These latter points can be characterized as approximating a “freedom of expression” or “political voice” aspect to democracy. We found the World Bank governance metric of “Political Voice and Accountability” to represent Diamond’s notion well in that it measures “perceptions of the extent to which a country’s citizen’s are able to participate in selecting their government, as well as freedom of expression, freedom of association, and a free media” [66]. Like all the World Bank Governance metrics, it was built from surveys and other sources of data within each country. These sources give an idea of “freedom of belief, opinion, discussion, speech, publication, assembly, demonstration, and petition” present in the country, although they do not measure the proliferation of channels of communication that Diamond enunciates. The World Bank governance indicators metrics are based on 35 data sources some of which yield “subjective or perceptionsbased data” including that from “household and firm survey

respondents, as well as thousands of experts working for the private sector, NGOs, and public sector agencies” [66]. In points eight through ten, Diamond gives a description of the role of law in a democracy. The World Bank has a governance metric that expresses some of this: “measuring perceptions of the extent to which agents have confidence in and abide by the rules of society, and in particular the quality of contract enforcement, property rights, the police, and the courts, as well as the likelihood of crime and violence” [66]. This approximates the World Banks “Rule of Law” metric and we propose it as an empirical measure of Diamond’s points eight through ten. The World Bank also has a metric measuring political stability: “perceptions of the likelihood that the government will be destabilized or overthrown by unconstitutional or violent means, including politically-motivated violence and terrorism” [66]. While not enumerated in either Diamond’s or Dahl’s lists, Amartya Sen postulated an empirical correlation between democratic regimes and political stability [67]. He notes both the “political incentives provided by democratic governance” to prevent crises and specifically that the “positive role of political and civil right applies to the prevention of economic and social disasters in general” [67]. Thus we investigate the World Banks “Political Stability” metric as another measure of democratic tendency. In measuring the relationship between ICT penetration and these democratic variables, it is clear that country wealth is a confounding factor that sound be taken into account: wealthier countries are both more likely to be democratic and to be the heavier users of both the Internet and the cell phone. We gathered gross domestic product (GDP) data from 2000 to 2006 from the International Monetary Fund. The GDP data is purchasing parity adjusted to be directly comparable between countries. Diamond notes that country size is highly correlated with regime type: “countries with populations under one million are much more likely to be both democracies and liberal democracies. Two-thirds of these countries are liberal democracies, while only 30 percent of countries with populations over one million are. Among the larger 150 countries, only half are democracies, while 70 percent of the small countries are. The countries with populations over one million are about twice as likely as small states to have an electoral authoritarian regime and half again as likely to have a closed authoritarian regime.” [64], [65]. Because of this, we included population in our models to control for country size. The population data for 2000 to 2006 was also gathered from the IMF.1 B. Limitations of The Data The ICT data is gathered from the International Telecommunications Union (ITU). The ITU requested the number of Internet and cell phone users from each country. This raises a host of questions about the reliability of the data since it 1 Both the IMF GDP and population data are available at http://www.imf. org/external/pubs/ft/weo/2008/01/weodata/

9

is self-reported by the country. Perhaps the country has an incentive to under or over-report? It is plausible that some countries have more reliable data collection mechanisms in place than others. With one report per country per year, the data are highly granular, but they do seem to follow steady trends, and steadily upwards in ICT adoption. The World Bank did not calculate governance metrics for 2001. We carried out simple linear interpolation to provide a data point for each country in this year. We made the decision to do this since there are only six time points from which to interpolate, and only one, the year 2000, falls before our missing data. This creates data that are “too smooth” for year 2001 since they have been created from our pre-existing data and this will make our results appear more precise than they in fact are. We argue that this effects is minor since the World Bank data are themselves aggregated from a large number of sources, and thus less subject to noise than using a single source would be. Other metrics of the level of democratic rights exist, such as the Freedom House “Freedom in the World” metric. Freedom House carries out an annual global survey of political rights and civil liberties. We choose not to use this as a measurement of democratic tendency for two reasons. We felt that the World Bank Governance Indicators could be well grounded in the theory of democracy as measures of democracy. Secondly, the Freedom House measures have some quantitative limitations. A certain amount of inertia is built into the measurements so that it is difficult for a country to move much from one year to the next. For both political rights and civil liberties a country is scaled from 0 to 7, giving only 8 possible outcomes for a country. Combining these two factors leads to a database that does not shift very much from year to year. In the years of our study, 2000 to 2006, of the 193 countries surveyed by Freedom House (after subtracting the 9 with missing values for both political rights and civil liberties for the entire time series), 105 had no change in their scores for political rights and 86 had no changes in their civil rights scores. The average variance of those that did exhibit some change from 2000 to 2006 was 0.48 for political rights and 0.32 for civil liberties. This means the majorities of countries, if they changed at all, changed by perhaps one point on the 8 point scale. Having more years of data, including 2007, would improve our modeling. At the time of this writing, the World Bank Governance metrics were not available for 2007. The IMF estimated some of the population number for some of the countries. It is likely this has the effect of providing population data that is smoother than it would otherwise be. Note also that both ICT measures, the Internet and Cell Phone use, are measures per 100 inhabitants. We emphasize that this must be carefully noted in interpretation of the regression results, since we use population as an independent variable. We also note that even though the ITU collects Internet use statistics for each country, what it really means to use the Internet can vary by country due to filtering, censoring, and other restrictions on access. The OpenNet Intiative at the Berkman Center for Internet and Society monitors the filtering

activity for 40 of the countries most actively engaged in repressing internet activity [68], [69]. Although not as extreme as the case of the Internet, cell phone use can be restricted by the government as well, and will differ from country to country. Zuckerman gives several examples of government crackdowns on mobile phone use: Belarus’s reported shutdown of their SMS network in March 2006, reports of Ethiopian cell phone blocking during the 2005 election protests, and Cambodian blocking of SMS for two days before their 2007 elections [7]. North Korea and Cuba were dropped from the study since official data is not reported for these countries. It is generally known that there is very little internet access in Cuba, and little to none in North Korea and these are both regimes with little democracy. If we had been able to include these countries in our estimation of the models, this would likely have bolstered our results.2 Details of the data cleaning and amalgamation process are on the study website at http://www.stodden.net/ ICTD. Our population data was obtained from the International Monetary Fund and contains a sparse number os missing values. The IMF has made estimates of their missing data to complete the dataset.3 III. E MPIRICAL M ODELING Our data comprises a panel containing N different times series each consisting of T observations. The number of countries, N , is 181, and T , the number of years in our study, is 7. A fixed effects model of our democratic measures’ effect on ICT penetration follows: ICTit = β0 + β1 RLit + β2 V Ait + β3 P Sit + β4 P OPit + β5 GDPit + β6 M Fit + γt T D + ξi CD + it i = 1, . . . , N, t = 1, . . . , T. In this paper we model the penetration of Internet or cell phone usage per 100 inhabitants, ICT , as a function of the World Bank democratic measures (Rule of Law, Voice and Accountability, and Political Stability)4 , country size, country wealth, and the male/female ratio in the country. ICT is one of “Internet Use” or “Cell Phone Use.”5 GDPit is the per capita Gross Domestic Product for country i at year t, adjusted for purchasing power parity.6 P OPit is the 2 Countries with missing values also tended to support our hypothesis: Afghanistan, Bhutan, Comoros, Kiribati, Serbia, St. Kitts and Nevis, TimorLeste, and Tonga. Cuba, Iraq, Montenegro, and North Korea simply did not furnish enough data for inclusion in the study. 3 A precise explanation of their data interpolation procedure was not readily available. See http://www.econstats.com/weo/V023.htm 4 Available at http://info.worldbank.org/governance/wgi/ 5 The data used in this study is available at http://www.itu.int/ITUD/ict/informationsharing/ 6 The data are available at http://www.imf.org/external/pubs/ft/weo/2008/01/weodata/index.aspx

10

population of country i at year t and M Fit is the male/female gender ratio.7 β0 is the intercept term for country i, and γt T D and ξi CD are a time effect and a country effect, respectively. The effect of time is controlled for by dummy variables: T Dt is 1 for year t and 0 otherwise. Similarly CDi is a dummy variable that is 1 for country i and 0 otherwise, controlling for the differences between countries. Finally, it is a disturbance term with distribution N (0, σi2 ), which we assume to be uncorrelated across country cross sections. In this study there are N = 181 countries and T = 7 years. A. Autocorrelation in Panel Data A panel regression model of this type is subject to possible autocorrelation between subsequent observations because of the time series components. In a regression model as described above, it is possible to ’discover’ what Granger and Newbold [70] termed “spurious” relationships between the variables. That is, tests of significance on estimated coefficients may indicate a significant result, when in fact none is present. Granger and Newbold suggest economic time series data may be especially prone to autocorrelation since they tend to be non-stationary, in that it is not uncommon for the process generating the data to depend on the time it was sampled. For example, economic time series are commonly subject to seasonal or cyclical effects. Time series data that are not stationary will violate the assumptions of least squares regression since the variance of the error term will depend on time and thus introduce a bias into coefficient estimation.8 Granger and Newbold describe a high R2 value and a low Durbin-Watson statistic as warning signs that the estimated model may be yielding spurious results. Thus it is important to determine whether the time series data in this study are nonstationary. Our data are sampled yearly suggesting they may avoid annual cyclical effects. We analyze 181 countries from 2000 to 2006 (with 2001 interpolated for the World Bank variables) and thus have 7 values in each time series. As Granger and Newbold mention, finely sampled time series tend to exacerbate the cyclical effects and thus nonstationarity in the data. As is typical, the Augmented Dickey-Fuller test was used to assess nonstationarity in each of our time series. Since Internet and Cell phone use are increasing rapidly for the vast majority of the countries in our study, we measured the autoregressive structure of each time series as stationary around a trend line, and modeled with one lag. The null hypothesis is that the data are nonstationary. The test was applied to the ICT and World Bank data at the country level. For the Internet and mobile phone data the test rejected the null hypothesis of nonstationarity for 40 and 35 of 181 countries respectively, meaning that for around 20% of the Internet and cell phone penetration time series, nonstationarity is not evident.9 . When 7 The

gender ratio data was obtained from the Census Department’s International Database at http://www.census.gov/ipc/www/idb/tables.html 8 For a mathematically precise explanation see [70], p. 2. 9 The tests were performed at the 10% level

the same test was performed on the World Bank variables, 54, 62, and 60 of 181 rejected nonstationarity for Rule of Law, Voice and Accountability, and Political Stability, respectively (about one third of the data). Although there some evidence of stationarity, it appears that the majority of the time series included in this study are nonstationary, and it’s potential biasing of coefficient estimates is a concern.10 The typical remedy is the difference the data to remove the nonstationarity. Running the regressions in the above equation allowed us to carry out tests on the residuals directly to evaluate the level of autocorrelation. There are two regressions to be run in this study, modeling Internet penetration and cell phone usage. Typically the Durbin-Watson test with one lag is used to test for autocorrelation in the structure of the regression residuals and the regression with Internet use as a dependent variable was found to have autocorrelation present, and the cell phone penetration regression was not, with Durbin-Watson values of 1.86 and 2.01 respectively.11 The Durbin-Watson statistic is created by calculating d = PT (t −t−1 )2 t=2P , where t is the tth residual from the regrest 2 1 t sion. It follows that 0 < d < 4. A value of 2 indicates no autocorrelation. To test whether the Durbin-Watson test statistics could be considered equivalent to 2, the test in the R statistical software package was used [71]. The adjusted R2 values were 0.9246 and 0.9199, seeming to fit the Internet regression squarely into Granger and Newbold’s area of caution: a low Durbin-Watson statistic and a high R2 value, and also casts some suspicion on the cell phone regression. Granger and Newbold offer that until “a really satisfactory procedure is available, we recommend taking first differences of all variables that appear to be highly autocorrelated.” (p. 8.) We carried out this operation on both the Internet and cell phone use regressions because of the high R2 values and the nonstationary data in both regressions, even though the Durbin-Watson statistic did not suggest autocorrelation among the cell phone regression errors. The plots of the residuals for both regressions indicate possible heteroskedasticity. This suggests running the following differenced model: ∆ICTit = β00 + β10 ∆RLit + β20 ∆V Ait + β30 ∆P Sit + β40 ∆P OPit + β50 ∆GDPit + β60 ∆M Fit + γt0 T D + ξi0 CD + it i = 1, . . . , N, t = 2, . . . , T.

Running this model for differenced Internet penetration and differenced mobile phone use did not improve the Durbin10 Note that a combination of nonstationary time series may in fact be stationary. This is termed cointegration. 11 The p-values for the Durbin-Watson test are generated via a bootstrapping method and can fluctuate. In this case the p-values were 0.022 for the Internet regression and 0.962 for the cell phone regression.

11 TABLE I

TABLE II

S UMMARY S TATISTICS FOR 2000 DATA

S UMMARY S TATISTICS FOR 2006 DATA

Variable Internet Use per 100 Cell Use per 100 Rule of Law Voice & Accountability Political Stability GDP (PPP per capita) POP (millions) Gender Ratio (M/F)

Mean 7.25 16.02 -0.07 -0.06 -0.08 8,998.10 33.20 100.4

Min 0 0 -2.02 -2.05 -2.73 229.36 0.04 85.3

Max 45.58 81.73 1.95 1.67 1.54 55,248.25 1267.43 212.3

Variable Internet Use per 100 Cell Use per 100 Rule of Law Voice & Accountability Political Stability GDP (PPP per capita) POP (millions) Gender Ratio (M/F)

12

Watson statistics. They became 2.16 and 1.92 respectively. The adjusted R2 values were reduced to 0.3686 and 0.4059. Both Durbin-Watson statistics reject the null hypothesis of no autocorrelation at the 5% level. A common method of controlling for autoregression is using a 2-stage least squares approach [74], [75]. In the first stage, the autoregressive structure in the residual is estimated using a model postulating that the autocorrelation has a single lag structure, specifically:

Mean 21.66 53.19 -0.07 -0.07 -0.08 12,147.70 35.67 100.4

Min 0.03 0.42 -2.00 -2.28 -2.31 195.43 0.051 84.3

Max 92.52 138.06 2.03 1.72 1.60 76,537.15 1314.48 218.5

TABLE III R EGRESSION C OEFFICIENT E STIMATES FOR G LOBAL I NTERNET P ENETRATION Variable Rule of Law Voice & Accountability Political Stability GDP POP Gender Ratio

Estimate 0.9018 0.9122 -0.3783 0.0004 -0.0488 0.8212

Standard Error 1.1193 0.8514 0.5345 0.0002 0.8123 0.9189

p-value 0.4207 0.2844 0.4793 0.0972* 0.9521 0.3718

i = ρ ∗ i−1

A. Modeling Global ICT Penetration as a Function of Democratic Tendency

where i is the ith residual from the initial regression. The first stage allows us to find an estimate of ρ, ρˆ using a least-squares model. In stage 2, ρˆ is used to remove the autocorrelation in the variables. Each variable, represented as V AR in the next equation, is then adjusted to create a new variable, adjV AR, using the formula:

We estimated two panel regressions of ICT penetration with controls for autocorrelation as discussed in the preceding section (differencing and the Durbin-Watson correction). Internet and cell phone use were modeled as functions of demographic variables along with control variables:

adjV ARi−1 = V ARi − ρˆV ARi−1 Since we are analyzing panel data with both time and country dimensions, implementation of the autocorrelation was applied at the country level using different estimates of ρ for the Internet and cell phone regressions. Carrying this out on the differenced data increased the Durbin-Watson statistics to 2.24 and 2.18, and autocorrelation is still detected by this test at the 5% level. The adjusted R2 statistics were 0.4292 and 0.3599. This was our final model analyzed in the following section. Since autocorrelation has not been eradicated from the data we interpret our results cautiously and look for corroboration. IV. R ESULTS AND F INDINGS We used the R statistical package to estimate these models (version 2.7.2) [71]. The complete set of code and data used in this study can be found at http://www.stodden.net/ICTD. Tables I and II present summary statistics of the dependent variables (Internet and cell phone use per 100 country inhabitants) and the World Bank metrics for the countries in 2000 and 2006 respectively. 12 Testing the residuals for autocorrelation using the Durbin-Watson test is the typical procedure, when the sample is large. It is also possible to test for cointegration: whether the combination of time series is stationary. See [72] and [73].

∆Internetit = β00 + β10 ∆RLit + β20 ∆V Ait + β30 ∆P Sit + β40 ∆P OPit + β50 ∆GDPit + β60 ∆M Fit + γt0 T D + ξi0 CD + it i = 1, . . . , N, t = 2, . . . , T. The regression coefficients for Internet penetration are given in Table III. The coefficients on the individual country and time dummy variables are not included for space reasons.13 The most significant coefficient was GDP, and none of the democratic measures were significant. The positive coefficient on GDP confirms our intuition that wealthier countries have higher levels of Internet use. The same regression was run for cell phone penetration, and the coefficient estimates are presented in Table IV. Cell phone use appears not to associated with wealth as Internet use is, and the male to female gender ratio in the country is strongly associated with increases in cell phone use, as is the level of political stability. The influence of the gender ratio may be driven by outlier countries: Most countries had a male/female gender ratio of a little less than 100, implying slightly more females than males in the population. The coefficient of 4.88 implies that 13 The complete regression results can be found online at http://www. stodden.net/ICTD.

12 TABLE IV R EGRESSION C OEFFICIENT E STIMATES FOR G LOBAL C ELL P HONE P ENETRATION Variable Rule of Law Voice & Accountability Political Stability GDP POP Gender Ratio

Estimate -1.4120 -2.458 2.2823 -0.0002 0.1225 4.8839

Standard Error 2.0333 1.5946 1.0044 0.0005 1.4327 1.7342

p-value 0.4876 0.1236 0.0233** 0.7061 0.9319 0.0050***

TABLE V T OP 10 M ALE /F EMALE R ATIOS , 2000 AND 2006 Rank 1 2 3 4 5 6 7 8 9 10

Country United Arab Emirates Qatar Kuwait Oman Bahrain Saudi Arabia Maldives Bhutan Jordan Djibouti

2000 212.3 198.1 150.3 131.3 129.9 125 117.3 112.3 109.9 107.1

Country United Arab Emirates Qatar Kuwait Maldives Bahrain Oman Saudi Arabia Bhutan Jordan Grenada

2006 218.5 202.5 152.3 127.3 126.7 124.7 120.5 111.0 110.2 108.1

as the gender ratio increases by about 5, cell phone usage per 100 inhabitants will increase by one phone. Throughout the years studied, roughly 7 of the 10 countries with the highest male/female ratio each year were located in the middle east, and the ratios at that end of the distribution dwarfed the other countries’. As displayed in Table VI, it is plausible some of these values are extreme enough to have a large impact on the regression fit, although why this did not occur in the internet regression is not clear. It is also possible the very high gender ratio values represent a ’middle east effect’ since a number of those countries are highly represented in the top 10 gender ratio values. This implies that the coefficient on the gender ratio variable could represent a high growth in cell phone use in the middle east. Table VII gives the 2000 and 2006 cell phone data for these countries. The coefficient indicates that as the rate of change in the proportion of men increases, so does the rate of change in cell phone use. Interestingly, the greater political stability and the lower the perceived threat of violence, the greater cell phone penetration. This may represent infrastructural stability if associated with political stability and thus a measure of investor’s confidence. It is not clear why this factor would not therefore also be associated with an increase in Internet use. Perhaps cell phones are easier to proliferate than access to the Internet and so a smaller increase in political stability encourages cell phone increase before Internet increase. As shown in Tables I and II, the average cell phone penetration in 2000 was about 16 phones per 100 inhabitants and in 2006 it was about 53, nearly a three-fold increase. This is a high rate of increase but, notably, the countries listed in Table VI (those with the highest male to female gender ratios) had much higher than average growth in cell phone penetration. This is quantified in the significant coefficient in the regression in Table IV, while allowing for the included

TABLE VI C ELL P HONE U SE PER 100 I NHABITANTS FOR H IGHER G ENDER R ATIO C OUNTRIES , 2000 AND 2006 Rank 1 2 3 4 5 6 7 8 9 10

Country United Arab Emirates Qatar Kuwait Oman Bahrain Saudi Arabia Maldives Bhutan Jordan Djibouti

2000 43.98 19.90 21.74 6.63 30.61 6.40 2.83 0 7.72 0.04

Country United Arab Emirates Qatar Kuwait Maldives Bahrain Oman Saudi Arabia Bhutan Jordan Grenada

2006 118.51 109.6 91.49 87.88 122.88 69.59 78.05 9.77 74.4 44.59

confounding factors. The World Bank measure for political stability is also significantly positively correlated with increased cell phone use. This finding suggests that political instability is related to the mass diffusion of cell phone usage. In other words, an increase in cell phone availability could increase the perceived likelihood that the government will be destabilized or overthrown by unconstitutional or violent means. In their statistical analysis, Mansfield and Snyder find that the process of democratization itself is indeed a destabilizing one [76]. “Certainly, the virtues of working democratic structures do not translate into a carefree path to the stabilization of democracy” [77]. This finding also supports the arguments presented by Rosenau and Johnson, as well as Shirky, who opine that the Internet can be used as a tool by civil society to organize collectively to effect political change [25], [78]. B. Modeling ICT Penetration as a Function of Democratic Tendency Among the Most and Least Affluent Countries Examining ICT penetration for different strata of wealth may help isolate effects that are characteristic of those groups. Since wealth is a driver of investment one would expect GDP to play a role in the country’s readiness and ability to adopt new communication technologies [79], [80], [81]. As established in the literature we also found increases in GDP to be associated with increases in Internet use. We choose to examine ICT penetration in both the top and bottom 20% of countries by 2006 GDP more closely. Our focus on these groups, in particular the bottom quintile, is driven by Mansfield and Snyder’s work theorizing the instability of emergent and transitional regimes and the existence of the global digital divide [82]. Table VII lists the countries that fall into each of these groups. As in the previous section we fit a model with an ICT penetration measure as the explanatory variable, and measures of democratic tendency and controls as independent variables for a panel regression over years 2000 to 2006. For the top wealthiest quintile of countries our model did not yield statistically significant results for Internet penetration. Table VIII gives the coefficient estimates. Since these countries are exceptionally wealthy and relatively stable politically it may not be a surprise that GDP is not a driver of Internet use, and

13 TABLE VII T OP AND B OTTOM 20% OF C OUNTRIES BY GDP IN 2006 Top 20% Countries Qatar Luxembourg Brunei Darussalam Norway Singapore United States Ireland Switzerland Hong Kong, China Kuwait Iceland Canada Netherlands Austria Denmark United Arab Emirates Sweden Australia Belgium United Kingdom Finland Germany Japan France Bahrain Italy Spain Taiwan, China Greece Cyprus New Zealand Slovenia Israel Bahamas Korea (Rep.) Saudi Arabia Czech Republic

Bottom 20% Countries Zimbabwe Congo (Dem. Rep.) Liberia Burundi Guinea-Bissau Afghanistan Sierra Leone Niger Central African Rep. Ethiopia Malawi Eritrea Mozambique Togo Rwanda Uganda Myanmar Mali Madagascar Guinea Comoros Tanzania Nepal Burkina Faso Lesotho Bangladesh Gambia Haiti Zambia Ghana Sao Tom´e & Principe Benin Kenya Senegal Cambodia Chad Cˆote d’Ivoire

Variable Rule of Law Voice & Accountability Political Stability GDP POP Gender Ratio

Coefficient -0.3709 4.771 -1.0344 0.0003 1.6791 -1.7979

Standard Error 5.4055 4.0981 2.2760 0.0004 5.5724 3.7568

Coefficient 0.3702 -0.8115 -0.2583 -0.0002 -0.0309 0.4026

Standard Error 0.4081 0.3656 0.1822 0.0014 0.2784 0.3235

TABLE X C ELL P HONE P ENETRATION , T OP 20% Variable Rule of Law Voice & Accountability Political Stability GDP POP Gender Ratio

TABLE VIII I NTERNET P ENETRATION , T OP 20% OF GDP (2006) Variable Rule of Law Voice & Accountability Political Stability GDP POP Gender Ratio

TABLE IX I NTERNET P ENETRATION , B OTTOM 20% OF GDP (2006)

p-value 0.945 0.246 0.650 0.450 0.764 0.633

nor are the measures of democratic tendency, even though our subsets contains countries with varying levels of autocratic control. Among the least wealthy quintile we fit the same panel regression model as above. Table IX gives the regression coefficient estimates. Our predictors did not yield highly significant coefficient estimates with the exception of the World Bank voice and Accountability metric. Voice and Accountability is negatively correlated with Internet penetration: implying that when countries notch up in the Voice and Accountability ranking, the use of the Internet increases. This seemingly paradoxical finding may be explained when note that our analysis is restricted the the lowest quintile of country in wealth. These countries experience disproportionately greater

Coefficient 2.354 4.012 5.071 -0.0008 -2.896 2.8300

OF

p-value 0.3660 0.0281** 0.1585 0.9053 0.9117 0.2155

GDP (2006)

Standard Error 6.161 4.736 1.887 0.0005 6.483 4.3690

p-value 0.7029 0.3984 0.0612* 0.1071 0.6558 0.5182

political turmoil and it may be the case that countries with higher Voice and Accountability rankings have been reluctant to permit the growth of the Internet in their milieu. Table X gives the coefficient estimates from the panel regression for cell phone penetration for the top 20% of wealthiest countries. There is a statistically significant effect in the World Bank metric of Political Stability: greater Political Stability is associated with an increase in cell phone use per inhabitant. This result seems intuitive as political stability is historically associated with greater investment in communications infrastructure and is consistent with our earlier regression on cell phone use. This suggests the wealthiest countries may be driving the correlation between political stability and cell phone penetration. None of the other variables were found to have a statistically significant relationship with cell phone penetration in the wealthiest countries. Among the poorest countries, growth in Voice and Accountability had a statistically significant negative effect on growth in cell phone penetration. Although consistent with the Internet penetration regression results for this group of countries, the paradox remains as to the increase in per capita cell phone use as Voice and Accountability decreases. It is plausible citizens desire newer forms of ICT when Voice and Accountability is restricted. Another explanation may be that although not all countries in the bottom quintile are autocratic, a significant proportion are and cell phone use may facilitate the mobilization, organization and coordination of resistance against autocratic rule. Interestingly, the voice and accountability metric is not a significant predictor globally, yet is significant for types of ICT among the poorest countries. V. C ONCLUSIONS AND F UTURE R ESEARCH This paper is the first to our knowledge that uses recent Internet and cell phone use data in an empirical study of their relationship to democratic tendency. Previous studies uses measures of Internet use that ended in 1993 [26], [38]. Best and Wade’s data reached only to 2002. This paper is also the first to the best of our knowledge that measures ICT diffusion

14 TABLE XI C ELL P HONE P ENETRATION , B OTTOM 20% OF GDP (2006) Variable Rule of Law Voice & Accountability Political Stability GDP POP Gender Ratio

Coefficient 0.3702 -0.8115 -0.2583 -0.0002 -0.0309 0.4026

Standard Error 0.4081 0.3656 0.1822 0.0014 0.2784 0.3235

p-value 0.3660 0.0281** 0.1585 0.9053 0.9117 0.2155

as a function of democratic indicators. Previous research has focused on whether ICT use predicts democratic measures. These results support Bimbers assertions that the structure of political power has not been revolutionized or transformed into a new epoch of democracy [12]. Evidence can be found for both the populist thread in the literature and the notion that ICTs may act as an extension of the ruling class. We found a statistically significant positive relationship between the rate of diffusion of the cell phone and the World Bank’s “Political Stability” measure capturing perceptions regarding the likelihood that a government will be destabilized or overthrown by unconstitutional or violent means. Political Stability continues to have a significant positive relationship with the rate of cell phone use among the most affluent 20% of countries, but that relationship does not hold among the wealthiest countries. The “Voice and Accountability” indicator which measures perceptions of the extent to which citizens are able to participate in selecting their government, as well as freedom of expression, association and of the media, was a significant negative predictor of the rate of cell and Internet diffusion among the poorest quintile of countries. The diffusion of ICT access did not have any significant influence on “Rule of Law” while the increasing availability of cell phones were shown to have no influence on any of the three World Bank indicators. Recall that the Rule of Law metric measures perceptions of the extent to which individuals have confidence in and abide by the rules of society—in particular the quality of contract enforcement, property rights, the police, and the courts as well as the likelihood of crime and violence. This measure was originally included in the analysis based on Diamond’s research on the characteristics of democracy. However, the findings here suggest that the increase in “Rule of Law” perceptions does not influence ICT penetration, either positively or negatively. In other words, perceptions regarding the “Rule of Law” may be framed and influenced by factors other than widespread ICT use. The populist school of thought believes ICT diffusion will decentralize access to communication and information while increasing citizen access [12], while Hill and Hughes claim that perhaps the Internet will bring about “a wider democratic revolution in the world” [22]. In our modeling Internet diffusion was not predicted by our measures of democracy, implying this revolution has not yet arrived. The rate of cell phone diffusion was predicted by higher rates of the “Political Stability” metric. Recall that cell phone are much more widely

used globally than the Internet, so this may be a function of cell technology’s earlier foothold than the Internet’s. As Internet diffusion catches up to that of cell phone, the democratic metrics may be found to be predictors of this diffusion. This is not necessarily inconsistent with the populist thread in the literature as, according to Bimber, other scholars believe that the Internet may have a “transformative potential” because of the “one-to-one interaction among citizens and between citizens and government” it creates [12]. A paradox is created for the populist school in the finding that “Voice and Accountability” is negatively correlated with cell phone diffusion. This can be interpreted in favor of the argument that ICT diffusion can be centralized by government turning them into social control mechanisms [43]. As the rate of Voice and Accountability increases, the rate of diffusion of ICTs decreases among the poorest and least developed countries. Rates of cell phone use increase globally with decreases in political stability, lending further support to the thesis of ICTs as an extension of ruling class control. This paradox may be resolved is we consider Page’s view in 1995, that ICT diffusion may be still too nascent and insufficient to generate a well functioning system of political communication and deliberation [48]. It appears we are still too early to expect a close relationship between a vibrant public sphere and ICTs globally. As Bimber states, the “central theoretical problem for the populist claim is the absence of a clear link between increases in information and increases in popular political action” [12]. This paper provides evidence of the existence of this problem and the need to develop our understanding of this dynamic further. It would be interesting to tie this research more closely to development, in line with Sen’s reasoning that “Developing and strengthening a democratic system is an essential component of the process of development,” could extend the empirical analysis in a fruitful direction. This might mean specifically testing whether the order in which political and civil rights are extended as a country emerges from an autocracy affects the rate of development. This could provide a setting in which to test the “Lee Thesis,” that political rights should be withheld until economic development is achieved. It may be valuable to explore empirical issues further. Modeling the autocorrelation structure with more than one lag may help reduce autocorrelation. It would also be interesting to test for cointegration among these variables. Certainly documented feedback loops exist between our independent variables, such as GDP and measures of democracy, and taking this explicitly into account may improve the modeling [83], [26]. It is also plausible that feedback loops exist between ICTs and democratic measures and future modeling could accommodate this. Further research into the modeling aspects could estimate models including variables that control for the different manifestations of cell phone and Internet use in different countries. A more comprehensive model might explore possible non-linearities: whether countries with low ICT adoption rates have different patterns of democratic tendency than those with high adoption rates.

15

ACKNOWLEDGMENT The authors would like to thank Michael Best and Colin Maclay for helpful discussion and the anonymous reviewers for constructive criticism. Sadia Ahsanuddin and Kristen Lovin provided outstanding research assistance. R EFERENCES [1] M. Best and K. Wade, “The internet and democracy: Global catalyst or democratic dud?” June 2006, sam Nunn School of International Affairs. [2] B. Barber, “Three scenarios for the future of technology and strong democracy,” Political Science Quarterly, vol. 113, no. 41573-589, 1999. [3] P. Schmitter and T. Karl, “What democracy is . . . and is not,” Journal of Democracy, vol. 2, no. 3, 1991. [4] P. Norris, “Political protest in fragile states,” July 2006, international Political Science Association World Congress. [5] F. Zakaria, “The rise of illiberal democracies,” Foreign Affairs, November 1997. [6] Huber, D. Rueschemeyer, and J. Stephens, “The impact of economic development on democracy,” Journal of Economic Perspectives, vol. 7, no. 3, pp. 71–85, 1993. [7] E. Zuckerman, “Draft paper on mobile phones and activism,” 2007. [Online]. Available: http://www.ethanzuckerman.com/blog/2007/04/09/ draft-paper-on-mobile-phones-and-activism/ [8] D. Kidd, Indymedia.org: a new communications commons. Routledge, 2003, pp. 47–69. [9] A. Scott and J. Street, “From media politics to e-protest,” Information, Communication and Society, vol. 3, no. 2, pp. 215–40, 2000. [10] S. Andrew, “The control revolution: How the internet is putting individuals in charge and changing the world,” The Century Foundation, Tech. Rep., 2000. [11] R. Clarke, “Information technology: Weapon of authoritarianism or tool of democracy?” August 1994, iFIP World Congress. [12] B. Bimber, “The internet and political mobilization - research note on the 1996 election season,” Social Science Computer Review, vol. 16, no. 4, pp. 391–401, 1998. [13] R. Dahl, Democracy and its Critics. New Haven, CT: Yale University, 1989. [14] H. Rheingold, Virtual Reality. New York, NY: Summit Books, 1991. [15] R. Anderson, T. Bikson, S. Law, and B. Mitchell, “Universal access to e-mail: Feasibility and societal implications,” RAND, Tech. Rep., 1995. [16] A. Etzioni, The Spirit of Community: Rights, Responsibilities, and the Communitarian Agenda. New York, NY: Crown Publishers, 1993. [17] J. Pavlik, New Media Technology: Cultural and Commercial Perspective. Boston, MA: Allyn and Bacon, 1996. [18] L. Grossman, The Electronic Republic. New York: Viking, 1995. [19] G. Browning, Electronic Democracy: Using the Internet to Influence Politics. Wilton, CT: Online Inc., 1996. [20] N. Negroponte, Being Digital. New York, NY: Vintage, 1995. [21] J. Snider, “New media: Potential information and democratic accountability: A case study of governmental access community media,” August 1996, american Political Science Association Convention. [22] K. Hill and J. Hughes, “Computer mediated political communication: The usenet and political communities,” Political Communication, vol. 14, pp. 3–27, 1999. [23] A. Corrado and C. Firestone, Elections in Cyberspace: Toward a New Era in American Politics, A. Corrado and C. Firestone, Eds. Washington, DC: Aspen Institute, 1996. [24] C. Steele and A. Stein, Communications revolutions and international politics. Albany, NY: State University of New York Press, 2002. [25] J. Rosenau and D. Johnson, Information technology and turbulence in world politics. State University of New York Press, 2002. [26] C. Kedzie, “Communication and democracy: Coincident revolutions and the emergent dictator’s dilemma,” RAND, Tech. Rep., 1997. [Online]. Available: http://www.rand.org/publications/RGSD/RGSD127 [27] S. Lipset, “Some social requisites of democracy: Economic development and political legitimacy,” American Political Science Review, vol. 53, pp. 69–105, 1959. [28] R. Jackman, “On the relation of economic development to democratic performance,” American Journal of Political Science, vol. 17, pp. 611– 21, 1973. [29] K. Bollen, “Political democracy and the timing of development,” American Social Review, vol. 44, pp. 572–87, 1979.

[30] R. Burkhard and M. Lewis-Beck, “Comparative democracy: the economic development thesis,” American Political Science Review, vol. 88, pp. 903–10, 1994. [31] J. Longdregan and K. Poole, “Poverty, the coup trap, and the seizure of executive power,” World Politics, vol. 42, pp. 151–83, 1990. [32] ——, “Does high income promote democracy?” World Politics, vol. 49, pp. 1–30, 1996. [33] B. Geddes, “What do we know about democratization after twenty years?” Annual Review of Political Science, vol. 2, pp. 115–44, 1999. [34] F. Webster, Globalization, information and change. State University of New York Press, 2002. [35] R. Reich, The Work of Nations: Preparing ourselves for 21st century capitalism. Vintage, 1992. [36] T. Friedman, The Lexus and the Olive Tree: Understanding Globalization. Anchor, 1999. [37] L. Waverman, M. Meschi, and M. Fuss, “The impact of telecoms on economic growth in developing countries,” Centre for Economic Policy Research, Tech. Rep., March 2005. [38] D. Richards, Making the national international: Information technology and government respect for human rights. Albany, NY: State University of New York Press, 2002. [39] D. Barney, Prometheus Wired: The Hope for Democracy in the Age of Network Technology. University of Chicago Press, 2000. [40] A. Mazrui and R. Ostergard, Technology transfer in the computer age: The African experience. State University of New York Press, 2002. [41] D. Strienstra, Gender, women’s organizing, and the Internet. Albany, NY: State University of New York Press, 2002. [42] S. Kalathil and T. Boas, Open Networks, Closed Regimes. Carnegie Endowment for International Peace., 2003. [43] W. Neuman, The future of the mass audience. Cambridge University Press, 1991. [44] T. Eyck, “Does information matter? a research note on information technologies and political protest,” Social Science Journal, vol. 38, pp. 147–160, 2001. [45] N. Postman, Amusing ourselves to death. New York, NY: Vintage Books, 1985. [46] H. Schiller, Culture Inc. Oxford University Press, 1989. [47] W. Lippmann, Public Opinion. New York, NY: Macmillan, 1934. [48] Page, Who Deliberates? Mass Media in Modern Democracy. Chicago, IL: University of Chicago Press, 1996. [49] G. Cherian, “The limits of singapore’s ‘light touch’ web regulation,” 2008, politics 2.0 International Conference. [50] R. Deibert, J. Palfrey, R. Rohozinski, and J. Zittrain, Access Denied: The Practice and Policy of Global Internet Filtering. The MIT Press, 2008. [51] S. Oats, “Internet and democracy in russia,” April 2008, politics 2.0 International Conference. [52] R. Singel, “Seeking tighter censorship, repressive states target web 2.0 apps,” Wired Magazine, March 2008. [Online]. Available: http://blog.wired.com/business/2008/03/etech-what-happ.html [53] J. Zittrain, The Future of the Internet—And How to Stop It. Yale University Press, 2008. [54] S. Mydans, “Myanmar junta unplugs internet,” The International Herald Tribune, October 2003. [Online]. Available: http://www.iht.com/ articles/2007/10/04/asia/04info.php [55] J. Goldsmith and T. Wu, Who Controls the Internet? Illusions of a Borderless World. Oxford University Press, 2006. [56] L. Lessig, Code: Version 2.0. Basic Books, 2006. [57] D. Drezner, “The internet’s effect on state-society relations,” March 2006, international Studies Association Convention. [58] T. Price, Cyber Activism: Advocacy Groups and the Internet. Washington, DC: Foundation for Public Affairs, 2000. [59] J. Goldsmith, “Regulation of the internet: Three persistent fallacies,” Chicago-Kent Law Review, vol. 73, December 1998. [60] R. Beilock and D. Dimitrova, “An exploratory model of inter-country internet diffusion,” Telecommunications Policy, vol. 27, pp. 237–252, 2003. [61] B. Mesquita and G. Downs, “Development and democracy,” Foreign Affairs, September 2005. [62] D. Scheufele and M. Nisbet, “Being a citizen online: New opportunities and dead ends,” The Harvard International Journal of Press/Politics, vol. 7, no. 3, pp. 55–75, 2002. [63] H. Milner, “The digital divide: The role of political institutions in technology diffusion,” August 2003, american Political Science Association.

16 [64] L. Diamond, “elections without democracy: Thinking about hybrid regimes,” Journal of Democracy, vol. 13, no. 2, April 2002. [65] ——, Developing Democracy: Toward Consolidation. John Hopkins University Press, 1999. [66] D. Kaufmann, A. Kraay, and M. Mastruzzi, “Governance matters vii: Aggregate and individual governance indicators 1996-2007,” World Bank, Tech. Rep., 2007. [67] A. Sen, Development as Freedom. Oxford University Press, 2001. [68] “Opennet initiative.” [Online]. Available: http://opennet.net/ [69] R. Faris and N. Villeneuve, Access Denied: The Practice and Policy of Global Internet Filtering. MIT Press, 2008, ch. 1. [70] C. Granger and P. Newbold, “Spurious regressions in econometrics,” Journal of Econometrics, vol. 2, pp. 111–120, 1974. [71] “R.” [Online]. Available: http://www.r-project.org/ [72] C. Granger, “Some properties of time series data and their use in econometric model specification,” Journal of Econometrics, vol. 16, pp. 121–130, 1981. [73] ——, “Co-integration and error correction: Representation, estimation and testing,” Econometrica, vol. 2, pp. 251–176, 1987. [74] T. Amemiya, “The nonlinear two-stage least squares estimator,” Journal of Econometrics, vol. 2, pp. 105–110, 1974. [75] R. Cumby, J. Huizinga, and M. Obstfeld, “Two-step two-stage least squares estimation in models with rational expectations,” Journal of Econometrics, vol. 21, pp. 333–355, 1983.

[76] E. Mansfield and J. Snyder, “Conceptualizing and measuring democracy:evaluating alternative indices,” International Organization, vol. 56, no. 2, pp. 297–337, 2001. [77] H. Barnes, “The contribution of democracy to rebuilding postconflict societies,” The American Journal of International Law, vol. 95, no. 1, pp. 86–101, 2001. [78] C. Shirky, Here Comes Everybody: The Power of Organizing without Organizations. Penguin Press, 2008. [79] C. Argevou, “Ols bias in a nonstationary autoregression,” Information Technology for Development, vol. 8, no. 1, pp. 15–28, 1998. [80] B. Oyelaran-Oyeyinka and K. Lai, “Internet diffusion in sub-saharn africa: A cross-country analysis,” Telecommunication Policy, vol. 8, no. 1, pp. 507–527, 1998. [81] J. Gibbs, K. Kraemer, and J. Dedrick, “Environment and policy factors shaping global e-commerce diffusion: A cross country comparison,” The Information Society, vol. 19, no. 1, pp. 5–18, 1993. [82] N. Kshetri and N. Dholakia, ENCYCLOPEDIA OF INFORMATION SCIENCE AND TECHNOLOGY, 2nd edition. Idea Group Publishing, 2008. [83] P. Norris, Digital Divide: Civic Engagement, Information Poverty, and the Internet Worldwide. Cambridge University Press, 2001.

17

A review of the research on mobile use by micro and small enterprises (MSEs) Jonathan Donner and Marcela Escobari

Abstract— The paper offers a systematic review of 14 studies of the use of mobile telephony by micro and small enterprises (MSEs) in the developing world, detailing findings about changes to enterprises’ internal processes and external relationships, and findings about mobile use vs. traditional landline use. Results suggest that there is currently more evidence for the benefits of mobile use accruing mostly (but not exclusively) to existing MSEs rather than new MSEs, in ways that amplify existing material and informational flows rather than transform them. The review presents a more complete picture of mobile use by MSEs than was previously available to ICTD researchers, and indentifies priorities for future research, including comparisons of the impact of mobile use across subsectors of MSEs and assessments of use of advanced services such as mobile banking and mobile commerce.

Index Terms—Business Economics, Communication, ICTD, Mobile Communication, MSE, Social Factors

T

I. INTRODUCTION

his paper presents a systematic review of fourteen studies of the use of mobile telephony by Micro and Small Enterprises (MSEs) in the developing world [1-14]. The majority of non-agricultural enterprises in the developing world have ten or fewer employees [15, 16]. These MSEs employ up to 25% of working-age adults in some countries [16], and while the contribution of MSEs to aggregate economic growth remains a matter of debate [17], their importance to household livelihoods and poverty alleviation is undeniable. Thus, MSEs are the focus of programs at many of the world’s largest development institutions [18]. Since the year 2000, the spread of mobile telephony across the developing world has raised hopes among policymakers that MSEs will benefit from easier access to telecommunications. The successful entrepreneur, suddenly enabled by his mobile phone, has been given a prominent role in the global development narrative and become a semiregular fixture in both the popular press [19, 20] and practitioner media [21, 22]. Manuscript version received February 20, 2009. J. Donner is with Microsoft Research India, [email protected] M Escobari is with the Center for International Development, [email protected]

Like landlines, mobile phones allow people to communicate at a distance and exchange information instantaneously. Thus, there is significant potential for mobile use to increase MSE productivity. However, since the dynamics underpinning this potential are nuanced, and since current supporting evidence is scarce and methodologically heterogeneous [9], it is important to rigorously examine mobile use by MSEs. For example, there is a difference between using a mobile to serve existing customers more effectively, and using it to start a new business. There is a difference between using a mobile to check market prices and using it bypass a middleman who carries goods to market. Popular press and practitioner reports generally fail to make these distinctions. Fortunately, a small but methodologically diverse set of research studies have examined mobile use by MSEs in detail. This paper offers a systematic review of this existing literature, identifying known patterns of mobile use, as well as some important gaps in the research. The review employs distinct foci. First, it offers an assessment of how mobile use influences the internal process of an enterprise, using Porter’s value chain model [23]. Second, it offers a corresponding assessment of how mobile use influences the network of relationships external to the enterprise—the value system [23] of producers, traders, wholesalers, retailers and end-customers. Finally, it explores two elements unique to mobile communication—the increased spatial and temporal mobility afforded by wireless devices, and the resulting blurring of the personal and the professional spheres—to assess how MSE mobile use differs from landline use. II. MICRO AND SMALL ENTERPRISES (MSES) Working definitions of MSEs vary from country to country and from researcher to researcher [24]. This analysis defines an MSE as any non-farm i enterprise, formal or informal, with less than 50 employees, including sole proprietorships, parttime businesses, and home-based businesses. The size thresholds draw on Mead and Leidholm, [16], who note that the absolute majority of such enterprises in the developing world are sole proprietorships, and that firms with less than 10 employees substantially outnumber larger enterprises. A number of factors distinguish the term MSE (micro and small enterprise) from SME (small and medium enterprise). The terms MSE and SME are acronyms, each combining two distinct sizes of enterprises into a single reference. However,

18

with no commonly accepted definitions of the thresholds between micro, small, and medium, there are often implicit conceptual overlaps between the acronyms. Unlike SMEs, the majority of MSEs are informal enterprises. Once again, however, there is no universal standard to determine what makes an enterprise informal vs. formal [6]. In many cases, even the term entrepreneur may be a bit of a romantic misnomer. Evidence suggests that among MSEs, only a small minority of enterprises are poised for growth; most remain small or struggle to survive, and yield a low return on labor and capital [17, 25]. Though significantly less growth-oriented and productive (on average) than SMEs and other larger firms [17], MSEs share a basic similarity with all enterprises; each combines investments in capital with some labor (their own, their families’ or their employees) in the hopes of yielding a product or service whose market value exceeds the cost of those inputs. Thus, there has always been a thread in the ICTD literature that seeks to understand how various technologies could be used advantageously by MSEs [26-28]. Prior to the widespread introduction of the mobile into the developing world, the landline’s importance in this regard was already clear: Phones are the information-related technology that has done the most to reduce costs, increase income and reduce uncertainty and risk. Phones support the current reality of informal information systems, they can help extend social and business networks, and they clearly substitute for journeys and, in some cases, for brokers, traders and other business intermediaries. They therefore work “with the grain” of informality yet at the same time help to eat into the problems of insularity that can run alongside. Phones also meet the priority information needs of this group of communication rather than processing of information [27: 18] The quotation focuses directly on the basic tasks of running a business—reducing costs, increasing income, managing risk—and links them to core functions of mediated communication technologies, particularly the substitution for journeys. As demonstrated elsewhere, [26] the key is increased productivity. III.

STUDIES ON MOBILES AND MSES

Recently, studies have emerged that directly address how MSEs in the developing world are using mobiles rather than landlines or other ICTs. The studies are not as numerous as the enthusiasm in the popular press might suggest. They are a tiny fraction of the total literature on mobile use in the developing world [29]. They have emerged from different disciplines, and, as relative contemporaries, often do not cite each other. This section presents two studies representing distinct methodologies and conclusions, to provide an example of the range of available perspectives and to set the stage for the systematic review.

Particularly focused and powerful evidence appears in Jensen’s [10] research on the fishermen of Kerala. Working with five-year time series data at three fish markets in coastal India, Jensen and his team found that “the adoption of mobile phones by fishermen and wholesalers was associated with a dramatic reduction in price dispersion, the complete elimination of waste, and near-perfect adherence to the Law of One Price. Both consumer and producer welfare increased.” [10: 879] Soon after the introduction of mobile coverage, fishermen bought mobiles and accumulated lists of up to 100 buyers in their handsets’ address books; subsequently, while still at sea, fishermen could call a range of possible landing points and buyers in order to determine the best price and best place to sell their catch. By contrast, Jagun, Heeks, and Whalley’s [9] examination of the mobile’s role in mediating supply chains in the Nigerian market for traditional hand-woven ceremonial cloth is broad in scope. It offers a multidisciplinary literature review, a conceptual framework articulating effects at multiple levels, and a detailed case study. They describe “process” benefits to mobile use, as calls at a distance reduce the time of trades and replace costly journeys. They also describe “structural” impacts; finding no disintermediation of traders, but rather an intensification of their role. Traders are more likely to have mobiles than the less prosperous weavers in the supply chain, and thus are better positioned to coordinate with a wider range of downstream customers and to maintain a more dynamic and responsive set of relationships with weavers. For example, weavers previously had to pay cash to get their supplies. Mobiles give weavers access to credit by enabling calls on their behalf to fabric vendors by traders, who vouch for the veracity of weavers’ orders, and promise to cover the costs of the fabric in advance of the completion of the weavers’ work. IV. METHODS AND CODING PROTOCOL Many of the studies of mobile use by MSEs are qualitative, and do not report statistical findings. Even among quantitative studies, there is little agreement in terms of dependent and independent variables under scrutiny. Thus, a statistical metaanalysis would not be applicable [30]. Similarly, a method designed specifically for comparing ethnographies, such as reciprocal translation [31] would be unlikely to bridge qualitative and quantitative studies. The analysis draws instead on a systematic review methodology [32] to aggregate findings across the available studies. By using a standardized protocol, coding each individual study for the appearance or absence of certain assertions, the review assesses and parsimoniously represents what the research literature, in aggregate, suggests about mobile use by MSEs. The exercise relies on clearly articulated eligibility criteria to select studies and on standardized questions to evaluate them. These two levels of standardization, agreed upon before the formal review commenced, separates the exercise from a conventional listbased or thematic/narrative literature review.

19

A. Selecting studies Papers were initially identified by online literature and database searches (using keyword combinations of mobile, cellular, microenterprise, MSE, etc.), and by snowball references from the bibliographies of studies already in hand. Next, these studies were assessed against a series of eligibility criteria: to be included, studies had to be specific to both mobiles and MSEs, report generalizable findings, and contain detailed primary data about mobile use in everyday conditions. What started out as a reasonably large body of studies was trimmed back significantly. In order to provide additional resources to other researchers, however, this section lists those excluded papers, along with the rationale for the decisions. To be included in the review, papers had to be specifically focused on mobile phones, which excluded some excellent research on landlines or payphones and MSEs [28, 33, 34]. The papers also had to be about MSEs, not SMEs. Papers that did not explicitly include sole proprietors and informal enterprises were excluded [35-37]. We made a more difficult decision to exclude papers that were not generalizable to a wide range of MSEs. An important line of research explores how many individuals earn livelihoods in the mobile business itself, by selling airtime, fixing handsets or operating village phones [38-40]. However, these studies treat mobiles as products and services, rather than enablers of general business processes. To fit into the evaluation protocol, papers had to offer sufficient details around the use of mobiles to illuminate their role in these business processes. A few surveys that were otherwise topically correct did not yield information of this kind [41], or blurred the lines between mobiles and other ICTs to the point where assertions about mobiles in particular were difficult to extract [42, 43]. Reviews without new primary data were excluded [44, 45]. When multiple papers drew on the same set of data [46, 47], only one paper was retained. Finally, the review focused on analyses of mobile use in everyday settings, rather than proposals for or evaluations of new pilot technologies [48-51] or programmatic interventions by NGOs [52]. The development of such technologies and programs is central to the ICTD field, but such initiatives yield different forms of evidence about mobile use than those that examine MSEs operating on their own. This limiting exercise forced a trade-off: the remaining papers clearly describe some element of the use of mobiles by MSEs in developing countries, but the population of such studies is relatively small. Thirteen papers and one book were retained. B. Evaluation protocol The process of developing the evaluation questions was iterative. It was based mostly on an initial reading of the documents by the researchers, while also integrating current narratives in the popular and practitioner literatures. An original goal was to code studies according to subcategories of MSEs (to discern differences in mobile use between traders

and producers, for example), but it became clear that the population of existing studies is too small to support that inquiry. The final protocol employed three distinct foci. First, it assessed the impact of mobile use on the internal process of an enterprise, using Porter’s value chain model [23], Fig. 1. The value chain comprises the activity inputs into a product or service: inbound logistics, operations (production), outbound logistics, marketing and sales, and after-sales service. The value chain also includes supporting functions: firm infrastructure, human resources, technology development (knowledge developed or owned by the enterprise), and procurement. Together these activities can create customer value in excess of the costs to provide it, yielding profit. Porter [23: 168], argues that information and communication technologies can be used to improve almost any of these primary and supporting activities. Although the value chain framework was developed with larger enterprises in mind, it can be applied to MSEs, since in small firms the same individual can carry out different business-related activities during the day. (Indeed, even in larger firms there is often not a perfect mapping between the activities and functions in the value chain and distinct people or departments). Both researchers coded individual papers for mentions of the mobile’s role in any of the primary or supporting functions. FIGURE 1: PORTER’S VALUE CHAIN

note: image released to public domain as per http://en.wikipedia.org/wiki/Image:ValueChain.PNG The second analysis used another Porter framework, the value system [23] ii , to offer a corresponding assessment of how mobile use influences the network of interdependencies and relationships external to the enterprise, including producers, traders, wholesalers, retailers and end-customers. An initial reading of the papers identified four categories of potential impacts. Some research stresses (a) the increased availability of information in the network; other studies stress (b) the entry of new actors, particularly buyers and sellers, into markets. Both factors tend to increase competition, but do

20

so in different ways. One focuses on the actors in the network, the other on the information those actors exchange. Nevertheless, the two changes do not necessarily move in tandem; it is one thing to assert that the same set of actors exchange information at lower cost and higher frequency, another to say that markets have expanded. This informationvs.-actors cut highlights this distinction. Two other categories of network impacts can be expressed as assertions that (c) mobiles help enterprises cut out middlemen and that (d) mobiles help individuals start new businesses. Unlike the value chain analysis (coded for affirmative mentions only), the value system analysis coded for both affirmations and negations of the four potential impacts. After the initial reading, we elected to track negations since some of the sources made a point of arguing against one or more of the assertions from the popular and practitioner literature. The third analysis explored how the impact of mobile use on small enterprises is or is not different from the impact of landline telephony on those same kinds of businesses [28, 33, 34]. Technological properties of mobile communication make it inherently more prone to adoption by MSEs than landlines: it is cheaper to build towers than lay cable, prepay accounts have no startup costs, and inexpensive/used handsets are readily available. However, in this case the third analysis focused narrowly on two differences in use rather than cost or access. First, studies were coded for mentions of mobility. Mobility is a crucial difference between mobiles and landlines—while landlines connect places to places, mobiles generally connect people to people, wherever they are and regardless of the time and situation. This mobility leads to increased individual addressability, and can change how people structure social and economic activity [53-55]. Most relevant to this analysis, mobility may enable the rise of roaming businesses, just-intime service and what Townsend [56] has called the “real time city”. And yet some evidence suggests that in the developing world many mobiles are purchased as substitutes for landlines, rather than complements to them [57]. Second, the nature of the mobile as a portable, personal device means it is particularly easy to use for both personal and business functions during the same day. Thus, studies of the role of mobiles in the lives of MSE operators are often different from studies of the role of the device in the businesses themselves. The analysis coded for studies that explore these social functions. Once the protocol was established, each researcher re-read the papers, coding them in isolation. We then compared our codes and resolved any discrepancies through discussion. The resulting codes are less prone to reflect the bias of a single reader. Of the 112 cells on the matrix requiring codes, 16 required discussion to resolve coding discrepancies between the two researchers.

V. RESULTS A. Enterprise value chain Most studies mention the core processes of marketing and sales [1, 2, 4, 6, 7, 9-13]. Analyses ranging from Jensen’s [10] model of searching for the best price for fish to Kamga’s [11] description of improvements to the local laundry services in Cote d’Ivoire asserts that mobiles help connect vendors and buyers, often at a distance and usually at lower cost than an in-person journey. Esselaar et al. [6] report results of a survey of SMEs, including 1/3 microenterprises, conducted across 13 countries. “Mobile phones are used more often for keeping in contact with customers and clients” (p 92). This is the highly visible, intuitive role of mobiles for small enterprises. The picture is sparser for other core processes within the value chain. Three studies mention inbound and outbound logistics [1, 9, 13], particularly Abraham [1], who details how fishermen can now use small supply boats (dispatched via mobile) to stay out fishing longer. Overå [13] describes how traders in Ghana can time harvests (inbound) and change the terms of delivery financing (outbound) because of the mobile. Operations receives two mentions, again by [1], who notes that fishermen use mobiles to coordinate the timing and location of when to drop nets and search for fish. Similarly, [9] describes how weavers call customers mid-process to revise plans for the garments they are creating. Only studies by Frempong et al. [7] and Molony [11] mention after-sales service. Molony describes how Tanzanian exporters of carved wood use the mobile to elicit feedback and built trust with buyers after (and ideally between) sales. In terms of crosscutting functions, five studies reference procurement [1, 6, 7, 9, 13] and address price search by buyers of inputs (or by traders). There is little evidence to date for the mobile’s role in transforming the proprietary technology, infrastructure or HR functions of MSEs, perhaps because these enterprises are too small to invest in these assets. Although studies outside the review [36, 58] provide anecdotes of small employers giving mobiles to employees, this infrastructure function is of limited utility for tiny firms and sole-proprietorships. B. Industry value system The second analysis turns the lens outside the enterprise, towards its location in a network of relationships. The most common finding links mobile use to an increase in the flow of information between actors in the value system [1, 2, 6, 7, 914]. The two primary sub-themes are more frequent or wideranging exchanges of price information [1, 2, 13], and a more generalized discussion of increased communication with customers [6, 12, 14]. These findings are reflections of the frequent references to marketing and sales and procurement activities in the previous value chain analysis. Reference [9] mentions an increase in the completeness of the information, but notes that they saw no increase in quality

21

TABLE I: SUMMARY OF THE MSE AND MOBILE STUDIES IN THE REVIEW

Enterprise Value Chain Studies Core Processes

Industry Value System

Add Cross-cutting InformFunctions ation

Uses

Add Buyers/ Sellers

Bypass Middlemen

Start businesses

Mobility

Social

Inbound & Outbound Logistics, Operations, Marketing & Sales.

Procurement

Yes

Yes

No

--

Yes

Yes

[2] J. C. Aker, “Does digital divide or provide? The impact of cell phones on grain markets in Niger”

Marketing & Sales

--

Yes

Yes

--

--

Yes

--

[3] J. Donner, “Microentrepreneurs and mobiles: An exploration of the uses of mobile phones by small business owners in Rwanda”

--

--

--

--

--

--

Yes

Yes

[4] J. Donner, “The use of mobile phones by microentrepreneurs in Kigali, Rwanda: Changes to social and business networks”

Marketing & Sales

--

--

Yes

--

--

--

Yes

[5] J. Donner, “Customer acquisition among small and informal businesses in urban India: Comparing face to face, interpersonal, and mediated channels”

--

--

--

No

--

--

--

Yes

[6] S. Esselaar, C. Stork, A. Ndiwalana, and M. Deen-Swarra, “ICT usage and its impact on profitability of SMEs in 13 African Countries”

Marketing & Sales

Procurement

Yes

--

--

--

--

Yes

[7] G. Frempong, G. Essegbey, and E. Tetteh, “Survey on the use of mobile telephones for micro and small business development: The case of Ghana,”

Marketing & Sales; Service

Procurement

Yes

Yes

--

--

--

--

--

--

--

--

--

No

--

Yes

Inbound & Outbound Logistics, Operations, Marketing & Sales

Procurement

Yes

Yes

No

--

--

--

[10] R. Jensen, “The Digital Provide: Information (Technology), Market Performance, and Welfare in the South Indian Fisheries Sector”

Marketing & Sales

--

Yes

Yes

--

--

Yes

--

[11] O. Kamga, “Mobile phone in Cote d'Ivoire: uses and self-fulfillment”

Marketing & Sales

--

Yes

--

--

--

Yes

Yes

[12] T. Molony, “‘I don't trust the phone; it always lies’: Trust and information and communication technologies in Tanzanian micro- and small enterprises”

Marketing & Sales; Service

--

Yes

No

No

--

--

Yes

[13] R. Overå, “Networks, distance, and trust: Telecommunications Development and changing trading practices in Ghana”

Inbound Logistics, Outbound Logistics, Marketing & Sales

Procurement

Yes

Yes

--

--

Yes

Yes

--

--

Yes

Yes

--

Yes

Yes

Yes

[1] R. Abraham, “Mobile phones and economic development: evidence from the fishing industry in India,”

[8] H. Horst and D. Miller, “The Cell Phone: An Anthropology of Communication.” [9] A. Jagun, R. Heeks, and J. Whalley, “The Impact of Mobile Telephony on Developing Country Micro-Enterprise: A Nigerian Case Study”

[14] J. Samuel, N. Shah, and W. Hadingham, “Mobile Communications in South Africa, Tanzania, and Egypt: Results from Community and Business Surveys”

22

While it is one thing to say that mobile use accelerates the flow of information in existing value systems, it is another to say that mobile use brings new customers or suppliers into the market. This is the first element in the systematic review in which there is some disagreement between the primary studies. Numerous studies present evidence that mobile use expands markets by allowing MSEs to reach new customers [1, 2, 4, 7, 9, 10, 13, 14]. Of the Keralan fishermen, [10: 891] explains “while almost all sales before mobile phones were conducted via beach actions, fishermen with phones, often carrying lists with the numbers of dozens or even hundreds of potential buyers, would typically call several buyers in different markets before deciding where to sell their catch”. Similarly, Aker finds that “Grain traders in markets with cell phone coverage search over a greater number of markets, have more contacts and sell in more markets. This underscores the fact that the primary mechanism by which cell phones affect market efficiency is a reduction in search costs and hence transaction costs.” [2: 4-5] In the study, mobile use lowered price dispersion by 21%, increased profits by 29%. And yet two other studies specifically argue that the phones have done little to introduce new buyers. Donner [5] finds that MSEs in urban India are much more likely to recruit customers via face-to-face channels, rather than via a phone call. [12] argues that mobiles help accelerate and strengthen trusting relationships but only among parties that have already established a face-to-face bond. Two other general assertions about the impact of mobile use on MSE value systems receive less support from the studies. None of the studies asserts that mobiles help MSEs bypass middlemen. Indeed three of the papers focus specifically on middlemen, wholesalers or traders as enterprises, [2, 9, 13] describing how mobiles allow them to perform their roles more effectively. Another specifically emphasizes how producers work with existing middlemen in their industries, rather than routing around them. Rather than radically restructuring these marketplaces, Molony argues, “mobile phones can be seen as a facilitating technology for existing, trust-based relationships” [12: 78] Similarly, there is relatively little evidence for the assertion that mobiles help people start new businesses. Only Samuel et al. [14] make this case, reporting that among a sample of MSEs in Egypt and South Africa, 26%-29% of businesses attributed their start to the availability of the mobile. Taking the opposite position, Horst and Miller [8: 164] argue that despite some isolated examples to the contrary (taxi drivers and musicians), “there is no new spirit of enterprise based on either the cell phone or the internet” among the Jamaican households in their study. Nevertheless, they argue that despite a dearth of new enterprises, the mobile is essential to the economic survival of those households. By allowing individuals to leverage broad networks of informal social and financial support through a process Jamaican mobile users call “link up,” “the phone is not central to making money, but is vital to getting money.” [8: 165] In sum, in value systems where mobile telephony is

introduced, there is more evidence for changes in degree (more information, more customers) than for changes in structure (new channels, new businesses). C. On attributes of the mobile vs. the landline Roughly half of the studies described use cases that take advantage of mobility. Clearly, fishermen take advantage of wireless telecommunications [1, 10] to place and receive calls while on the water. This is not only an advantage for determining which markets to target, but [1] points out that it also enables fishermen to feel safer while at sea. Traders [2, 13] use the mobile to be individually addressable wherever they are. Reference [11] illustrates the responsiveness of businesses that can serve the customer, 24 hours a day, while [13] portrays “availability as comparative advantage”, and argues that this more frequent interaction builds trust between suppliers and customers. Given that MSE operators often carry their mobiles throughout the day and into the evening, a blending of mediated communication for social and instrumental purposes often occurs. While some of the papers in the review focus exclusively on the business functions, others [1, 3-6, 8, 11-14] illustrate this blurring. Blurring occurs at the aggregate level, —a survey by Donner [4] found that roughly 1/3 of calls made by MSE owners in Rwanda were business-related. It also occurs within individual calls—non-business (“chit-chat”) exchanges increase trust between clients and customers [12, 13]. Finally [8] describes the “link up” process in Jamaica, in which individuals retain a roster of numbers of friends, family and acquaintances that can be tapped periodically for loans or small cash gift transfers. This process intermingles social and economic functions of mobile use. VI. DISCUSSION This paper offers a systematic review of the current research on the impact of mobile use on MSEs, applying both an internal (value chain) and external (value system) perspective. The review finds a pattern of evidence suggesting that mobiles increase the information available to MSEs. Some [2, 10] provide quantitative evidence for how this information translates into reduced price variability and higher profits per actor. The current studies suggest mobiles are most useful for streamlining marketing and sales (downstream) and procurement (upstream) with existing business contacts. In some cases, studies suggest that mobile use expands the size of markets by bringing a larger number of buyers and sellers into the marketplace. However not all studies found evidence that new customers were acquired. Far fewer studies present evidence that mobiles enable the creation of new businesses, or that mobile use re-organizes value systems to allow producers to bypass middlemen. Indeed, middlemen are positioned to take advantage of mobiles themselves. To summarize, the review of the evidence offered across the thirteen studies suggests that within the MSE sector, benefits of mobile use accrue mostly (but not exclusively) to existing enterprises, in ways that amplify and accelerate

23

material and informational flows, rather than fundamentally transform them. This summary does not diminish the positive utility of mobiles to MSEs, but it places that utility in context and in doing so echoes assertions by Castells [54] and Harper [59]. Mobile use by MSEs can be understood as an extension of the information society, not its restructuring [60]. A second theme in the analysis involves mobiles as complements or substitutes for landlines. Evidence here remains mixed; while some case studies highlight enterprises that take advantage of “availability as comparative advantage” [13], there is insufficient data to determine whether these are isolated cases or representations of a more universal condition. The majority of MSEs may take greater advantage of place-to-place connectivity, than mobility, but this point merits further study. A. Generalization and segmentation Clearly, mobiles offer distinct benefits to MSEs— everything from more accurate price information and access to new customers to better after-sales service and procurement, from increased responsiveness to the opportunity to build trust at a distance. However, it is currently difficult to determine whether the various impacts and benefits enumerated in this review accrue equally to all MSEs. Looking across a diversity of MSEs, across nations, industries, and different locations in value systems, the current research points toward a multiplicity of intertwined and sometimes contradictory impacts of mobile use, rather than the universal and rather uncomplicated benefits which have characterized the popular rhetoric. To guide future policy or institutional interventions, it would be helpful to deploy future research against a set of open questions: Which kinds of MSEs are gaining the most return on mobile use? Which kinds (and what proportion) of MSEs are poised to find new customers and expand their markets, and not simply be more responsive to the ones they already have? Which kinds (and what proportion) are unlikely to reap any benefit from mobiles, or actually may be threatened by changes in mobile use elsewhere in their value systems? These questions remain open because most studies to date have been either sector-specific explorations or broad aggregate surveys; few studies specifically identify differences in mobile use or impact among subpopulations of MSEs. By contrast, recent studies of mobile use by farmers are identifying factors which differentiate between subgroups, for example, between growers of perishable and nonperishable crops [61], by distance from local markets [61], or according to different levels of infrastructural constraint [62]. An important path for further study would apply similar comparative analyses to assess and predict the impact of mobile use by different classes of MSEs. These are hefty quantitative tasks. Future designs will require increased attention to the factors that distinguish subgroups of MSEs as well as careful measurement of desirable outcomes such as productivity, market participation, or revenue growth. However, this review helps to identify a

range of variables for both the independent and dependant sides of such analyses. Indeed, a quantification of mobile use by subpopulations of MSEs may begin to close the gap between micro-level case studies and research on the impact of mobiles on macroeconomic growth [63]. B. Enterprises, livelihoods or lives? The conclusions of this review naturally depend on the methods and theoretical frameworks employed at the primary and secondary stages of analysis. Ethnographies such as [11] and [9] tended to discuss a broader range of uses and impacts than focused quantitative tests [2, 5, 10]. Similarly, the lenses chosen for this systematic review highlight some dimensions of MSE use over others. For example, a livelihoods framework [64], instead of enterprise-specific frameworks from Porter [23] would have emphasized different patterns. The dichotomous treatment of new vs. existing businesses has its limitations. Reference [9] describes the evolution of some weavers to “coordinator weavers,” suggesting a transformation in the structural location and internal processes of some enterprises, but this was an exception among the studies, most of which looked only to existing enterprises for their sample. The methods used by [8] can identify occupational multiplicity—holding down more than one job at once—in a way that studies focused specifically on existing MSEs cannot do. Similarly, [7] is able to assert that few households start new enterprises thanks to mobile use only because households rather than MSEs are its primary units of analysis. These examples reveal that insights about use of mobiles in MSEs can come from studies that focus not exclusively on enterprises, but rather on the individuals who manage them. Focusing on individuals also allows for increased linkage to research on social uses of the mobile. C. New applications on the mobile platform Studies have documented how mobiles can enable information search and improve communication between MSEs, customers and suppliers. However, there has been little evidence to date that suggests mobiles are being used for information storage or processing. As was the case with landlines [27], MSEs value voice calls more than any other function on the mobile, and use the calls to augment, rather than replace, face-to-face ties. As [6: 99] explains, mobiles “cannot be used to track inventory, provide cash flow and income statements, or even more basically, produce formal letters, marketing campaigns, or brochures”. Recently, however, various systems have appeared that go beyond the voice and peer-to-peer texting functions on the handset. These make the handset approximate a PC (with processing happening on the handheld), or as the client in a client-server model, with primary processing happening elsewhere on the network. These latter models take advantage of everything from basic SMS [65] to voice to full-blown mobile internet browsing experiences. Two of the more promising applications are distributed marketplaces, such as

24

Cell Bazaar, Manobi, and Tradenet, and mobilebanking/mobile-payments initiatives such as Kenya’s popular M-PESA [66]. In addition, a variety of promising pilots are underway, such as [48-51], which promise to yield further functionality, for example, in supply chain management. Full evaluations of the use and impact of these services on MSEs are not yet available, and are urgently needed. This baseline analysis can help clarify which impacts these systems may be having. Findings that suggest, for example, that mobile trading platforms help MSEs bypass middlemen, would be even more impressive given this synthesis’ conclusion that there is limited evidence that basic voice calls can have this effect. Similarly, findings that suggest MSEs are using m-banking or m-payments applications to transform credit relationships or otherwise change the procurement and sales functions could be assessed more accurately against the baseline of the voice-based behaviors found so far. VII. CONCLUSION This paper has summarized fourteen primary research studies examining mobile use by MSEs; the research generally concurs with conventional wisdom—mobile use helps many MSEs become more productive, particularly but not exclusively via improvements to sales and marketing and procurement processes. That said, the review suggests that not all enterprises will prosper from increased access to telecommunications, and among those that do, their uses of mobiles will vary across industries and positions in value systems. As mentioned above, current evidence suggests that the benefits of mobile use accrue mostly (but not exclusively) to existing enterprises, in ways that amplify and accelerate material and informational flows, rather than fundamentally transform them. On balance, MSEs are likely to remain unproductive relative to larger enterprises [17]. However, the improvements to productivity associated with mobile use do seem to be improving the livelihoods of many individuals in the MSE sector. The results of this review are helpful to the ICTD research community in at least three ways. First, by disaggregating and identifying distinct impacts of mobile use, the review provides a more nuanced and more accurate representation of the value of mobile use to MSEs than was previously available. Second, the review identifies a skew (in both sample and implied impacts) towards existing enterprises that should be noted by policymakers who may expect mobiles to create new businesses and new employment. Finally, the review identifies two priorities for future research: a segmentation and further quantification of impacts by subsectors of MSEs, and an assessment of the use of new non-voice advanced mobile services (such as mobile banking and mobile marketplaces) by MSEs REFERENCES [1]

R. Abraham, “Mobile phones and economic development: Evidence from the fishing industry in India,” Proceedings of the

[2]

[3]

[4]

[5]

[6]

[7]

[8] [9]

[10] [11]

[12]

[13] [14]

[15] [16] [17] [18] [19] [20] [21] [22]

International Conference on Information and Communication Technologies and Development, (ICTD 2006). Berkeley, CA: IEEE, pp. 48-56, 2006. J. C. Aker, “Does digital divide or provide? The impact of cell phones on grain markets in Niger,” BREAD Working Papers. vol. 177, Durham, NC: Bureau for Research and Economic Analysis of Development, 2008. J. Donner, “Microentrepreneurs and mobiles: An exploration of the uses of mobile phones by small business owners in Rwanda,” Information Technologies and International Development, vol. 2, pp. 1-21, 2004. ___, “The use of mobile phones by microentrepreneurs in Kigali, Rwanda: Changes to social and business networks” Information Technologies and International Development, vol. 3, pp. 3-19, 2006. ___, “Customer acquisition among small and informal businesses in urban India: Comparing face to face, interpersonal, and mediated channels,” The Electronic Journal of Information Systems in Developing Countries, vol. 32, pp. 1-16, 2007. S. Esselaar, C. Stork, A. Ndiwalana, and M. Deen-Swarra, “ICT usage and its impact on profitability of SMEs in 13 African countries,” Information Technologies and International Development, vol. 4, pp. 87-100, 2007. G. Frempong, G. Essegbey, and E. Tetteh, “Survey on the use of mobile telephones for micro and small business development: The case of Ghana,” Institutes of the Council for Scientific and Industrial Research, Science and Technology Policy Research (STEPRI), Accra, 2007. H. Horst and D. Miller, The cell phone: An anthropology of communication. Oxford: Berg, 2006. A. Jagun, R. Heeks, and J. Whalley, “The impact of mobile telephony on developing country micro-enterprise: A Nigerian case study,” Information Technologies and International Development, vol. 4, pp. 47-65, 2008. R. Jensen, “The digital provide: Information (technology), market performance, and welfare in the South Indian fisheries sector,” Quarterly Journal of Economics, vol. 122, pp. 879-924, 2007. O. Kamga, “Mobile phone in Cote d'Ivoire: Uses and selffulfillment,” Proceedings of the International Conference on Information and Communication Technologies and Development, (ICTD 2006). Berkeley, CA, pp. 184-192, 2006. T. Molony, “‘I don't trust the phone; it always lies’: Trust and information and communication technologies in Tanzanian microand small enterprises,” Information Technologies and International Development, vol. 3, pp. 67-83, 2006. R. Overå, “Networks, distance, and trust: Telecommunications development and changing trading practices in Ghana,” World Development, vol. 34, pp. 1301-1315, 2006. J. Samuel, N. Shah, and W. Hadingham, “Mobile communications in South Africa, Tanzania, and Egypt: Results from community and business surveys,” in Africa: the impact of mobile phones. vol. 2007, Vodafone, Ed. London: Moving the Debate Forward: The Vodafone Policy Paper Series #3, 2005, pp. 44-52. International Labor Organization, “Key indicators of the labour market,” International Labor Organization, Geneva 2003. D. C. Mead and C. Leidholm, “The dynamics of micro and small enterprises in developing countries,” World Development, vol. 26, pp. 61-74, 1998. R. la Porta and A. Shleifer, “The unofficial economy and economic development,” Brookings Institute, Washington, DC 2008. D. Snodgrass, “Inventory and analysis of donor-sponsored MSE development programs,” Development Alternatives Inc., Washington, DC 2005. Economist, “Calling an end to poverty,” The Economist. vol. 376, 2005, pp. 51-52. S. LaFraniere, “Cellphones catapult rural Africa to 21st century,” in The New York Times New York, p. 1, 2005. USAID, “Using cellular phones in Uganda for rural income generation and more,” Issue 7 ed. vol. 2006: DOT-COM Alliance E-Newsletter, 2004. A. Bhavnani, R. Won-Wai Chiu, S. Janakiram, and P. Silarszky, “The role of mobile phones in sustainable rural poverty reduction,”

25

[23] [24] [25]

[26] [27]

[28] [29] [30] [31] [32] [33]

[34]

[35] [36] [37] [38]

[39]

[40] [41] [42] [43] [44]

World Bank Global Information and Communications Department, Washington, DC 15 June 2008. M. E. Porter, Competitive advantage: Creating and sustaining superior performance. New York: Free Press, 1985. D. C. Mead and C. Morrison, “The informal sector elephant,” World Development, vol. 24, pp. 1611-1619, 1998. R. Duncombe and R. Heeks, “Enterprise development and information and communications in developing countries: Supporting ‘ICT flyers’,” Manchester: IDPM, University of Manchester, 2001. R. J. Saunders, J. J. Warford, and B. Wellenieus, Telecommunications and economic development, 2 ed. Baltimore, MD: Johns Hopkins University Press, 1994. R. Duncombe and R. Heeks, “Information and communication technologies and small enterprise in Africa: Findings from Botswana." vol. 2006 Manchester, UK: Institute for Development Policy and Management, University of Manchester, 1999. ___, “Enterprise across the digital divide: Information systems and rural microenterprise in Botswana,” Journal of International Development, vol. 14, pp. 61-74, 2002. J. Donner, “Research approaches to mobile use in the developing world: A review of the literature,” The Information Society, vol. 24, pp. 140-159, 2008. R. J. Light and D. B. Pillemer, Summing up: The science of reviewing research. Cambridge, MA: Harvard University Press, 1984. G. W. Noblit and R. D. Hare, Meta-ethnography: Synthesizing qualitative studies. Newbury Park, CA: Sage, 1988. J. H. Littell, J. Corcoran, and V. Pillai, Systematic reviews and meta-analysis. New York: Oxford University Press, 2008. D. Souter, N. Scott, C. Garforth, R. Jain, O. Mascararenhas, and K. McKemey, “The economic impact of telecommunications on rural livelihoods and poverty reduction: A study or rural communities in India (Gujarat), Mozambique, and Tanzania,” Commonwealth Telecommunications Organisation for UK Department for International Development, 2005. R. Duncombe and R. Heeks, “Information, ICTs and small enterprise: Findings from Botswana,” Small Scale Enterprises in Developing and Transitional Economies, H. Katrak and R. Strange, Eds. New York: Palgrave, 2002, pp. 285-304. J. O. Adeoti and A. I. Adeoti, “Easing the burden of fixed telephone lines on small-scale entrepreneurs in Nigeria: GSM lines to the rescue,” Telematics and Informatics, vol. 25, pp. 1-18, 2008. R. A. Boadi, R. Boateng, R. Hinson, and R. A. Opoku, “Preliminary insights into m-commerce adoption in Ghana,” Information Development, vol. 23, pp. 253-265, 2007. W. Mei and L. H. Yun, “Perpetual contact for business: Mobile phones and Fujian entrepreneurs,” Proceedings of ANZCA08: Power and Place, Wellington, NZ, 2008. S. Aminuzzaman, H. Baldersheim, and I. Jamil, “Talking back: Empowerment and mobile phones in rural Bangladesh: A study of the village pay phone of Grameen bank,” Contemporary South Asia, vol. 12, pp. 327-348, 2003. H. Ilahiane and J. Sherry, “Joutia: Street vendor entrepreneurship and the informal economy of information and communication technologies in morocco,” The Journal of North African Studies, vol. 13, pp. 243-255, 2008. J. Lugo and T. Sampson, “E-informality in Venezuela: The 'other path' of technology,” Bulletin of Latin American Research, vol. 27, pp. 102-118, 2008. B. F. Chogi, “The impact of mobile phone technologies on medium and small enterprises/jua kali (MSEs),” Proceedings of CPRsouth2 Chennai, 2007. R. O. Opiyo and O. A. K'Akumu, “ICT application in the informal sector: The case of the Kariokor market MSE cluster in Nairobi,” Urban Forum, vol. 17, pp. 241-261, 2006. E. D. Moyi, “Networks, information and small enterprises: New technologies and the ambiguity of empowerment,” Information Technology for Development, vol. 10, pp. 221-232, 2003. K. O. Davis, Cosmas. , “ICTs as appropriate technologies for African development” IFC/FT First Annual Essay Competition: Business & Development: The Private Path to Prosperity,

[45]

[46] [47] [48]

[49]

[50]

[51]

[52] [53] [54] [55]

[56] [57] [58]

[59] [60] [61] [62]

[63]

[64]

Washington, DC: Financial Times Limited / International Finance Corporation (IFC), pp. 43-54, 2007. A. T. Rashid and L. Elder, “Mobile phones and development: An analysis of IDRC-supported projects,” Electronic Journal of Information Systems in Developing Countries, vol. 36, pp. 1-16, 2009. T. Molony, “Running out of credit: The limitations of mobile telephony in a Tanzanian agricultural marketing system,” Journal of Modern African Studies, vol. 46, pp. 1-22, 2008. ___, “The role of mobile phones in Tanzania’s informal construction sector: The case of Dar es Salaam” Urban Forum, vol. 19, pp. 175-186, June 2008. P. S. Javid and T. S. Parikh, “Augmenting rural supply chains with a location-enhanced mobile information system,” Proceedings of the International Conference on Information and Communication Technologies and Development, (ICTD 2006). Berkeley, CA pp. 110-119, 2006. A. Kumar, N. Rajput, S. Agarwal, D. Chakraborty, and A. A. Nanavati, “Organizing the unorganized - employing IT to empower the under-privileged,” Proceedings of the 17th international conference on World Wide Web, Beijing, pp. 935-944, 2008. D. Chakraborty, K. Dasgupta, S. Mittal, A. Misra, A. Gupta, E. Newmark, and C. L. Oberle, “Businessfinder: Harnessing presence to enable live yellow pages for small, medium and micro mobile businesses,” IEEE Communications Magazine, vol. 45, pp. 144151, 2007. S. Biswas and S. Roy, “A shared ICT infrastructure for Indian SME clusters,” in 2nd IEEE/ACM International Conference on Information and Communication Technologies and Development (ICTD2007), Bangalore, pp. 221-230, 2007. Food India, “Inter-city marketing network for women microentrepreneurs using cell phone,” Food India 2002. J. E. Katz and M. Aakhus, “Perpetual contact: Mobile communication, private talk, public performance,” Cambridge, UK: Cambridge University Press, 2002. M. Castells, J. L. Qiu, M. Fernández-Ardèvol, and A. Sey, Mobile communication and society: A global perspective (information revolution and global politics). Cambridge, MA: MIT Press, 2007. C. Licoppe, “'connected' presence: The emergence of a new repertoire for managing social relationships in a changing communication technoscape,” Environment and Planning D: Society and Space, vol. 22, pp. 135-156, 2004. A. M. Townsend, “Life in the real-time city: Mobile telephones and urban metabolism,” Journal of Urban Technology, vol. 7, pp. 85-104, 2000. J. Hamilton, “Are main lines and mobile phones substitutes or complements? Evidence from Africa,” Telecommunications Policy, vol. 27, pp. 109-133, 2003. J. Donner, “The social and economic implications of mobile telephony in Rwanda: An ownership/access typology,” in Thumb culture: The meaning of mobile phones for society, P. Glotz, S. Bertschi, and C. Locke, Eds. Bielefeld, Germany: Transcript Verlag, 2005, pp. 37-52. R. Harper, “Are mobiles good or bad for society?,” in Mobile democracy: Essays on society, self and politics, K. Nyíri, Ed. Budapest, Hungary: Passagen Verlag, pp. 185-214, 2003. J. Donner, “Shrinking fourth world? Mobiles, development, and inclusion,” in Handbook of mobile communication studies, J. Katz, Ed. Cambridge, MA: MIT Press,2008, pp. 29-42. M. Muto and T. Yamano, “The impact of mobile phone coverage expansion on market participation: Panel data evidence from Uganda,” Japan Bank for International Cooperation, Tokyo 2008. S. Gandhi, S. Mittal, and G. Tripathi, “The impact of mobiles on agricultural productivity,” in India: The Impact of Mobile Phones, The Vodafone Policy Paper Series #9 London: Vodafone, 2009, pp. 21-33. L. Waverman, M. Meschi, and M. Fuss, “The impact of telecoms on economic growth in developing nations." vol. 2007 London: Moving the Debate Forward: The Vodafone Policy Paper Series #3, 2005, pp. 10-23. R. Duncombe, “Using the livelihoods framework to analyze ICT applications for poverty reduction through microenterprise,”

26

[65]

i

Information Technologies and International Development, vol. 3, pp. 81-100, 2007. R. Veeraraghavan, N. Yasodhar, and K. Toyama, “Warana unwired: Mobile phones replacing PCs in a rural sugarcane cooperative,” Proceedings of the 2nd IEEE/ACM International

The analysis includes two studies of fishermen but excludes farmers ii The term ‘value system’ is sometimes used interchangeably with ‘industry value chain’. This review uses the Porter nomenclature to distinguish between the intra- and extraenterprise systems

[66]

Conference on Information and Communication Technologies and Development (ICTD2007), Bangalore, pp. 89-98, 2007. O. Morawczynski and G. Miscione, “Examining trust in mobile banking transactions in Kenya: The case of m-PESA in Kenya,” in IFIP WG 9.4-University of Pretoria Joint Workshop, Pretoria, South Africa, 2008

27

An Evaluation of the use of ICT within Primary Education in Malawi David Hollow and Paola Masperi

Abstract— The paper demonstrates how appropriate technology, when combined with quality curriculum-based content, has the potential to have a positive impact on primary education within developing countries. It documents an evaluation of portable learning technology from the Ministry of Education, Science and Technology in Malawi, providing theoretical background and educational context before detailing the methodology and findings of the study. Significant impacts from the introduction of the program were increased school attendance, reduced dropout rates and improved student and teacher enthusiasm. Alongside this, the audio and video material and interactive learning techniques offered potential pedagogical benefits in combining learner-centered and outcome-based activities with continuous assessment, helping children to retain information more effectively and record higher test scores. Major challenges were also identified regarding implementation and sustainability. These centered around the need for rigorous teacher training, classroom integration, appropriate deployment, maintenance, sustained impact and overall cost-effectiveness. In closing, the paper emphasizes the need for such programs to be driven by educational concerns and recognizes the similar challenges faced in many related initiatives. Index Terms— Education, Evaluation, Malawi, Technology

I

I. INTRODUCTION

N setting the context for the subsequent evaluation the paper begins with an overview of two distinct spheres, firstly the current education context in Malawi and secondly the potential role of technology within education. These two seemingly disparate themes are brought together through the case study of the Interactive Learning Program. An overview of the program is provided, followed by an explanation of the methodological approach employed in the monitoring and evaluation exercise. The findings are then categorized into four sections of analysis which inform the concluding comments and recommendations. The Interactive Learning Program in Malawi comes under

Manuscript received September 21, 2008. David Hollow is a PhD candidate with the ICT4D Collective in the Geography Department at Royal Holloway, University of London. e-mail: [email protected] Paola Masperi is an independent development consultant, based in London. e-mail: [email protected]

the authority of the Ministry of Education, Science and Technology (MoEST). Funding for the initiative was provided by a British company who also developed the educational software for the portable devices. II. CONTEXT A. Education in Malawi Free primary education was introduced in Malawi in 1994 as a result of the commitment made by the country at the World Conference on Education for All (EFA) in 1990. Widespread agreement exists regarding the vital place of education within poverty reduction efforts, capacity building and growth strategies of developing countries. In Malawi, numerous policy, budgetary and multilevel commitments have contributed to significant progress in the delivery of primary education since 1994. This includes enrolment figures rising from 1.9 million to 3.2 million, the construction of 1,000 new classrooms and roll out of the new curriculum to 5,500 schools [1]. However, Malawi is one of the poorest countries in the world, with a GNI per capita of $230. Life expectancy at birth is 48 years and 63% of the population live on less than $2 per day [2]. Within such a context of extreme poverty, the rapid advancements in access to free schooling have put the national education system under considerable strain. The increase in enrolment has led to significant pressure upon primary schools in Malawi, most notably in regard to increasing class sizes, a lack of fully qualified teachers, limited teaching materials and inadequate infrastructure. It is estimated that an extra 8,000 teachers are currently required to meet the MoEST target pupil to teacher ratio of 1:60 [3]. In addition, and despite national net enrolment levels of 91%, national drop out rates remain high with only 44% of those that enroll in Standard 1 completing Standard 5 and less than 30% reaching Standard 8, the final year of primary education [4]. Linked to this and demonstrating the ongoing challenge in regard to teaching capabilities and educational outcomes is a declining performance in national examinations with a failure rate of over 40% [1]. The importance of ensuring good quality education alongside improved access [5] has been long recognized in Malawi and in 1999 it was decided that a comprehensive

28

reform of the curriculum was required due to the developments that had occurred since 1994 [3]. The relevance of topics such as democracy, human rights, gender and HIV/AIDS were increasingly acknowledged and these have been systematically incorporated into the new primary school curriculum in the form of life skills education. B. Technology in Education There is significant attention focused upon the potential for ICT to assist in leapfrogging educational challenges within developing nations [6] [7] [8], with much enthusiasm for a possible technology-enabled ‘breakthrough in learning’ [9]. However, much debate surrounds the question of whether the infusion of technology into education has actually instigated more than incremental changes to the field. Proponents assert that the last decade have resulted in the emergence of a new landscape for education across the globe [10] with technology positively effecting student motivation [11] and, when initiatives are implemented with fidelity, leading to a significant increase in learning [12]. This is challenged by those who argue that achieving structural technological change in schools takes much longer than anticipated [13] with no substantial evidence that the introduction of technology has yet caused any fundamental change in either a developed or developing context [14]. Wagner [15] notes a wide variety of outcomes from ICT for education projects in developing countries, with significant negative impacts including the reinforcing of dependencies, imposition without community involvement and collapse due to lack of funding or political commitment. Within this contested environment, the range of technologies theoretically available for deployment within developing country educational contexts is rapidly expanding, with a transition towards increasingly portable, powerful and adaptable tools [16]. The new technologies available have potential to mark a significant transition beyond conceptualizing e-learning through conventional static computer laboratories [17], emphasising instead the place of anytime, anywhere computing [16] and mobile learning [18]. Such a shift creates opportunity for major changes in the application of educational technology, with key potential benefits noted as increased enthusiasm, cooperation, communication and student ownership [19]. Despite the potential benefits of learning technologies [20] it is important to recognize that not all solutions which have contributed to educational advancements in the developed world can simply be transferred to a developing world context. In order for the potential of ICTs to be realized in improving the quality of education it is important they are applied with cultural understanding, local knowledge and sensitivity [11] [21]. Indeed, ‘effective use of ICTs must be tied to the needs of developing countries and challenge the one size fits all approach of many programmes’ [19;7]. A recognition of this fact emphasises the need for primary focus on appropriate software development linked to curriculum-

based content and sustained classroom integration. III. INTERACTIVE LEARNING PROGRAMME Having outlined the overall context of education within Malawi and the theoretical role of technology within education, attention now turns to the Interactive Learning Program. It considers whether the initiative constitutes a suitable application of portable technology to support the provision of basic education within the country. To begin, the specifics of the program are explained, including the background, technology and pedagogical rationale. A. Background to Program The Interactive Learning Program was introduced in 2006 and following initial positive feedback the MoEST requested, in early 2007, that the initiative be scaled up to incorporate 50 primary schools. The schools that the MoEST selected for participation ranged from Karonga District in the north through to the southernmost districts of Mulanje and Phalombe. A total of 520 custom-made handheld interactive learning aids were distributed to the participating schools. This totaled ten devices for each school, except for two test schools which received 20 devices for the purposes of targeted monitoring and evaluation. B. Technology Utilized The device in question is a handheld interactive learning aid (from herein referred to as the learning machine or gadget as named by the participating children), slightly larger than a mobile phone and able to play video and audio through either a loudspeaker or headphones. Positioned next to the screen are a selection of buttons which can be pressed in response to questions asked in the lesson being watched. The device has an in-built rechargeable lithium battery with power for between four and six hours of continuous play. A total of 25 lessons in Chichewa and 40 in English are preinstalled and stored on two gigabytes of internal Flash memory, leaving additional room for newly developed lessons to be incorporated. The lessons are designed for use by children in Standard 3, 4 and 5, each lasting approximately 20 minutes and covering a range of curriculum based subjects such as General Studies, Social Studies, Science, Mathematics, Geography, Life Skills and English. With less than 5% of primary schools across Malawi estimated as having access to a reliable supply of mains electricity [3] the initiative was dependent upon solar charging systems. Each of the project schools were equipped with a 20Watt solar panel connected to a deep-cycle sealed battery, generating enough energy to power 20 of the learning machines. Teachers were given training regarding appropriate charging and usage, and instructed to leave the devices to recharge overnight once every two days.

29

C. Pedagogy The digital content on the learning machines is designed in order to enable and promote interactive and outcome-based learning, linked to the Malawi national curriculum and actively supporting the government Primary Curriculum Assessment Reform [22]. The lessons are designed to be user friendly, with the audio-visual content accessible to illiterate or semi-literate users. Instructions from lip synchronized cartoon characters explain the functions of the various buttons on the learning machine. Once a lesson title has been selected the methodology becomes interactive and learner-centered. Having listened to brief teaching points the learners are tested on their understanding of the information supplied through the posing of multiple-choice questions answered by pressing buttons on the device. At the end of the lesson learners are given opportunity to test what they have learnt through undertaking an overall quiz of ten questions where they are congratulated according to the score attained. This outcomebased approach provides incentive for the children to work conscientiously and allows the teacher to monitor progress. The decision to include lessons in both Chichewa and English was taken on the basis that early learning content is most effective when communicated in the vernacular language [23] [24]. The dual language approach equips children for the challenging transition from Chichewa to English as the official medium of school instruction which takes place in Standard 5. The most conducive learning environment for using the devices is with groups of between four and six students. In this context team work, group participation and the development of leadership skills can each be encouraged. IV. METHODOLOGICAL APPROACH In regard to the overall context of technology within education, Kozma [25;21] concludes after assessing a wide variety of studies that there is ‘no consistent relationship between the mere availability or use of ICT and student learning’. Beyond such generic assertions, there are significant knowledge gaps remaining regarding what kind of initiative works and what does not. As InfoDev [11;5] summarize, ‘despite thousands of impact studies, the impact of ICT use on student achievement remains difficult to measure and open to much reasonable debate’. In light of such observations all partners in the Interactive Learning Program initiative recognized the necessity for a comprehensive monitoring and evaluation exercise, the key objectives of which were to; • assess the impact of the program on primary education • assess the feasibility of the program in each test school • identify program weaknesses for future refinement • develop both teacher and organizational capacity • provide feedback to partners regarding suitability of scaleup and sustainability A common theme from ICT-enhanced education programs within developing countries is the marginalized place of

monitoring and evaluation. Where undertaken, there is often strong emphasis placed on hardware-based input indicators [25] rather than more complex learning outcomes. In order to avoid this simplification and demonstrate an alterative approach, a rigorous monitoring, evaluation and impact assessment structure was built into the program throughout implementation. This involved three field visits from the monitoring and evaluation team in September 2007, November 2007 and March 2008. In addition to being marginalized, monitoring and evaluation of ICT for education programs has rarely embraced creative processes of assessment [26]. Recognizing this, a systemsbased approach was utilized in designing the methodology [27] [28] [29]. This involved adopting a mixed- and multimethod approach which focused on process and aimed to engage with multiple stakeholders so as to gain a plurality of perspectives regarding program impact. It also ensured that data gathered was not solely anecdotal but credible, dependable and confirmable [30]. Within this overall aim for increased creativity and rigor, the research remained subject to the standard complexities and constraints of limited time, budget, data and personnel [31] [32]. It light of the constraints it was decided by the MoEST that five representative test schools would serve as a sample of the 50 participating in the program. The schools selected by the MoEST were Chin’gombe, Mwatibu, Mthentera, Mbinzi and Dzenza. None of these schools had any form of ICT incorporated into the curriculum prior to the introduction of the program. A total of 15 days were spent in these schools due to a recognition of the benefits of prioritizing classroombased research [33]. A combination of qualitative and quantitative methods were utilized and these are now outlined alongside the rationale for each. A total of 15 lessons in which the learning machines were being used were observed throughout the process [34]. Teachers were asked to conduct the lesson as they would normally, without altering classroom arrangements. Observing usage of the technology in a normal environment provided an ideal foundation for the subsequent methods, giving opportunity to discuss successes and suggest ways to work around challenges with the children, teachers and headteachers. Some 15 group interviews were held with children in the five schools, in order to hear what difference the program had made to their lives. The children were asked what they liked and disliked and improvements they would make for the future. The interviews followed a semi-structured, guided approach [35] [36] facilitated by a teacher and conducted in Chichewa. Participants were selected at random from Standard 3 and 4 classes, were aged between 7 and 15, and had equal gender representation. Following these group interviews, more detailed conversations were held with individual children in order to learn about particular experiences or perspectives that they had expressed. Such unstructured and story-based conversation was useful in identifying unanticipated program

30

impacts [37]. In addition to this, a three stage individual interview was conducted with the head-teacher of each school and also with four related officials from the MoEST. These interviews provided opportunity to obtain input regarding the program feasibility, impact and future direction. Conducting focus groups was considered to be the most appropriate method to gain detailed feedback from teachers regarding their view of the program and the difference it had made to their lives, including positive and negative impacts and potential improvements. A total of 15 focus groups were conducted in this safe environment [38] which facilitated robust dialogue [39] but ensured no one felt pressurized to share personal experiences [40]. Building on these, the learning octagon was a research tool developed specifically for the program in order to stimulate more detailed discussion between teachers in the focus groups. The octagon providing a pictorial representation to assess the strengths and weaknesses of eight different dimensions to the program by drawing on a combination of the Octagon tool [41] and Most Significant Change approach [42] [43]. This enabled the teachers to collaborate on creating a visual representation of the impacts of the initiative. In addition to these methods, and in order to enable ongoing monitoring on a weekly basis, one teacher from each of the five schools was selected to complete a diary documenting their experiences of using the learning machines in their lessons. The diaries were semi-structured and designed to provide a continuous record of particular strengths, weaknesses and challenges encountered. Baseline tests were conducted at stage one and three of the research. The objective of the test was to provide a quantitative assessment of the impact of the program on the attainment of the children regarding both curriculum and lifeskills. In each of the five test schools there were 12 children randomly selected from Standard 3 and 4, providing a total of 60 children, 30 boys and 30 girls. Five additional schools were selected to act as a control group with an additional 60 children tested. Alongside focusing on the test schools, evaluation questionnaires and equipment feedback forms were distributed to all 50 participating schools in order to gain a broader understanding of program impact. The questionnaires gave opportunity for feedback regarding patterns of usage and challenges encountered in implementation. The equipment feedback forms enabled the documentation of technical problems encountered throughout the program. Each of the methods outlined above enabled a detailed perspective to be developed regarding the strengths and weaknesses of the program. The monitoring and evaluation methodology employed was participatory throughout [44] and engaged with children, teachers, headmasters, community leaders, government officials and civil society representatives. Each school was visited at least three times throughout the six month research period and this allowed for a progressive approach and the development of good relationships with a number of stakeholders.

V. ANALYSIS The monitoring and evaluation exercise was focused on assessing three interconnected spheres which cut across the five previously identified research objectives. These were impact upon students, impact upon teachers and effectiveness of the technology. Analysis of each sphere is based on the data gathered using the range of methodological approaches previously outlined. The analysis is illustrated throughout with quotations from children, teachers, head teachers and MoEST officials. A. Impact upon Students In assessing the overall impact upon students the study considered the specific impact on four main areas of student attendance, enthusiasm, attainment in curriculum subjects and attainment in life skills subjects. A significant and universally agreed impact of the program was the increase in school attendance. Teachers reported a large increase in class sizes with fewer children absent than prior to the intervention. One student from Standard 4 explained why this was the case: Before the gadgets more pupils were absenting themselves from classes but now we encourage our fellow pupils to come to school and tell them, today if you absent yourself, you will miss using the gadgets. I used to absent myself 50% of the time before the gadgets came, now I come to school everyday. This is a significant achievement within a national context of high drop-out rates and absenteeism. For any educational intervention to succeed it is necessary for learners to be attending school on a regular basis and the motivating influence of appropriate technology is clearly a contributing factor in achieving this desired end. Despite strong anecdotal evidence, assessing the statistical significance of the change in attendance was hampered by a lack of daily attendance records maintained in each school, meaning it was impossible to track exact attendance patterns prior to the current term. Children in the group interviews reported that they were enthusiastic about coming to school now that they were using the learning machines. Several students reported having shared the experience with family and community members who actively encouraged them not to miss out on the opportunity to learn with the new technology. As one student from Standard 3 explained: They [parents] said to me that you have to work hard and make sure you do not run away from lessons when you are using the computer. I used to run away from class but now I have changed my behavior … because I am attracted to the gadgets and if I miss the chance to use it, it will never come again. Other children emphasized what they had learnt through the lessons and how this related to what they hoped to do in the future. One student from Standard 5 reported the impact of a particular lesson regarding counting:

31

Now I know how to add and subtract – I could see the pictures and I was able to subtract. This makes a foundation for me to be able to work in a bank – this is what I want to do in the future. The learning machines are perceived by the children as a mixture between mobile phones, games devices and video players. This makes them attractive to the children, who quickly become familiar with them and enjoy the process of being congratulated by the device for answering questions correctly. Observed learners appeared to enjoy working in groups and taking turns in pressing the buttons. Teachers also noted a similar impact and emphasized the improved listening skills of the students: They are able to explain things now – and the gadgets really help them with listening skills – if they do not listen then they miss what has been said and they cannot answer the question – and it makes them be fully attentive. Increased attentiveness in class and greater motivation to attend school also had an effect on the attainment of the children in both curriculum and life skills based lessons. Both students and teachers reported that the use of audio and video, as well as the continuous assessment quizzes, increased retention and affected attainment. One teacher reflected on tests completed the previous week: We had the mid-term tests last week and more pupils did well than before – when I asked them why they said it was because of the gadgets – ‘we just remember what we have learnt on them’. The increase in attainment has gone up by about 30% on average from what they normally achieve. Despite such assertions it was difficult to assess quantitative impact on student attainment due to the short period of time since the beginning of the intervention. A comparison of scores between the baseline and second test demonstrated that certain sectors of the curriculum had considerably more correct answers after using the learning machines. However, the majority of students had only completed 15% of available lessons at the time of the second test and the overall improvement in curriculum attainment was limited to 1.5%. Similarly, improved attainment in life skills was difficult to quantify as test scores related to knowledge gained rather than necessarily to behavioral and lifestyle change. The overall impact on life skills attainment in the baseline test was a 3% improvement. Again, the questions which were answered correctly indicated an increased understanding of the content for those lessons which the teachers had chosen to use most regularly. Although quantitative change proved difficult to assess, many students interviewed were able to recall accurate information on the life skills lessons they had used and expressed satisfaction to have learnt skills useful to their daily lives. Learners explained how they had put into practice what had been taught in the lessons, indicating potential behavioral change: [my favorite lesson was…] Preventing malaria – because it affects many children in the village. I learnt that we must sleep under treated bed nets and must not play with stagnant

water. I did not know this before the gadgets. Children were also able to talk about culturally sensitive subjects such as HIV/AIDS and explain what they had learnt from the lessons. Teachers stressed the value of the interactive lessons in a cultural context where such discussions are often considered taboo. The children requested that new content be added to the devices so that they could learn about a greater variety of topics. The most commonly requested subjects were Agriculture, Mathematics, English, Religious Education, Science and Technology, Physical Education and Music. Similarly, the teachers requested that new content would remain linked with the curriculum reform in order to provide them with support in this transition. The impact of the initiative on the students is intrinsically linked to the way in which the teachers are affected, and this is now considered. B. Impact upon Teachers In assessing the overall impact upon teachers the study considered two factors, teacher enthusiasm and teacher workload. The majority of teachers were enthusiastic about the introduction of the program into their school and several commented that they felt honored to have been selected as a school for the pilot. They were pleased that MoEST officials were visiting their school and showing an interest in their work. They were keen to learn about technology and could see the positive effects on themselves and the learners. They also noted how the introduction of the learning machines had helped them in altering their teaching style, adopting new and innovative approaches: We love it – before the gadgets it was just talk and write for some subjects – but now the kids can see the pictures. Some of the children can be sitting doing one thing while others are doing the gadgets. Young trainee teachers particularly enjoyed using the technology and often engaged in using it without apprehension. This confidence had a clear effect on the manner in which the children approached the devices and made use of them in the classroom. The enthusiasm of the students also affected the teachers: …because if the learners are enthusiastic then it makes us enthusiastic. If the gadgets help the students to learn then the teachers have to be happy that the students are doing well. However, this enthusiasm was not felt universally. Mature teachers often showed less interest in the program, were fearful of using the technology and reluctant to let the children use it independently. In one of the test schools after six months of usage there were only three teachers still using the learning machines. All the other teachers in the school had become uninterested in the program, considering the additional workload to be too much of a burden. The teachers reported a variety of different experiences regarding the impact of the program on their workload. This was dependent upon the manner of implementation in each

32

school and the degree to which the learning machines were incorporated within the lesson schedule. The overall feeling from the teachers was that the introduction of the gadgets had increased the workload in certain areas and decreased it in others: There are two things, lesson preparation - for this it might take more effort - and lesson delivery - for this it might take less time. It has made us do more work – we have to guide the students. And there are not enough gadgets – so if we had enough gadgets then it would be ok and then it would not increase our workload. Since the intervention was a pilot project the majority of schools chose to use the gadgets after the school day had finished so as not to interfere with the daily timetable. This meant that the teachers were required to stay in school for an extra hour every day to facilitate the lessons. Most were willing to do this but some expressed reluctance due to preexisting after-school commitments. It was also commented in two schools that the initiative resulted in more preparation work for the teachers because they needed to become fully familiar with the devices themselves before making use of them in lessons with the children. In the two schools where the teachers had chosen to incorporate the gadgets into the school day then a decrease in workload was reported. In these cases the gadgets helped with effective facilitation, the management of large classes and provided support in preparing teaching and learning materials for the lessons. In light of this, the program would benefit from the development and distribution of a pedagogical guide for teachers regarding the effective integration of the learning machines into the learning routine of the children. Teachers should also be encouraged to utilize the full range of lessons that are available on the devices. This would complement the development of a schedule for teachers to use the machines during school time, giving clear explanation how it can most appropriately support the curriculum. The increased ability of the children to memorize what had been taught was also recognized to make the job easier by one teacher from Standard 4: Gadgets have simplified the work for the teachers – when the child sees something he remembers – when he just hears he forgets. When they see it, it stays in their mind. Others noted that the impact of the learning machines in encouraging so many children to return to school had caused class sizes to expand significantly, resulting in more work attempting to manage the children. A common concluding comment from teachers was that the workload would decrease significantly if there was a lower pupil to gadget ratio and if the lessons on the gadgets were fully compatible with the syllabus and incorporated into the curriculum. The program would therefore benefit from an increased allocation of 20 learning machines per school. This would facilitate a more concentrated impact and allow a greater number of children in smaller groups to benefit from the content on the devices. In light of the challenges presented, it is clear that regular training sessions for the teachers would assist in maintaining

the momentum of the program and improving teacher motivation. The limited budget of the MoEST makes it a constant challenge to deliver adequate training, especially in remote areas. The teachers requested opportunity to ask questions and provide feedback, requiring effective communication channels between the stakeholders. As highlighted by one official: If you come back frequently then the schools will know you are coming and they will keep using the gadgets but if they think no one is coming to visit them then they will forget to use them. This is because when they are monitored they will feel that somebody is appreciating their job – they are happy that they can be involved and they see that somebody cares. The feedback from teachers demonstrated that sustained monitoring from MoEST officials constituted a significant capacity building process. The impact was especially pronounced in marginalized rural areas which were inaccessible and rarely visited. In light of this is it clear that a continuing focus on teacher training which both develops skills and instills value is intrinsic to the success of this and any ICT for education initiative. The ability of the teachers to utilize technology with confidence and adapt it to the specific needs of the classroom is a central determining factor. Without this integration any initiative remains an appealing add-on but does not have significant effect on the culture of the classroom, pedagogy, or rationale for learning. C. Effectiveness of the Technology The success of the initiative, whilst not determined by the technology, was dependent upon it operating effectively within each school. Usability, durability and the reliability of the devices and charging solution were the three areas assessed in the study. The lesson observations demonstrated that the majority of children were able to operate the learning machines independently, understanding the purpose of the buttons and completing quizzes without supervision. The process of lesson selection occasionally required guidance from teachers but this was rarely necessary because the most capable children were seen to take a lead in the learning groups, demonstrating considerable confidence with the devices. However, a significant factor limiting usability was the low number of devices per classroom and subsequent high number of students per group, on occasion as many as ten learners. This was not conducive to learning as the machine is designed for small groups, with small screen and quiet audio output. In such situations, dominant children were observed bringing the machine to their ear in order to hear the lesson, meaning other children were unable to engage and became easily distracted. This was confirmed by a participating official from the MoEST who noted: Bringing such technology is very important - it is only that at the moment the gadgets are very few – there needs to be more. It is important that you have the right number of gadgets for the number of children because it is difficult for so

33

many children to share. Due to the nature of such a pilot program it was anticipated that there would be significant problems regarding device durability. However, after six months of usage only 8% had developed problems, the majority being due to software issues easily resolved through reformatting. Two of the devices stopped working due to the on/off switch breaking, these were repaired in situ and left fully functioning. A major concern in this respect was that teachers often assumed they had broken the devices when all that was actually required was for them to be charged fully. The solar panels operated effectively in each of the five test schools throughout the program with all of them maintaining adequate power to charge the devices as often as necessary. Several program schools outside the five included in the test did report problems including two faulty batteries, one loose connection and one stolen panel. In addition to these, one school adopted the practice of disconnecting the battery from the panel to remove it for safe keeping during the night and this prevented the battery from charging fully. VI.

CONCLUSIONS

The analysis of impact on students, teachers and technological effectiveness has demonstrated the educational potential of the Interactive Learning Program. The introduction of the devices had significant positive impact upon school attendance and levels of enthusiasm. Also noteworthy was the increased value placed on teachers and the role of education within the community. However, alongside this, considerable challenges were encountered including the negative impact on teacher workload, lack of classroom integration, significant infrastructural constraints and the necessity for further teacher training. The future success, sustainability and potential for project scalability is dependent upon engaging with each of these identified challenges. The observations made in the analysis speak directly to the program efficacy but are also applicable lessons for ICT for education initiatives more widely. The introduction of technology into education systems in the developing world is a complex procedure with considerable potential for failure and therefore it is vital that rigorous monitoring and evaluation is incorporated throughout all such initiatives. The price of portable educational technology is constantly decreasing, making initiatives such as the Malawi Interactive Learning Program increasingly cost effective. The September 2008 price of each learning machine was circa US$55 and as each device can be used by several children the per child cost is considerably lower than other widely publicized solutions. Significant future potential also lies in the opportunity for integrating the educational content with new generation mobile phones and other handheld devices becoming progressively more available across the region. Linked to this, increased ability to send and receive data in standard formats (for example Macromedia Flash, xml, mp3 or mp4) will allow

for the development of user interaction. Market influences on the development of mobile technology also ensure sustained improvements in hardware capabilities and the subsequent emergence of new educational possibilities. However, within all such initiatives, it is necessary that the application of technology is conceptualized as a tool in facilitating the overall aim of catalyzing a more fully effective approach to education. In reaching this goal many of the challenges documented remain the same regardless of the presence of technology. It is not overly surprising that the introduction of ICT into primary schools in Malawi caused a dramatic increase in attendance figures. The significant question for ongoing research remains one of long term impact on approach to schooling, once initial enthusiasm surrounding the program has subsided. Whilst a significant achievement, increasing school attendance figures is not the most significant long-term challenge facing education in Africa [5]. Instead, it should constitute an initial building block which can be capitalized on through adopting an integrated approach to developing the capacity of the education system. One dimension to this will be the effective utilization of curriculum enhancing technology, such as the Interactive Learning Program. For full benefits to be realized, such programs require transition from their position of peripheral curiosity to one of being integrated as a sustainable teaching tool. Future research is required to assess the degree to which this is occurring in Malawi. A variety of different solutions are required to address the challenge of providing equitable access to good quality education in the developing world. The increasing availability and decreasing cost of portable devices ensures that they are likely to play a prominent role in the future, however this should not lead to such technologies being presented as an educational panacea. For programs to be of maximum educational benefit in Malawi, as in many such initiatives, the critical issues for consideration and action remain pedagogy, classroom integration and teacher training. REFERENCES [1] [2] [3] [4] [5] [6]

[7] [8] [9]

MoEST, Link for Education Governance. Education Sector Performance Analysis. Lilongwe, Malawi. 2007. World Bank, Data and statistics: Malawi. 2006. www.web.worldbank.org Accessed 21/08/08 MoEST, Primary Curriculum and Assessment Reform (PCAR). Implementation Plan 2007-2010, Lilongwe, Malawi. 2008. UNESCO, Institute for Statistics, country data 2008. www.uis.unesco.org Accessed 21/08/08 EFA, Global Monitoring Report: The quality imperative. 2005. www.portal.unesco.org/education Accessed 21/08/08 D. Pye and J. Stephenson, Using ICT to Increase the Effectiveness of Community-based, Non-formal Education for Rural People in SubSaharan Africa: the CERP Project Final Report (DFID Educational Paper 50). London: Department for International Development 2003. V. Tinio, ICT in Education. E-Primers for UNDP, New York 2003. J. Leach, DEEP IMPACT: an investigation of the use of information and communication technologies for teacher education in the global south. Researching the Issues, 58. DFID 2005. D. Wagner, Monitoring and evaluation of ICT for Education: An Introduction. Chapter 1 in: D. Wagner et. al. 2005. Monitoring and

34

[10] [11] [12] [13] [14]

[15] [16] [17] [18] [19]

[20] [21] [22] [23] [24] [25]

[26]

[27]

[28] [29] [30] [31] [32] [33] [34]

Evaluation of ICT in Education Projects – A Handbook for Developing Countries, InfoDev 1-19, 2005. D. Keats, The genesis and emergence of Education 3.0 in higher education: the potential for Africa. First Monday 12, 3, 2007. InfoDev, Knowledge Maps: ICTs in Education. World Bank 2005. C. Fadel and C. Lemke, Technology in Schools: What the research says. Metiri Group, Commissioned by Cisco Systems 2006. L. Cuban, Oversold and Underused – Computers in the Classroom. Harvard: USA 2001. T. James and J. Miller, Developing a Monitoring and Evaluation Plan for ICT in Education. Chapter 4 in: D. Wagner et. al. 2005. Monitoring and Evaluation of ICT in Education Projects – A Handbook for Developing Countries, InfoDev. 57-77 2005. D. Wagner, B. Day, J. S. Sun, Recommendations for a pro-poor ICT4D non-formal education policy. Final Report for, Imfundo: Partnership for IT in Education, DFID, 2004. M. Sharples, Disruptive Devices: personal technologies and education. Educational Technology Research Paper Series 11, The University of Birmingham, 2000. H. U. Hoppe, R. Joiner, M. Milrad and M. Sharples, Wireless and Mobile Technologies in Education. Journal of Computer Assisted Learning, Guest editorial, 19, 3, 255-259, 2003. F. Lehner and H. Nosekabel, The Role of Mobile Devices In E-Learning -First Experiences With A Wireless E-Learning Environment. WMTE 103-106, 2002. J. Leach, T. Power, R. Thomas, X. Fadani and A. Mbebe, 4D Technologies: appropriating handheld computers to serve the needs of teachers and learners in rural African settings. Centre for Research and Development in Teacher Education, Open University: UK, 2003. C. Pontefract, Learning to Share, Insights Education 1 Missing the Connection – using ICTs in Education, February 2003 R. Dhanarajan, Learning Technologies, where is the challenge? Education, Communication, Information 1, 1 Spring 2002 MoEST, Primary Curriculum and Assessment Reform (PCAR). Literature Review of Primary Education in Malawi, Lilongwe, Malawi, 2002. G. Bunyi, Rethinking the place of African indigenous languages in African education. International Journal of Educational Development 19, 4-5, 337-350, 1999. EFA, Global monitoring report: Literacy for Life, 2006. www.portal.unesco.org/education Accessed 21/08/08 R. Kozma and D. Wagner, Core Indicators for Monitoring and Evaluation Studies for ICT in Education. Chapter 3 in: D. Wagner et al. 2005. Monitoring and Evaluation of ICT in Education Projects – A Handbook for Developing Countries, InfoDev. 35-57, 2005. J. Taylor and S. Soal, Measurement in Development Practice: From the Mundane to the Transformational. Chapter 5 in L. Earle (ed) 2004. Creativity and Constraint: Grassroots Monitoring and Evaluation and the International Aid Arena. INTRAC Policy Series 18, 95-110, 2004. P. Morgan, Measuring the Development of Capacity: Is this Still a Good Idea? Chapter 2 in L. Earle (ed) 2004. ‘Creativity and Constraint: Grassroots Monitoring and Evaluation and the International Aid Arena’. INTRAC Policy Series 18, 47-62, 2004. D. Watson, Monitoring and evaluation of capacity and capacity development. ECDPM, Discussion paper 58B. April 2006 D. Cabrera, A Theory of Systems Evaluation. Systems evaluation and evaluation systems whitepaper series. New York: Ithaca, Cornell University, 2006. J. Baxter and J. Eyles, Evaluating qualitative research in social geography: establishing ‘rigour’ in interview analysis. Transactions of the Institute of British Geographers, 22, 4, 505-525, 1997. M. Bamberger, J. Rugh and L. Mabry, Real World Evaluation. Working Under Budget, Time, Data and Political Constraints. London: Sage, 2006. D. Gray, Doing Research in the Real World. California: Sage 2004. M. C. O’Sullivan, What is happening in the classroom? A commonsense approach to improving the quality of primary education in developing countries. Teacher Development, 9, 3, 301-314, 2005. J. W. Creswell, Educational Research: planning, conducting, and evaluating quantitative and qualitative research. New Jersey: Pearson Education, 2005.

[35] J. Burgess, The Art of Interviewing. In: A. Roger and H. A. Viles, (eds) The Student Companion to Geography. Blackwell: Oxford, 242-249, 2003. [36] M. Q. Patton, Qualitative Research and Evaluation Methods. California: Sage, 2002. [37] R. Chambers, Participatory Workshops – a sourcebook of 21 sets of ideas and activities. United Kingdom: Earthscan, 2002. [38] G. Valentine, Being seen and heard? The ethical complexities of working with children and young people at home and at school. Ethics, Place and Environment. 2, 141-155, 1999. [39] D. Stewart and P. Shamdasani, Focus Groups: Theory and Practice. Newbury Park: Sage, 1990. [40] P, Bhattacharjee, Stepping Stones – A participatory tool to integrate gender into HIV/AIDS work. Development in Practice, 10, 5, 691-694, 2000. [41] Sida. The Octagon. A tool for the assessment of strengths and weaknesses in NGOs, 2002. [42] P. Sigsgaard, Monitoring without indicators: an ongoing testing of the MSC approach. Evaluation Journal of Australasia. 2, 1, 2002. [43] J. Dart & R. J. Davies, A dialogical story-based evaluation tool: the most significant change technique, American Journal of Evaluation. 23, 2, 137-155, 2003. [44] M. Huberman, The Many Modes of Participatory Evaluation. Chapter 7 in J.B. Cousins and L. M. Earl (eds) 1995. Participatory Evaluation in Education. Studies in Evaluation use and organisational learning. London: Routledge, 103-114, 1995.

35

Claim Mobile: Engaging Conflicting Stakeholder Requirements in Healthcare in Uganda Melissa R. Ho, Emmanuel K. Owusu, and Paul M. Aoki

Abstract—Claim Mobile is a platform designed to support a project that subsidizes healthcare by reimbursing health service providers in Uganda for treatment of patients with sexually transmitted infections. As with many development projects, the Uganda Output-Based Aid (OBA) project involves a number of stakeholders: the service providers, the project implementers, the financiers, and the Ugandan government. Design of an appropriate solution requires meeting the various and conflicting requirements of all of these stakeholders. In this paper we detail the rapid design and testing of a pilot implementation of a mobile and web-based system for processing claims forms, based on two prior field visits to Uganda. Based on a comparative device study, semi-structured interviews, health clinic surveys, and a brief deployment, we affirm the selection of the mobile phone as a platform from the health clinic perspective, and further suggest that effective design for development requires more than addressing requirements of the the “users” of the mobile phones but also all the other stakeholders involved, who may have conflicting requirements. Index Terms—mobile phone, ICTD, health, participatory design, Africa, HCI

I. I NTRODUCTION Mobile phones are frequently touted as being the appropriate and sustainable platform for rural healthcare in Africa. They are relatively cheap, durable, consume less power than laptops and desktops, and incorporate a battery that makes them more amenable to use in places with intermittent or no power. Commonly proposed uses are for data collection [1], [2] and decision support for rural health workers [3], [4]. Some projects also use mobile devices as a platform for information dissemation as well as data gathering [5]. However, these are all generally “closed loop” systems in which researchers are able to control all aspects of the system design and operation, focusing their research primarily on the rural health workers that will be using the mobile phones. Other applications have even more potential for large-scale impact. In the agricultural sector, we have observed how the introduction of transparent market prices and subsequent hiring of “middlemen” to purchase from farmers has reduced Manuscript received October 1, 2009. This work was supported by the Blum Center for Developing Economies and the U.S. National Science Foundation Summer Undergraduate Program in Engineering Research at Berkeley (SUPERB) under Grant No. 0453604. Melissa R. Ho is with the School of Information at the University of California, Berkeley, 94720, USA. (phone: +256 777 723 786; email: [email protected]). Emmanuel K. Owusu is with the Computer Engineering Department at Iowa State University, 50011, USA. (email: [email protected]). Paul M. Aoki is with Intel Research, Berkeley, CA, 94704, USA (email: [email protected])

fraud and transformed supply-chain management for the Echoupal project [6]. While health information is critical to the improvement of healthcare in developing regions, financing healthcare also remains a significant unsolved problem. Can we take lessons from e-Choupal and apply them in the healthcare sector? The design of usable, reliable, and fraudresistant tools for management of these aid flows is an area with potential for very significant impact. However, in the case of healthcare, the financial models are very different from commercial markets – financing of healthcare typically comes through transnational aid agencies like the World Bank and International Monetary Fund (IMF), and is implemented by non-governmental organizations (NGOs) and the local government. Since the NGOs are typically experts in health, not technology, data processing is often outsourced to third-party information technology (IT) vendors. Relationships between the vendors, the NGOs, the local governments, and the transnational aid agencies are not always smooth - and limitations in communications infrastructure means that the information flows between them are scattered at best. In this paper we suggest that the “closed loop model” generally used by researchers in deployments of mobile health applications does not map onto the financial and political realities of the mainstream of healthcare provision in Africa, and limits the ability of pilot programs to increase their scale and impact. We describe an innovative, IT-based, NGO-run healthcare access program in Uganda, and our experiences designing and deploying Claim Mobile, a mobile-phone based system intended to address inefficiencies and help the program scale to additional districts. We argue that in addition to addressing the needs of the primary users in the system, the health workers, our design must consider the requirements, motivations and concerns of the other stakeholders: the IT vendors, the NGOs, the government, and the aid agencies. Our designs must consider the larger order ramifications of how we may positively and negatively impact both the “users” who will be generating the data, and the entities that will be engaged in managing and using the information in the resulting database. Just as the e-Choupal project assimilated the middlemen by hiring them as kiosk operators, we propose that we can design applications structured to accommodate conflicting stakeholder requirements, while also alleviating information inequalities resulting from limitations in the system prior to the introduction of the information technology.

36

Mbarara, Uganda • HIV prevalence: 10% of adult population (15-49 years) • Syphilis prevalence: about 5-7% of adult population • 1 in 4 households had at least one phone. • 39% reported STI symptoms • only 1/3 sought care • 54% of respondents who sought any STI treatment reported using private clinics. Fig. 1. A HealthyLife voucher. The ‘M’ in the top left is a note written by the health service provider to indicate that the first client voucher on the left was given to a male client, and that therefore the partner vouchers on the right should be given to a female client.

II. BACKGROUND Providing effective health care in poor countries is an essential component to economic development and poverty reduction. Unfortunately donors supporting this endeavor often find that resources given are not matched by desired gains. The output-based aid (OBA) model of financing seeks to address this by paying healthcare providers directly for services rendered instead of paying for the service provision up front. However, OBA program management is information intensive, necessitating much paperwork to track and reimburse payment claims. Smartphones (mobile phones with advanced features such as the ability to run third-party software) have the potential to alleviate this burden. In collaboration with a local NGO and their partnering IT vendor, we have proposed to deploy a number of smartphones for use in an OBA project based in Western Uganda, with dual goals of reducing claim processing time and improving communication between the health care providers and the OBA management agency. The project is managed by the local branch of a multinational NGO and a for-profit health insurance company, in collaboration with the Ugandan Ministry of Health (MoH) and Ministry of Finance (MoF). The project is primarily funded by an aid agency based in Europe, with additional funding for the expansion coming from a separate transnational funding agency. Together, they run a voucher program called HealthyLife, which treats sexually transmitted infections (STIs), reimbursing providers for the diagnosis and full course of treatment only after the patient is seen. This program was implemented in response to the high burden of sexually transmitted infections in Uganda, and began in July 2006 in four districts of southwestern Uganda: Mbarara, Ibanda, Kirihura and Isingiro (See Table I). Patients buy treatment vouchers in pairs, one for the client and a second one for the client’s sexual partner (See Figure 1). Each voucher is good for one consultation (generally including a lab test to diagnose the STI) and three follow-up visits. During the consultation, the provider completes a claim form recording the client’s demographics, the examination and laboratory results, a diagnosis and details of the course of treatment prescribed (See Figure 5). Completed claims forms are sent to the voucher management office in the city of Mbarara, the main urban center of Western Uganda. Forms

TABLE I S OME BACKGROUND STI STATISTICS ABOUT M BARARA , U GANDA [7].

can take two weeks or more to move from the providers office to the management agency. The current data management system requires all claims to be submitted on paper forms to the management agency. At least another two to four weeks are spent reviewing each claim, cleaning data from improperly-completed forms, and verifying that the service took place among suspect claims. Two months or more can go by before the provider is reimbursed for service provision. In Uganda, private providers traditionally operate on a feefor-service model, receive prompt payment, and do not have a large operating margin. In many cases, payment is provided prior to service. Delays in payment result in delays in procurement of replacement prescriptions and medical supplies, often leading to a temporary hiatus in service. Encouraging provider involvement in the OBA program requires a great deal of confidence on the part of the providers to participate. If a system to shorten claims processing could be devised, more providers could join the scheme and more patients could be provided the life-saving STI treatment voucher subsidy. The remainder of this paper details the system we are currently piloting, in which claims are submitted via Internet from a mobile phone directly to all the parties in the management agency. In addition to describing our user studies and how this has informed the design of the system, we discuss the problem of negotiating conflicting stakeholder requirements. We find that in projects with multiple stakeholders, the introduction of a system may disrupt balances of power, particularly around the flow of information and money. As a result, the design of this system, in order to secure positive support from all parties involved, must carefully balance stakeholder incentives. III. M ETHODS The research described here involved an iterative process of field research and prototyping. The fieldwork and deployments have been done over the course of three visits to Uganda: an initial two-week visit in Summer 2007 to establish a relationship with the project, in which we also conducted a survey of the clinics in the program; a followup visit for three weeks in November 2007; and a five-week pre-pilot deployment in August-September 2008. During all three visits we conducted semi-structured interviews with the various stakeholders, and directly observed claims form entry and processing. When given permission, we did audio and video recording of interviews and user study activities. In all, we

37

have approximately 30 hours of audio, and have done detailed interviews in seven of the 12 participating clinics (in addition to the initial survey of all of the clinics), as well as intensive observation in two clinics, a rural, high-claim-volume clinic with very little exposure to computers, and an urban lowclaim-volume clinic with its own computers. The last visit entailed a comparative user study as well as deployment of the mobile phones in the latter two clinics. A. Clinic Surveys The clinic survey was conducted in conjunction with a larger survey of available infrastructure at contracted clinics. We asked 14 questions, assessing familiarity with computers and mobile phones, but primarily gathering feedback from the health clinics on the claims process (e.g., how long it takes them to fill out the paper forms, and what their priorities might be for improvement of the process). We also collected various documents from the management agencies regarding the performance of each clinic, including all available financial reports on processed claims, and in November, we returned to seven of the clinics to do in-depth surveys and to follow-up on the survey findings. B. Rapid Development and Pre-pilot Deployment Initial prototyping occurred in early 2008, and we returned to Uganda in Summer 2008 to do a three-stage pilot deployment, first testing the functionality of our software, second reviewing the proposed claims process with the management agencies, and finally taking the phones to the health clinics to test the mobile phone interface in the field. During this time we also conducted another round of semi-structured interviews to gather information on changes in the claims submission process (for example, claims processing had moved from Mbarara to the national capital, Kampala). We did iterative development based on feedback from the various stakeholders, trying out features as they were suggested, and developing new tools as seemed merited by findings in our interviews. To gain a more in-depth understanding of health clinic life, we stayed overnight for three days in the rural health clinic, thereby supplementing the the semi-structured interviews with direct observation of actual practice. The primary purpose of this last field visit was to conduct a pre-pilot demonstration, using the mobile phones to submit actual claim data to the management agency, have it reviewed, and have the management agency provide feedback to the health clinics via the mobile phones. We simulated the proposed process, physically following the claims forms from the time the patient comes into the health clinic, through the preparation of the claims summary forms, physically transporting the forms to the management agency where we observed the claims approval, and data entry into the existing database. We simultaneously had the service providers submit the claims form via Claim Mobile, enabling the management agency to provide feedback to the service providers through the system. The pre-pilot is still operational, with mobile phones remaining in the two clinics, and the full pilot will

Fig. 2. This diagram illustrates the flow of money and information between selected stakeholders in the OBA project, both currently, and how it will be once the Claim Mobile system is fully deployed.

be conducted in May 2009, with mobile phone-based claims processing expanded to 8-10 additional clinics. IV. S TAKEHOLDERS In this section we describe the funding, management and service provider organizations to which we alluded in Section II. Fundamentally, all stakeholders want to improve STI treatment and reduce the prevalence of STIs. Each stakeholder also has a financial interest in the success of the overall project - staying afloat for the health service providers, and staying within the aid agencies’ target budgets for the management agency partners. The discussion highlights the ways in which the various stakeholders have competing as well as common interests. While we will detail several stakeholders in this section, there are six key stakeholders: the aid agencies who fund the OBA project, the financial management agency (FMA) which receives the funds from the government and disburses them, their program management office (PMO) in Mbarara which runs the program and interacts directly with the health service providers, the technical management agency (TMA) that manages the claims processing, the health service providers (HSPs), and us, the mobile platform developers (MPDs). Figure 2 illustrates some of the relationships between these entities which we will describe in detail in the remainder of this section, based on qualitative fieldwork and document analysis. A. Aid Agencies As the funder of the HealthyLife STI treatment program, the involvement of the European aid agency is more than apparent. Their role in the management of the program is more supervisory – a consultant goes to Uganda at irregular intervals to help with planning of the program, and they do some monitoring. They also have commissioned another nonprofit, affiliated with a North American university, to conduct an evaluation of the program. Ultimately, however, they control the flow of money to the financial management agency, which then pays the IT vendor to handle the technical aspects of the operation. In the past year, the European aid agency has worked with an additional transnational aid agency to fund the expansion

38

of the project into additional districts. While they may not have a direct impact on the information processes in the project itself, the funders’ internal actions have direct impact on the project as a whole. In one example, a delay in payment to the European aid agency resulted in a delay in payment to the two management agencies. As a result the IT vendor ceased processing of claim forms until payments were received. However, not only did the voucher program grind to a halt, but reimbursement to the participating providers for patients already seen was delayed as well; the management agencies ended up with a backlog of claims forms to process, exacerbating the length of time it takes to process claims and further delaying payment for services. B. Financial Management Agency (FMA) The local NGO partner that acts as the financial management agency (FMA) is the Uganda office of a multinational non-profit sexual and reproductive health organization with a goal of reducing unintended pregnancies and unwanted births through family planning and other methods. Their role in this project is to receive the funds from the aid agencies via the Ugandan government, using these funds to pay the health clinics and to pay for other program costs, including the database software development and management. The main office in Kampala runs this program (as well as others) and manages several clinics throughout Uganda, one of which was a participating clinic in the HealthyLife program until Summer 2008. In addition, they have a program management office (PMO) in Mbarara which is directly in charge of coordinating the the OBA project. In the claims process, the FMA disburses payments to each of the service providers, based on claim reports from the technical management agency (TMA).

for the PMO, and one for the TMA. However, the copy that remains in the PMO does not have the voucher number, a critical piece of information, and with stacks of hundreds of claims per month, the information is not in a format actually accessible to the program office until the TMA sends back claims summaries. However, even this is stripped-down and only includes the value of the claims, without any patient or diagnostic information from the claims. This poses a problem for the PMO staff’s interaction with the health clinics. They lack sufficient information to effectively counsel and train the clinics, and often feel like they do not know what is going on with the program because they do not have access to the claim data for the long claims processing cycles. In the version of Claim Mobile developed in early 2008, we intended to make the claims process more efficient by enabling the mobile phones to submit claims directly to the database (originally co-located in the PMO, now located in Kampala). In the Summer 2008 interviews it became clear that having the claims data bypass the PMO staff would deny them even more of the information they need in their interactions with the clinics. Based on this realization, we discussed the possibility of an intermediary application, a website in which the project coordinator in the PMO would be able to view claims as they are submitted, as well as any status updates. Furthermore, Claim Mobile could facilitate another of the project coordinator’s key roles in the OBA ecosystem: as the primary interface between the health service providers, the project coordinator would also be able to send messages to the service providers through Claim Mobile, either as individual messages, or broadcast announcements.

C. Program Management Office (PMO)

D. Technical Management Agency (TMA)

The HealthyLife PMO in Mbarara has five full-time staff, in addition to two people that help with cleaning and cooking, and the FMA staff that come in and out of Kampala for related programs. There are two computers in the office, one in the project coordinator’s office, and another in the finance office, shared by the Behavior Change Campaign (BCC) coordinators who go out into the field to run community radio advertising programs and to distribute vouchers. Their Internet connection was down when we arrived, but was repaired the same day and largely functional for the remainder of our four weeks there. They share a 56kbps dial-up connection over a local area network. It became clear through our interviews in this office that, while the PMO is the nominal clearinghouse for information between the TMA and the health clinics and is primarily responsible for communication with the health clinics, they actually have the least information of all of the stakeholders in the OBA program. At the point in which the database processing moved from Mbarara to Kampala, all of the claims information moved there as well. They have been able to change the claims process such that the health clinics submit two copies of each claim to the management agencies, one

The TMA is a for-profit health insurance agency based in Kampala, providing conventional employer-based health insurance for the formal private sector as well as conducting nonprofit health management for targeted low-income informal sector populations. That is, their work for the HealthyLife program is in addition to their private health insurance program, and is part of a company effort to help deliver quality affordable healthcare to the poor. Their business is highly technical, and they have a wholly owned software company based in Chennai, India. As the IT vendor, the TMA’s responsibility in the program is to provide the claims forms, and the Voucher Management Unit System (VMUS), the database implemented by their software company to cross-check the claims and to generate reports. The TMA also prints (through another agency) the glossy color vouchers that the patients purchase in exchange for subsidized care. Although the data entry clerks and the database engineer were initially located in the FMA’s PMO in Mbarara, they are actually employees of the TMA, and moved when the TMA shifted claims operations to their offices in Kampala in March 2008. They carefully enter each claim into the database, later updating its status with information from the clinical officer

39

Fig. 3. This is a sample summary sheet prepared for one health clinic, showing a partially paid claim (QC11=Wrong consultation fees), and another claim quarantined for having the wrong voucher. While these summary reports can be informative, most clinics are not familiar with the quarantine codes, and they often don’t reach the health clinic for several months after the original claim has been submitted, often too late for the clinic to rectify any errors indicated on the report.

(a doctor) who “vets” the paper claims1 . They then produce two reports: a summary report for all clinics, and an itemized report (see Figure 3) for each clinic detailing the status of each claim, as well as any quarantine codes (Table II) or rejection reasons for any partially paid or rejected claims. In addition, when required, the TMA produces reports (based on the information in the database) for the FMA, the European aid agency, and the aid agency’s evaluating partners. Although these reports were not part of the original specified mandates for the operation, they have proved necessary for the program’s external evaluation, and there has been much friction over the work involved in the creation of reports. The relationship between the TMA and the FMA in this program is highly contentious. While initially they were equal partners in the program, both reporting directly to the European aid agency, changes in funding have led to a situation in which the TMA reports to and is paid by the FMA. On top of this, the funding for the expansion of the program has been delayed a number of times, from October 2007 to April 2008, and again to September 2008. While the TMA has received some payment, both the TMA and FMA have been operating without pay (but with promise of pay) since April 2008, just to keep the program running while the aid agencies work out the details of the new grant and the expanded program. This is part of the reality of dealing with aid-funded projects – unexpected delays in funding are common, and projects are subject to the vagaries of arbitrary rebudgeting. While the FMA is often powerless to address the issue, in this case, the TMA often chooses to respond by cutting off program access to the database, ceasing claims processing and refusing requests for reports, until their problems have been resolved. Perhaps in response to these database shutdowns, but officially as part of the aid agency’s project policy and the Ugandan government’s policy on software developed for government-funded projects, there is an expectation that the TMA’s VMUS software should be turned over to the project. However, since the TMA outsources development of this software to its partner company in India, this IT vendor considers its software to be part of its key assets, and sees its role in the project as a software licensor and service provider, not a 1 The

clinical officer is employed by the FMA, and was terminated in March 2008 due to temporary lack of funds.

software vendor. Again, while this situation is being resolved, the TMA asserts its control over the project by processing the claims, but refusing to pass on the summary reports to the FMA. While the project continues running, and the service providers continue to see patients, this introduces additional delays into the claims process, and frustrates the health clinics, whose payments are delayed without explanation. E. Service Providers: Health Clinics/Hospitals Service Providers are selected on the basis of a number of factors (e.g., services offered, capacity, personnel, geographical location). In one respect, they are the origin of the primary information in the claims management process, producing the claims records, which are then used to determine reimbursement. At the same time, as is perhaps typical, they are information-poor, because they are not given tools to use this information effectively. At the point of claims submission, they are no longer agents in the process, and must wait passively for both payments and any feedback reports produced from the information in their claims. Code QC01 QC02 QC03 QC04 QC05 QC06 QC07 QC08 QC09 QC10 QC11 QC12 QC13 QC14 QC15 QC16 QC17 QC18 QC19 QC20 QC21 QC22 QC23 QC24 QC25 QC26 QC27 QC28

Description No indication of date of treatment No indication of time of treatment Wrong visit type: Consultation or follow-up, etc Wrong demographic information: no age and name of client Wrong/No syndrome, no diagnosis Wrong Clinical examination / not applicable to OBA Wrong/Poor diagnosis Wrong investigation/poor lab reporting Wrong drugs prescribed/invalid treatment Over prescription: more than enough Wrong consultation fees Wrong patient status: cured or not cured Next date of visit: wrong or not filled in Wrong voucher attachment/interchanged vouchers on followup visits Unclear claim/uncharged claim and treatment contradicts other visits Partner treated on client form Exceeded VMUS ceiling limit Treated syndromically and asymptomatically Unclear/wornout/blank attached vouchers Claim without patient thumbprint No voucher attachment No doctor’s signature Diagnosis contradicts clinical examination Used drugs not on OBA list Undercharged/overcharged drugs, double lab charged Patient free/normal from STI or cured not allowed for next visit Diagnosis not catered for by project Follow-up contradicts previous visits

TABLE II C ODES USED BY

THE MANAGEMENT AGENCY TO INDICATE REASONS FOR PARTIAL PAYMENTS .

Providers are expected to follow a rigorous course of diagnosis and treatment — they must select a lab test based on symptoms presented, and prescribe particular medications

40

Fig. 4. “All above denied b’cos rest of P[atien]t mgt n[ot]. ethical”: this is a sample medical advisor review of a claims summary, occurring often well after the original mistake has been made several times, before it could be caught and corrective measures could be made, as noted in the first line: “Cipro pricing b4 C[ontinuing] M[edical] E[ducation]”

on the basis of the results of the test. Any deviations from this treatment are penalized; the service providers are not paid for medications given that are not prescribed by the program. This is reasonable by public health and insurance standards, and necessary for the cost-effectiveness of the program. However, despite educational illustrative posters, training sessions, and on-site continuing education provided by the PMO’s clinical officer and project coordinator, the learning curve on the exact protocol to be used is high, and the subsequent errors are costly for the service provider. Figure 4 illustrates a particular problem in which costly errors are caught well after their first occurrence, often after the provider’s staff has have made the same mistake for a month. A few months into the program, there had been so many quarantined claims (claims that had been held for review due to discrepancies from the treatment protocol) that the management agencies and the service providers were required to do a financial settlement, in which the service providers were paid some percentage of the value of the disputed claims. Subsequently, the approval process was modified such that deviations from the protocol were partially paid (e.g., minus the cost of the incorrectly prescribed drug), and could be disputed in later reviews with the PMO’s clinical officer. Here are some figures on the value of the disputed claims for one of the rural service providers for a randomly selected month of the program: Claims submitted: 294 Approved: 259 (88%), $1379 Approved, but adjusted: 27 (9%), $149 Quarantined: none Rejected: 8 (3%), $51 Total Requested: $1642 Total Paid: $1526 (difference: $114)

Thus the claims submission process, while nominally a simple information flow between cooperating organizations, becomes a site of financial contention. Claims are disapproved for a number of reasons (see Table II), which could be disputed but in practice are not. However, due to the change in policy,

many claims are approved with incomplete data (e.g., missing demographic information for the patient). In any event, the claim form and its contents are the object of much dispute between the service providers and the management agency. Often, the service providers (especially more distant providers) feel disempowered to address this dispute and choose to accept the given variance in payment as the cost of participation in the program. The health clinic survey we conducted in July 2007 explains some of their extreme disengagement from the OBA program. Out of the twelve clinics surveyed, three said they had not gotten feedback from the OBA program at the time of the survey (July 2007), and 4 said they did not know how many claims had been rejected. While they were receiving payments on a regular basis (albeit late - 9/12 clinics defined “timely processing” as less than 15 days, half of the current processing time of 30 days), there was no mapping between the claims they submitted and the payments they received. I don’t know. I don’t know how we are performing. I don’t know how we are faring. And of course my staffs are also complaining. They are overworked, they dont get any benefit from the project, and of course it takes a lot of time. They need to be motivated as individuals. All that will depend on – are we making any profits?

This has deeper ramifications than inability to follow up on quarantined, partially paid, and rejected claims. This communications gap between the health clinics and the OBA program leads to continued errors in adherence to treatment protocol, a feeling of lack of control over health clinic finances, and discouragement on behalf of the participating service providers. At the time of the second field visit in November 2007, we were able to follow up with the PMO and the health clinics, and noted that this situation had improved. The then-informal practice of passing on copies of the itemized clinics reports to the service providers was formalized, and clinics are receiving more feedback on their claims. However, there are physical limits to a paper and in-person based communication system, and it remains to be seen how this practice will scale as more clinics join the program. It should also be noted that the service providers differ greatly, in number of clients, setting, and availability of resources. While some clinics have computers and use them regularly, in one clinic, our smartphones were greeted with enthusiasm because they were the “first computers we have seen.” F. Patients Patients are the real target beneficiaries of the HealthyLife program. They purchase vouchers from distributors (at a heavily-subsidized price) and then go to the service provider for diagnosis and treatment. After an initial consultation with the health worker, they are directed to the lab technician, who performs the requisite test and sends them back to the health worker with a slip of paper indicating the lab result. The health worker then fills out the remainder of the form, writes a prescription, affixes the appropriate stub from the

41

voucher, and has the patient sign and fingerprint the form, at which point their participation in the claims form process is complete, until they return for a follow-up consultation. For the follow-up, the service provider checks recovery progress and prescribes additional medication if necessary. In some clinics, patients are given a copy of the claim form, which they are directed to keep and bring back for the follow-up. However, most clinics do not depend on the patient copy of the claim form, and just go back through their time-ordered record book, finding the prior consultation manually. Sometimes patients either accidentally swap vouchers with those of their sexual partners, fraudulently give their own voucher or the partner’s voucher to someone else, or simply choose to go to a different clinic for follow-up. Claims are quarantined or rejected if any of these potential errors are detected, but not until the claim has been processed by the TMA, and the fraudulent patient has already been treated. Since the original voucher is attached to the submitted claim, the clinics do not always have a way of verifying these external aspects of voucher validity. Although their direct involvement in the claims process is minimal, it is their identity that is often contested in the vetting process. G. Mobile Platform Developer (researchers) As ICTD researchers, we are of course also stakeholders the claims management process – initially as outside observers, later as designers interested in using technology to measurably improve the process, and finally as researchers interested in watching the mechanisms by which the process changes over the course of the project. From an outside perspective, our role is most allied with TMA, the technical partner in the project; however, since the aid agencies and FMA are interested in replicating the mobile device system in other OBA projects, there is a vested interest in the new technology from other stakeholders as well. V. D ESIGN Claim Mobile is a two-part system, including a web-based PHP/MySQL application and a Java-based mobile application running on Palm (GarnetOS) phones. For the pilot program, the web-based application has a single level for all users, but the final implementation will be tiered, having appropriate access levels for service providers, management agency users, medical advisors, etc. Both the web and the phone applications require user login to protect patient data. The two applications are paired, designed such that the phone-based application uploads claims to the web site, and downloads configuration information (drug lists, status feedback, claim form backups) from the web site. Eventually, the web application will also connect to the TMA database, sharing the cross-checked and validated claims form data directly so the TMA’s staff do not have to do redundant data entry. To facilitate end-user training, both of the applications are based on the original claim form and largely retain the same structure, titles, and information. Figure 5 illustrates some of the mappings between the phone-based user interface

and a revised version of the claim form. In addition, all of the codes and tables in the web application database also include mappings to their equivalents in the TMA database, so the information can easily be transferred between the two databases. A. Claim Mobile The web application is designed primarily with three functions in mind: claims submission, feedback/communications process, and in-clinic claims management. Claims Submission: This is the bulk of where the service providers will spend their time. In this case, we adopt common design strategies such as (1) using pre-filled checkboxes to reduce the amount of required text entry, (2) limiting answers to valid options to reduce coding errors (see QC01,02, 04-09, 27 in Table II), (3) downloading logistical data such as drug prices into the application to eliminate pricing errors, and (4) calculating dependent values such as expected claim amount to eliminate arithmetic errors and save time. However, we must counter-balance potential fraud by also introducing crosschecks that are not clarified explicitly. That is, to encourage accurate clinical reporting (as opposed to clinical reporting that has been “fiddled” to make electronic claim submission more convenient or favorable), providers are allowed to submit inconsistent claims but are warned that they should clarify any discrepancies from normal OBA treatment protocol. Closing the Feedback Loop: Based on our primary finding from the clinic surveys and follow-up interviews, we have also included the ability for the clinics to send queries to the management agencies about particular claims and to receive live updates on the a claim’s status (e.g., whether it has been approved, the amount for which a claim has been approved, and explanations why the full amount may not have been approved – see Figure 6). Any changes to a claim’s status are included in this annotation audit trail, and anyone with access to the claim can respond to and receive queries.

Fig. 6. On this screen the service provider can view the current status of their claim, as well as any annotations or feedback from the management agency made in response to their queries.

In-Clinic Claims Management: In the phone application, the service providers can also link between consultations and follow-up visits, as well as between client and partner visits, so they can easily check to see if the valid voucher is being

42

Fig. 5.

From paper form to mobile phone: a mapping of the mobile phone interface equivalents for each section of the claim form.

Fig. 7. This is the list of claims currently entered or downloaded to the mobile phone. The first number is the claim form number, followed by the patient name, and then a number indicating the current status of the claim. (1-unsubmitted, 2/3-under review, 4-preliminary approval, 5-quarantined, 6approved, 7-rejected)

used by a returning patient, and that treatment of a partner or during a follow-up matches the medical history. For new phone installations, or if the claims data is lost, the mobile application will automatically download all prior claims data from the web application. Future versions of the application will also include financial summaries, outpatient statistics, and other reports that may be useful to the clinics. B. Claim Mobile Web The web application, having been commissioned in the middle of the fieldwork in response to program office findings (see Section IV.C) has two main functions: receiving claims and displaying them for review. Much of the claims receipt is invisible to the web application user, and written as a backend for the mobile phone application. The login user for the mobile application and the web application is the same – and the login information

Fig. 8. Claim Mobile Web: the web-based view of the submitted claim forms is also based largely on the original claim form, split into several sections, and ending with annotations for the claim, and a form for adding additional annotations.

given to the mobile application is used to authenticate with the application when submitting claims and annotation data. There are three primary views in the web application. The claims list can be filtered by service provider and is modeled on the claim summary report (Figure 3). From each claim in the claim list, the user can either (1) click on the claim number to access the individual claim (and annotation/status update functions, see Figure 8) or (2) click on a patient name to view all consultations and follow-ups for the both the client and

43

partner associated with that particular voucher. This allows the viewer to correlate treatments, lab tests, and diagnoses across visits. C. The Modified Claims Process In the modified claims process, the service providers continue to complete and submit the claim forms. However, in order to receive faster payment, as well as the immediate crosschecked feedback from the phones on claims completion, they also enter the data on the mobile phones, submitting each claim to the web application as it is completed. Prior to claims submission the status of the claim is “Unsubmitted (1),” after which it can progress through a number of stages. The service provider can verify that a claim has been successfully received if the claim status has been updated to “Under Medical Review (2)” or “Under Administrative Review (3)” for medical advisor review or database cross-checking (validation of voucher) respectively. If a digital claim has been verified, the TMA will set its status to “Preliminary Approval (4)” until the paper form with the voucher has been received. Once preliminary approval has been received, the clinic can be paid. If no voucher is received, or the wrong voucher is attached to the form, then the preliminary payment is subtracted from the next month’s payment until the error is resolved. In the meantime, the service provider can view status updates as they are made to the web application and sent to the mobile phones, and can send annotations on each claim, which then appear in the web application and in the status update window (Figure 6) when they are received.

inconsistency is in lab reporting – some lab tests require a value to indicate the result, and where not included, the data entry clerk just changes the lab test in the database to one which does not require a result. This is an error, which never gets communicated back to the service providers because only errors which accompany a payment change are reported in the claim summary sheet. The 18 claims from the urban clinic spanned dates from February 16, 2008 through August 25, 2008. During the process of simultaneously entering some of the claims into Claim Mobile with the service provider, we were able to identify some problems: missing personal information, missing next visit date, and wrong consultation fees. However, not all claims were entered into and reviewed via Claim Mobile, and, as can be seen from Figure 3, three paper claims were submitted with wrong consultation fees, an error that would not have occurred with an electronic submission. In addition, a fourth claim was submitted with the wrong voucher. We were able to catch this while entering the claim into the mobile phone, noting that the voucher number did not match the consultation type, but too late to change the submission and retrieve the correct voucher. As a result, the claim has been quarantined until the correct voucher is given to the program office in Mbarara.

VI. P RE -P ILOT R ESULTS Having detailed the claims submission process, our findings with respect to the various stakeholders in the OBA project, and the design of the Claim Mobile system, we now discuss some specific results from our user studies. The pre-pilot demonstration was designed to last one claims cycle, following one week’s worth of claims (submitted in parallel through Claim Mobile and on paper) for two clinics through the claims submission process. A total of 35 claims were submitted to the web application, including the full complement of 18 claims from the urban clinic, 12 out of the 86 paper claims from the rural clinic, and 5 additional claims from the urban clinic following the pre-pilot study. The claims from the rural clinic spanned August 9, 2008 to August 27, 2008. We observed three patient consultations during our two visits to this clinic, as well as the preparation of the summary sheets for the 86 claims, taking careful note of what the service provider verified on each form. Notably, although “syndrome” is a required field (see QC05 in Table II), it was left blank in almost all of the claim forms. In one case, the drug was entered correctly, but with the wrong reimbursement value, and in another case, a drug was entered, but no reimbursement was claimed either in the subtotal or the total. At the time of the claim approval process, they were not reimbursed for the drug, because it had not been claimed in the amount, although it had been listed. Another

Fig. 9. Rural clinic staff entering data from claim forms into two of the phones.

With regard to the digitally-submitted claims, we spent about a day training the staff in the rural clinic on how to submit the claims, and returned later to spend another day in training. They were very enthusiastic, and although only one person was actually responsible for submitting claims, they were all training each other (Figure 9). However after the the researchers’ departure they have still not submitted any claims. It is unclear whether this is from technical difficulties or lack of time to enter the claims into the phone. The urban clinic has continued to submit claims, with five new claims arriving in the two weeks since the pre-pilot study. An interesting outcome from our observation of the claims review process is that there are many errors that are made that affect the quality of the data, but are never communicated to the service provider, in part because they have no attached financial consequences. The annotation feature (see Section V.A) enables attachment of quarantine codes to any claims that were in error without affecting the payment of the claim, providing feedback to the service providers on how to better complete the claims in the future.

44

One concern that emerged from this proof-of-concept study, however, was with the reliability and the speed of Internet access in the TMA office where the claims processing was occurring. Accessing individual claims took a long time, and the online database was completely inaccessible when the Internet connection was down, which occurred infrequently but noticeably often. It may not be desirable for the claims submission process to introduce a dependency on Internet connectivity where it is unreliable. Unfortunately, the financial and claim review aspect of the pre-pilot was halted early due to administrative and political reasons, the result of which is that payments in the OBA program as a whole have been halted; so, while the technical feasibility of the system has been demonstrated, the logistical details are still in process. We found that while the TMA’s database entry staff were enthusiastic at the prospect of spending more time reviewing claims and less time doing just data entry, their participation in the pre-pilot was limited by two factors: the press of other claims that still needed to be processed, and pressure from the TMA to be secretive about the data being processed until certain political issues had been sorted out. VII. D ISCUSSION A. Understanding Delays in the System Delays can occur in a number of places in the claims process, not all of which can be accounted for by the introduction of mobile phones. However, there are three key bottlenecks: 1) the delay in the health clinic between when the health clinic sees the patient and when the claim is submitted, 2) the time it takes to process the claims, entering each on into the database, and 3) administration of feedback to the health clinics, especially in case of errors. Claim Mobile is able to address all three of these cases by 1) encouraging providers to submit claims as they see patients, 2) reducing the data entry burden through the use of digital claims, and 3) eliminating the possibility of a number of errors, and providing a digital feedback mechanism to supplement the infrequent in-person feedback. However, another source of delay is the administrative and political dynamics by which program administration halts, although health distributors continue to sell vouchers, and health clinics continue to see patients. During these times payments are delayed unexpectedly for undetermined lengths of time, as can be observed from the early termination of our pre-pilot study. Understanding this particular delay is key: the TMA halts the program by withholding information, specifically claim reports, from other stakeholders in the system. This is possible because the database is owned and controlled entirely by the TMA. What happens when another outside database is introduced, with independent control? In this case, the data was not so much the key as the data entry staff that were responsible for approving the claims and validating the vouchers. At the same time, it is unclear where Claim Mobile Web fits in with the political strategy of the TMA.

B. Information Poverty In addition to trying to address delays, we have also tried to address information asymmetry and information poverty within the system, identifying where stakeholders are disadvantaged by lack of information, or lack of tools with which to use the information. This past year’s move of the data processing from Mbarara to Kampala especially has further exacerbated the gap between the people that have the information and the people that can make use of it. While limited remedies have been made to rectify the situation, with a paper-based process, these remedies have been ineffectual, leaving the PMO without access to necessary claims data, including voucher numbers. Through extensive stakeholder interviews, in particular with the program office in Mbarara, we have identified the need for a transparently accessible database, with the ability to generate reports based on the submitted claims data. While control over access to the database is a key means by which the TMA asserts itself in the OBA project, this practice is highly disruptive to the OBA program, causing deep difficulties for the health clinics and the program office, rather than affecting the financiers or the FMA. The initial design of Claim Mobile, reflected the paperbased process, and directly submitted claims from the mobile phones in the health clinics to the TMA, bypassing the PMO entirely. In response to our findings, we developed Claim Mobile Web as a means of re-engaging the staff of the PMO in the mobile claims process. The integration of the new web application database is specifically meant facilitate resolution of information gaps, not only sharing the information with the people that can make use of it, but also giving them the tools they need to make sense of the information. Likewise, for the mobile-phone application, we also specifically do not design one-way system in which claims data is going out and only money returns. Instead, the claims data created in the clinic is also used within the clinic to help them improve patient care, as well as the accuracy of future claims. C. Related Work There have been a number of recent technical projects on the use of ICTs for healthcare in Africa [8], [9], [10], [5], and specifically on mobile devices for healthcare in Africa [3], [11], [12]. However, many of these projects are design-focused and technology-driven, reflecting on designing a working technological solution to complete a particular task, rather than reflecting on the role of the technology in the system and how various solutions or approaches might affect social processes within the system. Braa describes two action research projects to deploy the Health Information System Program (HISP) in Cuba [13] and in South Africa [8], using Actor Network Theory (ANT) to think about how human and non-human (e.g. documents, events, software, standards) interact. He specifically addresses the challenges of designing for the multiple levels of entities involved in district health information systems, able to compare deployments across Mozambique, India, South Africa,

45

and Cuba. However, these entirely government-based contexts are much more hierarchical than the highly disparate multiorganizational context described here. VIII. C ONCLUSIONS

AND

F UTURE W ORK

There is clearly much additional work to be done, in which the lessons learned from this pre-pilot study will be applied in the development of a new version of Claim Mobile for a full pilot in Spring 2009. The outcomes from this study were three-fold. Firstly, the choice of the mobile phone as a platform was affirmed by the health clinics, for reasons of battery life, design for readability, portability (susceptibility to theft), and ease of data entry. Where we were concerned about introducing “qwerty” keyboards to novice users, our fears were alleviated, and all of our users assured us that “we can learn,” which they did, quickly. Secondly, the mobile platform is not a sufficient solution for this program, and alone has the potential to exacerbate information asymmetries between the stakeholders. To address this issue, we complement the mobile phone platform with a web application. However, Internet accessibility issues may require further development to enable local hosting and synchronization of the web application [10], [14]. Our final point is related – we consider the plethora of stakeholders in this project, and note that as technology providers we are not coming in as naturally neutral players. Our projects are necessarily disruptive, and equally potentially disrupted by other dynamics within the program as a whole. As a result it is necessary for us as researchers to position ourselves and our designs carefully, making sure to take into account the needs of all of the stakeholders, and not just our primary users. IX. ACKNOWLEDGMENT We thank Ben Bellows for his contributions to this project, and feedback on the paper. In addition, we would like to thank all the staff at our partnering organizations and in the health clinics for their assistance and patience in conducting

the fieldwork and setting up this pilot, as well as their honest feedback on this paper. R EFERENCES [1] Karen G. Cheng, Francisco Ernesto, and Khai N. Truong. Participant and interviewer attitudes toward handheld computers in the context of hiv/aids programs in sub-saharan africa. In CHI ’08: Proceeding of the twenty-sixth annual SIGCHI conference on Human factors in computing systems, pages 763–766, New York, NY, USA, 2008. ACM. [2] Cynthia Casas and William LaJoie. Voxiva: Case Study. December 2003. [3] Brian DeRenzi, Neal Lesh, Tapan Parikh, Clayton Sims, Werner Maokla, Mwajuma Chemba, Yuna Hamisi, David S Hellenberg, Marc Mitchell, and Gaetano Borriello. E-imci: Improving pediatric health care in lowincome countries. In CHI ’08: Proceeding of the twenty-sixth annual SIGCHI conference on Human factors in computing systems, pages 753– 762, New York, NY, USA, 2008. ACM. [4] E.S. Berner and M.J. Ball, Editors. Clinical Decision Support Systems: Theory and Practice. Springer-Verlag, 1998. [5] Henry Lucas. Information and communications technology for future health systems in developing countries. Social Science & Medicine, 66:2122–2132, May 2008. [6] ITC - e-Choupal. http://www.itcportal.com/sets/echoupal frameset.htm. [7] 2006 Venture Strategies and Mbarara University population survey. http: //www.oba-uganda.net. [8] Jørn Braa and Calle Hedberg. The Struggle for District-Based Health Information Systems in South Africa. The Information Society, pages 113 — 127, 2002. [9] Tessa Tan-Torres Edejer. Disseminating health information in developing countries: the role of the internet. BMJ (British Medical Journal), pages 797—800, 2000. [10] Rowena Luk, Melissa Ho, and Paul M. Aoki. Asynchronous remote medical consultation for Ghana. In CHI ’08: Proceeding of the twenty-sixth annual SIGCHI conference on Human factors in computing systems, pages 743–752, New York, NY, USA, 2008. ACM. [11] Tapan Parikh. Position Paper: Mobile Phones may be the Right Devices for Supporting Developing World Accessibility, but is the WWW the Right Service Delivery Model? In W4A at WWW2006. ACM, 2006. [12] Tapan Parikh and Edward D Lazowska. Designing an architecture for delivering mobile information services to the rural developing world. In Proceedings of WWW2006. ACM, 2006. [13] Jørn Braa, Ola Hodne Titlestad, and Johan Sæbø. Participatory Health Information Systems Development in Cuba the Challenge of Addressing Multiple Levels in a Centralized Setting. In Proceedings of Participatory Design Conference 2004. ACM, 2004. [14] Michael Demmer, Bowei Du, and Eric Brewer. Tierstore: A distributed file-system for challenged networks. In Proceedings of File and Storage Technologies (FAST), 2008.

46

Computer Games in the Developing World: The Value of Non-Instrumental Engagement with ICTs, or Taking Play Seriously Beth E. Kolko, Cynthia Putnam  Abstract—This paper argues that it is important to study noninstrumental uses of ICT, including computer games. Specifically, the article presents the results of qualitative and quantitative work spanning eight years of investigation in Central Asia focused on computer gaming in public Internet cafes as well as private spaces. The results presented demonstrate that people do indeed play games in resource constrained environments. The paper demonstrates that games constitute a significant portion of the ICT ecology in developing regions and provide a pathway to people’s “first touch” of a computer, that gamers have more frequent interaction with technology than basic Internet users, that games bring more diverse users to computers by providing a pathway to ICT use for people with lower levels of education, and that games can motivate innovation in the technology space. Additionally, our findings indicate that both genders engage in game playing. The article makes the case that games can be a source of informal learning about ICT, and as such, games and gaming culture in the developing world merit further study. Index Terms—international development, Central Asia, ICTs, digital games, gender, Internet cafes, technology use, entertainment

online traffic absorbed by gaming related activity. There is no ranking of popular games in each country, or listing of most popular gaming cafes. There‟s no comparison of how much it costs to play Counterstrike or how many youth under the age of 15 are estimated to have consoles in their homes. The HDI, in short, ignores games as an element of information and communication technology infrastructure within a country. Games, in fact, are invisible to the UN, as they are to just about every non-governmental organization and multi-lateral organization that has worked to create information and communication related projects in the past decade or more. The goal of this paper is not, actually, to argue that the HDI should include computer games as a measure of development. However, the purpose is to establish that games lead to learning computer skills and that, indeed, games are played by people throughout the world. Games are in fact a pivotal piece of a country‟s computerization, how its population gains information and communication technology (ICT) related skills, and how ICTs themselves begin to diffuse in developing world contexts. II. GAMES, LEARNING, AND CULTURE

I. INTRODUCTION The Human Development Report is published by the United Nations every year. It‟s an amalgam of facts and figures about infrastructure, health, agriculture, policy and other characteristics of most countries in the world that provide an overview of a country‟s stage of development. The Human Development Index (HDI) is extracted from the report, and includes, among other things, measures for landlines per capita, Internet nodes, and mobile phones. The HDI, however, has no column for gamers per capita. Nor is there an entry for percentage of GDP spent on playing games, or amount of Manuscript received September 22, 2008. This work was supported in part by U.S. National Science Foundation grants #0326101 and #0219350, and by a 2000 Fulbright Scholar award. B.E. Kolko is an Associate Professor with the Department of Human Centered Design & Engineering at the University of Washington, University of Washington, Seattle, WA, 98195 USA. (e-mail: [email protected]). C. Putnam is a PhD candidate with the Department of Human Centered Design & Engineering at the University of Washington, University of Washington, Seattle, WA, 98195 USA. (e-mail: [email protected]).

There is a robust academic community dedicated to exploring the connection between games and learning, drawing heavily from both education and psychology. Games are an increasingly central topic in educational research, with scholars researching games as part of informal learning [28], key skills like collaboration that people learn while playing multi-player games [29], psychological and reaction time skills gained from games [36], and the creation of actual educational games designed to teach complex skills [6,7], [9,10] [30]. As a companion piece to academic work, significant attention and dollars have been targeted at exploring the issue of games and informal learning in the US (e.g. the MacArthur Foundation‟s 5-year, $50 million project on Digital Media and Learning and the Robert Wood Johnson Foundation‟s $8.25 million program in Health Games Research). There are also organizations dedicated to NGO-like activity around games, including the Serious Games initiative, Games for Change, and Games for Health; these groups focus on developing games with explicitly pro-social goals in mind. Much of the research in the games and learning area has focused on the learning that occurs while people engage with

47

games. Developing world initiatives include work by groups such as the South Africa-based Mindset Network which has developed mobile phone-based games to teach math skills to girls (Mathstermind and Fashion Network), and literacy and numeracy games for disadvantaged youth developed by Pratham in India. These specific pro-social gaming projects similarly focus on in-game content and what people can learn as a result of playing. This paper, however, is more interested in how engagement with games serves as an ICT usage entry point. In other words, not what do people learn because of a specific game, but, rather, what do they learn because of the specific activity of gaming. To that end, this paper focuses attention on the commercial game space and how it diffuses throughout and affects nascent computer users in developing regions. This question of effect can be situated within a theoretical framework tied to cultural theory that investigates issues of identity and agency. Generally, cultural theory provides another lens through which we can view the importance of digital artifacts that transcends their literal or functional meaning [32]. Again, in the most broad terms, cultural theorists such as Bruno Latour [35] provide a framework against which we can examine technological artifacts not for what they are, but for what they enable. That is, games and non-instrumental uses of technology are important in the developing world not (only) because they teach people to collaborate or improve language skills, but, as this paper will demonstrate, because they provide an alternative mechanism by which many people experience their first “touch” of a computer. Additionally, we found that games allow users with less education and English language skills to interact with computers, and that they foster innovation and creative engagement with technology.

III. METHODOLOGY Our research is the result of longitudinal qualitative and quantitative work done as a component of a larger project on the effect of information technology on society. The Central Asia Information and Communication Technology Project (CAICT) is a multi-year study of ICTs. The project goals include investigation of how ICTs diffuse within societies, how cultural issues affect technology usage, and how patterns of trust and confidence in media and institutions change over time as technology diffuses in diverse communities. A. Survey The project includes a yearly survey of four countries in Central Asia: Kazakhstan, Kyrgyzstan, Tajikistan, and Uzbekistan. The quantitative results presented here are based on the nationwide survey of 1000 respondents in each of these four countries age 15 and older, administered in 2006, 2007, and 2008, for a total of 1000 respondents per year per country, and a total of 12,000 respondents overall. Survey sample was based on census information for age, gender, ethnicity, and geographic location as released by each country‟s government. The survey includes multiple urban and rural sampling

locations, and it was administered in Russian and other regional/local languages. In addition to the general sampling scheme, a three-stage process was used to select respondents that included Probability Proportional to Size sample of Primary Sampling Units (PSUs); consecutive random sampling of households in each PSU, and selection of a household respondent using a Kish Grid method. The survey instrument was designed by a team of researchers from the University of Washington, pilot tested in each country in conjunction with local researchers, and then revised based on analysis of pilot data. BRIF Research Group, based in Kazakhstan, translated the survey instrument from English to other languages. The University of Washington team back translated the completed Russian translation. Likert scale measures and other question formats were developed in response to initial open-ended interviews, ethnographic field notes, pilot tests and research on performing surveys in postSoviet contexts. Several steps were taken to guarantee high quality fieldwork including: (1) approximately 30% of interviews were checked through a back visit to the respondent‟s home; (2) interviewers were trained through workshops and practiced in a pre-testing phase; and (3) statistical analysis of logical inconsistencies were double checked with the original paper questionnaires and eliminated if necessary. Given the low rate of current Internet penetration in Central Asia, the survey also focuses on pre-existing patterns of information use, information seeking behavior, and levels of trust in various producers and sources of information. Since Internet usage rates in the region tend to be low, the survey targets the general population rather than Internet users; the findings include overall attitudes towards technology from a wider audience rather than simply usage patterns from a much smaller group. B. Ethnography Ethnographic results are from the CAICT project as well as earlier fieldwork conducted by the first author during a sixmonth residence in the region in 2000. Fieldwork data collection methods follow standard ethnographic format for participant-observation and include field notes and photography. In addition, interview studies have been conducted with various groups to further investigate issues relevant to survey findings. Qualitative data reported in this paper are drawn from multiple separate studies conducted in 2005 and 2006 including interviews with Internet users, interviews with mobile phone users, interviews with computer gamers, and a design ethnography. In addition, ethnographic observations are based on field notes collected during separate data collection trips in 2000, 2002, 2003, 2004, 2005, and 2006. IV. BEYOND STEREOTYPES OF GAMES AND GAMERS Before addressing the specific ways in which games provide an important pathway to ICT engagement, it is important to deconstruct some of the stereotypes of games and

48

gamers. Recent research has begun to establish that gamers in developed countries are not all male and not all teenagers [3334], that in fact the audience for games is diverse in terms of gender and age [1][22-24]. Studies of complex multi-player gaming environments, for example, demonstrate that although male players outnumber female players, it is those female players who actually spend more time online playing [25-26]. Games themselves come in many flavors, including the violent shooting games that gain so much press attention, but they also include puzzle games and simulations. Game systems such as the Nintendo Wii or games with alternative input modes like Rock Band or Dance Dance Revolution have similarly changed the cultural activity of games and broadened the audiences to which they appeal [27]. While not identically diverse, gamers internationally also belie some stereotypes. While gamers are still more likely to be male and under 30, in our research population, female respondents and those over 30 were as likely to play games as they were to use the Internet, See Figure 1. This finding indicates that depending on the population segment, gaming is either a more used pathway to ICT use than the Internet, or it is at least equivalent. 10%

Figure 2: Photograph of Starcraft competition announcement outside Kazakh game club

30% 25%

20% 15%

9% 8%

10%

7%

5%

6% 0%

5% 4% 3%

2% 1% 0%

Male

Female

Under 30

Over 30

Gamers

8%

5%

9%

4%

Internet users

5%

5%

6%

4%

Figure 1: Demographics of gamers vs. Internet users V. SETTING THE STAGE: GAMING CONTEXT IN CENTRAL ASIA Public gaming centers in Central Asia meet many of the preconceptions of computer game culture, but they are also nuanced environments. The crowds of young boys that are seemingly ubiquitous are reminders of gamer stereotypes, although our survey results indicate that women do indeed play games (see Figure 3), although women tend to play at home rather than in public cafes. Many of the popular game titles are familiar, but the style of play and the mechanics of getting the games to work in different infrastructure environments are unfamiliar. Counterstrike (CS) and Starcraft are literally everywhere, and kiosks on the streets sell countless numbers of CS mods, the cd covers in English and Russian. Starcraft competitions can be found in many of the capitol cities in the region, see Figure 6, and the World Cyber Games are a coveted destination (and Central Asian countries have placed quite high in the World Cyber Games in recent years).

Kyrgyzstan

Kazakhstan

Tajikistan

Female

8%

18%

5%

Uzbekistan 8%

Male

17%

26%

18%

19%

Figure 3: Male and Female Gamer Populations in Central Asia, 2008 Internet cafes vary in quality of equipment and furniture. Some businesses will have recent equipment, flat panel LCD screens, and glossy office furniture; others have older PCs, CRT monitors, and plastic chairs. Often the ones the feel more like a business center are located in the center of the city and cater to a mixed clientele. However, businesses located in neighborhoods similarly can serve a diverse clientele, although not necessarily at the same time. And games exist in a context of ICT diffusion overall, with Internet growth creeping slowly and mobile use skyrocketing, see Figure 4. 70% 60%

50% 40% 30% 20% 10%

0%

Mobile Users

Computer Users

Gamers

Internet users

2006

20%

25%

13%

9%

2007

37%

24%

13%

8%

2008

64%

30%

14%

12%

Figure 4 Relative Rates of ICT Usage 2006-2008

49

A. A Tale of Two Tajik Internet Cafes Plazma is an Internet café located in the center of Dushanbe, the capitol of Tajikistan. Plazma is on Ryudaki Street, the central boulevard in the city. It is located in a twostory standalone building, with a very large sign on the side advertising a bar and café on the first floor, and Internet center, computers, fax, IP telephone, and other services on the second floor. Plazma is typical of centrally located Internet centers in the cities of the region in that it offers a wide range of services. However, it is also a particularly good example of a business that has specifically not bifurcated into two separate establishments in order to accommodate both gamers and a growing population of Internet users. Plazma has, instead, blended the two business models and offers a central room with ten computers with Internet access arranged on long tables but separated by small tabletop dividers to allow privacy, 3 IP telephone booths, a central desk where items can be photocopied, text typed, faxes sent, and – very welcome in the Tajik summer – air conditioning. There is also a sign posted in each mini workstation providing some guidance about usage that reads in both Russian and English: “It is categorically not allowed to visit sites with pornographic photos and video materials. In case of visiting above mentioned sites, administration has a right to fine you from 80 up to 300 somani,” the equivalent in 2006 of US$23-86. The sign serves not just as instruction not to visit porn sites, but also as a not-so-subtle reminder that one‟s online activity is being constantly surveilled. What should be obvious, of course, is that the specific sites or what constitutes pornography is left open to interpretation. That signage echoes signs in Uzbekistan cafes that warn about “inappropriate content” which can be either pornographic or political. On the same floor, but in a separate glass enclosed room is the gaming center with 28 game stations preloaded with games such as Grand Theft Auto, Cossacks, Need for Speed, Call of Duty, FIFA, and Half Life/CS and equipped with high quality headsets. On the wall in the main room where the business services are offered is a large embossed sign that runs the length of the room that has the site‟s web address as well as their tagline in Russian and English: All Our Life is a Game. The manager reports that Counterstrike and Starcraft are among the most popular games at the café. During the school year the café is crowded, but during summer vacation students visit less often – presumably because they spend less time in the center of the city. The customer base for the gaming stations ranges from schoolchildren to adults, and the adults tend to call ahead to see if a station is free and what games are in session before they come to the café. Using the services in Plazma costs two somani an hour to play games and four somani an hour to use the Internet. The ratio – games costing half what Internet access costs – is in line with general trends in the region. As argued above, when Internet access is metered and slow, the price differential between playing games and browsing makes it increasingly attractive to see games as the open-ended entertainment activity associated with computer technology rather than aimless browsing or opportunistic link following.

Gigant 2 is located on the edge of Dushanbe in a residential neighborhood. It is located in a one-story strip of local businesses, There is a bread store on one side and a small dry goods market on the other. On the sidewalk in front are several women selling seasonal fruit. Gigant 2 has streetside signage advertising Internet services, prepaid IP telephone cards, DVD/VCD and VHS rentals. The café/club rents movies, burns DVDs and CDs, and offers multimedia services. There are 24 stations in the main room of Gigant. There is also a back room that is attached to the main room and separated only spatially. There are six stations in the back which serves as a VIP room and includes two couches and a large floor fan. Gigant is nominally air conditioned, and heavy plastic sheets hang as a barrier at the front door to keep the cooler air inside, but in the height of summer it is very warm inside. There are two IP telephone booths in the café, and a site-wide LAN for watching movies. Dozens of games were preloaded onto the machines, but Lineage, CS, Need for Speed, and Cossacks were reported to be most popular. The business remains open all night, and from 9pm to 6am the only services offered are Internet access and Lineage. CS was not offered during the overnight because it was too difficult to find many other players online during that time frame. Gigant has a three tired pricing structure, similar but not identical to Plazma, see Figure 8. Playing games costs one somani per hour, being “online” also costs one somani per hour, and “Internet” costs 2.5 somani per hour. The distinction between online and Internet is something that appears with some regularity in developing regions, and it is a further motivation for individuals to develop usage patterns that downplay the importance of open-ended web browsing. In the case of Gigant, online activities include chatting, but Internet means web browsing since ISPs often charge by the kilobyte. Under those circumstances, then, browsing can be extremely costly and, perhaps even more of an inhibiting factor, not particularly transparent in terms of expense.

Figure 5: 2006 pricing structure at Gigant 2: Games 1 somani, Online 1 somani, Internet 2.5 somani Gigant is a café that melds general Internet services and gaming into one space. In the afternoon, it was packed exclusively with schoolage boys. The manager asserted that in

50

the evening, when adults returned home from work, the clientele turned over. Indeed, around 6 pm, some of the boys drifted out, but increasingly adults – including women, began coming into the business. These two profiles capture many of the business environments where people play games in developing world contexts. Their public nature, idiosyncratic policies, and tiered pricing structures are common elements and evidence of variability across venues.

VI.

FINDINGS:

GAMES AS A SIGNIFICANT PART OF AN ICT ECOLOGY AND POTENTIAL SITE OF A USER‟S FIRST “TOUCH” OF A COMPUTER

Throughout this project, we have maintained a focus on the study of diverse users‟ information landscapes -- their ICT ecology. In other words, this work takes particular interest in how specific pieces of ICT form unique mosaics in different usage contexts, giving individuals and communities access to variable modes of interaction, communication, and information-seeking. What has become clear over the years is that games and gaming provide an important – if often overlooked – piece of this puzzle [15]. Often dismissed as irrelevant to capacity-building projects, overlooked as a measure of a nation‟s ICT sophistication, and prey to countless stereotypes about users, games are a largely invisible component of the ICT landscape in developing regions. However, from the standpoint of users gaining facility with ICTs, and allowing ICT-related industries to gain an early foothold, games play an important role as they are often the first attractor – the first ICT with stickiness – for novice users. In other words, games are an alternative pathway to some users‟ first „touch‟ of a computer. On a functional level (as well as philosophical level), ICTs mediate distance and time [2][3][19]. In a country such as Kenya or Cambodia, nations with poor road infrastructure, the ability to transfer information quickly and over large distances has a transformative effect on the kind of information environment within which people live. On the other hand, there is also the conceptual mediation ICTs play as they bring fractured components of the globalization narrative to small communities that might be otherwise have very limited exposure to outside media influences. In Central Asia, ICTs are a linchpin of economic development as well as a symbol of modernization. Computers in schools projects bring technology literacy to youth, e-government initiatives slowly bring elements of government online and into some version of transparency, and within certain professional sectors computers streamline practice and make international collaborations more effective. However, when such top-down initiatives are bracketed, and instead focus is placed on bottom-up patterns of technology diffusion, ICT usage follows a different narrative. Adoption and usage patterns that emphasize ICTs for communication, gaming, and entertainment emerge. From movies to games to ringtones, ICTs become attractive to everyday people for non-work purposes. This pattern runs

somewhat counter to assumptions made about users in resource-constrained environments, but the so-called frivolous uses of technology are, indeed, often what brings people through the door [31]. The sheer amount of game activity and gaming culture as revealed by longitudinal ethnographic work in Central Asia establishes the importance of games as a part of ICT ecology in this resource-constrained region. And when schools are not wired and home access rates are low, game cafes are likely places for people‟s “first touch” with computers. A. Nurturing public Internet access sites in early years: the role of games In the past several years, a number of studies of Internet cafes have demonstrated the importance of such public access sites for users in emerging markets and other developing regions; many of these studies have focused on NGOsponsored telecenters rather than commercial sites [45][8][[11-13][18][21]. Generally speaking, though, in the global south, access to the Internet is more often in public, shared space than it is in private homes, or even the workplace or school. There are several characteristics of public access usage, however, that mean people‟s relationship to the Internet develops a particular shape. [17] Public access generally brings an awareness that one‟s usage is economically constrained, metered by time and often by kilobyte, inhibiting the link-following that characterizes much Internet usage in broadband contexts. Internet cafes, for the purposes of this discussion, are publicly accessible commercial or noncommercial sites where people gain access to a variety of ICTs, including Internet, IP telephone, photocopying, faxing, etc. Many commercial sites augment their income by providing some café-like amenities, but often the name “Internet café” is conceptual only. But Internet cafes are also often the first places that ICTs come to communities in developing world contexts. Whether established by an NGO or a local entrepreneur, Internet cafes advertise novel services to a community that is largely technologically illiterate, and they then face the challenge of convincing people they have a need for services they do not know how to use. Indeed, the capabilities of the Internet are themselves somewhat opaque to individuals who have not been exposed to a new media infused environment, and games are often a way to first draw people through the door. The research conducted in Central Asia from 2000-2007 pointed to an evolutionary pattern for commercial public access sites. To illustrate, take the example of Tashkent, the capitol of Uzbekistan. Tashkent is a city of approximately three million; it was the fourth largest city in the Soviet Union. In December of 2000, there were twelve operating Internet access points in Tashkent. This list included two sponsored by NGOs or large multi-lateral organizations, one funded by a local cultural center, and nine commercial cafes – most of which did not actually serve food or drink. That number has since grown to the hundreds, and the actual number is impossible to estimate given the fact that many operate without being fully licensed by the government, signage can

51

be missing or misleading, and there is no reliable central directory of all functioning businesses in the city. However, it is safe to state that the Internet business in Tashkent as of 2007 was thriving. Home access is growing, numerous ISPs offer local dial-up service, others offer DSL, and schools and businesses are increasingly getting Internet access. In 2000 (the early years of Internet diffusion), it was not immediately clear to individuals why they should spend about US$1 per hour to use something that had linguistic and technological barriers, particularly in a country where the average monthly salary was US$20-30. Interviews with novice users over the course of five months in 2000 pointed to such confusion and illustrated the cultural gaps that acted as additional barriers to entry for new ICT users. There was an absence of external information resources in people‟s lives, so using analogies to describe the Internet as, for example, like yellow pages or encyclopedias was not particularly resonant. Keyboards were often in English yet people spoke Uzbek or Russian and needed the Cyrillic alphabet. Operating systems and software were often in Russian which was an additional hurdle to Uzbek speakers (Uzbek is a Turkic language, not Slavic). Printer drivers didn‟t have the Uzbek character set, there was very limited content online that was relevant to local inhabitants, and keyboarding or typing skills were not common. Literacy rates themselves were quite high, although the multi-lingual nature of life could make reading and writing a bit more complicated depending on primary and secondary language ability. [14][16] However, because copyright laws were loosely if at all enforced, there was no shortage of entertainment media available. The latest games could be bought for about US$1 in kiosks on the street, and VCD movies were about the same price. At the same time, games could be played either standalone on individual computers or over a café-based LAN. Alternatively, movies could be watched on the computers by an individual or a group. Cafés developed a two-tiered pricing structure where it cost roughly half as much per hour to play games. What emerged, then, was a pattern of usage in these Internet cafes where the majority of customers came in to either use computers to type documents (also much cheaper than Internet access), or to play games. During the day, schoolchildren would come to play a variety of games, and often café owners would let the youngest children hang around and watch movies or play games if the café was otherwise empty. Playing games provided young people with their first experiences touching computers. Playing demystified the technology, allowed them to gain keyboarding skills, taught them how to navigate operating systems and GUI menus, and generally build comfort and facility with computers. In those early days, in 2000, it appeared that revenue from activities other than Internet usage allowed Internet cafes to remain open in their communities. Although it was impossible to get owners to disclose actual revenue figures, in conversation and observation over the course of six years of field visits, it became clear that for many businesses, the revenue generated by game players helped sustained these early adopter businesses. Consequently, when

a community member did need occasional access to the Internet, there was in fact a local site to visit. However, without games and other non-instrumental uses of the computers, the businesses would have had a much smaller regular customer base on which to draw. We would argue, then, based on our interviews and observations that games played a pivotal role in the ICT diffusion within that country. And once people started playing games, particularly LANbased games, they began chatting and utilizing other communication tools. Their usage of ICTs expanded beyond games, but it was the gaming activity that allowed them to enter the so-called information society. B. The appeal of games as a pathway to ICTs compared to other technologies While initially games might have been appealing because they were an inexpensive way for users to begin experimenting with computer technology, there was also a functionality argument in favor of them. In many developing world contexts, including Central Asia, Internet access tends to be fairly slow. Often an Internet café will have a dozen or more computers sharing one dial-up or DSL line. Such bandwidth limitations play out, for example, in a contemporary web page from a more developed country taking two or three or five minutes to load. In and of itself that may not seem a hurdle, but the time adds up. Using a basic free webmail application like Yahoo or Hotmail becomes an exercise in patience. To write an email in Yahoo mail, for example, takes about six page loads: one to load mail.yahoo.com; a second to enter a username and password and wait for login authentication, a third to click on the inbox and check mail; a fourth to click on compose to write a message; a fifth to click on send and wait for the message to go out; a sixth to click on check mail again and be returned to the inbox. Do the math, and suddenly it‟s 20-30 minutes to write one email – and that‟s without having to check someone‟s email address by loading the addressbook. Because these are public Internet sites, pop mail or SMTP is of limited utility, and so webmail quickly became the mail application of choice. However, people were forced to use webpages created in 2000 with access speeds more approximate to 1995. The Internet, then, was simply not perceived as terribly efficient; indeed, it was not at all efficient. Opportunistic web browsing was not a habit easily adopted when web pages took so long to load. Expensive and slow, the Internet was a window to frustration rather than a window to the world. Added to this limited utility, however, was a slightly more acute problem of surveillance and censorship.. Many Internet cafes had signs posted warning that accessing inappropriate content would result in fines or arrest. The managers or administrators would sit at a desk and monitor traffic; users were crowded elbow to elbow with very little privacy for the items on their screen. Inappropriate content was rarely if ever defined. Sometimes it was specified as political information, or inappropriate political information; other times pornography was banned. But what constituted inappropriate

52

political content could change; a website with acceptable regional news one day might be overly critical of the president the next and suddenly be on the banned list. The habit of censorship and surveillance for one‟s Internet usage provided another tamping mechanism for Internet use at the Internet cafes, and another motivation for those who wanted to learn about computers to gravitate toward games. Amidst all this gaming in the region – activity that was economically as well as socially appealing, gaming-specific centers gradually emerged. While game playing allowed Internet cafes to remain open in the early days of ICT diffusion, eventually the industry matured substantially enough that gaming centers split off from Internet cafes. There were enough Internet users producing Internet-only revenue that many businesses could focus on one or the other. Many crossover businesses remain, especially in countries with lower percentages of Internet users (Uzbekistan, Tajikistan), but one can see that as the sector matures and a business can take in enough revenue from Internet access to remain sustainable, that game cafes separate themselves from Internet centers. It is not a zero sum game, though, and as the numbers of Internet users grows, gaming remains a vibrant activity. Indeed, the presence of multiple gaming centers throughout the region – particularly in the capitol cities -- makes it clear that games are being played, both players and games are diverse, and gaming provides a social interaction platform. 70% 60% 50% 40% 30% 20% 10% 0%

Home ***

General InterFriend's Access net cafe house Point ** ***

Kyrgyzstan N = 121

49%

Kazakhstan N = 215

62%

Uzbekistan N = 115

27%

Tajikistan N = 122

29%

3%

6%

School **

Work

27%

15%

8%

15%

2%

5%

22%

10%

12%

10%

30%

17%

20%

11%

30%

17%

18%

11%

we wanted to identify those users who overcame the barriers to entry for either activity. Our goal in this exploration was to investigate if the two groups differed, and if so, how they differed and what the differences would suggest about potential barriers. Two direct logistic regression analyses were performed using SPSS to predict gaming and Internet use outcome from eight predictors: (1) age; (2) gender; (3) years of schooling; (4) living in a rural or urban environment; (5) mobile phone use; (6) ability to speak and read English; (7) ability to speak and read Russian; and (8) socio-economic status reported on a scale of one to three. A. Comparing predictors of game playing to Internet use First, we analyzed game playing. A test of the full model with the set of predictors against the null model with no predictors was significant, 2(9, N=12000) = 2268.52, p<.001, Nagelkerke R2 = .317, indicating that the set of predictors reliably distinguishes between individuals who play games and those who do not. The approximate variance in predicting game playing accounted for by the set of predictors is 32%. According to the Wald criterion all eight variables reliably predicted computer gaming--listed here in order of influence: (1) age; (2) ability to speak and read Russian; (3) owning a mobile phone; (4) living in an urban environment, (5) gender; (6) ability to speak and read English; (7) years of education; and (8) SES. Next, we analyzed Internet use. A test of the full model with the set of predictors against the null model with no predictors was significant, 2(9, N=12000) = 2852.51, p<.001, Nagelkerke R2 = .447, indicating that the set of predictors reliably distinguishes between individuals who use the Internet and those who do not. Internet use accounted for by the set of predictors is a striking 45% According to the Wald criterion, all eight variables reliably predicted Internet use--listed here in order of influence: (1) ability to speak and read English; (2) age; (3) years of education; (4) owning a mobile phone; (5) living in an urban environment; (6) ability to speak and read Russian; (7) gender; and (8) SES.

Figure 6: Sites of Gaming in Central Asia, 2008 *** = p <.001; ** = p <. 05 Statistics based on Chi-square test 600.00

VII. SURVEY FINDINGS: LESS ADVANTAGED USERS MORE

500.00

LIKELY TO GAIN INTRODUCTION TO ICTS THROUGH GAMES

400.00

As discussed in the previous section, games can provide a mechanism by which youth can be drawn into ICT training centers, and games also provide a revenue stream for owners of public access ICT sites. Additionally, our survey results demonstrate that games offer tangible assistance in overcoming barriers to entry for novice users by allowing users with less education and English language skill to gain experience with ICTs. In order to explore potential barriers to entry for novice users we wanted to first identify what types of people play games compared to those that use the Internet, In other words,

300.00 200.00 100.00 0.00 Age

Gender

562.34

97.72

Internet Use 248.22

12.41

Gaming

Years of schooling

Living in an urban /rural

56.36

112.10 113.51

Mobile phone use

English Russian

75.40

SES

122.21

44.02

196.40 142.06 171.21 339.05 135.45

11.19

Figure 7: Predictor effects on the models according to the Wald Statistic

53

While the sets of predictors are the same, there are notable differences. First, the set of variables does a much better job at predicting Internet use than game players, indicating there is more homogeneity in the Internet using population; however, both models were very good at predicting the outcome. Second, the importance of individual variables is different. Whereas the ability to speak English is very important to predicting Internet use, it was less influential on game playing. Additionally, education level is very important to predicting Internet use, but not as much for computer gaming. Conversely, age is a stronger predictor of gaming. Together, these items suggest that entry into the game playing realm is more accessible to a different segment of the population. If the goal, then, is to broaden participation in ICTs, games provide a pathway to ICT usage for users who would not necessarily have the education or English language ability to become Internet users. See Figure 9. This finding is significant in that it clearly establishes that gaming and Internet use attract different segments of the population, and that an individual‟s likelihood to gravitate to one version of ICT over the other is not only due to availability or interest, but is also due to societal factors such as education. VIII. FINDINGS: GAMES AS A GATEWAY TO AND MOTIVATION FOR INNOVATION The strong attractor force of games can also be a motivator for innovation. Many games encourage users to become active participants in a digital environment, and the enthusiasm many players feel for games can motivate them to learn new technical skills in order to facilitate their playing. A. World of Warstan Blizzard‟s World of Warcraft (WoW) is a subscriptionbased game that requires Internet access. It is a massivelymultiplayer game that has thousands of players on each server. Much of the gameplay relies on collaboration, and players form guilds; trade happens in auction houses, and the game requires a certain critical mass of players to run effectively. The technology and banking infrastructure required by WoW would seem to relegate it to an impossible game in the context of Central Asia, a region with relatively slow Internet access and essentially no credit cards. However, in at least two of the capitol cities in the region, local ISPs have devised a way to offer WoW to their subscribers. Locally run servers allow DSL subscribers to play innetwork, so, for example, there is a Tashkent-wide version of World of Warcraft. The game is in Russian but with a server attached to the provider. Most of the people are from the capitol given the pattern of DSL diffusion. Players recruit friends and classmates, but because of the low in-game populations, the basic mechanics of the games are different from standard play in developed world contexts. For about UDS$12-13 per month, players get unlimited hours online with both WoW and ICQ. All other Internet activity is metered by kilobyte, and MSN and Yahoo Messenger are metered as well. As for WoW, there are only a few hundred users online in the world, which means there are

very few guilds, and the auction house is not a particularly effective mechanism for trading goods. While economically viable in this context, Azeroth, the world of WoW, is also a largely empty world for these players. However, clearly the online game has a draw, as do many other game genres. The local hacks that people use to be able to play games are a testament to the appeal that games of diverse genres have for users. B. Games as motivation to gain technical expertise Similar local adaptation can been seen among local gamers in Bishkek. A group of, neighbors, residents of the same large apartment building, discovered that they all liked to play the same games, but they preferred to play at home because they saw it as cheaper and they had “more freedom.” So they ran a LAN down the outside of their nine story Soviet style flat, connecting eight neighbors together so they could play together. Gamers display a fair amount of creativity in ensuring they are able to remain connected with global gaming culture. Their narratives of themselves as gamers, however, emphasize the public and social aspects of their gaming which is not necessarily consistent with gaming in the US, but does resonate with gaming culture in Korea and China. One of eight Kyrgyz gamers interviewed in 2006, Yuri echoed the pattern of games as a motivator for learning more about technology. He started playing games in the 4th grade, and was introduced to games by playing Flight Sim at a friend‟s house, long before computers had been introduced to his public school. After Flight Sim, he began going to computer clubs where he fell into the world of Counterstrike. Now 19, he plays games over a LAN only in the clubs because, as he says, Internet is too expensive. After playing for years, he gained a variety of computer-related skills and now works as an administrator at a game café. He calls his friends and arranges for them to come after hours and gather together to play. Although none of the other gamers interviewed in Kyrgyzstan in 2006 had direct stories of their gaming leading to jobs, they did all emphasize the collective, shared knowledge of their gaming circles, the enjoyment of gathering in public places together to compete, share, and improve skills. Their stories also stand in contrast to the usual ICTD narratives about how newcomers become first acquainted with ICTs and the associated information society. Often, development projects that incorporate technology emphasize instrumental uses of technology. Games, however, are part of an open-ended and alternative exploration of technology. IX. CONCLUSION Much of the research on game culture concerns itself primarily with gaming in extremely developed contexts – where resources are abundant, and where gamers‟ relationship with technology has evolved in the context of that abundance. Game culture is just as interesting, and just as vibrant, and just as reflective of complex social dynamics when it takes place in resource constrained environments and when the gamers who adopt these technologies have very different relationships

54

with the technology. Central Asia is one such resource constrained environment, and while the specifics of the region give it particular political and economic constraints, the stories that emerge of games‟ importance in allowing ICTs to take hold during early stages of technology diffusion, as well as the persistent public nature of gaming activity help to broaden the picture of what games can be and what purpose they can serve within an overall ICT ecology. It may also be that gaming culture is the non-instrumental use of ICT that can parallel some of the adoption stories for mobile phones. Games are fun, and there‟s nothing wrong with fun, even in a development context. If talking to friends and family is part of the leverage that gets people to adopt mobile phones, then playing with friends may be part of the leverage that gets young people to develop further skills with computers. As we have argued in this paper, gaming culture is vibrant and thriving, and it provides a potential “first touch” with ICTs , especially for people who may not have access to computers at home, work, or school. Additionally, our survey demonstrates that games provide different barriers to access than Internet usage, broadening the reach of ICTs. Finally, games and the enthusiasm that gamers have, can motivate innovation and technological skill acquisition. Ultimately, the goal of this paper has been to bring attention to games and some of their possibilities as an element of ICT ecology in resource constrained environments. Games have been almost completely ignored by the ICTD discourse, and as the community seeks to better understand how and why people adopt ICTs into their lives, it seems important to include entertainment-related uses of ICT as part of that adoption pathway if we hope to truly comprehend the cultural and economic changes associated with technology. ACKNOWLEDGMENTS Our thanks to Mark Chen, members of the CAICT research team, including Erica Johnson, Emma Rose, and Odina Salikhbaeva. Thanks also to our local researchers, research participants, and blind reviewers. REFERENCES [1] [2] [3] [4] [5]

[6] [7] [8] [9]

Cassell J. and H. Jenkins (eds.), (1998). From Barbie to Mortal Kombat: Gender and Computer Games, Cambridge: The MIT Press. Castells, M (1996). The Rise of the Network Society, The Information Age: Economy, Society and Culture Vol. I. Blackwell: Cambridge, MA. Castells, Manuel (1997). The Power of Identity, The Information Age: Economy, Society and Culture Blackwell: Cambridge, MA Colle, R. (2005). Memo to Telecenter Planners. Electronic Journal of Information Systems in Developing Countries, 21(1), 1-13. Dagron, A. G. (2001). Prometheus Riding a Cadillac? Telecentres as the promised flame of knowledge. Journal of Development Communication: Special Issue on Telecentres 12(2). Dubbels, B. (2003) Video Games as Metaphor for Learning and Curriculum Design Egenfeldt-Nielsen, S. (2003). Review of the research on educational usage of games Fuchs, R. (1998). Little Engines that Did: Case Histories from the Global Telecentre Movement. Ottawa: IDRC. Garris , R. , & Ahlers, R. (2003). Games, motivation, and learning: A research and practice model

[10] Gee, J. P. (2003) What video games have to teach us about learning and literacy. Palgrave/Macmillan: New York, NY [11] Hudson, H. E. (2006). From Rural Village to Global Village. Mahwah, New Jersey: Lawrence Erlbaum Associates, Publishers. [12] Jensen, M., & Esterhuysen, A. (2001). The Community Telecentre Cookbook For Africa - Recipes For Self-Sustainability. Paris: United Nations Educational Scientific and Cultural Organization. [13] Khelladi, Y. (2001). What Works:The Infocentros Telecenter Model. Washington DC: The World Resources Institute. [14] Kolko, B.E. (2002). “International IT Implementation Projects: Policy and Cultural Considerations.” Proceedings from the Annual IEEE IPCC Conference, Portland, OR, September 2002. 352-359. [15] Kolko. B.E., Thayer, A. (2003).“Games as Technological Entry Point: A Case Study of Uzbekistan.” Proceedings of the Digital Games Research Association. Utrecht University. 19 pages. [16] Kolko, B.E (2006) “Cultural Considerations in Internet Policy and Design: A Case Study from Central Asia.” Critical Cyberculture Studies: Current Terrains, Future Directions. Ed. David Silver and Adrienne Massanari. New York: NYU Press. 145-157. [17] Kolko, B.E., Rose, E.J., Johnson E. (2007). “Communication as Information-Seeking: The Case for Mobile Social Software for Developing Regions.” Proceedings of ACM World Wide Web Consortium Conference. 863-872. [18] Mercer, C. (2006). Telecentres and transformations: Modernizing Tanzania through the Internet. African Affairs, 105(419), 243-264. [19] Poster, M. (1990). The Mode of Information. University of Chicago Press. [20] Proenza, F. J. (2001). Telecenter Sustainability - Myths and Opportunities. Journal of Development Communications, 12. [21] Proenza, F. J., Bastidas-Buch, R., & Montero, G. (2001). Telecenters for Socioeconomic and Rural Development in Latin America and the Caribbean. Washington DC: Inter-American Development Bank. [22] Schott, Gareth R. and Horrell, Kirsty R. (2000). “Girl Gamers and Their Relationship with the Gaming Culture,” Convergence, v. 6, n.4, 36-53. [23] Taylor, T.L., Kolko, B.E. (2003). “Boundary Spaces: Majestic and the Uncertain Status of Knowledge, Community and Self in a Digital Age.” Information, Communication & Society. 6:4, 497-522. [24] Yates, Simeon J. and Littleton, Karen. (1999). "Understanding Computer Game Cultures: A Situated Approach," Information, Communication, & Society, 2:4. [25] Williams, D., M. Consalvo, S. Caplan & N. Yee. (2009, in press). Looking for gender (LFG): Gender roles and behaviors among online gamers. Journal of Communication. [26] Williams, D., N. Yee & S. Caplan (2008). Who Plays, How Much, and Why? A Behavioral Player Census of Virtual World. Journal of Computer Mediated Communication. [27] Nadia Bianchi-Berthouze, Whan Woong Kim, Patel Darshak, "Does body movement engage you more in digital game play? And Why?", Proceedings of the Int. Conf. of Affective Computing and Intelligent Interaction, LNCS 4738, 102-113, Lisboa (Portugal), September 2007 [28] Stevens, R., Satwicz, T., & McCarthy, L. (2007). In-Game, In-Room, InWorld: Reconnecting Video Game Play to the Rest of Kids' Lives. In Katie Salen (Ed.), The Ecology of Games: Connecting Youth, Games, and Learning (pp. 41 - 66). Cambridge: The MIT Press. [29] Nardi, B. and Harris, J. (2006). Strangers and friends: collaborative play in World of Warcraft. In CSCW '06: Proceedings of the 2006 20th anniversary conference on Computer supported cooperative work, pages 149-158, New York, NY, USA. ACM Press. [30] Holland, W., Jenkins, H. & Squire, K. Theory by Design (2003). In Perron, B., and Wolf, M. (Eds). Video game theory reader. (pp. 25-46). London: Routledge. [31] Sandvig, C. (2006). The Internet at Play: Child Users of Public Internet Connections. Journal of Computer-Mediated Communication, 11 (4). [32] Wise, John MacGregor. (1997). Exploring Technology and Social Space. Thousand Oaks, CA: Sage. [33] Lenhart, Amanda, Kahne; Middaugh, Ellen, Rankin Macgill, Alexandra; Evans, Chris; Vitak, Jessica. (2008). “Teens, Video Games and Civics: Teens' gaming experiences are diverse and include significant social interaction and civic engagement”. Pew Internet and American Life report. September 16, 2008. Available at http://www.pewInternet.org/pdfs/PIP_Teens_Games_and_Civics_Repor t_FINAL.pdf [34] Lenhart, Amanda; Jones, Sydney; Rankin Macgill, Alexandra. (2008). “Adults and Video Games” Pew Internet and American Life Project

55 Data Memo. December 7, 2008. Available at http://www.pewInternet.org/pdfs/PIP_Adult_gaming_memo.pdf. [35] Bruno Latour. (2005). Reassembling the social: an introduction to Actor-network theory, Oxford, New York: Oxford UP. [36] C. Shawn Green andDaphne Bavelier. (2003). “Action video game modifies visual selective attention.” Nature. Vol 423. pp. 534-537.

56

Content Creation and Dissemination by-and-for Users in Rural Areas Sheetal K Agarwal, Arun Kumar, Amit Anil Nanavati, Nitendra Rajput IBM Research, India Research Lab 4, Block C, ISID Campus, Vasant Kunj, New Delhi, India. Email: sheetaga,kkarun,namit,[email protected]

Abstract—83% of the world population does not have access to Internet. Therefore there is a need for a simple and affordable interaction technology that can enable easy content creation and dissemination for this population. In this paper, we present the design, development and usage pattern of a VoiKiosk system that provides a voice-based kiosk solution for people in rural areas. This system is accessible by phone and thus meets the affordability and low literacy requirements. We present usability results gathered from usage by more than 900 villagers during four month of the on-field deployment of the system. The on-field experiments suggest the importance of locally created content in their own language for this population. The system provides interesting insights about the manner in which this community can create and manage information. Based on the use of the system in the four months, the VoiKiosk also suggests a mechanism to enable social networking for the rural population.

I. I NTRODUCTION The total rural population in 2005 is about 51% worldwide. If we look at less-developed nations, then this number is even larger, at 57%, which translates to more than 3 billion people. For India, the percentage of rural population is even higher, 71%. Based on the projections in [1], this number is not likely to change fast in the coming years (the percentage yearly change from rural to urban is projected to be around 1%). Even so, most of the HCI research has been focussed on the interaction with computers, to which this population does not have access. Therefore it is extremely interesting to look at sections of this large population base and identify the information and communication needs, the means to achieve these needs, and study the interaction modalities of this society. A majority of this 71% of rural population in India is illiterate or at least textually-illiterate [2]. Therefore though they can count currency and make phone calls using numbers, the ability to read or write is very limited. The means of interaction of this society is either completely voice-based or paper-based. Some people (esp. the money lenders, shopkeepers) make paper notes of their business. Most people (such as farmers) interact using voice, and get information through radio, television and other public announcement methods in villages. News papers continue to be another source of accessing information. However the common problem across these information sources for these villagers is the lack of locally relevant content. Most news papers, radio and television programs provide

information that caters to a wider section of the population, thus leaving out the details for specific villages. One of the common methods of providing local information (such as a health-related warning) is done through announcements using loud speakers in vehicles. Figure 1 shows one such vehicle being used by a government healthcare official to announce an upcoming health camp in the village.

TriC 1

Pyr.mea.IT

Fig. 1.

IBM India Research Lab

June 24, 2008

Announcement of a government organised health camp.

There is a lack of enough information sources that can create sufficient locally relevant content such as doctor visit organised by the government, change in school timings, delay of the train service, list of movies in the village community, etc. If there is a system that can enable this community to create their own content, the information and communication ecosystem can be significantly improved. Since not much is studied about the rural population, to build an information system that can be used by rural population to create and access content, it is extremely important to follow the following design steps: 1) Perform a needs-study to identify the information needs of this population. It is important to understand what does a community consider as information. To enable easy access of information, a meaningful categorisation is also needed. 2) Follow a participatory design process to develop a lowcost, low-literacy solution. Since a majority of this rural population in India lives below USD 2 per day [3], it

57

is important that the solution is affordable. Moreover, the low literacy considerations also need to be taken in developing the end-user interface of this system. Participatory design process is important to study a population that has not been exposed to interaction modalities beyond paper, radio and TV. 3) Observe usage pattern of the solution for the rural population. Since not much research has been performed in enabling local content creation, the usage pattern of this population will be interesting to observe and can provide important insights. In this paper, we follow these three principals to build VoiKiosk – a system for increasing reachability of information kiosks in developing regions. A VoiKiosk is a voice based service available on the Telecom network [4]. As opposed to PCs, the phone penetration in rural India has been significant and continues to grow. Also based on field studies and literature reviews semi-literate and illiterate people are more comfortable with speech-based interfaces to access information services [5]. Leveraging the increased mobile penetration and comfort with speech based interfaces, we present this alternate model to create and host voice-based kiosks (VoiKiosks). We identify the information needs of villagers (Section III) in south of India by visiting several villages and talking to villagers and officials of an NGO that operates in these villages for more than seven years. Based on the needs, we implement the VoiKiosk system (Section IV) and involve a limited number of users and the NGO (Section V) in the participatory design phase. We deploy the system live in one village and gather important usage statistics by four months of live field deployment of the system (Section VI). We conclude the paper by discussing important insights gained during the entire process of needs-gathering, participatory design, prototype testing and usage analysis. II. R ELATED W ORK Several studies have suggested that rural communities have very different information and communication needs and patterns [6] [7]. Moreover, it has also been studied that technological innovation opportunities in rural areas have not been studied in great depth [8]. In [8], the authors present field studies suggesting that use of a local stakeholder contributes significantly to a better design of the technologies for rural communities. While there are significant design challenges for the rural community, at the same time, the potential of technology to improve their lifestyle is huge [9]. This forms the motivation for the work presented in this paper. A. Information Access and Dissemination In the current world of globalization, access to information is key to the development of a society [10]. The increasing amount of information on the WWW [11] is a reflection of this fact. Not surprisingly, therefore, there have been several efforts to provide this information to people in rural areas. There are solutions in the network space to provide low-cost Internet access access for rural areas [12], [13]. Significant work has

also been done on improving the usability for mobile Internet interaction. In [14], the authors highlight the various usability issues by studying the different critical comments by mobile Internet users. Despite all these solutions, the usefulness of Internet information for people in developing countries is very limited due to the fact that the content that is relevant for a villager is not available on the Internet [15]. This paper positions itself in the space of creation and dissemination of locally relevant content and is supportive of the statement: access to relevant information is key to the development of a society. B. Alternate Models for Local Information Systems Community Radio systems have been studied as a alternative solution to provide information in rural areas. In [16], the authors present a study that evaluates the effectiveness of 10 community radio stations in different rural communities in South Africa and report the popularity of programs. A sustainability study of the community radio service is presented in [17] for Nepalese radio stations. Extensions to community radio service to involve audience participation have been presented in [18]. Among other things, the success of community radio lies in the availability of relevant content for the rural communities. The CAM system [19] provides an interface to capture local data using the mobile phone camera. It has been applied for microfinance data capture [20] and for procurement and quality control in a rural cooperative [21]. Using camera phone to capture paper content is another approach to create local content that can be used in several application domains. We derive the importance of relevant content presented in the cited work and use these concepts in this paper. C. Speech and Mobile based Systems for Low-literacy Users Since textual literacy is a major issue in developing countries, especially in rural areas, several interesting information systems for this community are speech based. The community radio systems mentioned above are completely based on audio information. Studies from a spoken dialog system for rural India [22] show that users are able to navigate the dialog system irrespective of their literacy levels. A speech-based system for providing health information to low-literate users has been presented in [23]. This paper reports that there is a need to educate the users by creating a video to explain and demonstrate the system for a new user. Information systems based on low-end mobile devices provide an affordable alternative to PC-based systems. A system for providing information to farmers using mobiles has been presented in [24]. The authors in [7] conclude that exploring mobile social software holds significant potential as an information tool in rural areas. The MobilED [25] and the AudioWiki [26] are recent solutions that address this space by providing information to this population through mobile devices. The work presented in this section leads us to believe that creation and dissemination of locally relevant content through

58

speech interface using mobile devices can be used as an effective system to provide information needs of the rural population in developing countries. System level technology development work in this domain has been presented in [27] and [28] where the authors present technologies for content creation and linking. A solution based on these technologies [29] proposes to organize the unorganized urban poor businesses. In this paper, we extend the concepts of the World Wide Telecom Web [30] and apply them for providing information solutions to the rural population in developing countries. III. N EEDS S URVEY In order to understand the information needs of a community, it is important to know the community for a significantly long time. Selective interviews are often restrictive and may not provide deep insights that can be gained with time. The understanding of the working and living methods can either be learned through an extensive ethnography study or by learning from experience of organizations who have done such study and have a presence in the community [8]. We relied on the latter approach. This was one of the reasons for us choose villages in the Andhra Pradesh state in southern part of India. A. The NGO Operations in the Field Byrraju Foundation [31] is an NGO that has been operating in 198 villages in 6 districts of Andhra Pradesh. The NGO has performed detailed studies to identify the needs of the villagers from a much broader perspective. Their operations cover basic amenities such as sanitation, drinking-water and healthcare at one end to advanced services such as distanceeducation, remote-healthcare and rural BPOs. Following are the main activities of the NGO in these villages: •

•

•

•

•

Education: The NGO adopt one primary and one secondary government school in every village. They provide training to teachers, conducts distance education courses, and conduct health check-ups for students. Healthcare: There is a health center for every village where a doctor visits for 2 hours daily. The Village Coordination Officer (explained later) acts as an assistant to the doctor. There is an ambulance service for all the villages. Sanitation: The NGO has facilitated garbage collection system for all villages. The garbage collector generates and sells manure from the garbage to make the system self-sustainable. Drinking water plants have been set in some villages, they provide door-delivery of water at 10 paise per liter (USD x per gallon). Ashwini Centers: 20 villages have computer centers. These are used to provide distance education to students at school times. Later they are used to provide painting and embroidery classes to women. Agriculture/Aquaculture Advice: Experts in distant locations provide personal advice to agri/aqua farmers based on the crop/fish conditions.

The NGO has a Village Coordination Officer (VCO) for every village. A VCO is usually a woman who has had 12 years of formal education. The VCO can only understand local language, Telugu. For every 3-5 villages, the NGO has positioned a Nodal Coordination Officer (NCO) to oversee activities in these villages. The NCO can understand English and is a degree holder in any subject. The NCO can operate the computer. Every village has a Gram Vikas Samitee (GVS) that is a committee of 18 members from the village who have volunteered to work with the NGO on the 18 different modules. B. Needs Gathering on the Field We visited four villages accompanied with the NGO field staff to understand the information needs of this community. The NGO staff provided the development history and details of the community for these villages. These four villages were identified based on the varying demographics to get a broader perspective. The first village (Vandram) has agriculture as the main source of livelihood and paddy is the main crop. Juvvala Palam is one of the larger villages and is geographically located as the central hub for nearly 4 lakh population in nearby villages. This village has a large population base that deals with renting of transport vehicles and so the associated business of mechanics, drivers is also very common. The third village (Cherukumilli) is primarily aquaculture based and has a number of fish ponds where aqua farmers cultivate fishes. Ibhimavaram is a very wealthy village where farmers have large lands and they usually employ labourers to cultivate their farms. Except for Cherukumilli, all the other three villages had a computer center (called as Ashwini Center). The chart in Table I summarises the important inputs gathered in the field. Each meeting in a village was with the GVS committee head, the NCO, the VCO and some members of the GVS committee of that village, all in one room. Most villagers share a phone within their family. Thus the cell-phone penetration rates are more than 50% considering the number of families that have a cell-phone. Ibhimavaram has 100% cell-phone penetration. Privacy of contacts or calls is not a concern in this open community. This culture enables easy sharing of cell-phones between family and friends. Almost all villagers mentioned that the mode of interaction with the cell-phone device is voice. People do not use text messaging, even though this is relatively cheaper. All cell-phones have Roman letter inscriptions and there is no local language support either on the keypad or on screen. All cell-phone users are number-literate and can recognize English language digits. However very few can understand the roman script for characters. The cell-phone signal is sufficiently strong in all remote areas. Although there are lot of electricity blackouts, but these intervals are short in duration, thus charging of cell-phones is not an issue. There are two major service providers (Airtel, BSNL), however there are about 5 different service providers available. Most officials and village committee members mentioned that people will be interested in knowing several schedule

59 TABLE I S UMMARY OF INFORMATION NEEDS IN FOUR VILLAGES .

Village

Population

Families

Mobiles

Main Operators Airtel/ BSNL/ Idea/ Vodafone

DTMF

Will pay?

Information Required

Remarks

Vandaram

2292

450

150

Depends on value of information. Voice is better

Not initially.

Program schedule of Ashwini, Information categorised on Modules, Pesticides, Medicines, Aqua information, sports updates, railway enquiry.

Yes for pilot

Juvvala Palam

3800

850

600

Airtel/ Tata/ BSNL/ Vodafone

Yes for Pilot

Airtel/ Idea

Voice

Idea/ Airtel

Voice

After realising the value, they may. Not initially. Yes

Community news, agriculture info, NO to train, no to Electricity, doctor visit, Teleconferencing information announcement, panchayat meetings, ankur channel program schedule. Hub village for 4 lakh population in nearby villages. information about village, Panchayat meetings, GVS meetings

Cherukumilli

4047

1000

500

Ibhimvaram

4600

3000

Tractor renting, electricians, etc. panchayat meeting, national fish rates, V-agri, V-agri, v-agri

GVS very interested for pilot.

related information such as distance education class schedule, village committee meting schedule and minutes. Other information needs were based on agriculture information, community news and crop rates. For the Juvvala Palam village, people were interested in knowing about electricians, mechanics and different rental options for farming equipments. Except for Ibhimavaram, other villagers mentioned that the villagers may not pay for the service initially, but once they can see value from the information, they have the capacity to pay for the information. Based on these meetings, we identified the following categories for which the information is required by villagers: •

•

• •

•

•

• • •

•

V-Agri: Provide agriculture and aquaculture advice to farmers on cell-phones. Job work: items and type required by merchants to that villagers can get business. Availability of transport. Health: Mostly static info such as doctor timings, date for the eye camp, student camp Aqua prices: Traders can upload prices of fish, prawns at which they will purchase. People: Static information about electricians, mechanics, carpenters of that area with phone numbers Entertainment: Name of movies in the nearby theater. Ashwini program schedule General info about the village: Population, availability of health center, schools, famous local food, etc. Community news: GVS meeting timings, meeting min-

same above

as

utes, sports achievements, etc. It was clear from the on-field interactions that a phonebased information portal that enables interaction through voice will be ideal from the end-user perspective. Such a system will ensure that the end-users will not need to learn any new interaction modality nor a new device. Given the significant cell-phone penetration, a cell-phone based system can be used to fulfill the information needs of these villages. IV. P ROTOTYPE

IMPLEMENTATION AND TESTING

Of the 10 different categories of information identified in the previous section, we built a prototype of the VoiKiosk system for four main categories. A VoiKiosk is a specific instance of a VoiceSite that can be used by villagers to create and access locally relevant content. A VoiceSite can be thought of as a parallel to a Website, but which can be accessed by dialing a phone number and information can be listened rather than being read or seen. Creation of a VoiceSite is made easy by the VoiGen system [27] through which anyone can call up the VoiGen system and interact with it through voice. This can enable any illiterate person to create her VoiceSite. Such a system enables easy local-content creation. All information in the VoiceSite is stored as audio messages that are recorded by making a phone call to the system. Since a VoiKiosk is a VoiceSite for the entire village, different type of users can update content in different sections. We explain the type of users and their interaction with the VoiKiosk in the next section. The kiosk operator (typically a NCO or a VCO) navigates through the VoiKiosk application

60

to configure it to offer various services for the village. Figure 4 shows a sample operational scenario for the VoiKiosk. Health Message Health Message Health Message Health Message

VoiKiosk

Kiosk-operator uploading information on VoiKiosk

Villagers calling the VoiKiosk for information

Fig. 2.

VoiGen

VoiGen allows creation and configuration of VoiKiosk

VoiKiosk Usage Scenario

A VoiKiosk act as information and service portal for a village. It can be a central point of access for a community where information relevant to the community can be posted and accessed directly by the users themselves. This solution doesn’t rely on Internet connectivity which is most often not available in the rural areas and most importantly, it allows end-users to directly interact with the services, thus removing the dependence on the kiosk operator. Following are the four main categories of information available on the deployed VoiKiosk system for the village: •

•

•

•

V-Agri: Farmers use this service to consult agriculture experts regarding their crop related problems. Currently a picture of the crop is taken and sent to an expert, who then sends a reply back to the farmer through the foundation. The turn-around time for this process is 24 hours. With VoiKiosk, the expert will be able to post his advice for the farmer on the VoiKiosk, reducing the turn-around time to 4 hours. VoiKiosk identifies farmers based on their caller IDs. Health Information: Information related to different health advisory and health camps is posted in the VoiKiosk. The schedule of the doctor visit to the health center is also posted on the VoiKiosk. The VCO can change the message if there is a change in the doctor visit timing. Ashwini Center Schedule: Information regarding new programs, schedule of daily classes or changes in the schedule are advertised by word of mouth, local newspapers or posters that are pasted at various spots in the village. Often people go to the the ashwini centers to get this information in person. The VoiKiosk has a Ashwini Center section where a kiosk operator can post the latest news related to the distance education program. Professional Services: In this section users will be able to record their personal advertisements. In the current practice, micro-businesses such as mechanics, drivers, daily wage skilled laborers use word of mouth advertising to reach out to clients. VoiKiosk provides an opportunity to increase their client base and increase business opportunities for them. Users call the VoiKiosk and record their advertisement which other villagers can browse by calling the VoiKiosk.

Create: Listen:

Welcome Message

Create: Listen:

Operator All

Fig. 3.

Operator All

Advertisement

Agri/Aqua-culture Advice

Ashwini Center Schedule

Advertisement

Ad

Ad

Operator All

Expert Farmer

All All

Information on VoiKiosk: categories and access

As shown in Figure 3, the information in these four categories can be modified by different kind of users. All information is accessible to everyone, except for the expert advice, which is specific to a particular farmer. We discuss about the different types of users in more detail in the next section. The call-flow for a specific interaction of the Kiosk operator with the VoiKiosk is shown in Figure 4. The Kiosk operator can choose to modify the welcome message for the VoiKiosk system, or modify information in any of the four different categories. Within a category, the operator is allowed to create a new information message, delete any existing message or rerecord an existing message. Kiosk Operator

Record welcome message

Edit Edit to edit the welcome message Services to configure services

Select a service to add: Health, Agriculture, Weather, Transport Agriculture Record service description

Back

Add to add information, Edit to edit the service description, Configure to configure the service, Back to go to service menu Configure Add an agri expert Browse registered experts Back to go to service menu

Add

Fig. 4.

Browse

VoiKiosk Usage Scenario

V. PARTICIPATORY D ESIGN

OF

VOI K IOSK S YSTEM

Given the four categories of information, based on the creator and the consumer of this information, there are four types of users of the VoiKiosk system:

61

The Kiosk Operator has the permission to modify any content on the VoiKiosk. He can call the VoiKiosk from his cell-phone and the system identifies his caller-id and provides the right permissions. • Experts are allowed to post expert comments on agriculture and aquaculture for specific farmers. They dial the phone number of the farmer for whom they have to post a message. Experts can not post comments in any other category. • Farmers call in the system to listen to any message from experts. They can listen to information in other categories too. • Villagers can post an advertisements, or listen to any other general information in the VoiKiosk. A kiosk operator is usually identified for the VoiKiosk. The kiosk operator has permissions to add different categories, and also to add and delete information in any category. We have developed a kiosk template for a group of villages in Andhra Pradesh.

cognitive load on the users is less if they are told what all they can say to the system.

A. Participatory Design Modifications

From the experts perspective, they wanted that any advice they post to the farmers should also be archived in the NGO IT infrastructure. The NGO believes that this database of specific advices can act as a knowledge resource in the future. We created a FTP site so that all the content in the V-Agri section would be pushed to the NGO server at midnight. The duration of any message to be posted on the VoiKiosk was restricted to 10 seconds in the initial design of the system. The system automatically detects silence and identifies the end of the utterance. However if there is some noise in the background, then the system is not able to detect the silence and the message recording continues. This can result in user frustration, both during message recording as well at the time of listening. The other option was to ask the users to press a specific key to explicitly let the system know the end of utterance. But this would involves a key input, which we wanted to minimize, given the profile of target users. So we had fixed the duration to 10 seconds. However this duration was too less to record some of the messages from the experts. Since experts were more advanced users in terms of the device familiarity, they could use the keypad while talking over the phone. So we increased the duration to 30 secs for the Experts. They now have to press the # key to indicate the end of the utterance in case of background noise. The participatory design process improved the system significantly. With the first version of the system, the Kiosk operator used to take about 10 minutes to reach the Ashwini Center part of the VoiKiosk and then update a message. Now it takes him (and any novice user) less than 2 minutes to update a message.

•

We developed the prototype system in consultation with the NGO and then took this to the field to get feedback from the four different type of end-users. Since the Kiosk Operator has the maximum number of options available on the VoiKiosk, we started our testing with 2 Kiosk Operators. The initial design of the system was such that when the system expected a user input, there would be a beep sound at the end of the system utterance. However the beep sound was present only when the system used to record user voice and not at the time when the system would do a speech recognition on the voice. Since this fact was not known to the users, they did not know when to start speaking. Sometimes they would respond too early and sometimes they were too late. This resulted in increased frustration and so to make sure that they are speaking in the interval when the system accepts their input, they kept on repeating the same utterance, such as “exit, exit, exit”. We modified the VoiKiosk interface so that after every system input, there was a beep sound. The users could then figure out that they are always supposed to wait for a beep sound before they respond. This improved the system performance significantly. There was a system prompt that asked the following from the users: “Do you want information on Health or Agriculture or Ashwini Center Schedule or you want to about the people in this village?”. We expected that people will respond by saying either of Health, Agriculture, Ashwini Center or People. However the users used to answer Yes or No. So we had to change the prompt to “Please say Health to know about health related information, or say Agriculture to ...”. Although these descriptive prompts are too lengthy and not so natural, but this community preferred to listen to prompts that clearly tell the user what to say. This was a very good insight from these participatory design experiments. Users in village have more patience and prefer simple descriptive system prompts as compared to more natural sounding prompts. Perhaps the

Fig. 5.

Initial prototype testing with the farmers.

VI. U SABILITY

OF

VOI K IOSK

The VoiKiosk was tested with the villagers for a period of four months. We present the usage statistics of the VoiKiosk system by analysing the data through two approaches. The first approach presents the analysis by looking at the data as a

62 TABLE II

whole. This analysis is presented to demonstrate the increasing acceptability of the VoiKiosk system in the village. In the second approach, we present the results on the backdrop of the changes that were made in the system over time. This will provide more details about the effects of the usability of the system with improved interactions. In this section, we will present the two results in more detail.

B. Overall System Usage The VoiKiosk system was live, 24 hours a day and all seven days of the week for four months. In these four months, the system received a total of 20499 calls from 976 villagers. We assume that a unique phone number would map to a specific villager because the trace to users is based on the caller identification in the VoiKiosk system. Table II shows the number of calls that went to the different services in the VoiKiosk. The advertisements section of the VoiKiosk was accessed maximum number of times. The third column in this paper shows the percent of the total number of calls that access this service. Since this number does not add to 100, not every call accesses a service. Users spent a total of 477 hours on the VoiKiosk system. The average call time was 83 seconds, with a maximum of 49 minutes and minimum of zero seconds. The number of calls have been increasing since the pilot was launched. The graph in Figure 6 hints at the increased acceptability of the VoiKiosk system over time. On an average,

Number 1640 2383 1749 7492

Percentage 8% 11.6% 8.5% 36.5%

while there used to be about 50 calls per day at the start of the pilot, the last week of the pilot has witnessed more than 300 calls every day. The interesting thing to note is that the NGO did not actively train the villagers to use the system nor did they advertise this service aggressively. We had started with a group of about 30 users who were initially contacted to start using the system and provide us feedback. The caller base increased from these 30 users just by word of mouth and has reached even the neighbouring villages. We have had villagers from the neighboring villages calling and creating their advertisements on the VoiKiosk. 700

600

Number of calls per day

The pilot was deployed for villagers in the Juvvala Palam village in South India. This village has a population of about 4000 people that form about 850 families. About 70% of these families have a cell-phone. The male/female ratio is 100/94 in this village. The main occupation of this village is agriculture, especially paddy crop. Transportation is the main business for this village. People rent their trucks, tractors and smaller vehicles for a few days to people in nearby villages. These are used by farmers in their field and for a specific social function such as a wedding. The village a health center where the doctor visits for two hours every day. There is one bank in the village where about 10% of the families have a bank account. The average monthly income for a family is roughly about 2000 rupees (USD 50). All cellphone users are comfortable dialing numbers and talking with other people. However the use of text-message is very low. As a rough estimate, only about 10% of villagers use text-message for communication on cellphone. This is due to the fact that the local language (Telugu) is not supported on the device and since only about 20% of people know very limited English, text-messaging is almost non-existent. The village has a secondary school and a primary school. The NGO facilitates English and Math classes through a distance education program in these schools. The NGO also facilitates distance education of MS Office, spoken English, embroidery and painting classes for adults. Villagers from the Juvvala Palam village participated in the pilot.

Description Agriculture Health Ashwini Center Advertisements

500

400

300

200

100

0 1

6

11

16

21

26

31

36

41

46

51

56

61

66

71

76

81

86

91

96 101 106 111 116

Days since the start of pilot

Fig. 6.

Number of calls per day, starting from the launch of pilot.

While most calls were during the afternoon and late evening times, we were surprised to see calls arriving at 2:00 am. There were a total of 99 calls received between odd hours of 11:00 pm and 5:00 am. Figure 7 shows the calling pattern during different times of the day. This is an aggregate number summed over the number of calls in the four months. 2500

2000

Number of calls per hour

A. User Background and Village Demographics

S UMMARY OF DIFFERENT SECTIONS VISITED BY USERS .

1500

1000

500

0 1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

24 Hours of day (starting from 12:00 midnight)

Fig. 7.

Hourly breakup of number of calls in the day.

22

23

24

63

There were a total of 2528 advertisements recorded by 463 different users. Many people recorded their advertisement multiple times over the period of four months. Although one person can record only one advertisement, people often call again either to improve their advertisement, or to record a fresh advertisement. One caller had recorded his ad 62 times. Though most callers had recorded it only once. The graph in Figure 8 shows the different callers and the number of times they have recorded their advertisements. 70

Number of advertisement recordings

60

50

40

30

20

10

0 1

18 35 52 69 86 103 120 137 154 171 188 205 222 239 256 273 290 307 324 341 358 375 392 409 426 443 460 Unique callers

Fig. 8.

Advertisement update frequency of callers.

C. Improvements with Changes in Interface During the four months of the live pilot, we made two changes to the system: 1) Enabled easy navigation of advertisements by providing options to jump to next advertisement through keypad input. 2) Enabled easy recording of advertisement by allowing users to record without them having to listen to any advertisement. The initial system was live for 70 days. After the first change, the system was live for 21 days and then after the second change, it was live for 30 days. Table III shows the number of calls, average call time, number of advertisements and unique callers in the three situations. TABLE III C HANGE IN V OI K IOSK USAGE WITH SYSTEM CHANGE . I NITIAL S YSTEM IS THE FIRST VERSION THAT WAS DEPLOYED . M ODIFICATION -1 IS THE SYSTEM THAT ENABLED NAVIGATION OF ADVERTISEMENTS . M ODIFICATION -2 ALLOWED EASY RECORDING OF ADVERTISEMENTS . System State Initial System Modification-1 Modification-2

No. of calls 6239 4437 9820

Average time 79 104 75

Unique callers 335 227 412

It is interesting to note that over a period of these four months, the number of advertisements have been increasing. Even then, due to improved navigation, the time spent by users has not increased significantly. In the first deployment of the system, a caller had to listen to all advertisements before she could post her own advertisement. The reason

behind this was that by listening to the advertisements, a novice user will know what to speak in her advertisement. When the number of advertisements increased beyond 20, we noticed that the users were still patiently listening to them and then would post their advertisement. This usage pattern provides (at least) one of the following two insights about the users: This population is more patient and could listen to a long list of information over phone as compared to what has been observed with the western population’s experience with spoken dialog systems. And/OR The villagers ascertain certain value to posting an advertisement and so are ready to wait for a long time in order to get a chance to post. When the number of advertisements increased beyond about 40, we provided option to jump to the next advertisement by pressing a key on the phone keypad. Perhaps the users were comfortable with navigating the VoiKiosk system by then, and therefore they were able to learn the keypad navigation. It should be noted at this point that during the needs-gathering phase, the users had clearly mentioned that navigation through keypad will not be easy and so a speechinput method should be provided to navigate the VoiKiosk. Even then, the average time spent per-call was higher (104 seconds) after Modification-1, it reduced to 75 seconds when we provided option to record without users having to listen to all advertisements. This is a clear insight that users were able to interact more efficiently with the VoiKiosk system when we provided keypad-based shortcuts for faster navigation. VII. D ISCUSSIONS The advertisement section attracted the most attention, and was accessed the most. Initially, we had a couple of users (a mechanic and a truck owner) upload personal (classifiedtype) advertisements to the VoiKiosk. Inspite of these initial straightforward examples, the villagers soon found some very interesting and innovative uses for this section: • An eighth grade student announced himself and his mobile number. • A man uploaded his profile for matrimonial purposes (to invite marriage proposals). He made several attempts till he was satisfied that all the details he wanted to capture were recorded. – Another person created a ”response advertisement” commenting on the above profile! • A pair of young parents recorded a message in their child’s voice for the child’s grandparents to hear. • A politician posted a thank you note after winning a local election. These examples are very demonstrative of several things: (a) The villagers really understood the technology enough to play with it, and find innovative uses for it, and (b) Even

64

though they may have never heard of social networking on the Web, the need for social networking is natural and compelling, and (c) they independently discovered the message/responseto-message interface as well as voice mail (as obvious as these might sound to us). Thus the advertisements section evolved to a message-board by innovative use of the interface by the villagers. As was mentioned in the previous section, in an earlier incarnation of the VoiKiosk, a villager had to listen to all the advertisements before posting his own, but that did not seem to act as a deterrent. The villagers appear to be extremely patient in listening to all the information and spending a lot of time with the system, which may be a reflection of culture or the lack of alternatives, most likely both. Of the 20499 calls received on the VoiKiosk system, only 2532 were for creating the content. Creating the content in the VoiKiosk is either by recording of an advertisement by a villager or by recording of other messages by the Kiosk Operator or the agriculture Expert. Therefore, more than 87% of the calls were for accesing the content on the VioKiosk system while only about 13% calls were made for creating the content. This indicates the usefulness of the content that was being created on the VoiceSite. In the previous section, we have provided more details about the advertisements rather than the agriculture, health and ashwini center. This is because the other three categories were mostly information upload from one specific person (either the Kiosk Operator or the Expert), and the access was to the villagers. Therefore these services, although important, were different from the advertisements because the latter had user generated content. By providing the ability for all villagers to create content, VoiKiosk becomes a much more participatory platform for rural areas. Although the statistical analysis proves the acceptance of the VoiKiosk system by the villagers, the next step is to get usability feedback directly from the users. We plan to talk to a sample of VoiKiosk users and seek their feedback in terms of the usability of the system and the value of the content. We also plan to provide more critical services such as daily crop prices to registered farmers and medication reminders to patients through the VoiKiosk system. It will also be interesting to observe the usage pattern of a different village and identify the similarities and differences and question whether they reflect the demographics of that village. VIII. C ONCLUSION In this paper, we present a information kiosk system for use in rural areas to create and access locally relevant content. We identify the information needs in rural areas by talking to villagers and the NGO. We develop VoiKiosk – a voice-based system that can be accessed by a phone to provide a cost effective solution that is easy to use by the less-literate people in rural areas. We test the initial prototype with the villagers to evolve a participatory design of the final system. The system was then deployed live in a village in South India and we captured user statistics for 4 months. More than 900 users used

the system over a period of four months. Users made about 20000 calls to this VoiKiosk system. We present the usage pattern obtained by different modifications in the VoiKiosk over these four months. We provide a detailed description of how some of the services were used beyond what they were designed for. This leads to an interesting insight into the social networking applicability of the VoiKiosk system. The increasing use of the VoiKiosk system for the different purposes leads us to believe that a voice-based mechanism for local content creation is a very powerful interaction modality to provide information and communication technologies in rural areas. ACKNOWLEDGMENT We thank Byrraju Foundation for supporting VoiKiosk in Andhra Pradesh villages and for providing valuable insights. R EFERENCES [1] “World Urbanization Prospects: The 2007 Revision Population Database,” http://esa.un.org/unup/.

[2] Ministry of Finance, Government of India, “Economic Survey 2001-2002,” http://indiabudget.nic.in/es200102/chapt2002/chap106.pdf. [3] Population Reference Bureau, “World Population Data Sheet,” no. ISSN 0085-8315, Aug 2006. [4] S. Agarwal, A. Kumar, A. A. Nanavati, and N. Rajput, “Voikiosk: Increasing reachability of kiosks in developing regions,” in Proc. Intl. Conf. on World Wide Web (WWW), China, April 2008. [5] T. S. Parikh and E. D. Lazowska, “Designing an Architecture for Delivering Mobile Information Services to the Rural Developing World,” in Proc. Intl. Conf. on World Wide Web (WWW), May 2006. [6] S. Seshagiri, A. Sagar, and D. Joshi, “Connecting the bottom of the pyramid: An exploratory case study of india’s rural communication environment,” in Proc. Intl. Conf. on World Wide Web (WWW), China, April 2007. [7] B. Kolko, E. Rose, and E. Johnson, “Communication as information-seeking: The case for mobile social software for developing regions,” in Proc. Intl. Conf. on World Wide Web (WWW), Canada, May 2007. [8] D. Ramachandran, M. Kam, J. Chiu, J. Canny, and J. L. Frankel, “Social dynamics of early stage co-design in developing regions,” in Proc. CHI, USA, April-May 2007. [9] C. K. Prahalad, “The Fortune at the Bottom of the Pyramid: Eradicating Poverty Through Profits,” in Wharton School Publishing, 2004. [10] E. Brynjolfsson and L. Hitt., “Paradox lost: Firm-level evidence on the returns to information systems spending.” Management Science, vol. 42, April 1996. [11] “Netcraft Web Server Survey,” http://news.netcraft.com/archives/web server survey.html. [12] K. Matthee, G. Mweemba, A. Pais, G. V. Stam, and M. Rijken, “Bringing internet connectivity to rural zambia using a collaborative approach,” in Proc. Intl. Conf. on Information and Communication Technologies and Development (WWW), India, December 2007. [13] U. Saif, A. L. Chudhary, S. Butt, N. F. Butt, and G. Murtaza, “Internet for the developing world: Offline internet access at modem-speed dialup connections,” in Proc. Intl. Conf. on Information and Communication Technologies and Development (ICTD), India, December 2007.

65

[14] G. Buchanan, S. Farrant, M. Jones, H. Thimbleby, G. Marsden, and M. Pazzani, “Improving mobile internet usability,” in Proc. Intl. Conf. on World Wide Web (WWW), Hong Kong, May 2001. [15] R. Heeks, “Information Systems and Developing Countries: Failure, Success, and Local Improvisations,” The Information Society, vol. 18, March 2002. [16] E. R. Megwa, “Bridging the Digital Divide: Community Radio’s Potential for Extending Information and Communication Technology Benefits to Poor Rural Communities in South Africa,” Howard Journal of Communications, vol. 18, October 2007. [17] F. Hussain and R. Tongia, “Community radio for development in south asia: A sustainability study,” in Proc. Intl. Conf. on Information and Communication Technologies and Development (WWW), India, December 2007. [18] S. R. Sterling, J. O’Brien, and J. K. Bennett, “Advancement through Interactive Radio,” in Proc. Intl. Conf. on Information and Communication Technologies and Development (WWW), India, December 2007. [19] T. S. Parikh, “Designing an Architecture for Delivering Mobile Information Services to the Rural Developing World,” Ph.D. Dissertation, University of Washington, 2007. [20] T. S. Parikh, P. Javid, S. K., K. Ghosh, and K. Toyama, “Mobile Phones and Paper Documents: Evaluating a New Approach for Capturing Microfinance Data in Rural India,” in Proc. CHI, Canada, April 2006. [21] Y. Schwartzman and T. S. Parikh, “Using CAM-equipped Mobile Phones for Procurement and Quality Control at a Rural Coffee Cooperative,” in MobEA V, Mobile Web in the Developing World, Canada, May 2007. [22] M. Plauche and M. Prabaker, “Tamil market: a spoken dialog system for rural India,” in Proc. CHI, Canada, April 2006.

[23] J. Sherwani, N. Ali, S. Mirza, A. Fatma, Y. Memon, M. Karim, R. Tongia, and R. Rosenfeld, “Healthline: Speech-based access to health information by low-literate users,” in Proc. Intl. Conf. on Information and Communication Technologies and Development (WWW), India, December 2007. [24] R. Veeraraghavan, N. Yasodhar, and K. Toyama, “Warana Unwired: Replacing PCs with Mobile Phones in a Rural Sugarcane Cooperative,” in Proc. Intl. Conf. on Information and Communication Technologies and Development (WWW), India, December 2007. [25] “MobilED Initiative,” http://mobiled.uiah.fi/. [26] T. Leinonen, F. Aucamp, and E. Sari, “Audio Wiki for mobile communities: information system for the rest of Us,” in Workshop on speech in mobile and pervasive environments, Mobile HCI 06, September 2006. [27] A. Kumar, N. Rajput, D. Chakraborty, S. Agarwal, and A. A. Nanavati, “Voiserv: Creation and delivery of converged services through voice for emerging economies,” in WoWMoM’07 Proceedings of the 2007 International Symposium on a World of Wireless, Mobile and Multimedia Networks, Finland, June 2007. [28] S. Agarwal, D. Chakraborty, A. Kumar, A. A. Nanavati, and N. Rajput, “HSTP: Hyperspeech Transfer Protocol,” in ACM Hypertext 2007, UK, September 2007. [29] A. Kumar, N. Rajput, S. Agarwal, D. Chakraborty, and A. A. Nanavati, “Removed for blind review,” in Proceedings of the World Wide Web, April 2008. [30] A. Kumar, N. Rajput, D. Chakraborty, S. Agarwal, and A. Nanavati, “Removed for blind review,” in SIGCOMM Workshop on Networked Systems for Developing Regions, Japan, Nov 2007. [31] “Byrraju Foundation,” http://www.byrrajufoundation.org.

66

E for Express1: “Seeing” the Indian State through ICTD Renee Kuriyan and Isha Ray

Abstract— This paper examines how, in their attempts to liberalize and modernize their operations, Indian states are using ICTD e-governance services to represent themselves in a new way to their citizens. It reveals how states come to be seen by their citizens through their everyday interactions at ICTD telecenters. The research finds that, with its e-governance services, the state is trying to recast its image to fit marketfriendly principles such as economic efficiency, accountability and effectiveness. Citizens simultaneously trust the government as credible and are disillusioned with it as inefficient. Telecenterprovided e-governance services are partially re-shaping the boundaries between state, civil society and markets. Index Terms—e-governance, partnerships

India,

state,

public-private

I. INTRODUCTION “Paying an electricity bill [in India] could easily involve a day’s wait at a government office where a cross official would demand a bribe for doing his job. The same was true for phone bills, water bills, taxes and all other interactions with government. Often the customer would first have to go to a bank to get a banker’s draft and then take it to a queue at the payment office. Even a small firm would need an employee whose sole task was to pay bills and deal with other aspects of officialdom. Now all of this can be done online.” (The Economist, Special report on technology and government, February 14, 2008) The Economist report cited here, entitled the “Electronic Bureaucrat,” highlighted India as a nation whose government was on the path to technologically “leapfrog the rich world” by putting its services online. It argued that the Indian government was able to serve its citizens more effectively through its pioneering efforts in e-governance, with online This material is based upon work supported by the National Science Foundation under Grant No. 0326582, the Technology and Infrastructure for Emerging Regions group, Microsoft Research India, and Intel Research. Renee Kuriyan is a Research Scientist with Intel Research, Beaverton, OR, USA (phone: 503 329 0688; e-mail: [email protected]). Isha Ray is an Assistant Professor with the University of California, Berkeley, Energy and Resources Group (e-mail: [email protected]). 1 The Economist, Special report on technology and government, February 14 2008

electricity bill payment and issuance of government certificates and forms. The article drew attention to Andhra Pradesh (AP) in particular as the “hot-bed” of e-governance since it started e-Seva 2 , a project which was upheld as a model of efficient government service delivery [1]. The cartoon accompanying the article, entitled “E for Express” symbolized this new form of service delivery and the accompanying changes in the Indian bureaucracy (Figure 1). It depicted a sleek Indian express train carrying passengers on top of its roof and inside its cars, holding laptops and cell phones. The Indian railways were once emblematic of the country’s bureaucratic and slowmoving ways. At the same time, often associated with Gandhi traveling on their rooftops, the always-overflowing railways were symbols of accessibility to the Indian masses. The passengers in this cartoon, however, wearing expensive business suits and holding laptops, were a far cry from usual images of the “common man.” The cartoon also highlighted the links between modern technologies and an Indian bureaucracy that was changing from being slow and “neither equitable nor efficient” 3 , to one that is fast-moving, capable, and accessible to all. It hinted that with the spread of egovernance services and state of the art technologies the public’s experience of the government was being transformed.

Figure 1: E-governance in India, “E for Express” (Source: Economist, February 14, 2008)

2 E-seva was launched in Hyderabad in 1999 with goals of ”looking at 'service' from the citizens' point of view” and redefining citizen services using state-of-the-art technologies. It is a public private partnership between the Government of AP and private service providers. It provides services such as: payment of utilities bills, government certificates, licenses, permits, transportation department services, bus reservations, passport services, and business services based on agreements with private businesses such as cell phone providers and banks (http://esevaonline.com/). 3 The Economist, Special report on technology and government, 2008

67

This paper examines the implied hypothesis that, through Information and Communication Technology (ICT) projects, and in particular those that provide e-governance services, the nature and boundaries of the Indian state are being redefined. We do this through an analysis of telecenters that provide egovernance (and other) services in two states in India: Kerala and AP. Both states have been enthusiastic adopters of such telecenters, and both states have formed public-private partnerships (PPPs) in order to promote their e-governance agendas. We ask: Do citizens “see the state” [2] differently in light of governance and development services provided through these entrepreneur-mediated telecenters? The term “see the state” is a twist on the pioneering work of James Scott, who, in Seeing Like a State (1998), examines the ways in which the state sees (and thus controls) its citizens. He argues that the state often attempts to make its populations “legible” through simplified, yet strategic and technical, processes (such as mapping, censuses, and various other standardized modes of representing the population). Conversely, Corbridge et al in Seeing the State (2005) examine how citizens see government agencies, through an ethnographic analysis of state-citizen interactions in eastern India. We build on Corbridge et al to further explore the shifting dynamics of state-citizen relationships through ICTD telecenters in southern India. We begin by highlighting some of the important ideas in the literature on governance, e-governance and their associated reforms in India. Then, using two vantage points -- the view of the state and view of the citizens -- we argue that in both Kerala and AP, the state governments are turning to the private sector as partners not only for the provision of egovernance services, but to associate themselves with a liberal market order and the modernization process more broadly. The state is trying to recast its image to fit market-friendly principles such as economic efficiency, accountability and effectiveness, all of which embody the “good governance” agenda in India and elsewhere. These efforts reflect the state’s attempt to reposition itself in the context of a liberalizing economy and to alter the way in which it is perceived by its citizens. We find that citizens simultaneously trust the government as credible and are disillusioned with it as inefficient. We find that e-governance through decentralized entrepreneur-mediated telecenters are partially redefining perceptions and expectations of the state, the lay citizenry and the private sector. This hybrid version of government is gradually reworking both the way the state sees itself and how citizens see the state. II. METHODS Using a combination of methods such as interviews, participant observation, and literature and document review, we explored the reworking of these relationships through the case of ICT for development (ICTD) telecenters. Primary data collection took place on several trips to India over a period of

3 years from 2004 to 2006. We examined four projects in the states of Kerala and Andhra Pradesh, all of which provided egovernance services to citizens via entrepreneur-mediated telecenters. These were Akshaya in Kerala, and three separate projects in AP called Rural Eseva, Rural Service Delivery Points (RSDP), and Rajiv Internet Village Centers (Rajiv). The telecenter projects offered a glimpse into the realities and perceptions of good governance principles and government policies on ICTD. The telecenters in AP and Kerala focused on e-governance services that 1) provided entitlements (including certificates, licenses), 2) provided information (sectoral, agricultural, or health), 3) provided redress for grievances, and 4) enabled government bill payments (electricity, taxes, utilities). In Kerala the telecenter project offered computer education services as well. We conducted 31 interviews with state actors within the Government of India (GOI) and within the AP and Kerala state governments, using a semi-structured interview protocol. These interviews explored each state’s strategies for delivering e-governance services, and how states perceived their roles and those of the private sector in the process. We conducted open ended, key-informant interviews with 16 local entrepreneurs in Kerala and AP from the four different projects. Through these interviews and shorter conversations with other entrepreneurs, we explored the role of entrepreneurs in delivering governance services and the position of the entrepreneur with respect to the state. In the telecenters of each project, we observed entrepreneurial behavior, and engaged in informal conversations with the users. We interviewed telecenter users and non-users in each district, using an open-ended interview protocol, exploring perceptions of both the project and the state with respect to their roles in development. Sixty-five interviews with household members were conducted in Kerala, and 70 in AP. All interviews were recorded, transcribed and coded using interview analysis software. Finally, we analyzed literature and policy documents on good governance, ICTD, and the modern Indian state. III. GOVERNANCE REFORMS AND E-GOVERNANCE IN INDIA The expression “governance” is ambiguous and has become a catch-all term in the study and practice of development. The academic literature on governance is varied, containing perspectives from institutional economics [3], sociology [4, 5], development studies [6] and governmentality-oriented theory [7]. From a purely technocratic perspective, governance can be understood as a set of administrative or managerial tools, which, properly applied, lead to “good governance.” Administrative reforms of Western governments in the 1980s and 1990s, implemented under the marketoriented framework of new public management (NPM,[8, 9] underscored concepts such as efficiency, open markets, accountability, customer service and decentralization. NPM models indicated that market-based mechanisms for service delivery were more competitive and therefore more efficient than traditional government-based provision. The overall philosophy was that governments should be encouraged to perform like incentive-driven private businesses and

68

entrepreneurs. Similar principles were later invoked in “good governance” reform programs in developing countries. Development practitioners in recent years have defined good governance principles as transparent policymaking, professionalism and accountability in government actions, and civil society participation. International development agencies have sought to operationalize good governance by restructuring and privatizing state bureaucracies, decentralizing state power through local government or nonstate actors, reforming legal systems, and implementing public private partnerships [10]. The rationale behind these reforms was that by “combating corruption, nepotism, bureaucracy, and mismanagement … aid would be effectively used to achieve the objective of reducing poverty” [11: 270]. Critics of good governance argued that it had a neoliberal bias towards market-led development due to the influence of donors. Critics also charged that the words ‘governance’ and “government” were being used interchangeably because: “Governance appears to be used in place of government as if “government” was a difficult word to sell in a privatized, market-oriented society. Governance is about a ‘reinvented’ form of government, which is better managed” [12: 18]. In India, the good governance agenda developed against the background of a long history of concern over state-society relationships rooting back to India’s anti-colonial struggles [2]. Debates about good government with respect to appropriate power for India’s citizens, corruption, and abuse of executive powers have been part of the national agenda from the 1960s. The reforms of the 1990s, which liberalized the economy to a great extent, were introduced largely to address India’s balance of payments problem [2], But a broader program of administrative and economic reforms was also supposed to tackle political challenges, or what Kohli called India’s “growing crisis of governability” [13: 23]. By this time, the centralized state had lost a great deal of legitimacy, and devolution through administrative reforms promised a range of benefits [14]. Prime Ministers from the 1990s onwards turned to a mix of deregulation, privatization, civil service reform, decentralization and PPPs to address a range of administrative “failures” in India. Decentralization in particular was an integral part of the governance reforms and was viewed as a way to reduce the role of the state by fragmenting its authority and making it more responsive and efficient. Local governments were supposed to have better information on local needs, and were considered closer to their constituencies and thus more politically accountable, than centralized government agencies. Decentralization would thus “expand service deliveries as authority goes to those more responsive to user needs”[15: 173]. However, even local government institutions could lack accountability mechanisms and be vulnerable to corruption or to being captured by local elites. This frequently frustrated the goals of equitable public provision of services to the general population (Ibid). In addition to decentralization, therefore, the administrative reforms included PPPs in service delivery. Since Indian states were often in fiscal trouble, with limited budgets to deliver government services, they also needed private sector partners

for help with financial contributions and the modernization process overall. During our research, several government officials talked about the financial constraints of the government and its inability to scale service delivery without private sector participation. For example: “The government is not having many funds. The private sector has funds and if the government supports [them] they can invest. In Kerala, [the] government is running out of money. We go for heavy loans from the Asian Development Bank... For the last ten years it’s been like that… There is a lot of work [to be done], but not much money.” (Interview, IT Mission, Kerala) Officials in Kerala indicated that inadequate technical capacity was another reason why the government needed to move towards a PPP model for service delivery: “There was no capacity with the government. That had to be built up and that is not something you can do overnight. In e-governance you find a much higher acceptance of the private sector as a player. The government acknowledges that they don’t have the technical or financial capacity in the government and they need to look to the private sector” These constraints and incentives led several Indian states to introduce decentralized e-governance delivery projects through PPPs. E-governance measures aimed to improve administrative processes by using ICTs, and to build connections to promote socioeconomic activity [16]. In accordance with the principles of good governance, egovernance was meant to emulate the private sector’s qualities of reliability, transparency, scalability and treating citizens more like customers. The mainstream press was generally supportive of the promise of fewer lines, administrative efficiency and the electronic bureaucrat (Figure 2). For example, the Indian Express newspaper in a 2005 article credited e-governance with bringing citizens to the center of service delivery: “E-governance has, in many cases, restored the choice to the citizen as to the quality and adequacy of services he is entitled to expect from public organizations. Citizen-centric governance meant government was for the people and the services were tailored to meet their requirements. 4 ”

44 “Giving to people their new right: The right to good governance” (http://www.indianexpress.com/india-news/full_story.php?content_id=76910)

69

Figure 2: The Electronic Bureaucrat (Source: Economist, February 14, 2008)

The focus on e-governance led many state governments to adopt “integrated citizen service portals” or ICTD telecenters. Telecenters (or kiosks) were meant to bring government and other services closer to citizens and businesses through partnerships with local entrepreneurs [17]. Telecenter entrepreneurs had market-based incentives to perform efficiently; were situated at a village level; and in most cases were members of the communities that they served. Consequently, it was felt, they were best suited to make egovernance services responsive to the felt needs of the local population. Despite the Indian states’ embrace of e-governance through telecenters, there have been a number of critiques of these projects. It has been argued that telecenter services were not reaching the poor [18], were unsustainable [19, 20], and faced political challenges in actually delivering e-governance services [5]. We now examine the influence of this new form of governance from the perspectives of the state and of the citizens that these initiatives are meant to serve. IV. CONSTRUCTING THE STATE AND CHANGING ITS IMAGES Our research aims to understand how the state represents itself to its citizens and how it comes to be seen by them through their everyday interactions at ICT telecenters. Several works of political anthropology depart from the conventional treatment of the state as an institution defined by its powers and operations, and instead examine the state as it is ‘imagined’, constructed and conceptualized through the experiences of its citizens [21-24]. These works analyze how the state becomes “socially effective through particular imaginative and symbolic devices”[25: 981]. Gupta (1995) argues that through ethnographic research on the state, for example with lower level officials and politicians, it is possible to “illuminate the quotidian practices of bureaucrats that tell us about the effects of the state” on the lives of ordinary people [24: 376]. What quotidian practices “construct” the state? Corbridge et al. (2005) argue that state-citizen interactions are based on

everyday flows of power, money, commodities and information. The poor in rural India form their understandings of the state through financial and legal transactions at local government offices [2]. They note that a poor person most directly experiences the state when he or she registers for birth or death certificates, receives a registration form, or picks up an entitlement. These are many of the e-governance services now offered by entrepreneurs in telecenters. With telecenter entrepreneurs in effect replacing the functions of local bureaucrats, we examined the ways in which the state is being “seen” or experienced by different groups of people accessing e-governance services through ICT telecenters. Our research indicates that at least some segments of the rural and periurban population have developed new images of the state through their encounters with telecenter entrepreneurs. They evaluate these experiences against their previous and often negative encounters with the local arms of the state. We argue that both Kerala and AP, through changing their governance initiatives, are deliberately re-working images of the state versus the private sector. The growing acceptance of service delivery through ICTs, the political reality of economic liberalization policies, and discourses of India as a technological leader in e-governance have all influenced the types of images the state tries to portray to its citizens. We find that how the state is then perceived by civil society depends on a combination of factors -- the ambience of telecenters compared to government offices; how entrepreneurs interact with the public; how the state brands the telecenters to citizens (whether it is considered a private or a government office); and the extent to which citizens both trust, and are disillusioned with, the government. Our observations in Kerala and AP support neither the strong proponents of the good governance agenda who advocate a minimized role for the state, or the strong opponents who fear that service through PPPs will, in effect, ‘privatize’ the state. We find that, rather than removing the state and supplanting it with the private sector, telecenter projects under the PPP model have created a space for the state to construct a better image of itself with respect to its citizens. These telecenters and their electronic delivery of services are being used as, and to some extent are becoming, symbols of responsiveness and of accessibility to all. The partnering of the state with private entrepreneurs is a key mechanism through which the state is trying to recast its image and be re-imagined by its citizens. A. Representations of the State in Kerala and AP Throughout India, at e-governance and telecenter conferences, and in our discussions with state officials at the central government and AP and Kerala state government levels, we found a similar discourse about the state’s need to change the way it treats, and how it was perceived by, citizens. There was a general consensus among government officials that the state needed to behave more like the private sector in service delivery. Officials in Kerala and AP indicated that one way to accomplish the change was to partner with private entrepreneurs who had an economic incentive to provide good customer service.

70

Entrepreneur-run telecenters represented a new concept of government service delivery that was fast, efficient, hassle free, and accessible to the “common man” (Figure 3). These attributes allowed states to renegotiate their standing with their citizens, a standing that was clearly low on account of previous encounters between the state and its citizens. A former AP official was clear on the expected differences: “The appearance is totally unlike a government office. The ambience created is more of a private sector feelingit has a uniform ambience built across all the service centers across the state- same design, same color. You have a feeling that you have not walked into a government office, but you’ve walked into something the private service sector would run. The beauty is that it starts at 8 in the morning and closes at 8 at night.”

Figure 3: Old versus new images associated with government offices and services. (Photographs shown in presentation at National e-governance conference in Kochi (2006) to show old and new forms of government service delivery)

Frequent reference to the telecenters’ appearance or ambience revealed that state employees were at least as concerned with their image as they were with making government services more streamlined. Several officials acknowledged that government offices had historically been very different in appearance than something that the private sector would run. They recognized that, unlike government offices in which officials often closed early, were not present at their offices, and were on the whole unaccountable, these new telecenters had much more responsiveness and flexibility. State employees explained that with telecenters, they were aiming for transparency, speed, and convenience for the citizen, and that wanted the government to be seen as more “common man friendly”. Our research showed that the business-like appearance of the telecenter actually represented a type of streamlining to users and also to the officials. Both states emphasized the business-like aspect of these telecenter projects, a feature that is at the core of the good governance agenda. State actors, especially in AP, insisted that the credibility of the telecenter and e-governance projects could only be achieved when they were managed by the private sector because of their good services, longer hours, lack of resource constraints and flexibility. But both state governments wanted to make sure that the government’s name and brand was associated with these “private” telecenters. By having the private sector deliver the services but branding the centers with the government name, the state had found a concrete way to represent itself as protecting the public good and doing so in a business-like manner.

Our conclusion that both AP and Kerala were quite concerned about their public images is consistent with Madon’s (2005) study of image formation in the state of Kerala. She found that, given the historical resistance to and suspicion of the private sector in Kerala, the government was careful to manage the image of the Akshaya project. Rather than advertising the Akshaya project to the public as a PPP at first, it initially portrayed the project as social development bringing e-literacy to all. After a few years the government began to emphasize the private sector’s role in delivering services to citizens through state-entrepreneur partnerships. Madon attributed this change to the growing optimism and acceptance of IT in governance reforms and increasing confidence in the private sector in Kerala. The Kerala government in this case judiciously managed the image of the role of public-private partnerships and governance reform. Despite their acceptance of private sector participation, state officials in Kerala and AP asserted that the credibility of the government brand was highly important for the PPP-based implementation of government services. One official we interviewed in AP stated, “The image is that these private telecenters are the government. If you look at the transactions- there are millions a month. Probably 99% are government to citizen transactions… they are branded as government. That is why the credibility is much better. Because people will think twice if somebody else [who is private] wants to collect your electricity bill and pay taxes. People will think twice!” (Interview, 2006). State officials asserted that despite its reputation for poor quality and slow services, some level of government association with the entrepreneurs created credibility for the telecenters. The images of the state were constantly being constructed and renegotiated not only by state and project officials, but by citizens themselves, as we show in the next section. B. “Seeing the state” through entrepreneurs Citizens “see the state” through the individuals who represent it. Because the personal characteristics of key individuals, such as a pro-poor officer or a corrupt official, often appear large in the minds of citizens who have experienced them, the views of government are likely to be fractured [2] The local bureaucrat has traditionally embodied the “state” in India and the state is then “seen” through the everyday interactions between ordinary people and this person-cum-state: “The manner in which these officials negotiate the tensions inherent in their location in their daily practices both helps to create certain representations of the state and powerfully shape assessments of it, thereby affecting its legitimacy.” [24:388] Corbridge et al argue that “encounters with the developmental state build up a dynamic picture of “it”, both as an idealized set of values and practices (the state as it should work) and also as its flawed but more commonly experienced

71

counterpart (the state as it does work)” (2005: 119). Today, the local bureaucrat is no longer the exclusive embodiment of the state. When accessing ICT telecenters, citizens’ everyday interactions with the state are now being mediated by smallscale entrepreneurs as they, and not local officials, provide a host of government services to the populace 5 . These entrepreneurs, even though they are not state employees, come to represent the state at least in part, and also affect its legitimacy. During our research many entrepreneurs emphasized their ability to deliver government services through the PPP strategy better than the government could deliver them. However, they relied on being “seen” as the government, particularly when they collected electricity bill payments or issued government certificates. Entrepreneurs confirmed what state officials had claimed, that being thought of as the government or having some government endorsement gives people the confidence to pay their bills at the telecenters. The association with the government also prevented people from thinking the telecenters were corrupt. An entrepreneur in AP admitted: “If people think we are working under private people they won’t come. I think and tell villagers that we got a loan from the government to establish this center. So it is a government center. For the most part, entrepreneurs providing e-government services to citizens benefited from the association with the government’s name. Entrepreneurs faced challenges, however, branding their centers as “government” institutions when they provided more than just e-governance services to citizens, such as computer education courses[19]. Rajalekshmi’s study of the Akshaya project also found that the most important reason for its acceptance was “the trust that people had in government as an institution and the fact that this project was spearheaded by the government” [17: 29]. Our study in both AP and Kerala corroborated this observation; citizens’ trust in government institutions allowed these privately-run payment and other e-governance systems to function. C. Citizens’ perspectives We now turn to the perspectives of the households. Several of our respondents said that the private sector, meaning the telecenter entrepreneurs, treated all customers with a level of respect that they did not receive from the government. Our interviews with households revealed that ordinary citizens, especially the poorer ones, resented the bureaucratic and rude manner in which government officials often treated them. One woman, living below the poverty line, complained: 5 Undoubtedly, the range of people’s encounters with the state extends much beyond their interactions with telecenter entrepreneurs. Nor do we wish to imply that telecenter entrepreneurs are the only intermediaries between the state and its citizens: field-level government employees and NGO representatives, for example, also play this role. But telecenter entrepreneurs are a novel form of intermediary in that they are simultaneously the state and not the state, at least when they are providing government sanctioned egovernance services.

“At government offices, people are not helping me. Even when we go to a government office, they won’t help us people like us. Even if you have a small job, they will not give any amenities, like ration card. They will have to give, but they won’t give it. We have to do everything on our own. Here [at the telecenter] there is more respect for people. “ While poorer citizens seemed concern with the need for more respect from the government, middle-class individuals complained that traditional service delivery methods were slow and inefficient. A middle-income elderly man described going to a government office compared to a private telecenter to pay his electricity bills: “At a Government office, a person has to wait 5-6 hours. A person uses Rural e-seva now. Before they had to travel for many hours to many offices. [Now] instead of 4 hours, I use 10 minutes. We use [Rural] e-seva for electricity bills and telephone.” Speed and convenience were not the only advantages offered by the telecenters. Household members seemed to think that dealing with private providers was easier overall compared to dealing with employees at a government office. In the words of a middle aged man: “It is easier to deal with these private people than the government - if you have 2-3 private centers, definitely the private person will grab people by providing offers, private institutions, providing amenities. They give you water. You won’t find water in a government office. They [the private centers] provide good amenities to you.” In this sense, it seemed that the government had outsourced the provision not only of services but also of customer satisfaction to the private sector. Thus we found that citizens had mixed feelings towards the government with respect to basic governance services. Ordinary citizens trusted the credibility of the government name but were dissatisfied with the quality of government services, and with having to put up with rude and bureaucratic treatment in an often-corrupt system. Citizens had similarly divided feelings towards the private sector. Civil society in India may be disillusioned with government provided services, but it was equally skeptical of the private sector as the protector of the poor. Hansen’s study of the Indian state as a guarantor of order stated that, “It may be well that ordinary Indians are less in awe of the state than a few decades ago, but it is still regarded as indispensable for public order and for recognizing communities, leaders or claims as legitimate.”[26: 37]. The resistance to private sector participation in service delivery among some segments of civil society was rooted in the belief that this sector was solely concerned with profit making and was frequently corrupt [27]. Thus, just as the state officials and entrepreneurs indicated, the citizens we interviewed agreed that, despite a reputation for poor services, the government brand was nevertheless accountable and credible, particularly for the delivery of governance services. These

72

divided feelings explain why the state sees the private entrepreneurs as indispensable to its new image and why the entrepreneurs see the state as equally indispensable to their ability to provide e-governance services. D. The hybrid state We found that through these telecenter projects a new form of government is being experimented with that attempts to combine the accountability of an elected government with the efficiency and customer service associated with the private sector. The state’s new vision of itself contrasts sharply with the images that its population might have had from previous encounters at conventional state offices. This hybrid version of government is gradually reworking both the way in which citizens with access to ICTD telecenters now perceive it and also the state’s perception of itself in relation to its citizens. Scholars of the Indian state have argued that, at the most local levels, the boundaries between state and civil society are often blurred [24]. Through neighborhood ICT telecenters, the lines between states and civil society appear to merge as telecenter entrepreneurs hold multiple positions as local community members, private actors, and also representatives of the state in the delivery of government services. By negotiating their multiple roles in their daily practices, entrepreneurs create representations of the state that affect both its legitimacy and their own. The running together of the public and the private realms is especially evident where profit-seeking entrepreneurs offer public services in a market setting. When walking into an ICT telecenter, citizens encounter painted signboards that mark the telecenters as both private and public. In the case of the Rajiv Internet centers, the signboards might advertise the brands of more than 7 different public and private entities (See Figure 4). The state’s name and authority still define the terrain on which private entrepreneurs have to operate. At the same time, entrepreneurs create new experiences that alter the way in which the state is seen by citizens. Everyday interactions with telecenter entrepreneurs contribute to citizens’ sense of how, and for whom, government operates. Rather than thinking of the state and market as distinct spheres, our research found that, in these part-public part-private telecenters, the very definitions of the state and the market were mutually constituted.

Figure 4: Signboard outside Rajiv Internet Village Center (2006)

E. Conclusion This paper analyzed how, in their attempts to liberalize and modernize their operations, Indian states are using ICTD to represent themselves in a new way to their citizens. Equally, it reveals how modernizing states come to be seen by the citizens through their everyday interactions at ICTD telecenters. We find that e-governance initiatives in Kerala and Andhra Pradesh are not only about delivering services efficiently (“e for express”) but also about updating the state’s image to one that is modern and market-friendly. This goal has partly been achieved via the delivery of government services through telecenters, implemented in partnership with local entrepreneurs. We find that citizens with access to these telecenters generally prefer them to conventional government offices. In particular, the poorer citizens feel they are treated with some respect at the centers, while middle class citizens appreciate the conveniences and amenities that these new centers offer. As the good governance agenda has taken hold in India, Kerala and particularly Andhra Pradesh have embraced the language and the ideal of the modern state conducting its business in a business-like manner. This embrace has been only partial, however. With service delivery via privatelymanaged telecenters, we do not find that the governments in Kerala or AP have been privatized or have withdrawn, as the critics of the “good governance” agenda frequently aver (or as the proponents of the agenda tend to promise). Rather, they play a critical role in managing and constructing their image to their citizens. Citizens’ simultaneous trust in and disillusionment with their governments, combined with their simultaneous admiration for and suspicion of private enterprise, have kept the states practically and discursively important to the success of e-governance. Given that changes in perceptions of the state (for the most part in a positive way) is one outcome of these projects, our research suggests that policymakers could explore how, and if, these changed perceptions can be used in evaluating projects. Given the widespread implementation of telecenters throughout India and the government’s commitment to implement 100,000 telecenters in the country, our research also raises the question: Will citizens in other parts of India encounter the state via telecenters in a similar way as the citizens of AP and Kerala? We note that, as of now, most Indians still do not have access to e-governance, and most government business is still not conducted electronically. We agree with those who argue that the transparency and efficiency expected from public-private partnerships do not always materialize, and with critics of the rush towards telecenters when the results of such projects have been mixed. But our research does indicate that, in Kerala and AP, these public-private telecenters have created a space for the state to renegotiate its role and image in public service delivery.

73 [24]

V. REFERENCES [1]

E. Lucas, "The Electronic Bureaucrat A Special Report on Technology and Government," The Economist, vol. February 16,2008, 2008. [2] S. Corbridge, R. Wiliams, M. Srivastava, and R. Veron, Seeing the State: Governance and Governmentality in India. Cambridge: Cambridge University Press, 2005. [3] O. Williamson, "The New Institutional Economics: Taking Stock, Looking Ahead," Journal of Economic Literature, vol. 38, pp. 595613, 2000. [4] J. Kooiman, Governing as governance. London: Sage Publications, 2003. [5] S. Madon, "Governance lessons from the telecenters in Kerala," European Journal of Information Systems, vol. 14, pp. 401-416, 2005. [6] R. Jenkins, "The emergence of the governance agenda: sovereignty, neoliberal bias, and the politics of international development," in The Companion to Development Studies, V. Desai and R. Potter, Eds. NY: Oxford University Press, 2002. [7] N. Rose, Powers of Freedom: Reframing Political Thought. Cambridge: Cambridge University Press, 1999. [8] A. Cordello, "E-government: towards the e-bureaucratic form," Journal of Information Technology, vol. 22, 2007. [9] L. Kaboolian, "The New Public Management: Challenging the Boundaries of the Management vs. Administration Debate," Public Administration Review, vol. 58, pp. 189-193, 1998. [10] S. Batterbury and J. Fernando, "Rescaling Governance and the Impacts of Political and Environmental Decentralization: An Introduction," World Development, vol. 34, pp. 1851-1863, 2006. [11] V. Nanda, "The Good Governance Concept Revisited," ANNALS, AAPSS, vol. 603 January 2006, pp. 269-283, 2006. [12] G. Stokker, "Governance as theory: five propositions," International Social Science Journal, vol. 50, pp. 17-28, 1998. [13] A. Kohli, Democracy and Discontent: India's Growing Crisis of Governability. Cambridge: Cambridge University Press, 1990. [14] P. Bardhan, "Decentralization of Governance and Development," The Journal of Economic Perspectives, vol. 16, pp. 185-205, 2002. [15] P. Bardhan, "Governance issues in delivery of public services," Journal of African Economies, vol. 13, pp. 167-182, 2004. [16] R. Heeks, "Understanding E-governance for Development," Institute for Development Policy and Management, vol. Working Paper No.11, 2001. [17] K. Rajalekshmi, "E-governance services through telecenters: The role of human intermediary and issues of trust," Information Technologies and International Development, vol. 4, pp. 19-35, 2008. [18] J. Thomas and G. Parayil, "Bridging the Social and Digital Divides in Andhra Pradesh and Kerala: A Capabilities Approach," Development and Change, vol. 39, pp. 409-435, 2008. [19] R. Kuriyan, I. Ray, and K. Toyama, "Information and Communication Technologies for Development: The Bottom of the Pyramid Model in Practice," The Information Society, vol. 24, pp. 93-104, 2008. [20] J. Pal, S. Nedevschi, R. Patra, and E. Brewer, "A MultiDisciplinary Approach to Studying Internet Kiosk Initiatives: The Case of Akshaya," Proceedings of the Global e-development Conference, 2004. [21] J. Scott, Seeing Like a State: How Certain Schemes to Improve the Human Condition have Failed. New Haven: Yale Agrarian Studies Series, 1998. [22] G. Joseph and D. Nugent, Everyday Forms of State Formation: Revolution and the Negotiation of Rule in Modern Mexico. Durham: Duke University Press, 1994. [23] J. Comaroff, "Reflections on the Colonial State in South Africa and Elsewhere: Factions, Fragments, Facts, and Fictions," Social Identities, vol. 4, pp. 321-361, 1998.

[25] [26] [27]

A. Gupta, "Blurred Boundaries: The Discourse of Corruption, the Culture of Politics, and the Imagined State," American Ethnologist, vol. 22, pp. 375-402, 1995. J. Ferguson and A. Gupta, "Spatializing states: towards an ethnography of neoliberal governmentality," American Ethnologist, vol. 29, pp. 981-1002, 2002. T. Hansen, "Governance and myths of state in Mumbai," in The Everyday State and Society in Modern India, C. J. Fuller and V. Benei, Eds. New Delhi: Social Science Press, 2000. R. Vyas, P. Small, and K. De Riemer, "The private public divide:impact of conflicting perceptions between the private and public health care sectors in India," Int Journal Tuberc lung Dis, vol. 7, pp. 543-549, 2003.

74

Evaluating the Accuracy of Data Collection on Mobile Phones: A Study of Forms, SMS, and Voice Somani Patnaik, Emma Brunskill and William Thies

Abstract—While mobile phones have found broad application in reporting health, financial, and environmental data, there has been little study of the possible errors incurred during mobile data collection. This paper provides the first (to our knowledge) quantitative evaluation of data entry accuracy on mobile phones in a resource-poor setting. Via a study of 13 users in Gujarat, India, we evaluated three user interfaces: 1) electronic forms, containing numeric fields and multiple-choice menus, 2) SMS, where users enter delimited text messages according to printed cue cards, and 3) voice, where users call an operator and dictate the data in real-time. Our results indicate error rates (per datum entered) of 4.2% for electronic forms, 4.8% for SMS, and 0.45% for voice. These results caused us to migrate our own initiative (a tuberculosis treatment program in rural India) from electronic forms to voice, in order to avoid errors on critical health data. While our study has some limitations, including varied backgrounds and training of participants, it suggests that some care is needed in deploying electronic interfaces in resource-poor settings. Further, it raises the possibility of using voice as a low-tech, high-accuracy, and cost-effective interface for mobile data collection.

I. I NTRODUCTION Mobile devices have shown great promise for improving the efficiency and effectiveness of data collection in resource-poor environments. Compared to a traditional process that relies on paper-and-pencil forms with subsequent transcription to a computer system, mobile devices offer immediate digitization of collected data at the point of survey. This allows for fast and automated data aggregation. It also improves adherence to complex or context-dependent questionnaires, as the device determines which questions should be answered or skipped. The benefits of mobile data collection have been demonstrated mostly in the context of personal digital assistants (or PDAs) [31], [10], [8], [2], [32], [12], [24], [9], [16], [4], [3], [15]. Given the recent explosion of mobile phones around the world, there is growing excitement in extending the successes achieved on PDAs to a phone-based platform. While high-end phones provide the same capabilities as PDAs, low-end phones lack features such as high-resolution displays and touch-screen capabilities. To empower the full population of nearly 4 billion mobile phone subscribers [26] with the capabilities of mobile data reporting, it will be important to establish usable interfaces that are portable to inexpensive phones, and there have been a number of recent efforts in this space (see for example [13], [1], [22], [25], [7]). Manuscript received September 22, 2008. Somani Patnaik and Emma Brunskill are with the Massachusetts Institute of Technology (email: [email protected], [email protected]). During this research, William Thies was affiliated with both the Massachusetts Institute of Technology and Microsoft Research India (email: [email protected]).

In migrating mobile data collection from PDAs to cell phones, a critical issue is ensuring the accuracy of data entry. In the context of healthcare, an errant entry may prevent lifesaving treatments from reaching patients, or may cause the prescription of unnecessary treatment that is costly and dangerous. In financial applications, entry errors may jeopardize the economic standing of communities that are already very poor. Due to the importance of this issue, several researchers have studied the error rates incurred as PDAs are deployed in developing regions. As detailed later (in Table II), the error rates are generally less than 2% (i.e., 2 errors per 100 entries) in programs where users received at least an hour of training [12], [24], [4]. However, in the context of mobile phones, studies of data accuracy are distinctly lacking. The closest work is by Parikh et al., where a hybrid system of paper forms and camera-equipped mobile phones has demonstrated error rates of less than 1% [28]. For standalone data collection on low-end phones, we are unaware of any previous study with a rigorous evaluation of data accuracy. This research opportunity is highlighted in Table I. In this paper, we provide a quantitative evaluation of data entry accuracy using low-cost mobile phones in a resourceconstrained environment. We evaluate three practical user interfaces for entering data on a mobile phone: electronic forms, SMS, and voice. Electronic forms consist of numeric fields and multiple-choice menus, and can be implemented in Java or a native phone platform. The SMS interface requires users to send a structured SMS messages to a server, with logical fields separated by delimiters in the message. The voice interface represents a normal telephone call, with a live human operator that enters the data into a centralized spreadsheet. We evaluated these interfaces in a study of 13 health workers and paramedical staff over a month-long period in Gujarat, India. Each participant was trained and evaluated on all of the interfaces. We focus on the collection of health data relevant to tuberculosis (TB), as we anticipate deploying an electronic system in a real TB treatment program. The data in this paper represent only simulated patient interactions. Our results indicate an error rate of 4.2% for electonic forms, 4.5% for SMS, and 0.45% for voice. These represent the fraction of questions that were answered incorrectly; as each patient interaction consisted of eleven questions, the probability of error somewhere in a patient report is much higher. For both electronic forms and SMS, 10 out of 26 reports (38%) contained an error; for voice, only 1 out of 20 reports (5%) contained an error (which was due to operator transcription). As detailed in Section VI, error rates

75 PDAs

Cell Phones

Published error rates

Malaria monitoring in Gambia [12] Clinical study in Gabon [24] Tuberculosis records in Peru [4] Sexual behavior surveys in Peru [3]

Other programs

SATELLIFE [15] DataDyne EpiSurveyor [31] EpiHandy [10] Infant health in Tanzania [32] e-IMCI project in Tanzania [8] Respiratory health in Kenya [9] Tobacco survey in India [16] Ca:sh project in India [2]

None?

Cell-Life in South Africa [13] Jiva TeleDoc in India [1, p.42] Pesinet in Mali [22] Malaria monitoring in Kenya [25] Voxiva Cell-PREVEN in Peru [7]

TABLE I P REVIOUS WORK IN EVALUATING THE ACCURACY OF MOBILE DATA COLLECTION IN THE DEVELOPING WORLD .

are distinctly higher for health workers than for hospital staff, though this difference may also be influenced by variations in our training environment. We were surprised and alarmed by these results. In our own treatment program, our original intent was to utilize electronic forms. However, we consider it to be an unacceptable risk that 38% of submitted forms – containing critical health information – may contain errors. For this reason, we have overhauled our plans and will implement a treatment program using voice rather than forms or SMS. While the cost of a live operator may be prohibitive in many countries, in India it proves to be very cost-effective. The increased cost of a human operator is more than compensated by the decreased cost of voice-only handsets, voice-only cellular plans, decreased training time, and decreased literacy requirements for health workers. We offer a more detailed analysis in Section VII. While the results of this study have changed our own approach to implementing mobile data collection, we caution the reader in extending the results of the study beyond its original context. In particular, we are focused on the scenario in which users have limited cell phone familiarity and there is limited time to perform training. If either of these variables changes, it may be possible to implement high-accuracy mobile data collection with electronic forms or SMS. Also, while the error rates that we report on mobile phones are 3-8x higher than those previously reported for PDAs, our data are unable to distinguish whether this difference is due to the devices, or due to other aspects of the study demographics, training, and evaluation. A future study could address this question directly by evaluating both phones and PDAs in the same context. Despite these limitations, our study is the first (to our knowledge) that evaluates data entry accuracy on mobile phones. Based on our results, we submit only that electronic forms and SMS may need further validation before gaining widespread deployment in accuracy-critical applications, and that voice may deserve more attention as a high-accuracy and low-cost means of data collection. The rest of this paper is organized as follows. We start by reviewing related work on mobile data collection (Section II). Then we consider the tradeoffs between electronic forms, SMS, and voice (Section III) and detail our implementation

of each interface (Section IV). We describe the setup of our user study (Section V) and the results obtained (Section VI), and we discuss the implications (Section VII). We conclude in Section VIII. II. R ELATED W ORK As summarized in Table I, there have been several initiatives to apply PDAs and cell phones for mobile data collection in the developing world. While a fraction of the studies on PDAs includes an experimental analysis of the error rate incurred, we are unaware of any study which systematically measures the accuracy of data entry on a cell phone. This is the principal novelty of our work. Lane et al. provides a review of nine randomized controlled trials that compare the effectiveness of PDAs and paper forms for data collection [21]. Six of the trials reported entry accuracy, with varying results: two studies found PDAs to be more accurate than paper [20], [29], three studies found the accuracy to be similar with both methods [17], [23], [36], and one study found that paper was more accurate [35]. None of the trials were in the context of the developing world (they took place in North America and Europe). Previous studies of PDA entry accuracy in the developing world are summarized in Table II. In cases where workers received at least an hour of training, error rates are under 2% (i.e., 2 errors per 100 questions). As early as 1991, Forster et. al evaluated the use of PDAs for a malaria morbidity study in the Gambia [12]. Employing secondary-educated workers who received five days of training, they report error rates between 0.1-0.6% and argue that the PDAs offer improved accuracy and efficiency over paper forms. Missinou et al. employed PDAs in a clinical study in Gabon, employing four clinicians who had no prior PDA experience and received 8 hours of training [24]. They report a 1.7% rate of discrepancy between PDAs and paper forms, and note that clinicians preferred the PDAs. Blaya et al. found that error rates improved from 1.3% (with paper forms) to 0.37% (with PDAs) in reporting tuberculosis bacteriology data in Peru1 [4]. The authors also argue that PDAs are cost-effective [5]. 1 Blaya et al. reports errors per form, rather than errors per entry [4]. Via personal communication with the author, we determined that there were an average of 7.5 entries per form, yielding the error rates quoted here.

76 Application Malaria morbidity [12] Clinical study [24] Bacteriology data [4] Sexual behavior [3]

Location Gambia Gabon Peru Peru

PDA Psion Organizer II XP Palm m500 Palm Zire Palm Zire

Education Level Secondary 3 M.D.s, one clinical officer Post-secondary (2-3 years) Secondary or less

Training 5 days 8 hours 16 hours 2-3 mins

Error Rate 0.1%-0.6% 1.7% 0.37% 14%

TABLE II E RROR RATES MEASURED BY PREVIOUS RESEARCHERS IN APPLYING PDA S FOR MOBILE DATA COLLECTION IN THE DEVELOPING WORLD .

Higher error rates have been reported in the case of self-administered surveys, when limited training is possible. Bernabe-Ortiz et al. evaluate the use of PDAs for surveys of sexual behavior in Peru [3]. To protect patient privacy, the PDAs were intended for use by actual subjects, rather than by health workers. As only some subjects had finished secondary education, and subjects received only 2-3 minutes of training, the authors observed a 14% discrepancy between electronic and paper forms. However, the error rate was substantially lower for subjects who had finished secondary schooling. Additional programs have applied PDAs for data collection in the developing world, but have not provided a rigorous analysis of entry accuracy. SATELLIFE uses PDAs for disseminating and collecting medical information in numerous countries [15]. There are anecdotal reports that the PDAs improved data quality [19], and the benefits of decreased error rates were estimated on a five-point scale [6]. Users of the system have also rated its usability [11]. However, we are unaware of a quantitative assessment of the error rates incurred. DataDyne EpiSurveyor [31] has been widely deployed for data collection in Sub-Saharan Africa; while it has been argued that the system is more accurate than paper forms [30], we are unaware of a controlled study. EpiHandy also provides tools for deploying electronic forms on PDAs and has been deployed in South Africa, Uganda, and elsewhere [10]. PDAs have also found application for gathering infant mortality data in Tanzania [32], for pediatric care (as part of the e-IMCI project) in Tanzania [8], for assessing respiratory health in Kenya [9], for surveying tobacco use in India [16] and for maternal and child health (as part of the Ca:sh project) in India [2]. These studies lack formal evaluations of entry accuracy. Cell phones have also found broad application for mobile data collection in the developing world. Cell-Life employs electronic forms on mobile phones to improve TB and HIV treatment in South Africa [13], [33]. Electronic forms are also used by Jiva TeleDoc for improving rural healthcare in India [1, p.42], and by Pesinet for monitoring infant health in Mali [22]. Mobile phones with forms are also being used to monitor malaria in Kenya [25]; while PDAs were also piloted, the authors note that phones are more intuitive due to worker familiarity. Voxiva’s Cell-PREVEN uses interactive voice response and voice recording to monitor adverse events amongst sex workers in Peru [7]. We are unaware of any quantitative evaluation of entry accuracy in these projects. To avoid the complexities of navigating electronic forms, the CAM framework offers a hybrid system in which paper forms are used for organization while phones are used for data entry [27]. Each field on the paper form is annotated with a barcode, which is recognized by a camera on the phone

prior to data entry. Users that lacked prior camera or computer experience were trained to a level of comfort within 5 to 15 minutes. A separate study measures error rates of 1% or below using the CAM system [28]. This represents an interesting and useful design point, especially in cases where paper forms are already ingrained into the workflow. We focus on solutions that are independent of any paper workflow, and which do not necessarily require a camera-phone (while Java-phones often have cameras, our SMS and voice solutions are suitable to the most inexpensive phones). While electronic forms have been widely deployed, there are fewer solutions that rely on user-constructed SMS messages for mobile data collection. One example is a system from Dimagi, Inc. which monitors water treatment plants in India [34]. We are unaware of other systems which rely on a cue card (as we do in our evaluation) for submitting a structured SMS message to a server. Others have considered broader issues in the contextual design of user interfaces for data collection in the developing world. Examples include interface design for Auxiliary Nurse Midwives in India [14] and a methodological framework for evaluating health devices [18]. Our focus is on assessing the entry accuracy for a range of standard interfaces. III. U SER I NTERFACES Three of the central modes on a cell phone that can be used to perform data collection are voice, SMS and an electronic forms application. Data collection performed by voice can be further split into systems that link the data collector with a live operator, those that connect to an automated interactive voice response system, and those that allow the user to record a message. We focus our discussion around live voice operators, SMS and electronic form based systems, and examine some of the strengths and weaknesses of these various approaches. We use SMS to refer to data collection systems that involve information entered by a structured text message: in particular we assume that the information is entered by following a small cue sheet with a flowchart that directs the collector how to enter the data. To our knowledge, using cue cards to guide data entry by text message has not been done previously. In contrast, electronic forms (particularly on personal digital assistants) have been widely used. In this paper, we use the term “electronic forms” to denote any external application that can be placed on a phone, and that automatically guides the user how to enter data, through the use of text, menus or other tools. In a voice operator interface, the user simply calls a live operator, who asks the user a series of questions to elicit the information needed. Figure 1 illustrates each interface as used in our particular experiment.

77

1

2

3

Electronic Forms Interface General Strengths Easy patient identification Ongoing cost is low (SMS or data plan) Can store visits when connectivity is poor

a)

General Weaknesses Requires programmable phones Requires basic literacy skills Hard to alter survey questions Hard to enter in free-form notes Application can be deleted by user Our Results: Accuracy & Efficiency We measured 4.2 errors per 100 entries The average interaction was 99 seconds

Electronic form screenshots 1. Start application 2. Select the patient 3. Select "new visit" 4. Enter the patient's temperature 5. Select severity of patient’s cough 4

5

SMS Cue Card

SMS + Cue Card Interface General Strengths Can be used with any phone Ongoing cost is low (SMS) Many workers familiar with SMS

b)

General Weaknesses Requires basic literacy skills Changing survey requires new cue card Hard to enter in free-form notes No confirmed receipt of data delivery Worker can forget or lose cue card Quite easy to fake visits (copy old SMS) Our Results: Accuracy & Efficiency We measured 4.5 errors per 100 entries The average interaction was 97 seconds

1. Create a new SMS Message Press Center Button Select “Messages” Select “Create Message” Select “New Short Message”

4. Enter a Space Press *

2. Switch to Numeric Input Mode Press Menu Button Select “Entry Mode” Select “Numeric”

11. Enter the Patient’s Cough No Cough - Press 1 Rare Cough - Press 2 Mild Cough - Press 3 Heavy Cough - Press 4 Severe Cough - Press 5 (with blood)

3. Enter the ID of the Current Patient Aamir Khan - Press 1 Abhishek Bachchan - Press 2 Aishwarya Rai - Press 3 …

21. Check Yourself Your finished message should be formatted similarly to the following: 10 372 62 68 4 1030007

Sample Voice Interaction

Voice Interface

c)

General Strengths Can be used with any phone No literacy required of workers Easy to change survey questions Easy to add in free-form notes Hard to fake a visit: operator can ask new questions

Patient

Worker General Weaknesses Ongoing cost of operator salary Voice plans often higher cost than SMS Awkward 3-way social interaction Our Results: Accuracy & efficiency We measured 0.45 errors per 100 entries The average interaction was 140 seconds

Fig. 1.

Operator

Operator: Hello. What is your name? Worker (to operator): My name is Lipika. I am calling to record a patient visit. Operator: What patient are you visiting? Worker (to patient): What is your name? Patient: Pavathi (reading from note note sheet) Worker (to operator): Pavathi. Operator: That’s Pavathi, right? Worker (to Operator): Yes (operator records name) Operator: What is her temperature? Worker (to patient): What is your temperature? Patient: 97.1 (reading from note sheet ) Worker (to operator): 97.1 degrees. Operator: 97.1 deg. (operator records temperature) …

The three user interfaces evaluated in this paper: a) electronic forms, b) SMS + cue card, and c) voice.

78

In general, there are a variety of factors that affect the choice of a data collection interface. These may be loosely categorized into operation, effectiveness and cost. Figure 1 supplements the below discussion by summarizing some of the strengths and weaknesses of each interface. A. Operation We use “operation” to refer to factors involved with the general infrastructure of the data collection system. Initially there is the investment of time to set up the system, and then train the workers who will be performing data collection on the system. For voice or SMS interfaces, the set up time for workers is minimal: each worker must simply be provided with a phone, if he or she does not currently own one. However, electronic forms require that the application be downloaded onto the phone, which requires either an Internetenabled phone in an area of good connectivity, or specialized development tools and an external computer. Training time for each application is an open issue, and is one of the factors we investigate further in this study. Worker education and worker cell phone familiarity are likely to affect how easy it is to set up each user with an interface, and train them how to use it. We expect that a voice interface requires the least amount of education and background to get users equipped to start performing data collection. In particular, a voice interface does not require that its users be literate. System coverage and reliability are also critical factors to ensure good data collection. Voice calls have priority over SMS, and there is the possibility of lost SMS messages. The delivery mechanism with electronic forms can vary: both GPRS and SMS can be used. GPRS has the advantage that there is an acknowledgment of whether the data was sent; however not all locations have coverage. From the user side, voice appears to be the most reliable and has the most far reaching coverage; however, this also requires that there exists a sufficient number of operators so that users can always reach a person when they call. If this is not always possible then there may be a reliability penalty as users may have to call back later (or wait for the operator to return their call). In addition to reliability, a good system should enable some degree of flexibility. Despite good initial prototyping, it may sometimes be important to be able to modify the data collection interface, fix an error, improve usability, or add or remove information to be collected. If users have an Internet enabled phone and are always working in areas of high connectivity, then updating an electronic form system is quite feasible. However, if this is not the case, then users must reprogram their phone using the same specialized tools needed for initial set up. SMS is similarly challenging to update since a new cue card must be distributed to direct the user to enter the data. In contrast, voice is trivial to update, as the operator can simply ask a new set of questions. B. Effectiveness In any data collection effort, one of the key considerations is the effectiveness of the program at obtaining high quality

data. High quality data can perhaps be characterized by two simple criteria: whether or not the data is intentionally faked by the user, and the accuracy of data that is not intentionally faked (which is the focus of this paper). Intentionally faked data can lead to incorrect conclusions and potentially lead to significant misallocation of resources when interventions are based on false data. There may be an incentive to fake data when users are busy and collecting real data is time consuming, due to the data recording itself or transportation time to reach the source of the data (such as visiting remote patients). Unfortunately in SMS systems it is quite easy to fake data, particularly for cell phone savvy users that can copy and paste prior SMS messages. Faking electronic forms is slightly harder as it requires the user to sequentially fabricate data across an entire form. It requires the most effort for users to fake data while speaking on the phone, as the operator can always ask a new question to try to ascertain if the user is fabricating the data. Voice also has the benefit that it is easy for users to convey additional information (not included in the original survey), whereas it is more challenging to spell out text using the keypad, particularly in other languages which may or may not be supported on a given phone. Voice is also likely to have fewer operational risks: users may accidentally delete the form application, or forget their SMS cue card, but since an operator can always call a worker directly, the voice system is fairly robust. Voice also makes it easy for users to correct previous visits, by simply calling back the operator. This is also easy to do by modifying and resubmitting a saved electronic form. However, it is also important to consider the speed of data entry, how much the user likes the interface, and the accuracy of data entry. To our knowledge there are no prior studies comparing the accuracy and speed of data entry using SMS, electronic forms and voice. Since we regard these as some of the most critical factors in choosing an interface, this is a large motivation for our current study. C. Cost One of the other important considerations is cost: the most beautiful, user-friendly, accurate interface may still not be practical if the cost overhead is too high for the particular problem. Costs consist of fixed one time costs as well as ongoing marginal costs. For all three interfaces users must have a cell phone. An electronic form requires a programmable phone (such as a Java-enabled phone or Windows phone) but both SMS and voice applications can be used with any phone. The ongoing cost for an SMS phone depends on the rate per message which is typically quite low. An electronic form can send data using SMS or through a data plan; typically SMS is cheaper depending on the amount of data that is being collected. Voice minutes are frequently more expensive than SMS. But most importantly, voice has the ongoing cost of the salary of the operator, which is an additional overhead not shared by electronic forms or SMS.

79

IV. U SER I NTERFACE I NSTANTIATION The prior section discussed some of the general factors important to consider when designing and selecting a data collection interface. We now discuss the context for our data collection effort and the interfaces we evaluated. A. Domain context Soon the authors, along with other collaborators, intend to conduct a trial that examines whether increased information and monitoring can improve heath outcomes and adherence during tuberculosis treatment in Bihar, India. Treatment will be conducted by having tuberculosis patients regularly visit health workers and receive drugs as part of a directly observed therapy (DOT) strategy. During these visits, health workers will collect data about their patients and report this information by mobile phone back to a central office. This information will be aggregated and analyzed to inform doctors and the trial manager about which patients may need to be visited, for example, if a patient is not improving or is experiencing adverse side effects. To support this effort we need a user interface that enables fast and accurate data collection. The data collected during a patient visit will include both identification and health status information. The worker will enter in information to identify both the worker name (done only once at the start of treatment, in the case of forms and SMS) as well as the patient name. In addition, the health worker will record the patient’s current temperature, weight and pulse, as well as the presence or absence of seven symptoms: night sweats, chest pain, loss of appetite, nausea, coughing with blood, yellow eyes and fatigue. These symptoms were chosen based on advice gathered from tuberculosis health experts. The worker will also record whether the patient’s current cough is absent, rare, mild, heavy or severe with blood. The trial intervention is centered around the hypothesis that better, more frequent data collected about patients can improve tuberculosis health outcomes and therefore high quality data collection is critical. However, even if an interface encourages high quality data, it is still essential that such a data collection method also be easy to use and affordable in order for such an intervention to have widespread applicability. Originally we were planing to use electronic forms for data collection. However, since there appeared to be a dearth of literature in evaluating mobile data collection accuracy, we decided to evaluate the accuracy, speed and usability of three mobile phone interfaces. The results of this evaluation influenced our choice of an interface for use in the treatment program. B. Electronic forms implementation We created a Java application which provides a sequence of electronic forms that guide the worker to request information from the patient. The worker identification number is encoded once into the phone and is included with each recorded visit. The worker has to either enter numeric data or make a selection from a multiple-choice menu to encode symptoms.

The electronic forms underwent several design iterations, including gathering feedback from a 3-day session with 22 health workers in Bihar, India, prior to this study. Based on feedback from the workers in Bihar, we choose to employ hybrid English/Hindi menus for some of the forms, since some medical terms are easier to understand in English, but others are easier to understand in Hindi. We also changed from using multi-select lists (with a checkbox per symptom) to using individual yes/no questions. Figure 1a shows a series of screenshots of the form interface used for the present study. The Java application can be set up to either relay this information via SMS or GPRS. This distinction is important for cost considerations but does not affect the interface testing considered here. C. SMS implementation For the SMS interface we designed a cue card that instructs the worker how to record information about the patient into a text message; Figure 1b displays a subset of the cue card used. All information is coded numerically; this is done to reduce the amount of cell phone familiarity necessary, as well as to increase the speed of data entry. Participants enter in data as prompted by the cue card and then send the text message at the end of the interaction. The final part of the cue card as displayed in Figure 1b shows a sample text message. D. Voice implementation For the voice interface the worker calls a live operator. The operator asks the worker a series of questions about the patient’s health, which prompts the worker to ask the patient that question. This means that workers interact simultaneously with an operator and a patient; we are unaware of previous programs that have taken a similar approach. Figure 1c displays a sample interaction. The live operator confirms answers with the worker; this adds to the length of each call but is done to increase accuracy. This can be particularly important when the phone connection is poor or there is background noise. V. S TUDY M ETHODOLOGY The user study took place in the Surat and Bharuch districts of the Indian state of Gujarat during July and August of 2008. A. Participants As detailed in Table III, the study participants consisted of six community health workers and seven hospital paramedical staff. The community health workers were associated with the Dahej public health center; five of the paramedical staff were at the Reliance Tuberculosis hospital; and the remaining two paramedical staff were at the dispensary of the Sardar Vallabhbhai National Institute of Technology. The study participants were recruited through contacts of the first author. Initially, we had hoped to perform the study entirely with community health workers, as they are often the primary agents of remote data collection (including in our upcoming tuberculosis treatment program). However, this turned out to be infeasible because some community health workers were

80

unable to travel to the Dahej public health center for training and testing, and it was not feasible for us to travel to each worker’s home. This prompted us to recruit participants from two other centers. There were also some logistical challenges in performing the studies due to adverse weather conditions and the bomb blasts occurring in July 2008 in the Surat area. The education level of the health workers ranged from 10 to 12 years, while the education of the hospital staff ranged from 10 years to a B.A. degree. The average age of the study participants was 26.4 years (range 19-35). Seven participants owned a cell phone, four participants had used but did not own a cell phone, and two participants had never used a cell phone previously. Eleven of the participants were native Gujarati speakers and all spoke Hindi. B. Training Participants were trained by at least two trainers in small groups of at least two. Initially, examples were presented on a whiteboard and participants were instructed to practice entering in the data on either electronic forms or as an SMS using the cue card. After this stage, a paper with a set of example patients was handed out, and participants were instructed to practice entering in this data. In the final stage, participants were instructed to practice role playing patient– worker interactions with each other. Participants received variable amounts of training, ranging from 45 minutes to 8 hours, depending on their experience and availability. The longer training sessions were not necessarily more effective, as they were performed in larger groups. While it would have been desirable to achieve more uniform training, this was difficult given the logistics of transportation and worker schedules. Prior to the completion of training, all participants had completed at least two perfect interactions on both electronic forms and SMS, and at least one perfect interaction on the live operator mode. Throughout the user study, we employed Motorola L6i cell phones for training and testing. This is the cheapest Java-enabled phone from Motorola (the source of our current development tools) that is available in India; see Appendix A-1 for a cost analysis. All interfaces and related tools (cue cards, etc.) were presented in Hindi, and the mobile phones used had dual Hindi menus. C. Testing Participants were tested in pairs, alternating who was being tested on data entry, and who was playing the fake patient for that data point. The order of the interfaces was randomized: for a given participant pairing, the order of voice, SMS, and electronic forms was alternated. For the voice interface, the first author acted as the operator and was located outside of the room testing was being conducted in; however, there was always an additional person associated with the experiment inside the room at all times with the participants. During testing, each participant performed two complete patient–worker interactions (in the role of the worker) for each of the forms and SMS interfaces. For the voice interface, the

six community health workers completed only one interaction, while others completed two interactions (we did not anticipate that voice would become a focal point of this study until halfway through our experiments). The lag time between training and testing was exactly one day for seven of the participants, and ranged between half a day and two days for the remaining participants. All participants received a brief refresher and supervised entry session immediately prior to testing. VI. R ESULTS The results of the user study are detailed in Table III. We present both the accuracy of data entry, as well as the time needed to interview patients and report the data. On average, electronic forms and SMS offered comparable error rates of 4.2% and 4.5% per entry, respectively. The voice interface proved to be approximately 10x more accurate, with an error rate of 0.45% per entry. While only one out of thirteen participants performed perfectly on both the forms and SMS interfaces, twelve out of thirteen participants performed perfectly on voice. A Student’s two-tailed, unpaired t-test revealed that voice had a significantly lower error rate than electronic forms (p < 0.01) and SMS (p < 0.01); no significant difference was found between the error rates of electronic forms and SMS (p = 0.84). It is important to note that our results indicate a bimodal distribution of error rates: participants 7-13 performed notably better than participants 1-6. While there are many compounding differences between these participants, including the manner in which we conducted training, we refer to them by their occupation in order to simplify the discussion; participants 1-6 are health workers while participants 7-13 are hospital staff. As summarized in Table III, health workers exhibited an error rate of 7.6% for forms and 6.1% for SMS, while hospital staff exhibited an error rate of 1.3% for forms and 3.2% for SMS. In addition, the only voice error occurred with health workers. Unfortunately, our data are insufficient to explain the differences observed between these two groups of participants. On average, the hospital staff were older, more educated, and more likely to own a cell phone than the health workers. It is plausible to suspect that these factors contributed to the higher accuracy achieved by hospital staff. However, due to logistical reasons, our training procedure also differed between the two groups: health workers were trained in a large group for 6-8 hours, while hospital staff were trained in small groups for 1-2 hours. Our trainers were also somewhat more experienced when working with hospital staff, as health workers were trained first. We re-iterate, however, that training continued until all participants were able to complete two perfect trials on forms and SMS, and one perfect trial on voice. To better understand the error rates observed using each interface, we tabulate the exact sources of error in Appendix A-2. We classify errors by their entry type (numeric, multiplechoice, yes/no). We also inspect whether each error could be detected, by a trained eye, using the submitted data only; in the future, such errors could potentially be flagged or

81

ID 1 2 3 4 5 6 7 8 9 10 11 12 13

Occupation Health worker Health worker Health worker Health worker Health worker Health worker Hospital staff Hospital staff Hospital staff Hospital staff Hospital staff Hospital staff Hospital staff

Accuracy of Entries Time per Interaction Owns Total Used (Wrong / Total) (Average) Cell Training Cell Education Level Age Phone? Phone? (Hours) Forms SMS Voice Forms SMS Voice pre-secondary (class 10) 25 X 8 1 / 22 3 / 22 0 / 11 2:00 1:45 3:07 pre-secondary (class 10) 25 X 6 2 / 22 1 / 22 0 / 11 1:55 1:12 2:29 pre-secondary (class 10) 30 X 6 1 / 22 1 / 22 0 / 11 2:15 2:05 2:50 secondary (class 12) 19 8 2 / 22 1 / 22 0 / 11 1:33 1:27 2:34 secondary (class 12) 19 X 6 2 / 22 0 / 22 1 / 11 1:45 1:27 2:12 secondary (class 12) 20 X X 6 2 / 22 2 / 22 0 / 11 1:35 2:10 2:00 pre-secondary (class 10) 30 2.5 0 / 22 2 / 22 0 / 22 2:25 1:40 2:05 secondary (class 12) 32 X X 2 0 / 22 1 / 22 0 / 22 1:42 1:17 2:35 secondary (class 12) 28 X X 0.75 0 / 22 1 / 22 0 / 22 1:30 1:17 1:55 post-secondary (B.A.) 35 X X 1.5 1 / 22 0 / 22 0 / 22 1:25 3:15 2:00 post-secondary (D. Pharm.) 26 X X 1 0 / 22 0 / 22 0 / 22 1:05 0:55 2:10 post-secondary (D. Pharm.) 24 X X 1 0 / 22 1 / 22 0 / 22 1:07 1:25 1:52 post-secondary (M.S.W.) 30 X X 0.75 1 / 22 0 / 22 0 / 22 1:10 1:15 3:15 Average (health workers only) g (hospital ( p y) Average staff only) Average (across all interactions) Std. Dev. (across all interactions)

7.6% 1.3% 4.2% 5.9%

6.1% 3.2% 4.5% 6.4%

1.5% 0% 0.45% 2.0%

1:50 1:29 1:39 0:28

1:41 1:35 1:37 0:45

2:32 2:16 2:20 0:28

TABLE III R ESULTS OF THE USER STUDY. A LL PARTICIPANTS WERE EVALUATED ON TWO INTERACTIONS WITH THE FORMS INTERFACE AND TWO INTERACTIONS WITH THE SMS INTERFACE . T HE COMMUNITY HEALTH WORKERS (1-6) WERE TESTED ON ONE INTERACTION WITH THE VOICE INTERFACE , WHILE THE PARAMEDIC HOSPITAL STAFF (7-13) WERE TESTED ON TWO INTERACTIONS . AVERAGES AND STANDARD DEVIATIONS ARE SHOWN AT BOTTOM .

automatically fixed using self-correcting forms. Finally, we tabulate whether each error is potentially dangerous (e.g., a severe cough reported as a mild cough would prevent a physician from delivering needed care). Electronic forms witnessed errors in each entry type; only three of the twelve errors are evident from the values submitted, while five errors may be dangerous. Surprisingly, eight of the errors were due to numeric entry problems on the electronic forms. Two errors were due to a mis-placed decimal point in the temperature entry; while our interface automatically places the decimal point if needed, the user failed to enter the right number of digits in the temperature. The SMS interface also witnessed errors in each entry type; out of thirteen errors, eight are detectable and seven may be serious. Three of the errors could perhaps be averted with a revision of the SMS cue card: to indicate the absence of a patient cough, many participants entered the code “0” rather than the desired (though perhaps less intuitive) value of “1”. Unlike the forms interface, workers sometimes entered the wrong patient identity when using SMS. The voice interface witnessed only a single error for the entire duration of the trial. We consulted a videotaped record of the interaction in question (we taped one interaction for each participant), and found that the error was incurred by the operator in translating the participant’s report into a spreadsheet. While such transcription errors could indeed occur in practice, it is encouraging that the participants were not responsible for any errors on the voice interface. While the voice interface offered the lowest error rates, it also led to the longest entry times. Electronic forms and SMS averaged 1:39 and 1:37 per interaction, respectively, while the voice interface required 2:20 on average (1.43x higher than

forms and SMS). One factor that contributed to the slower entry rates using voice was the cellular coverage in our study area; the connection between participants and the operator was highly unreliable. The audio quality was frequently degraded beyond recognition, and calls were occasionally dropped and re-started. While many resource-poor environments have excellent cellular coverage (including the area of Bihar that we are planning to target with our treatment program), the weak coverage in our study area nonetheless reflects a realistic hazard of voice in some environments. In addition to quantitative results, we also solicited qualitative feedback from each participant, asking them to rank the interfaces by their order of personal preference. The forms and SMS interfaces were most popular amongst the participants, with each receiving six votes as the most popular interface. Only one participant preferred the voice interface to the others. This feedback is indicative of the poor phone connections experienced during the trial; many found voice to be frustrating due to the bad call quality. We were surprised that any participants preferred the SMS interface, given the relatively cryptic message that is produced in the end; however, participants noted that fewer keys are required under SMS than under electronic forms (which requires scrolling and selection). We also note that 8 of the 13 participants preferred the interface on which they demonstrated the fastest entry time. VII. D ISCUSSION In addition to the factors examined in our experiment, cost is a critical variable for selecting a data collection interface. For the purposes of our own decision making with regards to selecting an interface for our tuberculosis treatment program, we performed a simple cost analysis. Details are provided in

82

Appendix A-1, but in summary, the expected cost for data collection for each patient during his/her treatment is US $7.89 using electronic forms, US $4.59 using voice, and US $2.99 using SMS2 . These results show the cost of voice is competitive with the cost of the other two interfaces. Though SMS is slightly cheaper, in order for tracking patient symptom status to be helpful, it is essential that the reported data be close to error-free. This data will be used to guide doctor intervention, and faulty data may lead to unnecessary visits or worse, missed visits when a patient is sick. The voice interface had close to perfect accuracy and was significantly more accurate than SMS or electronic forms. Voice also allows for additional, unscripted information to be easily collected, and provides a social dimension to the health worker’s job. We anticipate that this social dimension could potentially lead to higher performance and a lower turnover rate amongst workers, since talking to an operator is likely to increase the worker’s feeling of being supported and integrated in a larger project. Voice also allows for verification to be performed easily: operators can simply request the worker to verify the data entry just given, which can be particularly useful for unusual entries. In addition, a voice interface can be replicated very easily in other contexts– no special software or cue cards need to be developed, and any cell phones can be used. While voice requires longer entry times for workers, this represents a very small fraction of their overall working day. For all these reasons, we have now decided to use a voice interface for our upcoming tuberculosis treatment program. Despite the many advantages of voice, there are still several challenges that must be addressed in practice. In our upcoming treatment program, workers will be actively examining and collecting data from patients and must report this information back to an operator. Calling the operator and keeping him on the line as the worker examines the patient may lead to a slightly awkward social interaction. Another more general challenge for voice interfaces is how to handle scenarios in which a user calls and the operator line is busy. One potential solution for these two challenges is to have the worker write down the data on paper and then call the operator. This introduces an additional opportunity for transcription errors but has the side benefit of creating a paper trail that may be used for later verification. To handle missing calls the operator could be responsible for calling back workers, or workers could leave a message that would be transcribed by the operator. An alternative solution to these challenges would be to use an interactive voice recognition (IVR) system. IVR could also be useful when there is very frequent data collection or when each survey questionnaire is very long. Hybrid live-operatorIVR systems are also possible, such as directing the worker initially to an IVR system, but automatically transferring the 2 We use Motorola phones for the electronic forms due to our current set of development tools. Moving to the cheapest available Java-enabled phone would decrease the forms cost to $5.39. However, in practice the cost of voice phones can also be reduced by leveraging existing phones in the community. The cost of voice remains competitive with forms in most practical scenarios.

worker to a live operator if the patient symptoms entered are worrisome. We look forward to exploring solutions for handling these different tradeoffs, and considering IVR solutions, as part of our future work. VIII. C ONCLUSION Given the widespread excitement in using mobile phones for collecting and analyzing data in the developing world, it is important to establish that the data entered on these devices meets the strict accuracy requirements of health, finance, and other applications. In this study, we provide a quantitative evaluation of data entry accuracy on mobile phones using electronic forms, SMS, and voice interfaces in a resource-poor setting. Our results indicate that, within the context of our study, the error rates for electronic forms (4.2% of entries wrong) and SMS (4.5% of entries wrong) may be too high to deploy these solutions in a critical application. In contrast, the accuracy of the voice interface was an order of magnitude better (0.45% of entries wrong), with only a single error observed across all trials. This result has influenced us to overhaul our plans for an upcoming tuberculosis program in Bihar, India, to switch to a voice-only interface. Employing a voice interface requires the employment of an operator, and may not be cost-effective in all countries. However, in India, the cost of this operator is more than compensated by the lower cost of voice-only handsets, voice-only cellular plans, decreased training time, and decreased literacy requirements on health workers. While this study provides an initial data point for the accuracy of data collection on mobile phones, further research is needed to distinguish the factors that are responsible for the errors observed. In the case of electronic forms, we observed error rates that are 3-8x higher than previously measured on PDAs. Our data are insufficient to diagnose whether this difference is due to the devices themselves (screen resolution, touch screen vs. keypad, etc.) or due to other aspects of the evaluation (worker education, training duration, etc.). A future study could address this question directly by evaluating PDAs and mobile phones in the same focus group. However, it is not our goal in this paper to prescribe the optimum device for mobile data collection. Rather, we aim only to highlight that there exists at least one context in which electronic forms and SMS may be too error-prone for large-scale deployment in an accuracy-critical application. In this same context, there is evidence that a low-tech alternative (voice) provides an accurate and cost-effective solution. IX. ACKNOWLEDGMENTS We are very grateful to the many people who graciously facilitated and participated in our user study. We thank the hospital staff at the Sardar Vallabhbhai National Institute of Technology in Surat, and also at the Reliance TB Hospital in Hazira. We thank Joshnaben Godia of the Taluka Development office in Vagra, Gujarat, and especially Suprava Patnaik for all her help with our study. This work was supported in part by the MIT Public Service Center.

83

R EFERENCES [1] Achieving the Health Millennium Development Goals in Asia and the Pacific: Policies and Actions Within Health Systems and Beyond. United Nations Publications, Dec. 2007. [2] V. Anantraman, T. Mikkelsen, R. Khilnani, V. S. Kumar, N. R. Machiraju, A. Pentland, and L. Ohno-Machado. Handheld computers for rural healthcare: Experiences from research concept to global operations. In Workshop on Development by Design, 2002. [3] A. Bernabe-Ortiz, W. H. Curioso, M. A. Gonzales, W. Evangelista, J. M. Castagnetto, C. P. Carcamo, J. P. Hughes, P. J. Garcia, G. P. Garnett, and K. K. Holmes. Handheld computers for self-administered sensitive data collection: A comparative study in Peru. BMC Medical Informatics and Decision Making, 8:11, 2008. [4] J. Blaya and H. S. F. Fraser. Development, implementation and preliminary study of a PDA-based bacteriology collection system. Proceedings of the AMIA Annual Symposium, 2006:4145, 2006. [5] J. A. Blaya, W. Gomez, P. Rodriguez, and H. Fraser. Cost and implementation analysis of a personal digital assistant system for laboratory data collection. The International Journal of Tuberculosis and Lung Disease, 12(8):921–7, Aug. 2008. [6] Cost effectiveness study report for the PDA data capture and transmission. SATELLIFE, Sept. 2004. http://www.healthnet.org/coststudy.php. [7] W. H. Curioso, B. T. Karras, P. E. Campos, C. Buenda, K. K. Holmes, and A. M. Kimball. Design and implementation of Cell-PREVEN: A real-time surveillance system for adverse events using cell phones in Peru. Proceedings of the AMIA Annual Symposium, 2005:176180, 2005. [8] B. DeRenzi, N. Lesh, T. Parikh, C. Sims, W. Maokla, M. Chemba, Y. Hamisi, D. S. Hellenberg, M. Mitchell, and G. Borriello. E-IMCI: Improving pediatric health care in low-income countries. In CHI, 2008. [9] L. Diero, J. Rotich, J. Bii, B. Mamlin, R. Einterz, I. Kalamai, and W. Tierney. A computer-based medical record system and personal digital assistants to assess and follow patients with respiratory tract infections visiting a rural Kenyan health centre. BMC Medical Informatics and Decision Making, 6(1):21, 2006. [10] T. Engebretsen. Acceptance of information technology by health research projects in low-income countries: Intention to use and acceptance of using EpiHandy (IUAUE). Master’s thesis, Agder University College, Dec. 2005. [11] Evaluation of the SATELLIFE PDA project, Feb. 2003. Bridges.org. [12] D. Forster, R. H. Behrens, H. Campbell, and P. Byass. Evaluation of a computerized field data collection system for health surveys. Bulletin of the World Health Organization, 69(1):107–11, 1991. [13] R. Fynn. Remote HIV/AIDS patient monitoring tool using 3G/GPRS packet-switched mobile technology. In Appropriate Healthcare Technologies for Developing Countries, pages 129–138, 2006. [14] S. Grisedale, M. Graves, and A. Gr¨unsteidl. Designing a graphical user interface for healthcare workers in rural India. In CHI, 1997. [15] T. Groves. SatelLife: Getting relevant information to the developing world. British Medical Journal, 313(7072):1606–9, 1996. [16] P. C. Gupta. Survey of sociodemographic characteristics of tobacco use among 99,598 individuals in Bombay, India using handheld computers. Tobacco Control, 5(2):114–120, June 1996. [17] R. N. Jamison, R. H. Gracely, S. A. Raymond, J. G. Levine, B. Marino, T. J. Herrmann, M. Daly, D. Fram, and N. P. Katz. Comparative study of electronic vs. paper VAS ratings: A randomized, crossover trial using healthy volunteers. Pain, 99(1-2):341–7, Sept. 2002. [18] D. R. Kaufman and J. B. Starren. A methodological framework for evaluating mobile health devices. Proceedings of the AMIA Annual Symposium, 2006:978, 2006. [19] H. Ladd and N. Sewankambo. Uganda health information network: Technical report (submitted to the International Development Research Centre). SATELLIFE & Uganda Chartered HealthNet, Dec. 2004. http://www.healthnet.org/idrcreport.html. [20] S. O. Lal, F. W. Smith, J. P. Davis, H. Y. Castro, D. W. Smith, D. L. Chinkes, and R. E. Barrow. Palm computer demonstrates a fast and accurate means of burn data collection. The Journal of Burn Care & Rehabilitation, 21(6):559–61; discussion 558, 2000. [21] S. J. Lane, N. M. Heddle, E. Arnold, and I. Walker. A review of randomized controlled trials comparing the effectiveness of hand held computers with paper methods for data collection. BMC Medical Informatics and Decision Making, 6:23, 2006.

[22] Mali: Mobile service helps bring down infant mortality, 2007. Balancing Act News, Issue No. 364, http://www.balancingactafrica.com/news/back/balancing-act 364.html. [23] J. S. McBride, R. T. Anderson, and J. L. Bahnson. Using a hand-held computer to collect data in an orthopedic outpatient clinic: A randomized trial of two survey methods. Medical Care, 37(7):647–51, July 1999. [24] M. A. Missinou, C. H. O. Olola, S. Issifou, P.-B. Matsiegui, A. A. Adegnika, S. Borrmann, D. Wypij, T. E. Taylor, and P. G. Kremsner. Short report: Piloting paperless data entry for clinical research in Africa. Amer. Journ. of Trop. Medicine and Hygiene, 72(3):301–303, Mar. 2005. [25] Mobilizing against malaria. Expanding Horizons (by Nokia), pages 12– 13, Feb. 2007. [26] Nokia CEO message, 2008. http://www.nokia.com/A4942317. [27] T. S. Parikh. Using mobile phones for secure, distributed document processing in the developing world. Pervasive Computing, 4(2), 2005. [28] T. S. Parikh, P. Javid, S. K, K. Ghosh, and K. Toyama. Mobile phones and paper documents: Evaluating a new approach for capturing microfinance data in rural India. In CHI, 2006. [29] P. Quinn, J. Goka, and H. Richardson. Assessment of an electronic daily diary in patients with overactive bladder. BJU International, 91(7):647– 52, May 2003. [30] J. Selanikio. Advantages of using handheld computers for data collection. DataDyne. http://www.datadyne.org/files/BriefHandheldAdvantages.pdf. [31] J. Selanikio and R. Donna. Datadyne brief. DataDyne. http://www.datadyne.org/files/DataDyne brief.pdf. [32] K. Shirima, O. Mukasa, J. Schellenberg, F. Manzi, D. John, A. Mushi, M. Mrisho, M. Tanner, H. Mshinda, and D. Schellenberg. The use of personal digital assistants for data entry at the point of collection in a large household survey in southern Tanzania. Emerging Themes in Epidemiology, 4(1):5, 2007. [33] D. Skinner, U. Rivette, and C. Bloomberg. Evaluation of use of cellphones to aid compliance with drug therapy for HIV patients. AIDS Care, 19(5):605–607, 2007. [34] SMS data management. Dimagi, Inc., 2008. http://www.dimagi.com/ content/sms-data-management.html. [35] B. Tiplady, G. Crompton, M. Dewar, F. Boellert, S. Matusiewicz, L. Campbell, and D. Brackenridge. The use of electronic diaries in respiratory studies. Drug Information Journal, 31(3):759–764, 1997. [36] I. Walker, C. Sigouin, J. Sek, T. Almonte, J. Carruthers, A. Chan, M. Pai, and N. Heddle. Comparing hand-held computers and paper diaries for haemophilia home therapy: A randomized trial. Haemophilia, 10(6):698–704, Nov. 2004.

A PPENDIX A-1. C OST

ANALYSIS

In our basic cost analysis we first assume that the treatment pool is 1000 patients. In our treatment program each worker is responsible for 10 patients, so there is a total of 100 workers. Each worker must be equipped with a cell phone. Our current development tools for electronic forms are tied to Motorola, and require a Java-enabled phone. The cheapest such phone in India is the Motorola L6i which is 75 US dollars. In contrast, both the SMS interface and voice interface can be used on any cell phone, one of the cheapest of which is the Motorola Motofone F3 ($26). Voice calls are slightly more expensive than text messages: a 3-minute voice call, which is longer than the average time in our experimental study, is about 3 rupees (0.065 US dollars, Airtel carrier). SMS messages using Airtel are 1.5 rupees per message (0.0327 US dollars). The average call length in our user study is 2 minutes and 20 seconds; therefore conducting 100 calls would require slightly under 4 hours. We therefore anticipate that a 100 call load would be reasonable for an operator working 8–9 hours per day, in order to include a liberal number of breaks. Our program design involves each worker visiting each patient to record symptom information every two weeks. At this rate a single operator

84

working five days per week could handle the 1000 calls over the two week period. Based on our experience in hiring a qualified operator in Bihar for $100 per month, we choose a conservative estimate of an operator salary of $200 per month. The length of treatment is six months. Given the assumptions above, we calculate the total cost per patient over the course of the treatment for each interface. Note that we are only focusing here on the aspects of the interfaces that lead to different costs and we are not considering the salary of the workers or additional program overhead. The cost of the phone per patient is simply the cost per phone multiplied by the number of workers divided by the number of patients. The cost of an operator per patient is the salary of the operator per month ($200), multiplied by the 6 month treatment length, divided by the number of patients, yielding a cost of $1.20 per patient. Workers will upload health information approximately 12 times per patient (once every two weeks). Therefore the cost of communication per patient is equal to the cost for each data entry (either SMS or a voice call) multiplied by 12. Table IV displays the cost breakdown per patient. Due to the high cost of phones that can support external applications, such as Java-enabled phones, voice is cheaper than electronic forms over a single 1000-patient program, even given the ongoing cost of an operator salary. SMS is the cheapest since it requires no operator and can be used with any phone. Perhaps most important is that the cost for each interfaces is less than $10, a small sum compared to the total cost of approximately $90–100 needed to treat a tuberculosis patient in India. A PPENDIX A-2. D ETAILED L OG Error Number 1 2 3 7 4 5 6 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

Interface Mode Forms Forms Forms Forms Forms Forms Forms Forms Forms Forms Forms Forms SMS SMS SMS SMS SMS SMS SMS SMS SMS SMS SMS SMS SMS Voice

Entry Type Multiple-choice Multiple-choice Numeric Numeric Numeric Numeric Numeric Numeric Numeric Numeric Yes/No Yes/No Multiple-choice Multiple-choice Multiple-choice Multiple-choice Multiple-choice Multiple-choice Multiple-choice Numeric Numeric Numeric Numeric Yes/No Yes/No Numeric

Entry Name Cough Cough Temperature Temperature Temperature Temperature Temperature Weight Weight Weight Fatigue Nausea Cough Cough Cough Cough Cough Patient ID Patient ID Temperature Weight Weight Weight Yellow eyes Fatigue Weight

Interface Forms Voice SMS

Fixed Cost $7.50 $2.60 $2.60

Marginal (Ongoing) Cost $0.39 $1.99 $0.39

Total cost $7.89 $4.59 $2.99

TABLE IV A PPROXIMATE COST PER PATIENT INCURRED BY EACH USER INTERFACE AS PART OF A 6- MONTH TUBERCULOSIS TREATMENT PROGRAM IN I NDIA . F IXED COSTS COVER THE PHONE , WHILE MARGINAL COSTS COVER TRANSMISSION VIA VOICE OR SMS, AND , WHERE APPLICABLE , THE CALL OPERATOR SALARY. H EALTH WORKER SALARIES DO NOT DEPEND ON THE INTERFACE AND ARE EXCLUDED .

This cost analysis assumes that we continue to use the Motorola L6i Java-enabled phone for the electronic forms interface. There are some cheaper Java-enabled phones that we may be able to use in the future, such as the $50 Nokia 2626, but this would require us to obtain new development tools. This would change the cost per patient for electronic forms to be $5.39. This still means that voice is less expensive than forms in terms of cost per patient. Also, the cost of voice could be further reduced by leveraging existing phones belonging to the health workers. While the above analysis is conducted for a specific program in India, informal data suggests that in some other countries voice may also be worth considering. For example, the average salary of call center operators in Peru is approximately 150 US dollars per month. The biggest cost considerations when comparing interfaces in new locations are likely to be the operator salary, the cost of voice calls compared to SMS, and the expected frequency and duration of conversations between workers and the operator. OF

A LL DATA E NTRY E RRORS

Correct Entry "mild" "heavy" 100.3 100.8 98.5 98.7 100.2 62 68 68 Yes No "1" (none) "1" (none) "1" (none) "3" (mild) "5" (severe) "6" (Akshaye Khanna) "7" (Anil Kapoor) 1003 54 62 69 "6" "0000007" 69

Actual Entry "none" "mild" 103.0 108.0 98 98.687 100.0 empty 67 93 No Yes "0" (disallowed) "0" (disallowed) "0" (disallowed) "0" (disallowed) missing "5" (Akshay Kumar) "1" (Aamir Khan) 103 45 826 59 "2" "000007" 59

Error Detectable?

Error Dangerous? X X X

X

X X X X X X X X X

X X X X X

X X X X X

85

FOLKSOMAPS - Towards Community Driven Intelligent Maps for Developing Regions Arun Kumar, Dipanjan Chakraborty, Himanshu Chauhan, Sheetal K. Agarwal, Nitendra Rajput IBM India Research Lab., 4, Block C, Institutional Area Phase II, Vasant Kunj, New Delhi - 110070 Abstract— Many services taken for granted in the developed world are often missing from the developing countries. One typical example is that of map systems that form the basis of several location driven services. Its heavy reliance on content, provides a huge barrier towards building such systems. Further, in developing countries like India, the infrastructure typically has a history of unplanned development, leading to unorganized addresses and absence of standard naming conventions for roads. Detailed map systems such as online maps have only recently started becoming available but for major cities. Remote towns and villages remain out of reach till date. In this paper, we investigate a community-driven approach for creating maps in developing regions - following Web 2.0 principles, but not entirely relying on the existing Web. Our system, dubbed F OLKSOMAPS is an intelligent, community constructed map system, particularly architected with developing regions in mind. We present the design of F OLKSOMAPS, followed by an implementation of our proof-of-concept system. We present user studies aimed at understanding the uptake, usability and utility of F OLKSOMAPS. The results indicate a strong need for such a community-generated mapping ecosystem.

I. I NTRODUCTION As defined in Wikipedia: “A map is a visual representation of an area’s symbolic depiction highlighting relationships between elements of that space such as objects, regions, and themes. Maps may represent any space, real or imagined, without regard to context or scale” The map systems in developed countries have advanced to a state where users can view street level information in 3D and annotate the maps with their own personalized content1 . Such systems include Google Earth2 and MSN’s Live Maps3 etc. Due to the prohibitive development cost involved, offering such systems becomes viable if profitable services can be offered on top of the core content base. The existing map systems generate revenue from services such as driving directions, finding local businesses and advertising. In contrast, sparsely populated semi-urban and vast rural areas of developing countries such as India do not have detailed map systems built for most locations. Further, the semi-literate, low income, Non-ITsavvy population [1] residing in these areas cannot use such services even if they 1 http://www.wikimapia.org 2 http://earth.google.com 3 http://maps.live.com

were offered in the manner done today. Furthermore, lack of stable infrastructure including electricity, internet connectivity and the lower purchase power of people, also complicate the situation further. The urban metropolitan cities, however, are beginning to experience such services as the factors affecting semi-urban/rural areas are not applicable there. Examples in India include MapMyIndia.com and MapsOfIndia.com. These websites provide detailed mapping information for metropolitan cities but contain only high level content for other areas. Even in metropolitan cities (about 100 million people stay in Indian metropolitan cities [2]), the (online) maps often do not contain enough detail/content to get driving directions from door-todoor. Secondly, lack of structured addressing conventions and poor road signs makes it difficult to follow the maps4 . So, even people comfortable with maps, often need to ask people on the streets to find their way. To overcome the prohibitive cost of developing and maintaining such map services for semi-urban areas, as well as to address the limitations of using maps in urban cities, we propose utilizing the collective efforts of the community who would be motivated to populate, maintain and access content for their benefit. In this paper, we present F OLKSOMAPS a community driven map system that leverages Semantic Web 5 technologies to create and manage a community generated knowledge base and makes use of web and voice applications [3] to provide access to its services. It is non-trivial to build such a system since several issues crop up. For instance, due to unplanned, historic development over centuries, cities, towns and villages in developing countries typically do not have well structured naming of streets, roads and houses. For example, postmen in villages often need to know the inmates by name in order to reach their houses. This leads to imprecise directions and key landmarks become very important in specifying locations and directions. We make use of such insights among other obtained from a user study for the design of our system.

4 A tiny segment using GPS navigation might get along in the cities, however they become handicapped outside urban areas due to lack of maps. 5 http://en.wikipedia.org/wiki/Semantic Web

86

II. S URVEY

OF

C URRENT M ODELS

In this section, we present results of survey done with endusers who depend upon location/map-based information for their daily business and/or personal needs. We investigate their current models to identify how they manage and use location information. A. Survey Process We selected the subjects considering their technical background, so that we get a good variety. We targeted two categories of people: (1) ITsavvy: People for whom computers and Internet is part of life (2) Non-ITsavvy: People who do not use computers and Internet (reasons range from them being less literate to being economically challenged) but use low-end cellphones (primarily for communication). We believe that such a mixed set would be able to provide us the right insights to evaluate the need for a community-driven map for developing regions. With the Non-ITsavvy class, we conducted the survey in a face-to-face interview mode, where specific questions were asked, albeit in a very informal and interactive manner. On the other hand, we circulated our questionnaire to the ITsavvy class and requested them to fill it up independently. The questions were targeted to understand the subject’s current model for finding landmarks and directions in the city. There were 21 questions in total. At a high level, our aim was to understand the following: 1) How do people find out points of interest (ranging from very small mom-and-pop shops to popular landmarks) ? 2) How much do they rely on maps or people on the streets? Are all their information needs satisfied by maps? 3) How do they provide location information (information about landmarks, directions) to other people ? 4) Would they be interested in consuming and producing information for a community-driven map system ? We surveyed a total of 40 subjects, with 24 from the ITsavvy category and 16 from the Non-ITsavvy category. The average age of the ITsavvy subjects was 26.37 years, ranging from 21 to 34 years. The average age of the Non-ITsavvy subjects was 32.68 years, ranging from 21 to 62 years. Most of the Non-ITsavvy subjects either did not have a formal education or primary education (10th standard) and were working in the city (security guards, car mechanics, cab drivers etc). The ITsavvy set consisted of engineers, editors, reporters, business owners etc. We summarize the key findings in Table I. TABLE I C URRENT M ODEL : S UMMARY OF U SER R ESPONSES Responses Use maps Rely on people to find shops Rely on people for directions Want precise directions Will upload content

Non-ITsavvy 0% 68.75% 87.5% 35.7% 87.5%

ITsavvy 66.6% 75% 67% 58.3% 79%

B. Results of Survey with Non-ITsavvy Subjects As expected none of the subjects in Non-ITsavvy category used maps to find information about locations. In fact, many were not even aware of the concept of maps. A majority of them mentioned that they currently rely on asking people on the street to know the location of a shop in an area. They ask people nearby or ask their friends or colleagues about the location. The others mostly find a shop on their own. 3 subjects mentioned that the information that they receive from people on the street are often not correct and not accurate, and this leads to frustration. About 87.5% of these subjects rely on other people to provide them travel directions. However, only a small section of these felt the need for precise directions (needed only in congested residential areas). About 75% people preferred the choice of a phone based interface and were willing to upload content. About 56% also mentioned that they would be willing to pay for such a service. Key Insights : Use of maps for Non-ITsavvy segment is nil and they rely primarily on others for travel directions. Most prefer a voice based interface and many are willing to pay for the call. Also, precise directions are not necessary for this user segment. C. Results of Survey with ITsavvy Subjects As shown in Table I, a majority of ITsavvy subjects make use of maps for finding locations and directions to locations. The interesting part is that even with maps, a majority of these people rely on other people for location related information. This is primarily because the unstructured nature of city layout coupled with broken, missing, faded, hidden (behind posters, graffiti) or even inconsistent sign boards make it almost impossible to rely entirely on a map for travel directions. Many a times, the maps do not contain fine-grained information in the first place. 2 out of 24 subjects mention that maps help them to find exact destination, while 19 mention that they rely on maps, only for a rough idea of the direction or rely on public transportation and mostly ask people (if road signs are not enough) for the exact location once they reach nearby their destination. Most subjects mentioned it would be helpful to have a phone-based location and direction finding system in addition to the web based interface, and about 79% expressed interest in contributing to the service by uploading content either over phone or through a web-based portal. Key Insights : Even though this segment makes heavy use of maps, they still are forced to rely on other people due to various factors. Many people prefer rough directions in the beginning (as they probably know the city) of their journey, and want detailed directions only towards the end. People rely a lot on community information, and asking people on the streets for directions and location is a common practice. III. U NDERSTANDING T ECHNOLOGY R EQUIREMENTS F OLKSOMAPS harnesses user-generated content about locations and aims to provide map-based services that repre-

87

sent user’s intuitive way of finding locations and directions in developing regions. We conducted an additional survey with the aim of understanding the technological requirements for F OLKSOMAPS. In this survey, we primarily focused on understanding the ways in which users express directions and location information. In addition, to provide the subjects with an idea of the system, we created a paper prototype that shows how a user would typically access the services and populate content to it. The paper prototype was an audio recording of system prompts and user responses. We created audio prototypes for two scenarios 1) User calls F OLKSOMAPS to find directions to a particular location 2) User calls to add information about a location These prototypes were created using audio recording and editing tools and were used to illustrate the concept of F OLKSOMAPS to some users. We conducted this survey along with the working model survey with a total of 40 subjects - 16 from Non-ITsavvy and 24 from ITsavvy community. Here is a sample of the questions we asked: • How do you give directions to people on the road or friends? • How do you describe proximity of a landmark to another one? • How do you describe distance? Kilometers or using timeto-travel? A. Non-ITsavvy Community When asked about how they personally give directions to anyone who asks them, 12 out of 16 subjects said they make use of landmarks to explain the direction to the destination. They use names of big roads to describe a location, and use “near to”, “adjacent to”, “opposite to” relations with respect to visible or popular landmarks to point the destination. 5 subjects said they can provide exact directions within one kilometer of the destination. 4 said that they usually give directions up to the nearest landmark thereafter which people will need to ask again. 9 subjects felt confident about guiding a person to the exact landmark. Interestingly, 6 of 16 subjects said that they use time (only) as metric to measure the distance between any two locations. 12 out of 16 subjects mentioned that they use either time or kilometers. 3 subjects mention that they sometimes use “rough” distance measures in terms of kilometers. One subject did not have the notion of kilometer as a measure at all. TABLE II C OMMON R ELATIONSHIPS USED TO EXPRESS RELATIVE LOCATION Relationships A isNear B A adjacent to B A opposite B

Non-ITsavvy 93.75% 81.25% 75%

ITsavvy 67% 45.8% 75%

B. ITsavvy Community Only 2 out of 24 subjects tell people to use maps to guide them to their house. The rest either use landmarks on roads

to guide the person. This came as a surprise to us, as we were expecting this community to be more reliant on maps. 10 out of 24 subjects usually are able to give exact direction to a known landmark using relationships as shown in table II. 13 subjects mentioned that they guide the person to the nearest big landmark and then guide the person by phone or pick the person from the landmark. 14 out of 24 subjects felt confident that people should be able to follow the way they guide. Table II shows the most commonly used relationships to describe proximity of a landmark to another one. Typically, while giving directions, such relations are used to relate a less visible or a less known landmark with a more visible or popular one. Interestingly, 21 out of 24 subjects either use both time and kilometers as a measure of distance and only 3 subjects claimed to use only kilometers to represent distance. 17 out of 24 participants never use zip codes while 6 use them rarely. This study, coupled with insights obtained from the survey of current models, has been useful to design the ontology as well as the output of F OLKSOMAPS- differentiating it from traditional map-based systems prevalent today. IV. T HE T ECHNOLOGY Based upon insights gained from the surveys we developed F OLKSOMAPS – a community generated map system. F OLKSOMAPS is designed to be populated by end users for their own consumption. This section presents the architecture and design of F OLKSOMAPS system while highlighting design choices that differentiate it from the established notion of map systems. They key differences are listed below. • It relies primarily on user generated content rather than data populated by professionals. • It strives for spatial integrity in the logical sense and does not consider spatial integrity in the physical sense as essential. For instance, information such as “Building A is located near to circle J after taking first turn on the circle while arriving from location B.” is treated complete and correct for tracing path from B to A. In other words, the direction and distance parameters are not specified in precise terms. This is because, as evident from the surveys, the end users are not likely to specify physical data while populating geographical landmarks. • A visual representation is not essential to F OLKSOMAPS which is important considering the fact that a large segment of users in developing countries do not have access to Internet. • F OLKSOMAPS is non-static and intelligent in the sense that it infers new information from what is entered by the users. • The user input is not verified by the system and it is possible that pieces of incorrect information in the knowledgebase may be present at different points of time. F OLKSOMAPS adopts the Wiki model and allows all users to add, edit and remove content freely. From the established Wikis on the Web we expect that the community would actively remove or edit invalid content and keep

88

the maps up-to-date. However, to limit malicious intent, the system places two minor restrictions described in the next section. A. Conceptual Design We use the notion of a landmark as the basic unit of representing nodes in F OLKSOMAPS. A location represents more coarse grained geographical area such as a village, city, country etc., in addition to also representing a landmark. The core knowledgebase of the system needs to capture few key logical characteristics of locations that users are interested in specifying and making use of. These include the following: • Direction i.e. the positioning of a location relative to another one. From the surveys we found out that users are comfortable with providing relative information such as ‘towards left of’, ‘on the right side of’ etc. instead of absolute direction in the form of north, east, west, south compass points. • Distance i.e. the measure of amount of space between two locations. This can be represented as numbers along with units in which the distance is expressed. From the surveys, we learnt that F OLKSOMAPS needs to consider both time and metric units to represent distance. • Proximity and Reachability i.e. representation of information stating that one location is in close proximity to another or is reachable from another respectively. • Layer i.e. granularity of geographic area that a location name represents. It could be a division as big as a whole country or as small as a village. The notion of direction and distance from a location, are interpreted with respect to the layer that the location represents. In other words, direction and distance could be viewed as binary operator over locations of the same level. For instance, ‘is towards left of’ would be appropriate if the location pair being considered is or say <South Korea, Japan> but not if the pair is <Sheraton waikiki hotel, Mexico> where Sheraton Waikiki hotel is in Honolulu, Hawaii. We model the knowledgebase for representing and storing these concepts in two parts. The first one makes use of Web Ontology Language (OWL) 6 to model the categorical characteristics of a landmark, i.e. direction, proximity, reachability and layer. Use of a Semantic Web language to represent relationships between locations brings in the advantage that the system can reason on those and infer newer relationships not explicitly specified by users of the system. The second part makes use of a graph database to represent distance between landmarks which is numerical data. The two modules are used in conjunction to generate answers to queries submitted by users to the system. B. User Interaction The user interaction aspect is critical for the success of F OLKSOMAPS. This is especially true since users would tend 6 http://www.w3.org/TR/owl-ref/

to query the system either when they are stuck on road looking for directions or before starting on a trip and would be pressed for time. Further, the user set also consists of people who might be illiterate or semi-literate or not very ITsavvy. Considering these, we discuss three different modes of interaction that the system needs to support to cater to different user segments for different tasks. There are three main tasks that a user can perform with the system. First is to find a landmark/location by specifying its name possibly including some related information such as nearby places or enclosing area. Second, users can ask for tracing a path between two locations. Third, users could add to the knowledgebase by adding information about a location/landmark that they know of. In addition, some or all users may also be given the facility to edit or remove entries from the knowledgebase. We also consider three sets of users who would interact with the system. On one extreme, the users of F OLKSOMAPS are ITsavvy people who can access it over the Web. F OLKSOMAPS provides a web interface to these users for submitting queries as well as to update its knowledgebase by adding new locations and related information. On the other extreme, we have users who are illiterate or semi-literate and cannot afford to have high end devices but can use an ordinary low end phone for voice communication. Studies done earlier [3], [4], [5] suggest that a voice-based interaction works well for this user segment and for them F OLKSOMAPS supports a voice based interface for querying the system. The third segment of our users lies between the two extremes and consists of mobile people with low end devices who are familiar with SMS. F OLKSOMAPS allows SMS based querying and location updates in a constrained form for these users. C. System Architecture Figure 1 shows the architecture of the F OLKSOMAPS system. As shown, users can upload content into the knowledgebase through an SMS interface, a web based interface or through a voice interface. Similarly, the content delivery to the consumers also happens through these multiple interfaces. The knowledgebase consists of an ontology and a graph database. An ontology is used as the primary repository of the location information. This is because the user generated content cannot be expected to be complete. It is essential to be able to infer facts not explicitly populated by users in order to have a pragmatic map system. The graph portion of the database captures additional information that either cannot be expressed appropriately in the ontology or needs to be processed differently. This includes numeric data such as distances between locations. The central block of the figure forms the core of the runtime system of F OLKSOMAPS, acting as an intermediary between the consumers of the service and the knowledgebase. It consists of a module each corresponding to the tasks listed above, namely location insertion/removal, location finding and path finding. It provides a similar interface to the users across interaction modalities for information upload and retrieval.

Web App

SMS G/W

SMS

Fig. 1.

Location Finding Module

Ontology API

Voice

Path Finder Module Location Insertion Module

Graph API

Voice Browser

89

Location Removal Module

Ontology

Graph DB



The System Architecture

In the next section, we provide some details of the design of F OLKSOMAPS knowledgebase. V. K NOWLEDGEBASE D ESIGN The F OLKSOMAPS knowledgebase consists of two parts. A graph database and an ontology of locations. The graph database is primarily a graph data structure based representation of the locations. The locations are represented by nodes and the edges between two nodes of the graph are labeled with the distance between the corresponding locations. Given the insights gained from user surveys, precise distances (and exact directions) are not key components of a map for our target users. Therefore, Graph DB is an optional component and we do not discuss it in this paper. The other, more important, part that makes F OLKSOMAPS intelligent, is the ontology of locations that helps construct paths and retrieve information that no user may have explicitly entered. This is what we describe next. A. The Location Ontology Location

Country

CountryCapital

Union Territory

State Division StateCapital

Bootstrapped

District

Pargana Tehsil

Legend

City

PostOffice

PartOf isA

Town

Area

Village

SubArea

User-generated

Landmark

Fig. 2.

The Folksomaps Ontology Design

Figure 2 depicts the location ontology that we created. As shown, all concepts in the location ontology derive from concept Location. The highest level concept that the ontology currently represents is Country. The rest of the concepts are defined specific to India keeping in view the administrative structure of the country 7 . 7 http://en.wikipedia.org/wiki/Subdivisions

of India

Not shown in the figure is another concept labeled Space. It is defined as complementary and disjoint to Location. This became necessary since OWL ontologies follow the open world assumption8 which means that a relation not explicitly asserted in the ontology being reasoned upon cannot be concluded to be false since it may be specified elsewhere.



Fig. 3. •

OWL definitions of landmark class and partOf property

Relationships Each location can be related to other locations to logically represent the geographical relationship that exists between them in the world. As can be observed from the figure, apart from isA relation of all locations with the Location concept, each location concept is related to one or few location concepts through a partOf relation. This relation helps establish the layering between various geographic locations, as identified in Section IV. Landmarks lie at the lowest strata of the ontology. Since the administrative structure of a country remains relatively static and is generally well known, the top part of the ontology can be bootstrapped in advance and users can be allowed to contribute instances from the lower half consisting of landmarks and sub areas. Also, some key landmarks such as historic sites of national importance and key government offices such as the parliament could also be pre-populated into the ontology. The ontology supports several relationships in order to be able to specify the four key characteristics of a location defined in Section IV. In the interest of space we present here details of only two key relationships – nearTo and connectedTo that model the proximity and reachability characteristic respectively. Relation nearTo is a symmetric relation defined between two locations to express the fact that they are in close proximity to each other. This could intuitively mean a few hundred meters or a kilometer. The logical integrity of nearness can be applicable to locations other than landmarks. For example, two towns in the same district can be considered nearTo each other, compared to two towns in different districts. To capture these, for SubAreas a sameTown relation and for Areas a sameDistrict relation and so on are defined. However, for the purpose of this paper, we stick to the basic nearTo relation between landmarks. All the modules make use of this relation

8 http://en.wikipedia.org/wiki/Open

World Assumption

90

This subsection describes the API that we built for accessing the knowledgebase for finding path, location or for adding a landmark. 1) findLocation() : This method allows a user to search for a location specified by its name. Optionally, extra information can be supplied which includes the landmark’s relationship with another landmark or its attributes. 2) findPath() : This method takes source location name and destination location name along with a filtering criteria and returns a list of locations that represent the path traversal from source to destination. The location names can optionally be augmented with a list of nodes that represent meta information about the position of the source or destination node in the ontology. The filtering criteria specifies additional restrictions (e.g. only traverse paths connected by a nearTo relation) on the path traversal algorithm. 3) doesExist() : This method determines whether the supplied landmark name already exists in the knowledgebase. 4) addLandmark() : This method allows the user to insert a new landmark into the knowledgebase. It takes the name of the new landmark and also its immediate parent, i.e. its SubArea name. Optionally, its next parent, i.e. PostOffice is also supplied. If this landmark already exists, this

In the next section, we describe our prototype implementation that we used to conduct user studies. VI. P ROOF - OF -C ONCEPT I MPLEMENTATION We have implemented a prototype of F OLKSOMAPS and deployed it at our lab. The prototype’s knowledgebase includes the ontology module and does not have the optional Graph module. We used OWL to implement the ontology. We implemented the F OLKSOMAPS modules for finding a location, finding a path and adding a landmark. The ontology API used is JENA9 with Pellet reasoner10. We bootstrapped the system with data about New Delhi, the capital of India and initialized it with SubAreas located in a couple of Areas under South Delhi district. We implemented a Web based interface as well as a Voice based interface for this prototype. The Web based interface supports all the implemented modules and is developed using Java Server Pages (JSP)11 . On the other hand, the Voice based interface is accessible over a phone call and supports find location and find path modules. It is developed using JSPs and VoiceXML12 . Users are allowed to populate the F OLKSOMAPS system with new landmarks and associate them to the SubArea which they belong to. Additionally, users can also provide information about the landmark. This includes other landmarks located near to the one being added, and other landmarks that are connected to this by road etc. Figure 4 shows a partial snapshot of the populated Folksomaps knowledgebase. New Delhi (Capital)

India

South Delhi (District)

VasantKunj (Area)

Munirka (Area) VasantVihar (Area)

Sector B (SubArea) Badam Market

VasantLok Market

Sector C (SubArea)

partOf

ISID GDGoenka School

isNear

PVR Saket

Anchal Plaza

connectedTo Country/Capital/District /Area/subarea

Saket (Area)

Mahipalpur (Area)

PVRPriya

JNU

Malvianagar (Area)

GoldenDragon Hotel Sarvapriya Vihar

GrandHotel

Landmark

Fig. 4.

Bootstrapped Data

B. Knowledgebase API

new entry is rejected. Also, this landmark is added only within the context of its parent, i.e. it’s SubArea. If the SubArea supplied does not exist, then also the landmark is rejected. In addition to the landmark name, the user can optionally specify other meta information such as which all landmarks are located nearby and to which all landmarks is this new one is connected to. 5) editLandmark() : This method allows you to search for a landmark or a relation instance and allows you to delete it in the fashion similar to addLandmark().

A partial snapshot of populated ontology

9 http://jena.sourceforge.net/

10 http://pellet.owldl.com

11 https://java.sun.com/products/jsp

12 http://www.w3.org/TR/voicexml20/

User Populated Landmarks

•

to search for or add/update locations specified in user’s query. The connectedTo relation is a symmetric as well as a transitive relation. It expresses the fact that two locations are reachable from each other via one or more paths that can obtained from the ontology facts. Users may add connectedTo relation between location instances that they are familiar with. F OLKSOMAPS makes use of the ontology to infer new connectedTo relations based upon these individual assertions. The Path Finder module relies on this information to compute paths from the individual connections supplied by users. Bootstrap Process The F OLKSOMAPS system could be bootstrapped from existing databases to populate instances of location types in the upper part of the ontology. Two such sources of data in the absence of a full-fledged Geographical Information System (GIS) system come from the Telecom Industry and the Postal Department. While postal department is obvious, the telecom companies also maintain database of various circles that they operate in. Given that mobile phones have penetrated into the remote rural areas as well, the upper ontology can be populated from their data as well. While the actual GIS data benchmarking require significant efforts and cost on the field to map the spatial data, the telecom and post office data give a very good logical view of the locations. This complements our system’s design goal of providing a logical view rather than a spatial view to the users.

91

Users are also allowed to query F OLKSOMAPS for getting location information and directions. Figure 5 shows the screenshot of webpage for getting directions from a source to destination location. Figure 6 shows the flowchart for querying through the Voice based interface. Error steps are not shown in this flowchart. The results from F OLKSOMAPS reflect the ways in which people would essentially give directions.

Fig. 5.

Web UI showing the results of querying for directions

Welcome to

website. In addition to the bootstrapped data, the knowledgebase also contained some pre-populated locations. We requested them to populate content into F OLKSOMAPS while restricting the locations to a set of 6 Areas in South Delhi District. This was done so that the content populated does not get thinly spread out and is relatively rich for querying 14 . After populating some landmarks known to them, users then queried the system for finding information about other landmarks and travel directions to those. After this, we asked them a few questions about their view of F OLKSOMAPS system. For the Non-ITsavvy subjects, we briefed them about the purpose of the proposed system and gave an explanation of the prototype. We then walked them through the voice based interface, by querying for some location and requesting for a path to that location from another. This was followed by a question answering session. All subjects grasped the concept fairly quickly and were able to see the benefits they could derive from such a system. B. Results of Survey for Non-ITsavvy subjects We conducted a total of 22 surveys with Non-ITsavvy participants using the voice interface. The set of interviewed people consisted of porters, security guards, elderly people, draughtsmen, waiters and service staff. The results are tabulated below.

Folksomaps

Source Location?

Get Src_Name

Destination Location?

Get Dest_Name

Play Path Details

Path

Find 'Location' or 'Path' ?

Location

Location Name?

Get Loc_Name

Play Location Details

Sample User Interaction System: Welcome to Folksomaps. Using this service you can find information about locations in Delhi. To find information about a location say Information, to find a route between two locations say Route User : Route System: Please speak the source location User : IIT System: Please speak the destination location User : ParkBalluchiRestaurant System: You can go from IIT to SDA. From SDA you can go to GreenParkMarket From GreenParkMarket you can go to ParkBalluchiRestaurant

Fig. 6. Voice UI flow for querying F OLKSOMAPS. Red dots indicate voice recognition steps.

VII. S OLUTION S URVEY In this section, we present results of user studies we conducted to verify the benefit and acceptability of the proposed system. We further present insights that we obtained from users while conducting the survey. For the ITsavvy segment, we let the subjects try F OLKSOMAPS through the web based user interface of the system. For the Non-ITsavvy segment, we conducted the survey with the voice based interface of F OLKSOMAPS13 . A. Survey Process For ITsavvy survey participants, we started with a small introduction before giving them access to the F OLKSOMAPS 13 The voice based interface was in Hindi language and allowed a restricted set of landmarks to keep the speech recognition accuracy high for the prototype

Fig. 7.

Interviewing the Non-ITsavvy users

TABLE III S UMMARY OF N ON -IT SAVVY U SER R ESPONSES TO F OLKSOMAPS Questions Would you call to get directions? Prefer calling over asking people on the street? Ready to pay for call (else want ads)? Will upload content? Voice Interface preferred over SMS? Results need to be very accurate?

Yes 100% 82% 45% 73% 91% 86%

No 0% 18% 55% 27% 9% 14%

As can be seen from Table III, all the subjects surveyed were interested in using F OLKSOMAPS system. Most of them preferred the option of calling up a number for directions rather 14 We envision that the actual content in a deployed system will be much more richer than the content populated by the survey participants.

92

than asking someone on the street. Apart from convenience, this overwhelming response can be attributed to a couple of reasons. First, asking on the street does not always work. On a secluded road, there may not be anyone to ask. Several times people end up giving wrong directions (possibly to hide their lack of awareness or to not appear rude) leading to precious time spent traveling on a wrong route and recovering from it. Also, the point of view of street vendors who often get enquiries from passersby about directions is quite interesting. They candidly admitted that during the course of the day they get so many such queries for detailed travel instructions that it is easy to get irritated and shrug them off. Second, an important insight provided to us by our subjects was the sense of security that they would get with such a system. We were informed that even though asking for travel directions from strangers on the street is an option, it exposes the enquirer to criminal elements, who often take advantage of their lack of knowledge and use it to rob them, the activity sometimes resulting into a worse situation such as a murder. This is especially true for first time visitors to the city from remote towns and villages or female citizens traveling at odd hours of the day. Many people were willing to pay for the call even a small premium over normal charges as they saw value to having this information available to them at all times. Yet a majority of the subjects preferred the advertisement model where an advertisement played in the beginning of the call pays for the entire call. This is understandable, given this segment’s high sensitivity to cost. A few users suggested that the advertisement model was better since most of these users primarily have pre-paid SIM cards and often they do not have sufficient balance to make outgoing calls. Most people were willing to upload content, though a few refused as they were hesitant due to not owning a phone. Almost everyone preferred the voice based interface over SMS even though we demonstrated speech recognition errors during the study interviews. The primary reason for this cited by them was that many people are either not comfortable using SMS or not comfortable using a mobile phone itself. However, some users who were well versed with SMS preferred it over voice. In terms of accuracy of returned results, most people asked for full accuracy while a very few were okay with minor mistakes. The need for strong accuracy is driven by the fact that most of these people either use public transport, or use a bicycle or even walk to reach their destination. The cost of a wrong input for them is huge compared to a person driving in his own or rented vehicle. In fact, one of the main reasons for preferring a voice call over asking people for directions was to avoid wrong directions. This is an important feedback since we started with the assumption that we do not need strict controls over the content and the wiki model would work. But the tolerance for incorrect information is low and we need to factor this in. We also learnt that meta information is as important to NonITsavvy users as the landmarks themselves. For instance, in

cities, more than the road route from a source to destination, people from the underprivileged segment were more interested in knowing the bus route numbers that could take them to their destination. Road routes serve well those people who travel by their own vehicles but the underprivileged rely primarily on public transportation. Similarly, for rural areas that consist of remotely located towns and villages, what helps the underprivileged people is information regarding modes of transportation (train, bus, boat, cycle-rickshaw, taxi etc.) to take from source to destination, where to make a switch and estimated travel time. Time tables of these public transport mechanisms are another important feature for this segment that can become an essential part of F OLKSOMAPS. Key Insights : We realized that accuracy of the information is a key requirement and more the meta information available, merrier it would be for these consumers. Also, voice based interface is indeed a preferred mode for this user segment over SMS and the calls to the system should to be free of cost. C. Results of Survey of ITsavvy subjects For ITsavvy segment, we conducted the survey with a total of 15 subjects using the web based interface. We also told them that the system has a voice based interface available over a phone call and supporting similar API. The user list consisted primarily of software professionals apart from a couple of businessmen. As expected, the ITsavvy community had significant experience in using the current online maps of cities in Indian metropolitan cities and were able to carefully evaluate our approach, considering the map services that are already operating in metropolitan cities. We try to capture learnings from their feedback. TABLE IV S UMMARY OF IT SAVVY U SER R ESPONSES TO F OLKSOMAPS Questions Would you access it for directions? Prefer F OLKSOMAPS over asking people? Ready to pay for call (else want ads)? Will upload content? Prefer Web for upload? Results need to be very accurate?

Yes 93% 87% 67% 87% 92% 53%

No 7% 13% 33% 13% 8% 47%

As is evident from the results (Table IV), most survey participants mentioned that they would like to use this service and that it would certainly be more convenient than asking people around in the streets. Interestingly, a bulk of the ITsavvy community did not stress on getting fine-grained direction all the time. They were fine with getting high level directions involving major landmarks. Most people were fine with paying for the service when offered on phone. Most were also willing to upload content into F OLKSOMAPS but preferred to do so over a web based interface as opposed to SMS or a voice based interface. Accuracy of responses was important for this segment as well, though not as strongly as for the Non-ITsavvy segment. As discussed earlier, this segment typically used their own vehicles and need high level directions rather than precise route.

93

A few participants pointed out that while voice-based access is good for interactive session, they would still prefer the content is sent to them via SMS so that they can store it for future access, pointing out that it is easy to forget the directions if you just hear it. However, this segment had other expectations from the system. Some subjects mentioned that the system should adapt to the user’s request and have the ability to produce fine-grained details depending on whether the destination is a popular landmark, or an area or a house in a colony. Few others mentioned that the directions provided by F OLKSOMAPS should take into consideration the amount of knowledge the subject already has about the area, i.e. it should be personalized based upon user profile. One subject mentioned that the current prototype appeared more suitable for driving directions but not for pedestrians. Just like the Non-ITsavvy community, the ITsavvy community also reflects the need for meta information on such a community-driven map. One subject mentioned that he would be interested in uploading traffic information on routes and would benefit from community uploading such information on the system. A few subjects mentioned that frequent changes in road plans due to constructions should be captured by such a system - thus making it more usable than just getting directions. Key Insights : While accuracy and convenience score with IT-Savvy population as well, this segment turned out to be more ambitious in terms of deriving benefits from such a system. Based upon the feedback listed above, we learnt that the user interfaces of F OLKSOMAPS needs to be rich and adaptive to the information needs of the user when considering this community. It also appears to the authors that dynamic and real-time information augmented with traditional services like finding directions and locations would certainly add value to F OLKSOMAPS. VIII. D ISCUSSION F OLKSOMAPS we believe, presents a novel approach towards developing a self-sustaining map system, harnessing community input, particularly targeted towards developing countries, where there is a need for such a system. Reasoners using ontologies consume space and compute power. [6], [7], [8] reports ways through which spatial reasoning can be made faster. This is an issue with F OLKSOMAPS as well. However, F OLKSOMAPS as a system compliments this body of work as its focus is not on improving reasoning capabilities or address scalability needs of underlying ontology reasoners. Rather F OLKSOMAPS can benefit from this body of work by adopting the solutions suggested to improve the computational and reasoning efficiency. Given their preference towards voice based interface over SMS [9], [10], designing an efficient and user-friendly voicebased user interface for the masses is important for F OLK SOMAPS . For example, while finding directions, user interface should be designed in a way that facilitates users to specify the level of detail they are looking for, varying from source to the

destination. Voice interface also takes care of the language barrier since content can be delivered in local language as demonstrated in other systems [3]. Voice based interfaces are, however, constrained with the capability of speech recognition technology which is under slow but constant improvement. Our surveys indicate that most people would like to contribute to F OLKSOMAPS knowledge base. However, in reallife, there are several factors that provide impedence for a user to be an effective information producer. Reasons range from users becoming busy, loosing interest after an initial surge, etc. In a live deployment of F OLKSOMAPS, one needs to also consider pragmatic business models (such as bartering models, advertisements or incentives) using which an in-flow of information can be sustained to keep improving the quality of system responses. Accuracy of results being of primary concern universally, further research is needed to ensure that various modules of F OLKSOMAPS would guarantee correct and precise results given that the data input by the users is correct in the first place. Also, as the knowledgebase as well as the userbase grows, established scale-up techniques would have to be applied for real life environments. IX. R ELATED W ORK Two research areas that are very relevant for F OLKSOMAPS are the areas of research in use of IT for underprivileged in developing regions and semantic tools for geographic information systems. Apart from these, work in the area of intelligent user interfaces for masses in developing regions is also relevant. There is a lot of literature on means to harness available information and user generated content [11], [12], [13] to deliver useful services to underprivileged in developing regions. [14], [15] talks about voice-driven technologies (e.g. audio wiki) to capture user content from Non-ITsavvy masses. The Neighbourhood Mapping [16]15 initiative proposes the involvement of school students to gather community input in the context of building maps. The project used PDAs coupled with GPS to build an information repository that could be used of planning purposes. F OLKSOMAPS builds further along this direction and proposes to create alternatives to well established solutions in developed countries, solely through user generated content. Specifically, it focuses on creating a framework (exploiting ontological reasoning), where by geographic information can be captured, enriched, and funneled back to the masses - customized to the needs of developing regions. To the best of our knowledge, we are not aware of any such community-driven map system for developing regions. There has been considerable work on Place ontologies, retrieval and storage of geographical information using ontologies [8], [17], [18], [6], [7]. For example, [8] talks about limitations of OWL to support spatial reasoning, integrity rules, and proposes a combination of spatial data-based store 15 http://www.csdms.in/NM/

94

and ontology-based reasoning to better represent geographic information and resources. [6] further critically evaluates ontology-based approaches towards geographic information retrieval while [7] presents a spatially aware search engine, for semantic interoperability of distributed and heterogeneous GIS on the Internet. In contrast to this literature, F OLKSOMAPS does not focus on improving reasoning capabilities of OWL or address spatial and logical integrity issues of Place ontologies. Rather, we focus on designing concepts of an ontology that is suitable for capturing map information from communities, keeping only logical integrity in mind, and by exploiting currently standardized semantics of OWL. F OLKSOMAPS in a way is hence complimentary to this body of work and can benefit from it. A body of research focuses on intelligent user interfaces for developing regions [19], [10], [20] and infrastructures to take IT services to the economically challenged and NonITsavvy masses in developing regions [3], [21]. F OLKSOMAPS at its core can certainly benefit from user interface designs to facilitate upload and download of map information. The system, by having multi-modal front-ends fits well with the architectural principles outlined in [3], [21]. X. C ONCLUSION In this paper, we investigated feasibility of a communitydriven approach towards creating maps for developing regions. Our system dubbed F OLKSOMAPS, has the potential of providing an effective alternative to expensive map solutions using community input, making map-based services (finding directions, finding locations and landmarks) available to people in developing regions where such services are currently missing. F OLKSOMAPS builds on the current models adopted by users in developing regions and leverages their collective knowledgebase thus overcoming the huge cost barrier in developing such a system. It is specifically designed to provide content that is intuitive for the users. We conducted a total of about 77 interviews in the process of evaluating a need for such a system and testing our prototype for verifying usability and utility of F OLKSOMAPS. Our surveys suggest that the community is very receptive towards the concept of a community-driven map as that alleviates some of the problems (reliance on people, security, inaccurate directions, etc) they face in day-to-day life. We intend to build further upon this system and overcome its current limitations to bring it even closer to users’ expectation. ACKNOWLEDGMENT The authors would like to thank Amit Nanavati for suggesting the ’Folksomaps’ name. R EFERENCES [1] “The Use of ICTs by Small and Informal Businesses,” http://research.microsoft.com/ jdonner/ Papers/donner micros slides.pdf. [2] “World Gazetteer. India - Metropolitan Areas,” http://www.worldgazetteer.com/wg.php?x=&men=gcis&lng=en&dat=80&geo=104&srt=pnan&col=aohdq&msz=1500&va=&pt=a.

[3] A. Kumar, N. Rajput, D. Chakraborty, S. Agarwal, and A. A. Nanavati, “WWTW: A World Wide Telecom Web,” in ACM SIGCOMM Workshop on Networked Systems For Developing Regions, Aug 2007. [4] M. Plauche and M. Prabaker, “Tamil Market: A Spoken Dialog System for Rural India,” in Working Papers in Computer-Human Interfaces (CHI), 2006. [5] T. S. Parikh, “Mobile Phones may be the Right Devices for Supporting Developing World Accessibility, but is the WWW the Right Service Delivery Model?” in International Cross-Disciplinary Workshop on Web Accessibility (W4A), Scotland, May 2006. [6] A. Abdelmoty, P. Smart, C. Jones, G. Fu, and D. Finch, “A critical evaluation of ontology languages for geographic information retrieval on the Internet,” Journal of Visual Languages and Computing., pp. 331– 358, August 2005. [7] M. Kun and B. Fuling, “An Ontology-Based Approach for Geographic Information Retrieval on the Web,” in International Conference on Wireless Communications, Networking and Mobile Computing, Sept 2007, pp. 5959–5962. [8] A. Abdelmoty, P. Smart, and C. Jones, “Building Place Ontologies for the Semantic Web: Issues and Approaches,” in Proceedings of the 4th ACM workshop on Geographical Information Retrieval, Lisbon, Portugal, Nov 2007. [9] “Africa:Impact of Mobile Phones,” Vodafone Policy Paper Series, Mar 2005. [10] A. Kumar, N. Rajput, D. Chakraborty, S. Agarwal, and A. A. Nanavati, “Voiserv: Creation and delivery of converged services through voice for emerging economies,” in In Proceedings of the WoWMoM, Finland, June 2007. [11] R. Abraham, “Mobile Phones and Economic Development: Evidence from the Fishing Industry in India,” in IEEE/ACM International Conference on Information and Communication Technologies and Development (ICTD), Berkeley, USA, May 2006. [12] K. Ramamritham, A. Bahuman, C. B. S. Duttagupta, and S. Balasundaram, “Innovative ICT Tools for Information Provision in Agricultural Extension,” in IEEE/ACM International Conference on Information and Communication Technologies and Development (ICTD), Dec 2006. [13] P. K. Reddy, G. Ramaraju, and G. Reddy, “eSaguTM: A Data Warehouse Enabled Personalized Agricultural Advisory System,” in ACM International Conference on Management of Data, China, June 2007. [14] P. Kotkar, W. Thies, and S. Amarasinghe, “An Audio Wiki for Publishing User-Generated Content in the Developing World,” in HCI for Community and International Development (Workshop at CHI 2008), Florence, Italy, April 2008. [15] L. Wang, P. Roe, and B. Pham, “An Audio Wiki Supporting Mobile Collaboration,” in Proceedings of the ACM Symposium on Applied Computing (SAC), Brazil, March 2008. [16] R. Mallick, H. Kalra, and D. Banerjee, “Infusing map culture through participatory mapping,” GIS@development. http://www.gisdevelopment.net/magazine/years/2005/feb/infusing.htm, Feb 2005. [17] J. Kay, W. Niu, and D. Carmichael, “ONCOR: Ontology and Evidence based Context Reasoner,” in Proceedings of International Conference on Intelligent User Interfaces (IUI), Honolulu, Hawaii, Jan 2007. [18] G. Look and H. Shrobe, “Towards Intelligent Mapping Applications: A Study Of Elements Found In Cognitive Maps,” in Proceedings of International Conference on Intelligent User Interfaces (IUI), Honolulu, Hawaii, Jan 2007. [19] I. Medhi, A. Sagar, and K. Toyama, “Text-Free User Interfaces for Illiterate and Semi-Literate Users,” in IEEE/ACM International Conference on Information and Communication Technologies and Development (ICTD), Berkeley, USA, May 2006. [20] J. Sherwani, S. Tomko, and R. Rosenfeld, “Sublime: A Speech- and Language-based Information Management Environment,” in In Proc. ICASSP, May 2006. [21] A. Kumar, N. Rajput, S. Agarwal, D. Chakraborty, and A. A. Nanavati, “Organizing the Unorganized - Employing IT to Empower the Underprivileged,” in Proceedings of 17th ACM International Conference on World Wide Web(WWW), Beijing, China, April 2008.

95

HIV Health Information Access using Spoken Dialogue Systems: Touchtone vs. Speech Aditi Sharma Grover, Madelaine Plauché, Etienne Barnard, Christiaan Kuun

Abstract— This paper presents our work in the design of a SDS for the provision of health information to caregivers of HIV positive children. We specifically address the frequently debated question of input modality in speech systems; touchtone versus speech input, in a new context of low literacy users and a health information service. We discuss our experiences and fieldwork which includes needs assessment interviews, focus group sessions, and user studies in Botswana with semi and low-literate users. Our results indicate user preference for touchtone over speech input although both systems were comparable in performance based on objective metrics. Index Terms— Spoken dialogue systems, DTMF, Touchtone, Speech interfaces, Health information, HIV, Illiterate users, Semi-literate users, Low literate users, ICT, Developing regions, Information access, Africa. I.

INTRODUCTION

There is a widespread belief that spoken dialogue systems (SDSs) will have a significant impact in the developing world [1]. This belief is based on a number of factors. Firstly, illiteracy is predominantly a problem of the developing world (according to a recent estimate, about 98% of the illiterate people on earth live in the developing world [2]), and speechbased access to information may enable illiterate or semiliterate people to participate in the information age. Also, the availability of traditional computer infrastructure is low in the developing world, but telephone networks (especially mobile/cellular networks) are spreading rapidly [3]. (For example, a recent community survey in South Africa [4] found that 73% of households owned at least one mobile phone, but only 7% of homes had internet access.) A further factor is the strong oral culture that exists in many traditional societies, which is likely to render such systems more acceptable than text-based or graphical information sources. Finally, the availability of relevant services and alternative information sources is often low in the developing world. Manuscript received September 22 2008. This research was supported by OSI, OSISA and the NRF under the Key International Research Capacity (KISC) programme, UID no. 63676. (Corresponding author) A. Sharma, E. Barnard and C. Kuun are with the Human Language Technologies (HLT) Group in the Meraka Institute, CSIR, Pretoria, South Africa. (phone: +27128413028, email: @csir.co.za) M Plauché was a visiting researcher at the Meraka Institute during the project. (email: [email protected])

Based on this perceived value of SDSs in the developing world, a number of exploratory studies have been performed in recent years. Barnard et al. [5] report on preliminary experiments performed to assess the usability of a telephonebased information service for access to government information in South Africa. A kiosk-based SDS for agricultural information was developed by Plauche et al. [6], and evaluated with semi-literate users in rural Tamil Nadu, India. Nasfors [7] also developed an agricultural information service, aimed at mobile telephone users and deployed in Kenya. The most sophisticated speech technology in this category was employed in the telephone-based information service for community health workers developed by Sherwani et al. [8]; this was piloted in Sindh, Pakistan. Agarwal et al. [9] have implemented a telephone-based kiosk system, which they call “VoiKiosk”; this is being trialled in rural villages in Andhra Pradesh, India. Each of these pioneering studies was primarily aimed at assessing the feasibility of using speech technology in various settings in the developing world. However, in the process of determining feasibility, a number of practical lessons were also learnt. For example, it was found that user acceptance of such systems is proportional to the difficulty that users would have to access the same information through other mechanisms [6] (thereby confirming the concept of the “motivated user”), and two studies [5, 8] found that it may be preferable to use more verbose, less efficient user interfaces to guide inexperienced users for whom time pressure is not a primary concern. The current contribution similarly has a twofold aim: both to explore the use of an SDS in a new environment (namely, by caregivers of HIV positive children in Botswana, Southern Africa), and to contrast different input modalities used in such a system. In particular, we compare systems using key-presses (“DTMF or Touchtone”) with those that use automatic speech recognition (ASR) for user input in this application. To this end, we start with the most basic variant of speech systems (key-press replacement), and compare such systems with DTMF input. The motivation for this choice is twofold: on the one hand, key-press replacement is likely to be more acceptable in the developing world, where general numeracy is less common; on the other, such systems are much easier to develop than natural-language systems, and are therefore more attainable in the resource-constrained environments that typically characterize the developing world. To our current knowledge DTMF and ASR input modalities have not been compared systematically in the developing world. However, in a review of developed-world applications

96

comparing DTMF to ASR input by Lee and Lai [10], both user preference and performance are found to depend on the nature of the task, the personality of the user, and the capabilities of the speech-recognition system. With the exception of only one call-routing system [11], all studies found that simply replacing key-presses with speech does not improve user performance or perception. Conversely, wellengineered speech recognition systems with natural-language input are often preferred to DTMF for tasks that are not easily accomplished with DTMF. In terms of user preference, studies in laboratory settings [10], [12] report that users find speech input more interesting and enjoyable to use, but contrastingly in a recent informal poll of over a thousand users of realworld information-access systems [13], almost half of the users responded that they would prefer to use a speech input modality “as little as possible” and only 8% would do so “most of the time”. Despite these reservations, numerous applications that are completely reliant on speech input are currently in use in the developed world - examples are the health-management systems described by Migneault et al. [14] and commercial voice portal systems, such as the Tellme portal (which serves 40 million phone calls per month, according to its providers [15]). Below, we first provide background on the health-care application selected for our study and describe the system that was developed as well as the experimental protocol employed (Section II). Section III contains our experimental results, including user profiles, usability measurements and task completion rates. Finally, we discuss the scope and generalizability of our results, and conclude with thoughts on next steps to be taken along this research trajectory.

primary and specialty medical care for HIV/Aids, to catering for the psychosocial needs of HIV patients and their families [18]. Baylor provides treatment to over 2100 children infected with HIV and 260 families across Botswana. The centre is involved in a number of support activities such as community outreach programmes for patients in rural areas, servicing 20 communities outside Gaborone and a “Teen Club” to provide moral support and counselling to teenagers living with HIV [18]. A child is typically brought to Baylor to be tested for HIV/Aids by a caregiver. A caregiver is any individual who takes care of an HIV positive child; it may be the child’s parents (who themselves might be HIV positive), other family members or an unrelated community member. Children who test as positive receive free treatment from Baylor for the remainder of their infancy and adolescence. Caregivers of such children are also counselled and trained. Baylor provides free lectures for caregivers three times a week, where many aspects of HIV/Aids, antiretroviral (ARV) medication are explained and advice is given on how to live with the condition. Each caregiver on average attends two lecture sessions. The primary focus of the lecture sessions in Baylor is on adherence to ARV medication, with topics such as the principles of HIV, universal precautions, basics of ARV therapy, medication dosage, side effects and storage, and importance and strategies for adherence, being covered.

II. OPEN PHONE: HIV/AIDS HEALTH INFORMATION LINE IN BOTSWANA A. Background HIV/Aids is perhaps the gravest health pandemic to face the world, and Southern Africa has been the worst hit region. Of the 33 million people infected with HIV worldwide, approximately two-thirds are inhabitants of sub-Saharan Africa [16]. Within sub-Saharan Africa, Botswana has one of highest HIV prevalence rates, with 1 in every 4 adults being HIV positive [16]. The hardest hit in Botswana are women; nearly 40% of pregnant women (ages 25-39) are living with HIV and infection levels are increasing amongst pregnant women aged 30-34 years, with nearly one in two living with HIV [17]. Aids deaths have orphaned approximately 120 000 children (ages 0-17) and another 14 000 children (aged 0-14) are living with HIV in Botswana. During our study a partnership was established with The Botswana-Baylor Children’s Clinical Centre of Excellence (hence forth referred to as Baylor). Baylor is a specialised paediatric institute serving the Botswana capital, Gaborone and its neighbouring areas since June 2003. The centre is staffed collaboratively by U.S. and Botswana health professionals. The services provided by Baylor range from

Fig. 1. Baylor in Gaborone, Botswana.

B. Open Phone Development Open Phone is a pilot HIV/Aids community-oriented SDS service that makes use of language technologies to address acute informational needs of caregivers of children with HIV. 1) Preliminary Investigations Our initial investigations started in April 2007, where we conducted interviews and discussions with 2 doctors, 4 nurses, 9 caregivers, a social worker and the technical manager. We also accompanied 3 community outreach workers on visits to 2 caregivers’ homes. The intention of this investigation was to identify specific needs of the various user groups and their day-to-day tasks.

97

This trip also served to acquaint us with HIV domainspecific terminology used by the interviewees to describe their situations, tasks and other notable work processes. These ‘hidden’ pieces of information allowed us to build a profile of the individuals interviewed and make informed choices later in the process of application design.

These challenges and issues formed the basis of our motivation for the design of a health information SDS that provided not only adherence education from Baylor lectures but also general health information tailored to the needs of caring for HIV positive children. An SDS in the local language Setswana, toll-free and accessible at any time through a simple telephone call, could greatly support Baylor's services to caregivers of children with HIV.

This initial field trip highlighted a number of challenges and issues for consideration in providing health information to caregivers: ƒ

Caregivers often struggle to recall material covered in the lecture sessions.

ƒ

Caregivers often have questions regarding general health information, for example, how to deal with infections, nutritional needs, hygiene requirements and commonly held misconceptions about HIV.

ƒ

Travelling to Baylor to address health information queries is not always possible for caregivers due to family/work responsibilities, transportation costs and time constraints.

ƒ

The majority of caregivers are semi and low literate populations. No written material is used and thus there is a lack of reinforcement and support for remembering material learnt in the Baylor lectures.

ƒ

2) Content Development Our design process started with the identification of relevant content and development of a framework (Fig.2) which detailed the broad health topics that needed to be covered by the SDS. Interviews with Baylor staff (nurses, doctors, nutritionists) and the following printed sources were used to create locally relevant accessible HIV health related content:

Answers to C om m on Questions about HIV

Facts on Nutrition

Traditional Practices

Food Groups

ƒ ƒ

Fruits and Vegetables

M yths

M eat & Beans

The spread of HIV from M other to Child

Childcare

During pregnancy

Diarrhoea

Preparing and Storing food

2. Nutrition

Hom e-based care Danger Signs

Vom iting

Hom e-based care

Breathing Problem s

Hom e-based care

Learn about ARVs

M edication side effects How to store ARVs.

Danger Signs

Danger Signs Fever

Hom e-based care Danger Signs

Constipation

Hom e-based care

Fig. 2. OpenPhone content framework.

Keeping a clean household H ow AR Vs work ?

W hen and How to give m edication ? W hat to do if you m iss an ARV dose? Traditional m edication Drugs & Alcohol

W hen to give m edication? How to give m edication ? Planning ahead Vom ited dose Forgotten dose

W hat to do if gloves are not available How to clean up blood properly.

W hen to wash your hands

W ashing your hands

4. ARV M edication

W hen to use gloves

H ow to dispose infected gloves & waste

1. H ygiene & Cleanliness

G eneral Health H elpline Danger Signs

Dealing with body fluids & infected waste

Safe preparation of food Safe storage of food

During Breastfeeding

5. Facts about HIV 3. Com m on Sicknesses

M ilk and Dairy products

How HIV works How HIV is spread?

How H IV is spread

Cereals/Rice/ Bread

Infection Learn about HIV

Baylor Adherence lecture materials HIV Aids Care & Counselling, A Multidisciplinary Approach [19] Where there is no Doctor: A Village Care Handbook [20] HIV, Health & your Community, A Guide for Action [ 21]

ƒ ƒ

Most caregivers are uncomfortable with English, thus Baylor lectures and all interactions with caregivers are in Setswana. Baylor staff explains complex health information in accessible terms in local language. Cure

Although caregivers are encouraged to call Baylor with any questions they may have, most are reluctant (and unable) due to the high costs of mobile phone calls.

ƒ

W ashing your hands with an antiseptic W ater shortage & hand washing Cleaning household item s How to wash your linen.

98

We also held focus group sessions with 27 caregivers at Baylor on a second visit. The discussions were led by a Setswana facilitator and were aimed at soliciting input from caregivers on what kind of health information they typically require under the broad topics of the framework. The caregivers were first given a demonstration of a health SDS prototype and then asked what kind of information they would like the system to provide to them. For instance, the facilitator would introduce the topic of “Nutrition” and then ask e.g. “what kind of food does a child with HIV require? Caregivers’ responses were used to fuel discussion and direct further questions on sub-topics of the framework. Thereafter caregivers were asked to prioritize (rank) topics they had listed in order of need in a group consensus setting. In many cases caregivers debated and gave reasons for their choices. We found that the topics in our content were mostly inline with what caregivers proposed in the focus groups. However, the topics’ ranking done by the caregivers varied from ours, for example, they ranked Hygiene & Cleanliness as most important which we had thought was secondary to ARV medication. The rankings allowed us to determine the topics the SDS should focus on providing first and foremost and were later also used to determine to order of main menu options in the SDS. The focus group sessions allowed us to revise the content and framework to build a SDS that caters for the health information needs expressed by the caregivers themselves. 3) System Design Once the initial content had been created and approved, the dialogue flow and audio output of the SDS had to be designed in a user-friendly manner. We carefully crafted the primarily text-based content for spoken speech and simple language. For instance, words like “lower cranium” will have little meaning for semi & low literate users; such references were translated into more accessible language, using terminology and metaphors used by Baylor staff in their lecture sessions. For example, the white blood cells are the ‘soldiers’ of the body, and ARV medication is the ‘ammunition’ for these soldiers. Since the SDS was aimed at low literacy users and users who may never have used a SDS before, the design goal was to create an easy-to-use interface that placed a low cognitive load on the user. We grouped content under logical menus and sub-menus, so that the amount of audio output under each node was balanced with the number of menu options at any given point in time. The design team set the following constraints based on guidelines for speech user interface design [22, 23]: ƒ ƒ

The maximum depth of menus in the SDS should not be more than 3 levels. The maximum breath or number of options at any menu should be no more than 5.

A sample system-user interaction follows. User: [Dials number...] System (Introduction): Hello and Welcome to the Health Helpline, I am Nurse Lerato and I know you probably have many questions about caring for someone with HIV. System (Overview): I can tell you about Hygiene & Cleanliness, Nutrition, Common Sicknesses, ARV Medication, and Facts about HIV. If at any time during your call you want to start our conversation again, you can press 0. System (Main Menu): For Hygiene & Cleanliness, please press 1, for Nutrition, press 2, for Common Sicknesses, press 3, for ARV medication, press 4 or for Facts about HIV, please press 5. User: [Presses 2.] System: Eating a balanced diet of different foods helps people with HIV stay healthy and strong. A healthy diet does not have to be costly and contains food from all the different food groups. Healthy food is always prepared and stored in a clean environment... As a final step, the design team decided on the specific wording of all audio prompts in Setswana. This required careful consideration for the dialect, the register (informal or formal), the cognitive load (short audio prompts, but long enough to provide contextual “anchoring”), and use of appropriate user interface metaphors. For instance, a system that suggests to the user “say Main Menu” is unlikely to make sense to users who have never interacted with visual or audio interfaces before. Instead, we chose the metaphor of a “conversation” and asked the user to “Start our conversation again”. The Setswana-speaking linguist on the design team played an essential role in the selection of ASR keywords that were locally relevant and logical to users, yet acoustically dissimilar. 4) Final SDS Development All the prompts and the health content were translated using a registered translation service into the dialect of Setswana spoken in Botswana. We specified that the content would be spoken aloud and the intended audience would likely be lowliterate. Since user interaction with an SDS is based on audio modality, the voice of the system plays a crucial role. With our target audience we felt that this was an essential element in creating a persona that would not only make users comfortable in their interactions with the system but also make the user experience enjoyable. Thus, our ideal voice talent would: ƒ Sound like a caring nurse willing to answer questions. ƒ Be a mother-tongue Setswana speaker. ƒ Have a full, mature female, well-articulated voice. ƒ Instil a sense of confidence and trust.

99

Our recruited voice talent was a well-regarded local soap celebrity, which meant her voice would be familiar to many of the target users. All the system prompts and content recordings were done in a professional recording studio. The SDS was built using the Asterisk telephony platform, the current set up in Botswana is a free standing PC, connected through an Asterisk card to an ISDN line. The ISDN line allows up to two calls simultaneously and users dial a local telephone number to access the system. While the system is operational, all aspects of the calls are logged. From start to finish all key-presses are monitored and all audio is recorded. C. User Study The goal of the study was to compare the most basic variant of speech systems (key-press replacement) with DTMF input. Thus, we built two identical systems that differed only at the menu prompts in choice of input modality i.e. in one system the user would press a key to chose a menu option and in the other they would say a keyword or phrase. For example, a DTMF menu option would be; “to hear about Nutrition, press 1,” whereas the ASR menu option would say, “to hear about Nutrition, say Nutrition.” The ASR was simulated using the Wizard-of-Oz (WOZ) methodology [24] where a researcher played the role of the speech recogniser. The ‘wizard’ listened to the speech input of the user and chose the next state of the system on this basis. The ‘wizard’ only accepted the exact keyword or phrase that the user was allowed to say at a particular menu option, any other input was directed to a No-Match state. In the case of the DTMF system, key presses were handled by the back-end telephony platform. Both systems ran from a PC laptop which was connected to standard telephone through a voice over internet protocol (VOIP) gateway.

Fig. 3. DTMF vs. ASR experiment set-up.

The user study was held over a period of five days in April 2008 at the Baylor premises. A total of 33 caregivers were part of the study, of which 27 tried both the DTMF and ASR systems. The remaining 6 caregivers did not try both systems

due to either user’s time constraints or a technology failure experienced in system set-up. The experimental set-up included a facilitator, an observer and a WOZ operator. The facilitators were local graduate students who were trained by the authors to facilitate in the local language, Setswana. The observers took notes on user behaviour, number of verbal prompts needed by a user, any user comments, and general body language of the user. Each user was introduced to the system and asked to sign a consent form. Emphasis was placed on communicating that users were not being tested but rather the system for purposes of improvement. Thereafter, each user watched a five minute video showing how a caregiver could use the system to find typical health information that they might need (Fig. 4).

Fig. 4. User watching a full-context video.

The context of the video was carefully matched to depict typical scenarios where a caregiver could use the SDS to obtain health information. For example, in the first part of the video, a caregiver with a sick child who has just thrown up his ARV medication is unsure of whether to give the medication again. A friend of the caregiver arrives and tells her about a telephone information system that she can call and learn when and how the next dose of ARV medication can be given if a child throws up his medication. Research [25, 8] on designing interfaces for low-literacy users has shown that such a fullcontext video greatly improves user studies by offering not only a demonstration of how to use the interface, but also the source (and therefore trustworthiness) of the content, the context in which you might use such a service, and the potential impact in your day-to-day life. After this video, the facilitator gave a short demonstration to the caregiver on how to retrieve “Nutrition” information as shown in the second part of the video. For each task, the caregiver was asked to dial a number themselves to access the SDS. This simple action served as a quick check to verify the caregiver’s ability to recognise numbers and use a phone. In order to show the difference between the DTMF and ASR systems we used separate telephone sets for each system (Fig.

100

3). After the caregiver finished the tasks using the DTMF system, for example, the facilitator explained that they would now try a similar system (on the other handset) but one where they would now say a keyword to obtain information. An example of a task explanation is shown on the following page. Facilitator: What’s the name of any good friend in your neighbourhood? Caregiver: Kabelo Facilitator: Your friend Kabelo says she must wash her hands frequently to keep her family safe from disease but she has very little water at home so she doesn’t know what to do. You’ve heard that Baylor has a phone number (help line) that can answer many health questions. So you decide to call the phone number, can you help your friend? Since the study was a within-subject comparison, we refrained from using the same tasks in both DTMF and ASR to prevent bias of previous knowledge. Thus, we created two sets of tasks; Set A and Set B with two tasks each (an easy and a difficult task; Task 1 and 2). The tasks were designed to be similar across Sets and to require a user to make three correct menu selections (Menu levels 1, 2, 3) to reach the specified information. Recall that our HIV health helpline is only three menu levels deep. The correct paths for each task are shown in Table I. TABLE I TASK DESCRIPTIONS WITH CORRECT SDS MENU OPTIONS.

Task Description Task 1, Set A Find out how to wash hands during water shortage.

Menu Level 1

Menu Level 2

Menu Level 3

Washing Hygiene your and cleanliness hands

Water shortage and hand washing

What to do if Body Hygiene fluids & gloves are not and Find out how to protect cleanliness infected available hands when gloves are not waste available. Task 1, Set B

Task 2, Set A Find out how to care for a child with fever. Task 2, Set B Find out how to care for a child with diarrhoea.

Common Fever Sicknesses

Home Care *

Common Diarrhoea Home Care * Sicknesses

* Home Care can only be reached after the user has heard the Danger Signs of the selected illness. For this reason, Task 2 is slightly more difficult than Task 1.

Additionally, to minimize any possible bias due to the order of trial of the DTMF and ASR systems or the use of a Task Set (A or B) with a particular modality, we systematically ensured that our data covered all possible combinations of Order of Modality and Task Set (illustrated as Quadrants 1-4

in Fig. 5). For example, if a caregiver started with DTMF using Task Set A he/she would then proceed to do ASR with Task Set B (Quadrant 1). Each caregiver thus did a total of 4 tasks and users were approximately evenly assigned between Set A (15 caregivers) or Set B (18 caregivers). The permutations in the order of trials also help to counter the impact of subject fatigue and learning effects within the small sample size.

Fig. 5. Order of Trials and Task Sets in experiment.

After completion of each modality’s trial, a post questionnaire was administered verbally to the caregiver. It consisted of ten questions adapted from the PARADISE evaluation framework [26]. The facilitator recorded the response in a 5-point Likert scale format based on the strength of the user response. We then verbally interviewed the caregivers to gather demographic data on education levels, language, and occupation, telephone usage (mobile and landline). Caregivers were also interviewed on their familiarity with technology (computer, mobile phone, TV, radio, video/DVD machine) based on factors such as use, ownership, frequency of use, place of use, and reason for use. Data was also gathered on the number of children they take care of, how often they visit Baylor and how they usually resolve their information queries. At the end of each session caregivers were provided with small non-monetary incentives (juice, potato chips and fruit for the child and gloves & household disinfectant for the caregiver) to thank them for their participation.

101

Fig. 7. Methods of Health Resolution by caregivers.

Fig. 6. A user taking part in the study.

III. RESULTS In this section we present results from our user study, including a description of our users (Section A), their task completion scores, other usability metrics from the study and a comparison of their performances on the DTMF and ASR parallel systems (Section B). A. User Profile During our user study, we had 33 caregivers who participated, of which 27 tried both DTMF and ASR systems. The caregivers were all female with the exception of one male caregiver. The age of our users ranged between 22 and 61 years old with the average age of 34 years. The average number of years of schooling amongst our users was 9 years but 2 users had 0 years of schooling. All of the users could read and write the local language Setswana and approximately 79% of them knew some English. In terms of occupation, 47% were unemployed and of the 53% who were employed, the majority were in low-income occupations such as cook, cleaner, house maid, hair dresser, or security guard. Caregivers reported that they visit Baylor between 1-3 times a month, with average travelling distance and time at 28 km and 1 hour respectively, some travelling from as far as 130 km, with average cost of travel at 18 Pulas (approx. $3 USD). The average waiting time at Baylor reported by the caregivers was 2.5 hrs, also travel time to Baylor ranges from 30 minutes to 3 hours; together, these represent a significant portion of a working day and are a substantial burden to the caregivers. Caregivers were also asked what they usually do if they have questions regarding the child’s health and when the last time was that they had such a question. One third of the caregivers usually go the local clinic to resolve their health queries and another quarter go to Baylor for this purpose (Fig. 7). Forty percent of the caregivers had a query regarding the child’s health within the last 6 months, and another 21% had more recent queries; in contrast, only 12 % could not remember specific queries (Fig. 8).

In terms of technology familiarity, 30 out of 33 (91%) caregivers owned mobile phones and 85% of these knew how to load their mobile phones with ‘airtime’ (pre-paid phones which require users to load money by calling the network provider’s service number and entering a sequence of digits from the pre-paid calling card). Average mobile phone costs per month were 68 Pulas ($10.5 USD) with an average cost per call being reported as 4.5 Pulas ($0.75 USD). Only 30% of caregivers reported having access to a landline telephone and of these only 9% had the landline at home (Table II).

Fig. 8. Time since last Health Question by caregivers.

TABLE II SUMMARY OF TECHNOLOGY OWNERSHIP AND USAGE BY CAREGIVERS.

Technology

Use

Ownership

Mobile Phone

91%

91%

Landline

30%

9%

Computer

15%

3%

TV

76%

71%

Radio

91%

91%

DVD/Video Machine

41%

38%

ATM

35%

N/A

102

B. Broad Usability Metrics and Observations System usability was determined using objective (Task completion rate, response time, routing time) and subjective (user preference, Likert ratings) metrics [27]. 1) Response Time Response Time measures the time it takes someone to respond to the system for the first time. It is usually measured not from the system start, but from the end of the Main Menu prompt, which in our case was 29 seconds long. This results in three basic categories of responses: (1) people who barge-in will have a negative response time, (2) people who respond in the 4 seconds of silence after the end of the Main Menu prompt will have a response time between 0 and 4 seconds, and (3) people who listen to one or several timeouts before responding will have a response time greater than 4 seconds. The Mean Response Times of our users (Fig. 9) indicates many barge-ins, roughly corresponding to the time the correct option at the first menu level was played: Hygiene and Cleanliness for Task 1, Common Sicknesses for Task 2 (Table I). Although we expected that our users would exhibit some short term learning effects, we found that response time did not decrease from the first call to subsequent calls.

Fig. 10. Task Completion Rates for DTMF and ASR.

For this study, we provide the task completion rate at each menu level of the SDS (Fig. 10). Across both tasks and both input systems, 60% to 84.38% of caregivers selected the correct first menu option (Menu Level 1). Approximately half of those people were able to select the appropriate sub-menu option (Menu level 2) when the task was to find out about “Common Illnesses” (Task 2), with the ASR system yielding the highest task completion rate for this menu level (56.67%). Caregivers seemed to have had more trouble correctly selecting Menu level 2 options for Task 1, with ASR again yielding the highest task completion rate of the two systems (40.63%). Task completion rate for level 3 ranged from 2.85 % to 23.68%, with Task 2 causing the most difficulty. Rates did not vary significantly by input mode (DTMF vs. ASR) or by task set (A vs. B). 3) Routing Time

Fig. 9. Average Response Time (time from end of the Main Menu prompt until the user's first input).

2) Task Completion Rate Task completion rate measures how frequently users were able to reach the node that provided the correct information for their assigned task.

Routing Time measures the time it takes a user to reach the beginning of the node which contains the correct information for the assigned task. Fig. 11 shows the mean routing time from our user study. Users’ routing time was similar across both tasks and both systems for both Level 1 and Level 2. As expected, due to the forced loop through Danger Signs for Common Sickness, caregivers took longer at Level 3 for Task 2.

103

Fig. 11. Routing Time for completed tasks. Recall from Task Completion Rates (above) that the Level 1 mean response is averaged over approximately 75% of our users, whereas Level 2 and Level 3 times are averaged over fewer users, in some cases only one or two.

Only two users correctly completed Task 2 with the ASR system; both users took over 4 minutes to do so. With the DTMF system only one person completed Task 2, doing so in less than 3 minutes. All users that correctly completed Task 1 using the ASR system did so in about 60 seconds. For Task 1 on the DTMF system, most users completed the task in just over 60 seconds, but one user took over 6 minutes.

4) Other Usability Metrics Results from our user study show that user interaction with the ASR and DTMF system are very similar across a wide range of additional usability metrics, illustrated in Fig. 12. Use of the Main Menu global (press '0' or say 'Simolola') and the Exit global (press '9' or say "Fetsa"), were similar for both systems. There were almost the same average number of timeouts (when 4 seconds elapsed with no user response) and repeats (when user chooses to repeat an information node) for the ASR and the DTMF system. On average, caregivers used the barge-in function one more time when using the ASR system but the total number of turns taken by users was similar for both systems. Note, all the usability metrics are means for each call, whether the task was completed or not.

Fig. 12. Usability Metrics for DTMF and ASR systems.

5) User Preference Systems were rated separately using the PARADISE [26] framework after use. The Likert scores were found to be unreliable, however. Despite efforts to elicit honest, critical feedback to the system (for example, we had a different person in a separate room conduct the post-study evaluations), all caregivers gave the system the highest marks possible across all categories and were hesitant to provide any criticism. For those 27 (out of 33 total) caregivers who tried both the DTMF and the ASR system, we were able to elicit feelings of preference for one system over the other (Table III). Most caregivers (59%) preferred the DTMF system over the ASR one (19%) and 22% indicated no preference. Both the DTMF and the ASR systems were judged to be the faster system by those who preferred it. Our measurements of routing time and task completion, however, show the systems are comparable. TABLE III USER PREFERENCE FOR DTMF OR SPEECH INPUT

Preferred Num Reasons Given System ber

Example Remark

DTMF

16

Clearer instructions (7) Faster to use (4) More private (2)

"Its quick and the doctor gives you the instructions, you just have to follow them."

ASR

5

More accessible "Its faster and old people can also use (2) it. The button Faster (1) system takes too Clearer long. " instructions (1) Hands free (1)

Both None

or 6

Similar (2)

"They are similar. With one you press buttons, with the other, you say

104

things." 6) Social Factor Correlations We examined several social factors based on our users' responses to questionnaires to see if they correlated with either their interaction with the system, based on objective metrics, or their reported system preference (DTMF vs ASR). Employment and experience loading airtime on a mobile phone were significant factors in the overall task completion rate (p=0.09 and 0.02, respectively). Those users who were employed or who had experience loading airtime completed more tasks during the study. Age, education, use of landlines, mobile phones, or ATM machines were not significant factors. Only previous use of a landline was a significant factor in use of system globals (p=0.01). Use of landlines was also a factor, along with loading airtime, in whether a user barged-in during system use (p=0.1 and 0.1 respectively). Caregivers who use a landline and load airtime were more likely to barge-in during system use. No questionnaire responses were found to be significant factors for system repeats or overall response time. Employment, previous use of an ATM, and experience loading airtime on a mobile phone were significant factors in the overall correct response (p= 0.07, 0.09, and 0.03) 1 . Caregivers who were employed, used ATM machines and loaded airtime had more correct responses to tasks during the user study. Age, education, and amount of mobile phone use were not factors in correct response. Loading airtime was the sole significant factor in user preference of DTMF over the ASR system (p=0.1). Those people who load airtime regularly preferred DTMF over ASR. Employment significantly correlated (p<0.1) with Overall Task Completion and Overall Correct Response. Age and monthly mobile phone costs were not found to correlate significantly with user performance or user preference for either the DTMF or ASR system. 7) Other Observations We observed based on body language and explicit remarks that several caregivers were nervous at first, and then became more relaxed during the first few minutes of the study. Many caregivers (and their children) showed signs of fatigue while trying the second system (sometimes an hour later). Caregivers sometimes had trouble understanding the task. They would often try to find information about ‘Nutrition’, for example, which was the topic of the demonstration, instead of searching for information on their assigned task. Caregivers also often clarified with the interpreter what the task was, what the keywords were, and what they should press or say during their trial. The interpreter would nod or say 'yes' but 1

The following categorical groups (yes and no) were divided into two groups and the resulting dependent values analysed: Employment, previous use of an ATM, experience loading airtime, use of landlines, use of ATM machines. A p-value of 0.1 and less indicates a significant difference between the two groups, whereas a p-value higher than 0.1 indicate no significant difference.

would not help the caregiver any further. In some cases, they would ask to try again, which we allowed if time permitted. Most were very interested in the content and many referred to the voice they heard as ‘the doctor’. Only one caregiver recognized the voice of the SDS as the celebrity soap star although most commented that “the ‘doctor’ explained very nicely”. During the interviews, all of our users enthusiastically indicated that they would like to use the service again; many said that it would be very valuable for educating themselves and their family/friends on caregiving aspects for children with HIV. A SDS such as OpenPhone could also serve as persuasion tool for caregivers trying to educate others, as explicitly reaffirmed by a caregiver “now I can tell them at home that the doctor (SDS voice) says the same thing (referring to a HIV related topic) that I’m telling them”. IV. DISCUSSION From our pilot study, we found that there were no significant differences between task completion rates (ASR only performed slightly better) or other usability metrics for both systems. This agrees with a number of previous studies in the developed world [12, 28, 29] where no major differences were found in terms of performance. However, subjectively the majority of our users preferred DTMF (59%) over ASR (19%), which is in contrast to formal studies in the developed world [9]-[11] (where user preferences largely favour speech), but correlates with the observation that simple key-press replacement with keywords is generally not viewed favourably. The users who did prefer ASR did not as in developed world studies comment on the aesthetics of speech input, that “speech is more entertaining or enjoyable” but rather on the utility of speech “more accessible for older people or faster”. Our finding that users’ employment and experience loading airtime correlated with higher task completion but that education level does not, indicates that technological literacy is a more important factor in adopting new technology than literacy itself. This may also contribute to the finding that ‘loading of airtime’ was the sole significant factor found in DTMF preference over ASR. It is also interesting to note that our users who had minimal exposure to SDSs (except loading airtime), were relatively comfortable using our system for the first time, as indicated by their frequent, timely barge-ins. Also, our users noticed the value of speech (allowing hands-free operation, innumerate/older people being able to use it) and DTMF as well (provides privacy). This highlights that even if users may be technically inexperienced and unfamiliar with an ICT application, they have valuable and sound judgment on the utility of the interface. Also, both DTMF and the ASR systems were judged (subjectively) to be the faster system by those who preferred it, which indicates that time constraints, may also be of importance for low literacy users, in contrast to

105

earlier observations. The DTMF vs. speech input comparison could be improved in terms determining task performance by using a between-subjects experiment design; however this approach would not reveal user preference. Our preliminary investigations indicated that a good fraction of caregivers were in the low and non–literate range. In our user study though, we encountered that many of our recruits were semi- to low literate. Also, whilst our user numbers represent a significant sample size for a developing world user study, it may in comparison to studies in the developed world be on the ‘small’ end of the spectrum. The above-mentioned issues emphasize some of the challenges faced in research for the developing world; that the very users for whom an SDS could be most useful may be the hardest for us to reach and also that user recruitment in developing regions can present significant obstacles [30]. Whilst this pilot study illustrated that telephony services could in fact be easily used by semi and low literacy users, and that an SDS in local language can be a powerful health education tool, the decisive factor in widespread uptake is likely to be the cost incurred by the caller for the service. The majority of caregivers said that even though the service would be useful to them they would only be able to make use of it, if the service is toll-free. An average phone call in Botswana of 5-10 minutes to the SDS would cost a mobile phone user $1-2 USD. From our questionnaires, the average cost per month for mobile phone usage was $10.5 USD. Thus, a single phone call to the SDS would consume, 10-20% of a caregiver’s monthly mobile phone budget, making the case for a toll-free number all the more imperative. Notwithstanding our findings that show promise for SDSs for low literacy users, we did encounter challenges in introducing the concept of a SDS to our users, for instance a few users did not fully realise that the system was automated. For example, at the end of one call, a user proceeded to ask the ‘nurse’ (system persona) a question when prompted by the system to leave a comment (and waited for the answer). Another user repeatedly acknowledged what the ‘nurse’ was saying by responding with “Yes, yes” or “I agree with you”. An obvious solution here might be to use a text-to-speech (TTS) voice for the prompts. However, we run the risk of a mismatch between the target audience and the context (e.g. emotionally sensitive in terms of health care) or culturallybased communication norms of the community. This in turn may affect the willingness of users to interact with the system. A general design challenge for SDSs is to ensure a minimal cognitive load on the user. This magnifies more so in the case of information access applications where lengthy pieces of information need to be provided (e.g. in a health care context). In our case, all the users indicated in the subjective questionnaire that the length of the content was not too long; however, during the experiment some users did mention the need to concentrate in order “not to miss all the things being said by the ‘nurse’ ”. There is a need to address the abovementioned issues by developing the dialog to be more

interactive (perhaps with audio cues, getting intermittent user feedback) and conversational. In addition, the nature of educational information services tends to be rather exploratory; where a user may peruse various topics related to his/her general query (e.g. a user may want to, in general know about “Dealing with body fluids and infected waste” which has 4 topics all related to that option). This in our case translated to some users struggling to find the exact menu option related to the task and exploring related sub-topics. One user even singled out that she would like to know the mapping of the menu options beforehand to help her locate the information she is looking for. This experience highlighted challenges of using hierarchical menus [8] and the importance of paying attention to the taxonomy and vocabulary of the system to enable easy navigation for the user. SDS design for smaller languages also introduces challenges on other dimensions including prompting and persona. The prompts and content of a SDS application will typically be translated from a language such as English to the local language. Thus, great care has to taken in the prompt writing phase to ensure that intended meaning of the original prompt (English) is still preserved in the translated prompt (local language) and conveyed in the simplest and shortest way possible. Often, a concept described by a single word in English has no direct translation in another language. For instance, whereas a keyword in the English version of our application was “Safe food”, it became the phrase, “Dijo tse di siameng” after translation, in order to adequately describe the concept. Moreover, not only should the translated SDS prompts convey the intended meaning but the designer should ensure that the persona of the local language system is in line with cultural and contextual expectations of the intended audience. For instance we ensured our prompts not only had the right balance of formality and gravity appropriate for the message (HIV info) but were also understandable and conversational. Our experience in employing multiple data-gathering techniques in the needs investigation phase (interviews and discussions, observations, field visits, and focus groups) better equipped us in trying to comprehend the needs of our users. The interviews and discussions helped us establish a rapport with our users and stakeholders (Baylor staff), whilst also providing the flexibility of follow-up questioning and recalibration of interviewer terminology when needed. Observations on the other hand helped to further reveal the issues that users are unable or unwilling to articulate or express. Field work allowed us to pick up on the cultural nuances and social-economic context of the users and enabled us to gain a deeper understanding of the sensitive environment (HIV/health) we were working in. Finally, through focus groups we were able to observe the interaction amongst caregivers and most importantly it enabled us to obtain specific targeted design information in a group setting and correct our earlier assumptions on the ranking of SDS health topics.

106

REFERENCES V. CONCLUSION This paper has addressed the frequently debated question of input modalities of touchtone vs. speech in a completely new context; low literacy users and a domain (health) different from the usual business centric applications (call routing, voice mail systems or banking). Our pilot study also served to confirm the feasibility of SDS applications for semi and low literate populations. It is interesting to speculate on the applicability of our findings to other resource-poor countries. Our intuitive judgment is that the only characteristic of our user population that strongly influenced our results was their relative familiarity with mobile telephones; however, the matter requires detailed investigation. In future work we would therefore like to further explore the space of non-literate users and the suitability of SDS applications for them, as well as investigate the interaction of task types (linear vs. non-linear) and application domain – informational (Openphone) vs. transactional (tracking a social services payment) with the input modality. Once the system is free of charge and has been in use for several months, we would like to study whether OpenPhone leads to a change in health habits and improves the ability of a caregiver in providing care to children, like change of hygiene habits, better nutrition, and fewer misconceptions about HIV. Throughout the design and development process, we experienced that beyond usability and creating simple, accessible user interfaces for low literacy users, factors such as cultural and social context, establishing relationships with stakeholders and user communities, and localizing content, play a vital role in the success of an ICT intervention in the developing world. We intend to draw on these valuable experiences and carry our work forward in serving the information needs of citizens of developing regions with accessible and usable telephony based services. ACKNOWLEDGEMENTS The authors wish to thank the staff of Botswana Baylor Children’s Clinical Centre of Excellence for the generous manner in which we were received, in particular Dr Paul Mullan, for the amount of help and time given to us. We would also like to thank members of the HLT group at Meraka Institute who provided valuable contributions throughout the project; Victor Zimu, Jama Ndwe, Mpho Kgampe, Louis Joubert, Richard Carlson, Bryan Mcalister, Mark Zsilavecz, Marelie Davel and Alta de Waal. We also greatly appreciate the time given to us by Connie Ferguson, our celebrity voice talent. This research would not have been possible without the joint support of OSI and OSISA. The project was also in part supported by the NRF under the Key International Research Capacity (KISC) programme, UID no. 63676.

[1] [2] [3]

[4] [5] [6]

[7] [8]

[9]

[10]

[11]

[12]

[13]

[14] [15] [16]

[17]

[18] [19]

R. Tucker and K. Shalonova. "The Local Language Speech Technology Initiative", SCALLA Conference, Nepal, 2004. SIL International. (2001). Facts about illiteracy. [Online] Available at:: http://www.sil.org/literacy/litfacts.htm (last accessed 15 Sept 2008). ITU: International Telecommunications Union. (2003). Mobile overtakes fixed: Implications for policy and regulation. [Online]. Available: http://www.itu.int/osg/spu/ni/mobileovertakes/Resources/Mobileovertak es_Paper.pdf, (last accessed 5 September 2008). Statistics South Africa. (2007). Community Survey 2007. [Online] Available at: http://www.statssa.gov.za/publications/P0301/P0301.pdf (last accessed 15 Sept 2008). E. Barnard, L. Cloete, and H. Patel. “Language and Technology Literacy Barriers to Accessing Government Services,” Lecture Notes in Computer Science, vol. 2739, pp. 37-42, 2003. M. Plauche, U. Nallasamy, J. Pal, C. Wooters, and D. Ramachandran. “Speech Recognition for Illiterate Access to Information and Technology,” in Proc. IEEE International Conference on Information and Communications Technologies and Development ‘06, pp. 83-92, May 2006. P. Nasfors. "Efficient Voice Information Services for Developing Countries", Master Thesis, Department of Information technology, Uppsala University, Sweden, 2007. J. Sherwani, N Ali, S. Mirza, A. Fatma, Y. Memon, M. Karim, R. TRongia and R. Rosenfeld, “Healthline: Speech-based Access to Health Information by low-literate users”, in Proc. IEEE International Conference on Information and Communications Technologies and Development ‘07, Bangalore, India, Dec. 2007. S. Agarwal, A Kumar, AA Nanavati and N. Rajput, "VoiKiosk: Increasing Reachability of Kiosks in Developing Regions", in Proc. of the 17th international conference on World Wide Web, pp. 1123-1124, 2008. K. M. Lee and J. Lai “Speech Versus Touch: A Comparative Study of the Use of Speech and DTMF Keypad for Navigation” International Journal of Human–Computer Interaction, vol. 19, no. 3, pp.343–360, 2005. B. Suhm, J. Bers, D. McCarthy, B. Freeman, D Getty, K Godfrey and P Peterson, "A comparative study of speech in the call center: natural language call routing vs. touch-tone menus". In: Terveen, Loren (ed.) Proc. of the ACM CHI 2002 Conference on Human Factors in Computing Systems Conference, Minneapolis, Minnesota. pp. 283-290, April 2002. C. Delogu, A. Di Carlo, P Rotundi and D. Sartori, "Usability evaluation of IVR systems with DTMF and ASR", in Proc. of the 5th International Conference on Spoken Language Processing (ICSLP 98), paper 0320, Australia,1998. T. Pearce and M Bergelson. (2008). Alignment index for speech selfservice. Dimension Data Technical Report [online]. Available at http://www.dimensiondata.com/NR/rdonlyres/9191A848-5F35-459F82398D9D2248414E/8791/mainstreamspeechalignmentindexreport2.pdf (Last accessed 15 Sept 2008). J.P. Migneault, R. Farzanfar, J.A. Wright, and R.H. Friedman. “How to write Health Dialog for a Talking Computer,” Journal of Biomedical Informatics, vol. 39, no. 5, pp. 468 – 481, Oct. 2006. Tellme Networks Inc. Who uses TellMe? [Online]. Available at: http://www.tellme.com/about (Last accessed 15 Sept 2008). UNAIDS (August, 2008). Report on the Global AIDS Epidemic. Joint United Nations Programme on HIV/AIDS. [Online] Available at: http://www.unaids.org/en/KnowledgeCentre/HIVData/GlobalReport/200 8/2008_Global_report.asp (last accessed 7 Sep 2008). K. Seipone, W. Jimbo, F.d.l.H. Gomez, K. Ampomah, J. Othwolo, O. Kaluwa, M. Busisiwe. “Trends in HIV Prevalence Among Pregnant Women in Botswana 2001-2005” 16th Intl. Conf Aids. Toronto, Canada, Aug. 2006. Botswana-Baylor Children’s Clinical Centre of Excellence, Annual Report 2006, Gaborone, Botswana, 2006. A. Van Dyk, HIV Aids Care & Counselling, a Multidisciplinary Approach. 2005, Third Edition: Pearson Education, South Africa.

107 [20] D. Werner, C. Thuman and J. Maxwell. Where there is no Doctor, a Village Care Handbook.. 2007, New revised edition: Hesperian, CA, USA. [21] R. Granich and J. Mermin. HIV, Health & your Community, A Guide for Action. 2001, Hesperian, CA, USA. [22] G.A. Miller. "The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information," The Psychological Review. vol. 63, no. 2 pp. 81-97, 1956. [23] Suhm B. “IVR Usability Engineering using Guidelines and Analyses of end-to-end calls” in D. Gardener-Bonneau and H.E. Blanchard (Eds). Human Factors and Voice Interactive Systems. 2008, pp. 1-41, Second Edition, Springer Science: NY, USA. [24] N.M. Fraser and G.N. Gilbert, “Simulating speech systems”, Computer Speech and Language, vol. 5, no. 2, pp. 81-99, 1991. [25] I. Medhi and K. Toyama, “Full-Context Videos for First-Time, NonLiterate PC Users” ”, in Proc. IEEE International Conference on Information and Communications Technologies and Development ‘07, Bangalore, India, Dec. 2007. [26] M.A. Walker, D. Litman, C.A. Kamm and A. Abella. “PARADISE: A general framework for evaluating spoken dialogue agents” in Proc. of the 35th Annual Meeting of the Association of Computational Linguistics. ACL/EACL 97, 1997. [27] Nielsen, J., (1993). Usability Engineering. AP Professional, Boston, MA, USA. [28] Foster, J. C., McInnes, F. R., Jack, M. A., Love, S., Dutton, R. T., Nairn, I. A., et al. (1998). An experimental evaluation of preference for data entry method in automated telephone services. Behaviour & Information Technology, 17, 82–92. [29] Goldstein, M., Bretan, I., Sallnas, E.-L.,&Bjork, H. (1999). Navigational abilities in voice-controlled dialogue structures. Behaviour & Information Technology, 18, 83–95. [30] E. Brewer, M. Demmer, M. Ho, R.J. Honicky, J. Pal, M. Plauche, and S. Surana. “The Challenges of Technology Research for Developing Regions,” IEEE Pervasive Computing, vol. 5, no. 2, pp. 15-23, AprilJune 2006.

108

ICT4What? – Using the Choice Framework to operationalise the Capability Approach to Development Dorothea Kleine

Abstract— Identifying the specific contribution of the use of ICTs to specific development goals has proven to be extremely difficult. This paper argues that instead of trying to make ICTs fit with a linear conceptualisation of impacts and an often economistic view of development, ICT4D should be used as a prime example of a development process which has to be analysed in a systemic and holistic way. Amartya Sen’s capability approach offers a way of thinking about development not as economic growth, but as individual freedom. The Choice Framework is presented as a way of operationalising this approach and visualising the elements of a systemic conceptualisation of the development process. An individual case study, related to telecentres in rural Chile, is used to demonstrate the way the Choice Framework can be applied as a guide to a systemic and holistic analysis. Index Terms— ICT4D, Amartya Sen, capability approach, Choice Framework, telecentres, Chile

T

I. INTRODUCTION

he paradox is this: ICTs and particularly the internet are widely regarded as groundbreaking inventions that have changed the way millions of people live their lives, and yet researchers and practitioners in the field of ICT and development often struggle to prove specific impacts of the technology to funders. There may be specific reasons why particular projects fail, even some generalisable patterns of failure [1], but the overall degree to which the ICT4D community has to struggle when trying to legitimise its work to funders is astonishing in the context of a general discourse about how much these technologies have changed our lives. This paper tries to unravel the reasons behind this paradox by arguing two fundamental points: On a theoretical level, while there have been interesting alternative theoretical approaches to development, including Amartya Sen’s capability approach, the mainstream discourse’s conceptualisation remains heavily focused on economic growth, which is too narrow to capture the impacts of ICT. Secondly, and on a practical level, the common way of measuring impact by defining the intended Manuscript received September 22, 2008. This work was supported in part by the Dr. Heinz Dürr Fellowship Programme. Author’s Details: Dorothea Kleine is Lecturer in Human Geography at Royal Holloway, University of London, Egham, TW20 0EX and a member of the UNESCO Chair/ICT4D Collective (email: [email protected]).

development outcomes top-down and a-priori is unsuitable in the context of multi-purpose technologies which could empower individuals to attain development outcomes of their own choosing. The paper is structured in six parts. In section one I briefly introduce Sen’s capability approach before presenting, in section two, some important steps towards operationalising it. Building on this body of work, section three develops the Choice Framework as a further way of operationalising Sen’s approach. Methodological implications are explored in section four before in section five the Choice Framework is applied in a case study of one particular individual’s usage of the Internet in a telecentre in rural Chile. The final part points out limitations of the model, directions for further empirical research and calls for more theoretical work on the nature of the “development” element in ICT4D. The paper concludes by highlighting some concrete implications this theoretical work may have for practitioners. II. EVOLVING THEORIES OF DEVELOPMENT Research located in the contested intellectual space that is ‘development’ needs to be able to answer the fundamental question of what is understood as development. Broadly speaking, debates in development studies range from positions which equate development with economic growth (e.g. [2], [3], [4], [5]) through to critical perspectives stressing that uneven development, dependency and inequality are inherent in capitalist development (e.g. [6], [7], [8], [9]) to ideas of alternative, bottom-up development recognising social and ecological as well as economic goals (e.g. [10], [11]), and radical “post-developmentalist” critiques that often dismiss the entire “development project” altogether (e.g. [12], [13]). The most influential challenge to the mainstream growthfocused view of development has come from Amartya Sen’s capability approach (also known as the capabilities approach) in which development is defined as “a process of expanding the real freedoms that people enjoy to lead the lives they have reason to value” [14]. His understanding focuses on development as freedom of choice. While this understanding of development is a minority position within institutions such

109

as the World Bank [15], Sen’s approach has influenced the development discourse and it provides a means of building a bridge between those working in international development organisations and researchers in academia. The approach developed by Amartya Sen ([16], [17], [18], [19]) argues that development is about the freedom of choice in the personal, the social, the economic and the political sphere. In Sen’s approach, “functionings” are the various things a person may value doing or being, such as being adequately nourished, being healthy and being able to take part in the life of a community. In Sen’s terminology a person’s “capability” refers to the alternative combinations of functionings that are feasible for her/him to achieve [20]. The focus of development thus becomes increasing a person’s capability set, or her/his substantive freedom to lead the life she/he values. Functionings represent the “outcome” component, while capabilities are the “freedom” component in this approach. 1 In Sen’s more holistic view of development, economic growth plays an important, but not exclusive, role. Sen suggests ontologically focusing on human wellbeing and methodologically focusing on capabilities. While welcoming Sen’s approach at offering a more holistic view of development, scholars have been struggling to find a balance between its conceptual richness and its potential to be operationalised. Several scholars [21], [22], [23], [24] have attempted to operationalise the approach. Within this field the majority of studies use capabilities as a normative basis for the research while measuring functionings as a proxy [25], owing to the practical difficulty of measuring capabilities. Some authors (e.g. [26], [27]) have challenged Sen to draw up a general list of capabilities, but Sen has refused to do so, claiming that specific lists of capabilities ought to be drawn up for a given research or policy context [28] and, crucially, that the process of choosing capabilities should be left to the individual [29]. The dilemma which emerges is how to apply the capability approach to specific areas or sectors in a meaningful way while retaining openended development outcomes that do not presuppose individuals’ choices. II. OPERATIONALISING SEN’S APPROACH Sen intended his approach to be combined with other theoretical approaches [30]. The following section explains how, in order to operationalise the approach for ICT4D and other areas of development, elements have been drawn together from the literature on empowerment and on sustainable livelihoods to enhance the application of Sen’s approach.

A. Empowerment One of the most interesting attempts to operationalise Sen’s ideas is offered by Alsop and Heinsohn [31]. Writing for the World Bank, they link choice with their definition of empowerment 2 . They define empowerment as “enhancing an individual’s or group’s capacity to make effective choices and translate these choices into desired actions and outcomes” [39]. ICTs could be seen as useful tools in such processes of empowerment Alsop and Heinsohn see material and non-material assets, or resources, as the basis of individual agency which, together with the structural conditions frame empowerment processes. In their attempt to use empowerment as a middle-range theoretical concept to convert the development paradigm of choice into a construct that is of use to practitioners, Alsop and Heinsohn build a crude framework which connects “individual agency” with an “opportunity structure” from which follow the degree of empowerment an individual has to achieve development outcomes. The different “degrees of empowerment” are: existence of choice, use of choice and achievement of choice [40]. Individual agency is measured by an individual’s asset endowment, consisting of “psychological, informational, organisational, material, social, financial or human” assets [41]. These assets are listed, but not defined. An actor’s opportunity structure is said to be shaped by the “presence and operation of the formal and informal institutions” [42] and measured by the presence and operation of laws, social norms and customs. Alsop and Heinsohn have applied their framework in the evaluation of World Bank projects with women, on rural water supply and sanitation, on school decentralisation and with school dropouts. B. The sustainable livelihood framework Another literature which can be linked to the capability approach is the literature on livelihoods. Based on earlier work on livelihoods ([43], [44], [45] the Sustainable Livelihood Framework (SLF) used by the UK Department for International Development (DFID) [46] offers an analytical tool to understand in a systemic way the elements influencing the lives of the poor. Duncombe has demonstrated how the SLF can be applied to ICT4D research with microenterprises [47], while retaining the focus on poverty reduction through economic growth. The SLF includes the concept of an individual’s “capital portfolio” made up of five “capitals”: human capital, natural capital, financial capital, physical capital and social capital. In operationalising the SLF, human capital is measured by formal education and health indicators, but there has been a 2

1

For a more in-depth discussion of the capability approach, see also [32].

The concept of empowerment originated in work on gender relations and community participation (e.g. [33], [34]) and has been increasingly discussed in development studies (e.g. [35],[36], [37], [38]. There are several competing definitions of the term.

110

struggle to quantify “social capital” [48]. As a result, critics have argued that “everything social” gets packed into the social capital variable [49]. Individuals own or have access to this portfolio of capitals, their “livelihood assets” with which they negotiate “policies, institutions and processes”. They operate within a “vulnerability context” and develop livelihood strategies which then result in livelihood outcomes. Livelihood outcomes are defined a priori – In the DFID version, “more income” is listed at the top, even before “increased wellbeing”. The SLF offers a broad and systemic view of development processes, but its

A. Outcomes True to Sen’s statement that choice is both the aim and the principal means of development [51], the primary development outcome is choice itself. Secondary development outcomes depend on the individual’s choice as to what lives they value. These may include, for example, easier communication, increased knowledge, more income or time saved. Information and communication technologies might prove useful tools in achieving these outcomes. Just like other attempts to operationalise Sen’s work, here capabilities are not measured directly, though participatory research with individuals and groups may reveal them to some degree. Mainly, the outcome component will map or measure the

STRUCTURE institutions and organisations discourses policies and programmes formal and informal laws including: - norms on usage of space - norms on usage of time • access to ICTs - availability of ICTs - affordability of ICTs - necessary skills for ICTs

Key: ER = Educational Resources PsR = Psychological Resources In = Information FR = Financial Resources CR = Cultural Resources

• • • •

SR = Social Resources NR = Natural Resources MR = Material Resources GR = Geographical Resources He = Health

DEVELOPMENT OUTCOMES DIMENSIONS OF CHOICE

Principal: Choice

Secondary:

AGENCY

• easier communication

• sense of choice

• increased knowledge

• use of choice

• access to markets • business ideas

• achievement of choice

SR He

• existence of choice

CR

• increased income • more voice

Age

ER

NR

Gender

• higher job satisfaction

Ethnicity

PsR

MR

...

In

• time saved

•...

GR FR

Fig. 1: The Choice Framework set of capitals is limited and in it the development goals are predetermined and not up to the individual to choose. In this respect, the SLF fails to mirror the thinking behind Sen’s approach. III. THE CHOICE FRAMEWORK Based on Sen’s capability approach, inspired by Alsop and Heinsohn’s work on operationalising Sen’s work, taking elements from the SLF and informed by an in-depth research project with microentrepreneurs’ use of ICTs in Chile [50], the Choice Framework was developed. After presenting it in diagrammatic form (Fig 1), the following sections will in turn explain each of the key components of the framework.

achieved functionings resulting from an individual’s choices as a proxy to the capabilities. 3 An analysis based on the Choice Framework would then work backwards, from the outcomes, into the systemic relationships between agency, structure and choice, thus analysing how the outcomes were arrived at. B. Dimensions of Choice Alsop and Heinsohn’s dimensions of choice, which they call “degrees of empowerment” include, firstly, the existence 3 Two disadvantages of this method are that some of the individual’s capabilities are not captured in the achieved functionings and that it is difficult to trace the choices related to apparently negative outcomes. However, so far, capturing functionings is methodologically easier and more precise – in regards to both quantitative and qualitative methods - than capturing capabilities.

111

of choice – whether the different possibilities exist and are, in principle, attainable for the individual if the combination of their resource portfolio and the structural conditions would allow it. The second dimension, a sense of choice, not originally included by Alsop & Heinsohn, was added as a result of fieldwork experiences relating to ICT and development. Individuals were aware of some possibilities the new technology offered them, like email and online chat, but not of others, like Voice over IP. This was precisely because their educational resources (including computer skills) and the dominant discourse in the Chilean media stressed some usages over others. For any piece of research focused on a technology which is new to the respondents, the dimension of “sense of choice” will play a significant role. The “use of choice” dimension refers to whether or not an individual actually makes the choice and the “achievement of choice” refers to whether the outcome matches the choice expressed.

C. Agency Instead of using a terminology of capitals and asset or capital portfolios, Sen uses the term “resources” within the capability approach [52]. Resources can be interpreted as individual agency-based capability inputs which, together with structure-based capability inputs, can be converted into capabilities [53]. In the Choice Framework age, gender, ethnicity etc. are conceptualised as personal characteristics of an individual which may in a given social context become related to socially constructed axes of exclusion and influence the scope and scale of the resource portfolio. The resource portfolio consists of: Material resources: These sum up the material objects owned, such as machinery, computer hardware and other equipment. They are also essential inputs in the production process. Financial resources: These stand for financial capital in all its forms (cash, savings, shares etc.). The ability to obtain credit is a combination of the structural character of the banking rules and individual collateral. Natural resources: This includes issues such as geomorphological and climatic conditions in a locality and related aspects such as soil quality and the availability of or access to water as well as the attractiveness of the surrounding nature. Geographical resources: Covers the practical implications of location and relative distances, and also includes the intangible qualities of a location alluded to by writers from Marshall (who refers to the mysteries of the trade “in the air” [54]) to Storper and Venables (who describe the “buzz” of

face to face contact in the urban economy [55]). Human Resources: The term “human resources” has been used for decades in the economics and industrial relations literature. 4 In the Choice Framework, this term needs to be disaggregated into Health and Education and Skills (educational resources). Within Sen’s paradigm of development, good health is a prerequisite for a person’s ability to choose the life she/he values. Educational resources represent education and skills acquired through formal and informal means. Psychological resources: Alsop and Heinsohn [56] recognise the significance of “psychological assets” and give as an example “capacity to envision”. More broadly, psychological assets may include self-confidence, tenacity, optimism, creativity and resilience. Spiritual or religious beliefs stand in complex interrelation with psychological resources – they can strengthen or weaken an individual’s psychological resources. Information: Alsop and Heinsohn list informational assets as a key resource. Heeks [58] calls for putting information at the centre for analysis of ICTs and Development, and Gigler [59], adds “informational capital” to the capital portfolio. Access to information is the first step to knowledge acquisition, the process of filtering and transforming information into meaningful knowledge. Cultural resources: “Cultural capital” – which in the Choice Framework is called cultural resources – exists, according to Bourdieu [60], in three states: an embodied state (the habitus a particular person lives in); an objectified state (objects like paintings, instruments and monuments which only the initiated can use or appreciate); and an institutionalised state (prestige attached to, for example, academic titles). Social resources: “Social capital” – or social resources – is included in both the SLF and Alsop and Heinsohn’s work. It has been both immensely influential and highly contested in development discourse. For the Choice Framework, Bourdieu’s definition of social capital is used: “the aggregate of the actual and potential resources which are linked to possession of a durable network of more or less institutionalized relationships of mutual acquaintance and recognition – or in other words, to membership in a group – which provides each of its members with the backing of the collectivity-owned capital, a “credential” which entitles them to credit, in the various senses of the word.”[61] Membership of these groups can be defined by kinship, friendship, shared ethnicity or class, friendship or informal commonality ties. 4

For examples, see the collection by Fitzgerald and Rowley [57]

112

Thus, these ten types of resources – material, financial, natural, geographical, psychological, cultural, social, and educational (education and skills) resources; health; and information – represent an attempt to holistically map aspects of the agency element of the systemic framework. However, it is important to recognise that this resourcebased agency can only be realised within the confines of and in systemic interaction with a given structure. This aspect of the Choice Framework will be analysed in the following section. D. Structure Both the empowerment framework suggested by Alsop and Heinsohn and the SLF take into account not only individual agency, but also structures which aid or constrain this agency. Alsop and Heinsohn list “formal and informal laws, regulations, norms and customs” [62] as elements of this structure, while the DFID SLF includes these as laws and “culture” – the latter running the risk of being used as a kind of black box into which all locally specific aspects can be subsumed. The SLF includes not only laws, but also policies, institutions and processes. Rules, laws, norms and policies are embedded in, and often emanate from discourses, and hegemonic discourses can define the thinkspace in which policies, including ICT policies, can be conceived. Thus discourses are included as part of the structure element of the Choice Framework. In particular with respect to ICTs, relevant elements of the structure which influence an individual’s agency include dimensions of access, such as availability, affordability and capabilities needed for using different ICTs [63]. To avoid confusion with Sen’s use of the word, the term “skills” is used instead of capabilities. These dimensions of access are nationally and often locally specific, path dependent and embedded with other elements of the structure. Structural factors such as these stand in a complex relationship with an individual’s resource portfolio. For example, with the help of social resources an individual might have access to the internet (at a neighbour’s house) which might lead to frequent email contact with a distant family member, thus increasing occasions of, in Bourdieu’s terms, legitimate exchange with both the neighbour and the distant relative, in turn potentially increasing social resources. Similarly, a person with higher educational resources (skills and education) and information might find it easier to use the existing access facilities to enhance their skills and gain information. The interface between the opportunity structure and individual agency thus includes a host of reciprocal and cumulative processes. Structural constraints need to be recognized as being as important an element as individual agency. To reflect this, structure is placed above agency in the

diagram of the Choice Framework. The Choice Framework is an attempt to operationalise the capability approach in a holistic and systemic way, thus maintaining much of its conceptual richness. While it may prove particularly useful in the area of ICT4D, the framework could also be applied in other areas of development work.

IV. IMPLICATIONS FOR METHODOLOGY The basic challenge that the capability approach offers to the orthodox methodologies of development research, and ICT4D in particular is that, on a fundamental level, it questions the validity of outcomes that are defined a priori and without consulting the individual in question. Both the inclusion of a development goal and its position within a set of development priorities, however, relate to the question which kind of life people would choose to live and this, according to Sen, is what development is about. A funding institution or government may set, say, economic prosperity as the top priority. Once basic needs such as food and shelter have been met, however, an individual may value being close to family members above earning more money, or may value a healthy environment for themselves and their children over economic growth. In the practice of development projects, this means that before undertaking an intervention designed to improve people’s lives and later measuring its effectiveness, practitioners and researchers would have to ask individuals about their own development priorities and let these guide the planning, implementation, monitoring and evaluation of development projects and programmes. Work done in this area includes participatory monitoring and evaluation [64] and, in a broader context, initiatives around participatory budget planning [65]. Setting development priorities in a participatory way may make the process of development planning more complicated, but it comes with major benefits: Firstly, morally it is the right thing to do to engage the people themselves in the decisions that will affect their lives. Secondly, if the outcomes have been agreed upon in a participatory way, they are more likely to be locally and culturally appropriate and may reduce the rate of failure. Thirdly, a participatory process will lead to greater local buy-in to measures and therefore higher future institutional sustainability. Fourthly, such a process harbours the chance that the current overly economistic focus of development work can be broadened to include environmental, social and cultural aspects and thus better mirror the diversity of the kinds of things people value in their lives. Last, and perhaps not least, ICT and development practitioners work with multi-purpose technologies which offer far more significant changes to people’s lives than the economic impact they have been proven to have. Moving away from an a priori, top-down and often overly economistic

113

set of development priorities offers the chance to recognise the diversity of the contributions ICTs can make to the social, cultural, environmental and economic aspirations individuals may have for their lives. The Choice Framework is one way of conceptualising such choices by the individual. There are some specific implications for research design which result from the model: A focus on the individual’s own development outcomes means that the research needs to start from these wished-for outcomes, measure the degree to which they have been attained and work systemically backwards through structure, agency and choice to understand how these outcome have come about. The extensive list of resources covers six less tangible resources (social, cultural, educational, psychological resources, health and information) which pose challenges to measurement but need to be taken into account. On the structure side, access to ICTs is conceptualised as availability, affordability and necessary skills. ICTs are embedded in the wider set-up of institutions, policies, programmes, norms and discourses. As such they need to be analysed as firmly and historically engrained in the societies they affect. V. APPLYING THE FRAMEWORK: IMPACTS OF TELECENTRES IN RURAL CHILE The following example is part of an extensive ethnographic study of how state ICT policies affected microentrepreneurs in rural Chile [66]. When interviewing microentrepreneurs and their partners who were using a telecentre located in a public library in rural Chile, open-ended questions revealed that apart from business-related usages such as looking up prices of machinery on the internet (carpenters), looking up photos of furniture models (carpenters), communicating via email with a supplier (carpenters, spice vendor) or buyer (spice vendor), looking up recipes (cake vendor), and looking up guidelines for government business assistance (carpenters), there were several answers which reflected what in many studies of telecentre usage is described as “personal usage”. Respondents’ faces lit up when they described how they now could exchange emails or chat with relatives who were living abroad, children who were studying in a larger town or family members who were working as temporary labour on fruit farms or in the mines in the north of Chile. One respondent, a woman in her 50s whose household income was around 440 USD per month and who together with her husband ran a carpentry business, described how while the Word Cup was on in Germany in 2006 she visited the world cup site to find links and take virtual tours of some of the German cities she was not able to visit in person. As a young woman, she had had a pen friend from Kaiserslautern, and while he had come to visit her in Chile, her dream of visiting him had never been possible because, she said, the money she saved had been spent on her children’s education. Eventually, they had lost touch, but now, she told me with tears in her eyes, over 25 years later, she was finally able to “visit” Kaiserslautern, right

here in the telecentre. If one were to apply a typical questionnaire on telecentre usage to this case, this woman’s usage experience might be subsumed in the category “personal usage” or “other”. Yet the following section will offer a careful application of the Choice Framework to this case: A. Outcome The primary outcome was that the respondent had improved choice, in this case, between “no visit to Kaiserslautern” and “virtual visit to Kaiserslautern”. The secondary outcome achieved was defined by the individual: “to see more of the world” – which in this case, translated into “virtual visit to Kaiserslautern” - in Sen’s terms an “achieved functioning”. The aspiration was “visit to Kaiserslautern”, which since it is feasible, could be seen as a “capability” in Sen’s terms. The achieved functioning ”virtual visit” is not equal to the capability “visit in person” but it is an improvement in outcome over no visit at all. Studies of development outcomes, or more commonly of impacts, often operate with a set of impacts as defined by the funding body, government, international organisation or commercial sponsor. This set of impacts then acts as a checklist informing the construction of questionnaires and interview guides, possibly with some scope for “other activities”. Sen’s approach, with the individual’s choice as the primary outcome, however, would suggest that the analysis needs to start from the ground up, asking people about what lives they value and what outcomes they want to see. For this individual, one of the greatest impacts the telecentre had made was that it had given her the chance to virtually visit Kaiserslautern, something few policymakers or researchers would have predicted. Indeed, some might question whether this is a valid “development outcome” or “impact” for a telecentre. In Sen’s approach, expressed via the Choice Framework, it is. B. Agency The individual in question was a Chilean-mestizo 5 woman in her 50s, married with four children who were all grown up now and had left the home. Her material resources did not include a computer and internet access at home and her financial resources made it difficult for her to spend money on using a computer in the local cybercafés. However her social resources (contacts with friends) had helped her gain the information that there was free access to the internet available at the telecentre in the local library. Her geographical resources (the location of her house) and her state of health were such that she could easily reach the telecentre on foot. She had the cultural resources to not feel intimidated when 5 i.e. not considered part of the indigenous minority which had historically been discriminated against.

114

entering a space like a library and to know the behavioural code there. With the help of her social resources (knowing the librarian who was now also the director of the telecentre), her educational resources (literacy, rudimentary English) and her psychological resources (extrovert, willingness to ask questions) she quickly learnt how to use the computers. The information she gained online, together with her psychological resources (curiosity, tenacity) allowed her to understand the choices she had and find the site which offered the virtual tour of Kaiserslautern, thus achieving her chosen development outcome. C. Structure The agency of the individual is a shaper of, and is shaped by the structure in which it operates. In this case, as part of the national ICT policy, the Agenda Digital, the state of Chile had signed an agreement with the Bill and Melinda Gates Foundation for them to provide 9.2 million USD worth of hardware to be installed in public telecentres based in libraries around the country, running Microsoft software. The local library was an existing institution which was able to accommodate the hardware, delivered as part of the Biblioredes telecentre programme (availability of ICTs). The Chilean digital literacy campaign provided free IT courses to adults, and public discourse in Chile stressed the importance to become “digitally literate”, so the woman had taken the course (necessary skills). The formal rules for users of the telecentre stipulated that access was free (affordability of ICTs) but limited to 30 minutes per person at busy times, and the informal rules were that people were left to use the computers on their own unless they asked for help from the telecentre director. Norms on the usage of space made it easy for her to go to a library as a mestizo woman with a completed school education who was known in town. However, norms on the usage of time meant that she could only use the telecentre when she was not supposed to be home preparing meals (gendered norms on time) or when she was expected to attend to customers (business norms on time). D. Dimensions of choice In the Choice Framework, an individual’s resource-based agency can operate within a given structure to achieve degrees of empowerment, such as existence of choice, sense of choice, use of choice and achievement of choice. In this case, both the choice “travel to Kaiserslautern in person” and “take a virtual tour of Kaiserslautern” existed, the latter only since the links were offered via the World Cup website in 2006. In a capitalist market system, however, the former choice required an amount of financial resources which the individual felt unable to dedicate to this idea. The choice “take a virtual tour of Kaiserslautern” however, required a good internet connection, a computer, the knowledge that the tour was available via the website, the skills to find and run it, and time. The individual, thanks to among others, her social resources,

information, and psychological resources, knew that the telecentre offered a computer and a good connection, had acquired the skills to navigate the Internet and run an application in the free digital literacy courses offered at the telecentre. She felt that informal, gendered social norms allowed her to go to the telecentre during the morning before having to prepare lunch. Thus she developed a sense of choice, was able to choose (use of choice) and achieved her desired outcome (achievement of choice). VI. CONCLUSION Applying the Choice Framework to this particular case allows us to firstly, theorise the use of ICT in a systemic and procedural way which reflects the systemic and pervasive impact of ICT. The “impact of ICT” is not conceptualised in a cause- and effect chain, instead effects are carefully disaggregated and their systemic interrelatedness and cocausality is demonstrated. Secondly, the Choice Framework offers a way to operationalise Sen’s capability approach in the context of ICT. Sen’s approach is currently the most well known heterodox alternative to orthodox, growth-focused and often economistic conceptualisations of development. Given the enormous potential of ICTs to give individuals choices, and indeed a greater sense of choice, Sen’s approach is of particular interest to the ICT and development research community. There are three obvious limitations to the application of the Choice Framework, and this is where more theoretical work needs to be done: Firstly, the Choice Framework aims to be comprehensive in its modelling of the complex relationships between agency, structure, degree of empowerment and outcome, and this automatically entails a trade-off with the depth of theorisation of each element. Behind each of the terms included in the framework lies a wealth of theoretical literatures which may need to be synthesised for different research purposes and key issues brought to the attention of researchers in the development field. While for example, social resources can be theorised by linking to the wider debate on social capital (see [66]), which has been received in the development studies discourses, work on cultural capital (in Bourdieu’s sense) is hardly ever linked to development discourses in the South. Secondly, the Choice Framework is relatively easily applied in qualitative work on the micro-level of the individual. A further challenge will be how to apply the framework to groups of individuals, communities, or even nations. Within this and related to a theoretical tension evident in Sen’s original approach, there is a complex relationship between individual and collective choice which will have to be conceptualised carefully.

115

The example used here to illustrate the potential of the Choice Framework is part of a far more extensive ethnographic and longitudinal study of how the state ICT policies affected microentrepreneurs in rural Chile [66]. In this study, connections are made between individual and collective choice. For example, the local authority’s public procurement policy was analysed as an expression of collective choice, but related to individual’s views as to how their tax money should be used to create the community they wanted to live in. However, further empirical work is necessary to gather experiences in the applicability of the Choice Framework in other cultural and socio-economic settings. The relationship between individual choice and collective choice needs to be conceptualised carefully in these local contexts in order to allow for the empirical application of the Choice Framework at the more aggregate level such as socalled target groups and communities. The third limitation is a very practical one: funders prefer predefined and clearly measurable impacts. The Choice Framework, however, suggests that impacts of ICTs occur in a systemic, pervasive and transversal way, and that outcomes should be defined, in line with Sen’s approach, by the individual, based on their choices as to what kind of life they value. There are, however, some funders who are open to methods such as participatory evaluation and monitoring, and this means there might be hope for genuinely people-centred development work – and development theory. There are some key implications emerging from this theoretical work for practitioners of ICT. Firstly, while no technology is ever completely politically neutral [67][68], ICT4D projects can be placed on a continuum of “directional control”. At one end there are projects and programmes which focus on providing people with access to a technology which is recognised as multi-purpose, like some telecentre projects. On the other end of the continuum are projects and programmes which carry a much more narrow set of intentions, for example training microentrepreneurs to use a specific e-procurement system in order to “train them” to operate in a more competitive market environment under a specific set of rules [67]. The further down the directional control continuum a particular project and programme is located, the more risk there is that the intended outcomes of an ICT4D project diverge from the capabilities, or desired outcomes individuals in the so-called target group would choose. Thus, the more directional control is involved in the project or programme, the more participation of the set of individuals who are the intended group will be needed to reduce this gap. This would include conceptualising the development process as open-ended and the so-called target group as individuals empowered to choose the lives they themselves value. Participatory project design and participatory monitoring and evaluation techniques would be most appropriate.

Secondly, there are some macro-methodologies which reflect the ethos of giving people the power to choose. Voucher schemes are a good pragmatic way to monitor, in a heavily supply-driven development field such as ICT4D, what products (hardware, software etc), services (trainings, computer repair, communication etc) and content (economic, social, political, cultural etc) people would, after considering their options, actually choose. From the field of participatory urban planning come methodologies for participatory budget design, where communities get to debate and decide which of their desired outcomes to prioritise and pursue. This is a practical and democratic way to aggregate individual capabilities in order to enable collective decision making, and could also be used for ICT4D. Thirdly, practitioners may deduce that if the ideal is for development projects’ intended outcomes to reflect the individual’s choices, then the more individuals are aggregated to a group, the less probable it is that they can agree on a similar set of capabilities. From this follows that the further down the directional control continuum an ICT4D project is, the more sensitive/locally customised it has to be to the choices of a smaller number of people. Big, uni-directional development programmes with specific, a priori defined desired outcomes designed for a large number of people are most likely to be in contradiction to a people-centred holistic development process as proposed by Sen and expressed in the Choice Framework. Ultimately, researchers working on ICT, particularly the internet, need to consider the question: Should we try and fit a groundbreaking, multi-purpose and potentially liberating technology into orthodox notions of development – such as more ICT for higher GDP, more ICT for better school results et cetera – impacts which we may struggle to prove? Or can the field of ICT and Development serve as a test case and breeding ground for thinking about development in a more holistic way, putting the individual and their own choices at the centre of development? If the latter is the case, then we have plenty of work to do, but the Choice Framework may serve as one part of the big puzzle we have to begin putting together. ACKNOWLEDGMENT I would like to thank Diane Perrons, Robin Mansell, Alexandra Norrish, Macarena Vivent, Rodrigo Garrido, Tim Unwin, and four anonymous reviewers for their comments on earlier versions of this paper.

116

REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25]

[26] [27] [28] [29] [30] [31] [32]

R. Heeks, “Information systems and developing countries: Failure, success and local improvisations” in The Information Society, vol. 18, no. 2, pp. 101-112, 2002 W.A. Lewis, “Economic development with unlimited supplies of labour”, The Manchester School of Economic and Social Studies, vol.22, no. 2, pp.139-91, 1954 G. Myrdal, Economic Theory and Underdeveloped Regions, London, Duckworth, 1957 A. O. Hirschman, The Strategy of Economic Development, Yale, Yale University Press, 1958 W. Rostow, The Stages of Economic Growth: A Non-Communist Manifesto, Cambridge, Cambridge University Press, 1960 A.G. Frank, Capitalism and Underdevelopment in Latin America, London, New York, Monthly Review Press, 1967 T. Dos Santos, Dependencia y Cambio Social, Santiago, Universidad de Chile, 1970 M. Blomström and B. Hettne, Development Theory in Transition. The Dependency Debate and Beyond: Third World Responses, London, Zed Books, 1984 C. Kay, Latin American Theories of Development and Underdevelopment, London, New York, Routledge, 1989 M. Nerfin (ed.), Another development: Approaches and Strategies, Uppsala, Dag Hammarsköld Foundation, 1977 R. Chambers, Rural Development: Putting the Last First, Harlow, Longman, 1983 A. Escobar, Encountering Development: The Making and Unmaking of the Third World, London, Longman, 1995 J. Nederveen Pieterse, “After post-development”, Third World Quarterly, vol. 21, no.2, pp.175-191, 2000 A. Sen, Development as Freedom, Oxford, Oxford University Press, 1999, p.3 J. Harriss, Depoliticizing Development – The World Bank and Social Capital, London, Anthem Press, 2001 A. Sen, “Equality of what?”, in S. McMurrin (ed.) The Tanner Lectures on Human Values, Salt Lake City, 1980 A. Sen, Resources, Values and Development, Cambridge, MA, Harvard University Press, 1984 A. Sen, “Capability and well-being”, in M. Nussbaum and A. Sen (eds.) The Quality of Life, Oxford, Clarendon Press, 1993 A. Sen, Development as Freedom, Oxford, Oxford University Press, 1999 A. Sen, Development as Freedom, Oxford, Oxford University Press, 1999, p.3 M. Nussbaum, Women and Human Development: The Capabilities Approach, Cambridge, Cambridge University Press, 2000 S. Alkire, Valuing Freedoms, Oxford, Oxford University Press, 2002 D. Clark, Visions of Development, Cheltenham, Edward Elgar, 2002 I. Robeyns, “Sen’s capabilities approach and gender inequality: selecting relevant capabilities, Feminist Economics, vol. 9, no 2-3, pp. 61-92, 2003 I. Robeyns, “The Capabilities Approach: An Interdisciplinary Introduction”, Department of Political Science and Amsterdam School of Social Sciences Research Working Paper, University of Amsterdam, Amsterdam, 2003 R. Sugden, “Welfare, resources and capabilities: a review of Inequality Re-examined by Amartya Sen”, Journal of Economic Literature, vol. XXXVI, pp. 1947-1962, 1993 J. Roemer, Theories of Distributive Justice, Cambridge, Cambridge University Press, 1996 A. Sen, “Capability and well-being”, in M. Nussbaum and A. Sen (eds.) The Quality of Life, Oxford, Clarendon Press, 1993 A. Sen, “Maximisation and the act of choice”, Econometrica, vol. 65, no.4, pp.745-779, 1997 A. Sen, Inequality Re-Examined, Oxford, Clarendon Press, 1992 R. Alsop and N. Heinsohn, Measuring Empowerment in Practice – Structuring Analysis and Framing Indicators, Washington D.C.: World Bank, 2005 D. Kleine, Empowerment and the Limits of Choice: Microentrepreneurs, Information and Communication Technologies and State Policies in

[33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47]

[48] [49]

[50]

[51] [52] [53]

[54] [55] [56] [57] [58] [59]

[60]

Chile, unpublished PhD thesis, Department of Geography and Environment, London, London School of Economics, 2007 C. Moser, Gender Planning and Development, London, Routledge, 1991 W. Van Eyken, The Concept and Process of Empowerment. The Hague, Bernard van Leer Foundation, 1991 J. Friedmann, Empowerment: The Politics of Alternative Development, Oxford, Blackwell, 1992 N. Kabeer, “Resources, agency, achievements: reflections on the measurement of women’s empowerment”. Development and Change, vol. 30, no.3, pp.435-464, 1999 P. Oakley (ed.), Evaluating Empowerment, Reviewing the Concept and the Practice, Oxford, INTRAC, 2001 A. J. Bebbington, M. Woolcock, M. Guggenheim and E.A. Olson (eds), The Search for Empowerment. Social Capital as Idea and Practice at the World Bank, Bloomfield, CT, Kumarian Press, 2006 R. Alsop and N. Heinsohn, Measuring Empowerment in Practice – Structuring Analysis and Framing Indicators, Washington D.C.: World Bank, p.5, 2005 Ibid, p.6 Ibid, p.8 Ibid, p.9 R. Chambers and G. R. Conway, “Sustainable Rural Livelihoods: Practical Concepts for the 21st Century”, Discussion Paper 296, Institute of Development Studies, University of Sussex, UK, 1992 A. J. Bebbington, “Capitals and Capabilities: A Framework for Analyzing Peasant Viability, Rural Livelihoods and Poverty”, World Development, vol. 27, no.12, pp. 2021-2044, 1999 D. Carney, Sustainable Livelihood Approaches Compared, London, Department for International Development DFID, Sustainable Livelihoods Guidance Sheets, Department for International Development, London, 1999 R. Duncombe, “Using the Livelihoods Framework to Analyze ICT Applications for Poverty Reduction through Microenterprise”, Information Technologies and International Development, vol. 3, no. 3, pp.81-100, 2006 DFID, Sustainable Livelihoods Guidance Sheets, Department for International Development, London, 1999 A. Munasib, Social Capital at the Individual Level: A Reduced Form Analysis. Available: http://munasib.myweb.uga.edu/papers/SocialCapitalDemography.pdf, 2004 D. Kleine, Empowerment and the Limits of Choice: Microentrepreneurs, Information and Communication Technologies and State Policies in Chile, unpublished PhD thesis, Department of Geography and Environment, London, London School of Economics, 2007 A. Sen, Development as Freedom, Oxford, Oxford University Press, 1999 A. Sen, Resources, Values and Development, Cambridge, MA, Harvard University Press, 1984 I. Robeyns, “The Capabilities Approach: An Interdisciplinary Introduction”, Department of Political Science and Amsterdam School of Social Sciences Research Working Paper, University of Amsterdam, Amsterdam, 2003 A. Marshall, Principles of Economics, London, Macmillan, 1920, reprint edition 1961 M. Storper and A.J. Venables, “Buzz: face-to-face contact and the urban economy”, Journal of Economic Geography, vol. 4, no. 4, pp. 351-370, 2004 R. Alsop and N. Heinsohn, Measuring Empowerment in Practice – Structuring Analysis and Framing Indicators, Washington D.C., World Bank, 2005 R. Fitzgerald and C. Rowley (eds.), Human Resources and the Firm in International Perspective, Cheltenham, Edward Elgar, 1997 R. Heeks, “Information and communication technologies, poverty and development”. Development Informatics Working Paper No.5, Manchester, IDPM, 1999 B.S. Gigler, “Including the excluded – can ICTs empower poor communities? Towards an alternative evaluation framework based on the capability approach,” presented at the Fourth International Conference on the Capability Approach, Pavia, Italy, Sept. 5-7, 2004 P. Bourdieu, “The forms of capital”, in: J. Richardson (ed.), Handbook of Theory and Research for the Sociology of Education, New York, Greenwood Press, 1986

117 [61] Ibid, p.249 [62] R. Alsop and N. Heinsohn, Measuring Empowerment in Practice – Structuring Analysis and Framing Indicators, Washington D.C., World Bank, 2005, p.9 [63] R. Gerster and S. Zimmermann, Information and Communication Technologies (ICTs) for Poverty Reduction?, Bern, Swiss Agency for Development Cooperation, 2003 [64] I. Guijt and J. Gavena, Participatory Monitoring and Evaluation: Learning from Change, Brighton, Institute of Development Studies, 1998 [65] A. Shah (ed.) Participatory Budgeting. Washington D.C., The World Bank, 2007. Available: http://siteresources.worldbank.org/PSGLP/Resources/ParticipatoryBudg eting.pdf [66] D. Kleine, Empowerment and the Limits of Choice: Microentrepreneurs, Information and Communication Technologies and State Policies in Chile, unpublished PhD thesis, Department of Geography and Environment, London, London School of Economics, 2007 [67] D. Kleine, “The ideology behind the technology – Chilean microentrepreneurs and public ICT policies”, Geoforum, vol. 40, no.2., 2009 [68] D. Kleine, “Striking a Balance”, Engineering and Technology, February 2007, p. 30-33

118

ICTD for Healthcare in Ghana: Two Parallel Case Studies Rowena Luk, Matei Zaharia, Melissa Ho, Brian Levine, and Paul M. Aoki

Abstract — This paper examines two parallel case studies to promote remote medical consultation in Ghana. These projects, initiated independently by different researchers in different organizations, both deployed ICT solutions in the same medical community in the same year. The Ghana Consultation Network currently has over 125 users running a Web-based application over a delay-tolerant network of servers. OneTouch MedicareLine is currently providing 1700 doctors in Ghana with free mobile phone calls and text messages to other members of the medical community. We present the consequences of (1) the institutional context and identity of the investigators, as well as specific decisions made with respect to (2) partnerships formed, (3) perceptions of technological infrastructure, and (4) high-level design decisions. In concluding, we discuss lessons learned and high-level implications for future ICTD research agendas. Index Terms — Remote medical consultation.

I. INTRODUCTION ICTD research focuses primarily on matters relating to its target populations and their conditions, such as how individuals, communities and institutions interact with technology. For the most part, investigators only become part of the scene as objects of retrospective critique, usually in the context of having failed to see some vital point and achieve some critical goal. As with any kind of research, however, ICTD research is a product of social, cultural and professional influences on the investigators, influences that affect every decision made over a project’s lifetime. It is extremely difficult to reflect upon the effects of these influences because each project is so uniquely framed by the contexts of its investigators and its target setting that it is hard for analysts to imagine how “it could have been otherwise” [19]. In this paper, we address this reflective gap through a sideManuscript received September 22, 2008. This material is based in part upon work supported by the U.S. National Science Foundation under Grant No. 0326582. M. Zaharia is supported by the Natural Sciences and Engineering Research Council of Canada R. Luk is with AMITA Telemedicine Inc., Toronto, ON, Canada (e-mail: [email protected]). M. Zaharia is with the Department of Electrical Engineering and Computer Sciences, University of California, Berkeley, CA, USA (e-mail: [email protected]). M. Ho is with the School of Information, University of California, Berkeley, CA, USA (e-mail: [email protected]). B. Levine was with New York University, New York, NY. He is now with the Department of Obstetrics and Gynecology, New York Presbyterian Hospital, Columbia University Medical Center, New York, NY, USA. (email: [email protected]). P.M. Aoki is with Intel Research, Berkeley, CA, USA (e-mail: [email protected]).

by-side analysis of two completely independent projects which arose from similar objectives but resulted in two very different strategies. Two sets of researchers came to the same country to work with the same community, identified the same problems, and proposed two different solutions. Both projects were implemented over the course of 2007-2008, both aimed to tackle remote medical consultation among Ghanaian doctors, and both sought nationwide deployment over the course of the project lifetime. However, the Ghana Consultation Network (GCN) was initiated by a group of technologically-oriented researchers, while the OneTouch MedicareLine (ML) was initiated by a public health researcher and social entrepreneur. GCN’s solution consisted of a Web application hosted on a delay-tolerant network of computers running in each hospital and on the open internet. ML focused instead on a mobile phone program involving a combination of technological services and business innovation. We discuss the rationale behind decisions made by each party over the course of the project and the resulting outcomes. The main contribution of this paper is to reflect on some ways in which the context and background of researchers affect the structure of ICTD projects. A second contribution is in identifying relevant factors which will assist in the design and execution of ICTD projects in the future. In particular, we provide specific examples of how an investigator’s institutional context and identity affect not only the methodology used but also the interpretation of findings; how the partnerships chosen for co-development and codeployment have a fundamental role in the development and deployment of the technology; how ‘objective’ assessments of existing technology infrastructure are influenced by personal areas of expertise; and how projects can be designed with various degrees of technological ‘specificity’ and the resulting implications for their impact on various development indicators. We expect these findings to be of interest to the community of ICTD researchers and practitioners as a whole. The paper is organized as follows. We first present a background overview of some of the underlying issues of healthcare in Ghana and briefly introduce both projects. We then discuss in turn the framing of the research problem and how it affected the partnerships formed and resulting deployment strategies; the assessments of technological infrastructure by the project members; and the usage and appropriation of the two solutions. Finally, we discuss highlevel implications for ICTD research and present works related to each of these sections.

119

II. BACKGROUND In this section, we highlight some of the background issues of healthcare in Ghana and provide an overview of the two projects which are the focus of our case studies. A. Healthcare in Ghana Like many other countries, Ghana has a tiered healthcare system in which cases that cannot be handled by an institution at a given tier are referred to institutions above it in the hierarchy. All referral chains culminate in one of two teaching hospitals, both located in the South, where specialists have the training and resources necessary to carry out more complex procedures. Doctors, and in particular specialists in areas such as internal medicine, are highly concentrated in the urban South. This sometimes leaves only two or three doctors to serve in district hospitals in the rural North; predictably, rural doctors are confronted with heavy workloads (meaning they can spend only a few minutes face-to-face with a given patient per day) and isolated working environments (which prevent them from taking advantage of many of the educational and collaborative programs available to doctors in the South). Formal and informal consultation is highly integrated into the life of all doctors. Between hospitals, many doctors call personal contacts – friends, colleagues, classmates – to seek advice, and within hospitals this behavior is even more frequent. Such “curb-side” consultation has long been a common observed characteristic of medical practice [16] but it is of critical importance in environments where specialist expertise is spread thinly. Continuity of care is difficult to ensure. Patient records are almost universally paper-based. Despite attempts by some major hospitals to transition towards an electronic system, none of the 18 hospitals visited had a working system. Within a major hospital, there is nothing to guarantee complete patient records besides a doctor’s own discretion and multiple reminders from administration. It is not uncommon for records to be lost, confused, or incomplete. The communication infrastructure in the country is severely limited. Hospital landlines are frequently out of service, forcing many doctors to rely on personally-purchased mobile phones. Likewise, broadband Internet infrastructure is unevenly available, frequently unreliable where available, and dependent on hospital budget allocations. Other options are available throughout the country, such as dial-up, satellite, and mobile data plans, but these are generally expensive, unreliable, and/or slow as well. B. The Ghana Consultation Network (GCN) Solution GCN is an end-to-end, computer-based system providing doctor-to-doctor medical consultation on a network of servers implementing a distributed, asynchronously-synchronized database (Figure 1) [20]. The goal is to allow doctors throughout Ghana to consult with each other as well as medical professionals (in particular, the large Ghanaian medical diaspora) around the world. Doctors access the system through a Web-based UI (Figure 2), either by logging

Figure 1. GCN distributed, delay-tolerant server network (c) (c) new new thread thread

(a) (a) primary primary case case list list

(d) (d) navigation navigation menu menu

(b) other cases” ” list cases (b) ““other cases” list

Figure 2. Key elements of GCN’s welcome screen

into a local server (hosted at some participating hospitals) or by logging into one of the two public servers (hosted with Internet service providers (ISPs) in both Ghana and the U.S.). Providing local servers ensures availability and responsiveness to the users in the face of unreliable network connectivity and makes the task of synchronizing data between servers transparent to them. Synchronization is automated and carried out over a disconnection-tolerant messaging layer called OCMP [29] which draws inspiration from the research into delay-tolerant networking (DTN) [7],[12]. Because the doctors already view consultation as a matter of reaching out to personal contacts, the system is presented as a social networking platform – a forum for medical consultation with social and professional colleagues – and leverages social incentives using principles drawn from the HCI and CSCW communities [20]. The system supports two types of ‘conversations’: highly structured ‘consultations’ for specific patients (which work much like an electronic case history) and unstructured ‘discussions’ (which work much like online

120

forums). So, for example, a doctor unsure of how to treat a particularly resistant case of tuberculosis might fill out various fields of the ‘consultation,’ address it to a colleague, and wait for advice to be appended to that consultation, whereas another interested in general updates on malaria treatment might create a question under ‘discussions,’ addressed to any interested doctor. Recognizing the ubiquity of mobile phones, GCN also incorporates text message (as well as email) notifications of new content; yet the core of the interaction is designed for computers. This is based on feedback during the design process that the wealth of data required for patient management would demand a screen size larger than those present in mobile phones. To date, over 125 doctors have been enrolled from Ghana, the U.S., Mali, Nigeria, South Africa, and the U.K. and 39 consultations have been submitted. Methodology GCN is the product of a conventional user-centered design process. There have been four rounds of iterative design and fieldwork, starting with exploratory needs assessment (2005) and continuing in conjunction with design exercises (20062007), a pilot deployment (2007), and an ongoing deployment (2008). Overall, interviews or focus groups were conducted with 132 doctors in 15 hospitals throughout Ghana [20]. The most recent fieldwork in Ghana lasted six weeks in mid-2008. Exploratory interviews on mobile phones and computer usage were conducted with 35 internal medicine (IM) doctors at a major teaching hospital. Further, six evaluation interviews at one regional hospital and two interviews at one district hospital were conducted (both sites of the original pilot). Limitations of the methodology include an overrepresentation of internal medicine doctors from urban hospitals. While the initial rounds of fieldwork focused on rural hospitals which are more numerous in the North, the current live deployment is centered in the more accessible South, with plans to extend later. We have established public servers in both Ghana and the US, as well as local servers in 3 major hospitals and 2 rural district hospitals in the South. These district hospitals suffer from some of the same problems as in the North – a shortage of doctors and poor travel and communication infrastructure – although the severity of these challenges is much less. Our current strategy is first to recruit specialists from the major hospitals and to test our system in the district hospitals of the South before reaching out to more challenged environments in the North.

Figure 3. ML Phase 1 calls for free phone calls and text messages; the planned ML Phase 2 calls for MMS and data reports over SMS, and ML Phase 3 calls for free smartphones, reference tools, and custom applications.

C. The OneTouch MedicareLine (ML) Solution MedicareLine is a program currently offering free calls and text messages between any registered physician and/or surgeon within Ghana. Its current focus has been on reducing logistical and economic barriers to mobile phone use rather than on technological innovation. After submitting the required paperwork, a doctor registered with the Ghana Medical Association (GMA) receives a OneTouch GSM SIM which can be used with a privately-purchased mobile phone. Using this SIM, the physician can now call other program participants free of personal charge. For example, a physician can call a specialist in the capital or a friend in a rural town to ask or provide medical consultation. This can be a significant cost saving, especially given that airtime in Ghana is relatively expensive compared to many developed countries. The GMA also has a computer terminal that can send “blast” texts to all participants for updates and notifications. Future phases of the program envision new technological interventions (Figure 3). Phase 2, as yet uncompleted, calls for physicians to receive free MMS service so that they can augment their phone consultations with photos (e.g., of a skin condition or X-ray). Phase 2 will also allow the GMA and other government organizations to collect data from physicians via SMS. In Phase 3, ML anticipates partnering with hardware vendors to provide each physician with a smartphone preloaded with medical reference software. Phase 1 of this program has already experienced a very high rate of adoption. Approximately 1700 of 2000 doctors in the GMA have enrolled, with over 2 million calls made to date. Methodology ML arises from a five-week visit to Ghana in October 2007 and a two-week visit in March 2008. The purpose of the first

Table 1 – Comparison of GCN and ML Ghana Consultation Network (GCN) Computers Primary platform 89 doctors in 5 hospitals in Ghana Adoption >125 doctors total from around the world Network connectivity Problem addressed Doctors in Ghana and Ghanaian medical diaspora Target user Ministry of Health, GPSF, and KNet (ISP) Partners (see Table 2) Adequate internet connection quality and coverage Assumptions punctuated by regular power and network outages High – custom system software and Web application Specificity (see Section VI) Incremental (hospital by hospital) Deployment strategy

OneTouch MedicareLine (ML) Mobile phones >1700 doctors in Ghana Cost of cellular airtime Doctors in Ghana GMA and OneTouch (mobile operator) Adequate mobile phone quality and coverage Low – only generic mobile phone service National (available country-wide at launch)

121

Figure 4. Timeline of GCN and ML Projects, including fieldwork.

visit was to introduce and promote usage of a social networking Web system for international medical collaboration that had already been developed in the States; however, within two weeks of arriving in Ghana, the investigator decided that access to both computers and the Internet was a fundamental problem. Discarding the Webbased project entirely, he moved on to exploratory interviews with all stakeholders, including over 30 doctors, politicians, local businessmen, and administrators, focusing on the issue of barriers to communication, innovation, and current technology usage. He conducted site visits to five regional hospitals and one polyclinic in the urban South, and visited Tamale, a northern regional capital, for four days. Towards the end of that trip, the investigator orchestrated a “meeting of the minds” between the CEO of OneTouch and the head of the GMA, whereupon an initial understanding was reached. Implementation was left to the two partnering agencies, which announced the program the following month and launched it in January 2008. In March 2008, the investigator returned to conduct another round of unstructured interviews with over 20 individuals, again from diverse communities, in order to assess the program’s progress. In addition to ML’s own follow-up interviews, material for this paper also comes from 10 interviews conducted by the GCN investigators with doctors at one of the teaching hospitals, as well as their interviews with the ML investigator directly. III. THE ROLE OF PROJECT CONTEXT IN FRAMING PROBLEMS ICTD work frequently involves the interdisciplinary participation of various communities of research and of practice [5], but what are the implications of working within these communities, given similar projects with similar goals over similar timeframes? Both projects clearly did some amount of fieldwork in both the urban South and the rural North, both purportedly wanted to develop a solution that would work for all Ghanaian doctors, but GCN focused on connectivity in the rural North using innovative technology while ML focused on building communication in the urban South using proven technology. In this section, we examine the institutional, cultural, and personal contexts from which each project arose.

A. GCN GCN began as part of an ongoing collaboration between a U.S. corporate research laboratory and a U.S. research university. Specifically, it was a project of a joint research group which had worked extensively in the area of low-cost connectivity and delay-tolerant networking. Project funding came from U.S. government grants as well as the corporate sponsor. In the first brainstorming/conception phase of this project, the axes by which ideas were evaluated were defined as: (1) direct social impact (e.g., improvements in healthcare delivery), (2) medium-term impact on ICT adoption in developing regions (e.g., finding novel ways to make ICT more relevant in addressing local problems), and (3) long-term contribution towards HCI research. The primary short-term deliverables were software, real deployments, and research papers. As for the individual investigators, all came from a technology and research background, collectively with experience in systems, networks, and HCI. B. ML ML began as a project funded by a social entrepreneurship program and the international health program of the School of Medicine at a U.S. research university. As part of the international research and education component of a medical program, there were no short-term deliverables. The individual investigator comes from a medical program with prior experience in medical research and a personal interest in consumer technology. His personal goal was to identify a project with sustainable, wide-spread social impact that had the potential to be financially self-sufficient and reproducible in other developing countries. C. Discussion These differing contexts had far-reaching implications for how the problem was framed. First, coming from a more technological background, GCN investigators were in a better position to attempt technological innovations and, in particular given their knowledge of projects such as [9] and [22], on creative ways of addressing poor connectivity. The ML project’s lack of technological expertise limited the possibility of developing new technology. Second, ML, given its ties to a major U.S. hospital, had a natural predisposition to tackle problems internal to a specialist center, whereas GCN had no expertise in hospital administration and hence was more inclined to address the simpler logistics of the rural clinic. Third, ML’s agenda for large-scale social change predisposed it towards impacting the greatest number of doctors, who are of course concentrated in the urban South with better mobile phone coverage. GCN chose to focus design efforts on the problems of the North (such as those of the 14 doctors servicing half a million people in Ghana’s Upper West region), seeing this as the best use of limited resources. These different problem framings have various implications for the solutions developed and implemented, as we discuss in the following sections.

122 Table 2 – Healthcare institutions / partners Acronym Name and Description Ghana Medical Association GMA A voluntary association of 95% of Ghana’s doctors, representing their interests nationally Ministry of Health MOH The national health administration, responsible (through the Ghana Health Service) for all public hospitals Ghana College of Physicians and Surgeons GCPS Responsible for specialist education and certification Ghana Physicians and Surgeons Foundation GPSF U.S.-based non-profit organization promoting specialist education in Ghana Figure 5. Map of GCN current and anticipated partners.

IV. PARTNERING STRATEGY AND DEPLOYMENT The question of partner selection is a key one for ICTD deployments, as partners are the usual means by which technology makes it from the laboratory to the field. In this section, we examine the impact of local and non-local partners on technology design and deployment. A. GCN GCN chose partners based on the decision to deploy technology in hospitals to connect their doctors with the Ghanaian diaspora (Figure 5). The Ghana MOH was chosen as a local institutional partner because it had the central authority to allow the servers to be installed in public hospitals. With the initial approval of the MOH, GCN was able to work quickly with the public hospitals, conducting 121 interviews with relatively little trouble alongside iterative, incremental deployment of technology. Of course, the drawback of the centralized approach is that it requires convincing a risk-averse bureaucracy in advance, thereby running the risk of ‘over-selling’ the solution. In addition to the added effort, another unexpected difficulty of working through the MOH was overloaded communication channels. Advertising GCN to the doctors was a challenge because promotional material needed to work through the same mechanisms through which the MOH communicated, for example, minor procedural changes and optional seminars from pharmaceutical companies. So there were multiple instances when administration would send out notifications of GCN training sessions but only a small fraction of doctors would show up. GCN partnered with the Ghana Physicians and Surgeons Foundation (GPSF), a Ghanaian medical diaspora organization, to recruit medical consultants in the U.S. KNet, a small Ghanaian Internet service provider (ISP), provided local Web hosting. B. ML ML worked with a smaller number of partners, all nongovernmental (Figure 6). ML chose the GMA as a local institutional partner because it represented the interests of physicians and its leaders would immediately see the benefit of the project to the physicians as individuals, if not to the healthcare system as a whole. The program is framed as a GMA value-added member service (similar to other member benefits such as professional development seminars and quarterly publications). In contrast to the MOH information

Figure 6. Map of ML current and anticipated partners.

channels, doctors took personal initiative to read the GMA emails or check GMA bulletin boards; this lead to widespread awareness of ML. OneTouch was chosen as a technological service provider for political and pragmatic reasons. As the investigator stated: OneTouch was the national company. I wanted this to be Ghanaian – by Ghanaians, for Ghanaians. That says a lot more than ‘done by Ghanaians in concert with South Africans.’ [2] OneTouch had the means and the expertise to very quickly make the ML program a reality and the financial resources to support the program independently on an ongoing basis. Thus, while GCN’s deployment was incremental, ML’s was all-ornothing, a strategy consistent with ML’s social agenda for rapid, wide-scale impact. Of course, the drawback of working entirely through OneTouch was that the project was no longer under the control of the investigator but was now subject to the organizational vagaries of a for-profit corporation. That is, changes in OneTouch’s business priorities could result in the program being dropped as quickly as it was initiated. This question is immediately salient in light of the recent acquisition of OneTouch by a multinational carrier based outside of Ghana (in fact, in South Africa). C. Discussion In considering the non-local partners, GCN and ML bear striking similarities in that both were initiated in partnership with U.S.-based organizations hoping to foster communication with the Ghanaian medical community: GCN with GPSF and ML with a U.S. teaching hospital. However, GCN emphasized the role of GPSF throughout the course of its lifetime whereas ML stopped working with U.S. doctors in order to focus explicitly on the Ghanaian context. The investigator noted:

123

[B]y removing this component of having this international cross dialogue, and realizing that we needed to have intraGhanaian communication instead of inter-Ghanaian communication, I quickly came to the conclusion that I needed to do something to improve communication within Ghana. [2] The relevance of overseas medical consultants was ultimately one of the core reasons that ML could focus on mobile phones while GCN retained a focus on the Internet. With respect to deployment, this meant that GCN had the additional task of advertising at GPSF conferences in the U.S. and to other organizations in the West, diluting its efforts in Ghana. GCN’s partnership with the MOH, GPSF and KNet and ML’s partnership with the GMA and OneTouch were factors in the projects’ very different rates of adoption. Arriving in the country with equipment for the initial pilot deployment, GCN was deployed in four hospitals over the course of five weeks, garnering an initial user base of 73 doctors. Over a similar stretch of five weeks, MedicareLine went from being a conversation between two CEOs to a national program, and by four months later over 1600 doctors had used ML to make over a million phone calls across the country. On the topic of sustainability, the juxtaposition of these two projects raises some interesting questions surrounding the rhetoric of “organic adoption” and its impact on long-term sustainability of ICTD initiatives. That is, they reflect two different views of what ‘organic’ or ‘bottom-up’ adoption means. ML presented a more ‘organic’ adoption model in the sense that the program was announced and doctors could sign up according to their individual needs and interest. On the other hand, because one institution provides all of the technical and financial resources, there is less local ownership and control over the maintenance of each project. Essentially, there is no guarantee that OneTouch (or the GMA) will not unilaterally end the project. This creates a situation analogous to the many development projects which rely on inconsistent or limitedterm donor funding. GCN’s model involves a more laborious adoption process but benefits from complete in-hospital ownership. This is a different type of ‘organic’. This ‘organic’ involved more than just the end user; it involves the whole system of people and machines that need to be in place for this network to grow. Unfortunately, this also surfaces the issue of the lack of access to ICT expertise experienced by all but the largest urban hospitals; the benefits of decentralized ownership are compromised by the geographic concentration of ICT skills. GCN can be conceived as a centralized project (sponsored by MOH and adopted by hospitals) with decentralized deployment (core resources provided by hospitals) while ML is a decentralized project (adopted independently by individual doctors) with centralized deployment (core resources provided by OneTouch). Technology innovators (such as corporations) often exhibit a bias towards decentralized solutions distributed through markets. Yet there are many examples, of which GCN would be one, of a potentially valuable tool which does not become useful until it has sufficient infrastructure and

“network effects” to fulfill its potential. Balancing the costs and benefits of the approaches remains an open question. V. ASSESSING TECHNOLOGICAL INFRASTRUCTURE One might think that doing a baseline assessment of available technology infrastructure would be one of the most objective elements of an ICTD project, one that is a basic part of requirements analysis. In this section, we observe that GCN and ML drew on two very different sets of infrastructure assumptions and show that such assessments are highly influenced by personal and institutional context. A. GCN The GCN researchers designed the system around the pessimistic engineering assumption that the system must continue to function under the worst-case connectivity conditions in the rural North – i.e., that all options available are unreliable and often low-throughput – and optimistic assumptions about PC usage. The exploratory needs assessment fieldwork found that the quality of mobile phone connectivity was unacceptable in the North. It is generally so oversubscribed that if you are calling a mobile phone from another service provider, you need to dial 10-15 times in order to successfully connect. [1] GCN made no quantitative measurement of the quality and reliability of network connectivity, instead relying on qualitative descriptions of regular power and network outages in formulating design requirements. With respect to the technology baseline of users, the GCN needs assessment found that: About a third of the physicians interviewed access email regularly… whereas another third claimed that they had at some point accessed email regularly. [1] Thus, it was expected that only a minority of doctors would actually need to acquire computer literacy. B. ML The ML researcher made optimistic assumptions about mobile phone connectivity among its user base and pessimistic assumptions about PC usage. As with GCN, connectivity was assessed qualitatively, drawing on experiences during the four-day trip to the North wherein calls from a OneTouch phone were compared with calls from another provider. While there were occasions of calls dropped or text messages delayed, ML concluded that the quality was sufficient for most basic uses. With regards to computer literacy, ML concluded that: The daily routine of a physician in Ghana does not revolve around computers. I can’t go a day as a physician [in the U.S.]... without using a computer – taking orders to checking lab values – but in Ghana it’s all people ordered, from medical records to orders, operative notes… Everything is done by paper! [2] These observations are actually consistent with those of GCN, but what gives these findings an added dimension is the ability of the ML researcher to make a direct comparison of this

124

Figure 7. Urban Hospital (DSL), Wed-Tue

Figure 8. Teaching Hospital (VSAT), Thu-Wed

Figure 9. Rural Hospital (VSAT), Mon-Thu

Packet Loss Rates (%) and Round-Trip Average (ms.) for three hospital servers in Ghana. Left y axis shows median, 5th and 95th percentile of RTA.

environment with his experience in a leading hospital in the U.S. Thus, while GCN framed its findings as “most doctors do indeed have some experience with computers” and would use them more often given better access, ML concluded that “computers are not part of the daily routine of a Ghanaian physician” and so are not a useful option. C. Discussion Part of the difficulty in producing consistent assessments of conditions on the ground is that while metrics such as the frequency of power outages and the availability of DSL can be measured and mapped in great detail, it is up to each investigator to determine exactly what metrics need to be measured, how rigorous the measurement needs to be, and what quality of service constitutes something usable by the target community for the specific application. GCN had no resources or expertise with which to improve the mobile phone infrastructure, so it focused more on network infrastructure. ML had no expertise with network infrastructure, but instead saw an opportunity to address a very significant, non-technical barrier in mobile phone usage. A closer look at the details of connectivity suggests that both projects were somewhat optimistic in downplaying the infrastructural limitations. The qualitative assessments verified expected connectivity barriers, whereas additional issues remained hidden.

Figure 10. OneTouch GSM Coverage Areas. (Source: GSM Association)

GCN anticipated power outages and network disconnections occurring several times a day, but the regularity with which (nominally acceptable) bandwidth was unusably low was a surprise. For example, certain pieces of systems software on the local servers needed to be configured with estimates of the worst-case time needed to transfer an 8KB file chunk; the initial estimate of 20 seconds (~3.3Kb/s) was eventually increased to 5 minutes (~0.2Kb/s). Connectivity data obtained after the deployment of GCN show that the difference between best- and worst-case performance can be extreme. Figures 7, 8 and 9 illustrate the volatility of network performance at three sample hospitals. Figure 7 shows a DSL network connection with moderately variable packet loss rates and round trip times. However, Figure 8 illustrates the case of a satellite connection so overloaded that, on weekdays between 10:00 AM and 4:00 PM, packet roundtrip times to the public GCN server in Ghana consistently exceeded 10 seconds and loading google.com took 15 minutes. While the GCN software was successfully able to transmit doctors’ messages when congestion decreased in the evening, a doctor who requested a consultation in the morning would wait until the following day to receive a response (even if the consultant responded immediately upon receipt of a message). (The irony here is that this hospital, with the largest number of dedicated IT staff - 6 people - thus experienced the most apparent ‘outages’.) In contrast, Figure 9 shows the characteristics of a satellite connection that worked reliably and consistently. As ML is not based on Internet connectivity, an analogous examination of ML’s assessment of mobile phone coverage would require assessing the quality of service of OneTouch voice calls and text messages. While we do not have this information, we can look instead at geographical coverage (Figure 10). While OneTouch has more coverage than any other provider in Ghana, the coverage map shows that vast regions of the country remain out of coverage area, leaving doctors working in those regions at a significant disadvantage. Clearly, a trip to a single urban center in the North does not systematically gauge the limitations of coverage throughout the country, let alone quality of service within current coverage areas. To be clear, the issues illustrated above have not been a cause of project failure in either case. Further, exhaustive

125

quantitative assessment of all such issues in advance of deployment is not always cost-effective or even possible (for example, one cannot measure DSL links that are not installed). What we suggest here is that care must be taken to prevent “blind spots” in assessing infrastructure and that such blind spots can arise from the investigators’ backgrounds. VI. SOLUTION USAGE AND APPROPRIATION In this section, we discuss the usage of the technological solutions that arose from each project as well as the ways in which the solutions were appropriated for uses other than those they were intended to enable. We then discuss the role of what we call technological specificity in these processes. A. GCN As described previously, GCN adoption has been relatively slow; what is notable is that GCN experienced a better rate of adoption in the smaller hospitals than in the large ones. At the smallest deployment hospital, with only 2 doctors on staff, one of the doctors continues to log into the system and post cases on a roughly bi-weekly basis. In the mid-sized hospitals, the response was mixed, with a handful of doctors using the system every week or two but the majority losing interest. At the largest deployment hospital, however, after the initial flurry of activity surrounding the presentations and training, none of the doctors continued to use it on a consistent basis. Follow-up interviews shed some light on this disparity. One issue was computer access. In the smallest district hospital, the doctors shared one computer, but since there were only two doctors, having access to the computer was never a problem. In contrast, access to computers was an issue at the larger hospitals. One doctor said: Sometimes you go to the library, you see someone at the computer for 15 minutes; you don’t have that time to waste [waiting for the computer]. Another issue was the match between the system’s use case and the needs of the pilot participants. Through its design and pool of GPSF consultants, GCN had targeted general practitioners (GPs) in the North who wished to consult with urban and overseas medical specialists. This matched the needs of district hospital doctors in the South who had no ready access to specialist consultation and saw great value in the ability to connect with specialists (either in Ghana or overseas). The popularity of this system in the district hospitals is a very promising precedent for future deployments in the more rural North. In contrast, physicians at the larger Southern hospitals could speak and consult with other doctors and specialists more easily in-person than over the computer. GCN did not address the needs of urban specialists looking to (In tap into the global community of sub-specialists. evaluation interviews, many of the urban doctors requested a greater number of sub-specialists within the system.) GCN Appropriation Beyond the expected use for consultation, there were unanticipated uses of the system. First, 6 of the 39 ‘conversations’ observed were purely social in nature. This

was surprising in light of how ‘medicalized’ the investigators perceived the interface. Yet perhaps because their means of communicating with remote doctors are so limited, it appears the barriers between profession and person are much more fluid than anticipated. Second, many of the doctors discussed and requested functionality in the system for sharing literature and PowerPoint presentations, to the extent that these features were included in the upgrade from the pilot to the ongoing deployment. This is consistent with the findings mentioned earlier that a larger screen size was important in dealing with more information-intensive tasks, and also provides insight into the kind of tasks matched to a computer’s affordances. B. ML As previously noted, ML experienced an incredible rate of adoption, with 1700 of 2000 members of the GMA signing up within the first four months. While there are no statistics available concerning the fraction of usage that is related to consultation, there is little doubt (judging from the interviews and from multiple instances of observing doctors as they received calls) that ML is used frequently for consultation. There are multiple reasons for this popularity, some of which are hard to distinguish from the affordances and popularity of mobile phones themselves. (One doctor claimed his phone bill dropped from 150 USD per month to 8 USD per month after joining the program.) First, mobile phones are a popular medium for medical consultation because the realtime nature of voice calls is often critical to treating an emergency case – the three doctors who volunteered information on the breakdown of emergency/non-emergency cases reported that around 80% of cases that require further consultation are indeed emergency cases. One said: I prefer phone calls to SMS, because I prefer an immediate answer, and also so I can make sure the phone is on. If I’m dealing with a case right now, I want to know what to do when moving ahead as soon as possible. Second, phone skills are more widespread than computer skills. Because mobile phones are practically ubiquitous in Ghana, there is a lower learning curve as opposed to computers, which are owned by only one in three doctors. A few doctors claimed that texting on a mobile phone was easier than typing on a keyboard. While many of the junior doctors demonstrated great proficiency with both typing and texting – the vast majority sent multiple text messages a day – a few of the senior doctors demonstrated great difficulty typing during training sessions. Third, it promotes more tightly-integrated workflows. One doctor said: I use [ML] a lot and I think it is wonderful. If you want to talk to anybody concerning a case… concerning anything relating to your practice... it gives you a chance to relax and really talk. It’s so good. It’s a wonderful idea. Another said: There has been a move to ban mobile phones in certain hospitals. It is a very very big mistake, because all they are going to realize is that this is actually going to decrease

126

efficiency rather. Consultation is not going to be working as well as it used to. ML Appropriation ML was originally framed as a system to ensure continuity of care in long-term and referral patients, but after the system had been established for three months the ML investigator noticed its emergent effect in fostering camaraderie in the medical community. He said: I was talking to doctors and they were telling me. “Yeah, I’m reconnecting with classmates.” Indeed, many doctors were up-front about the fact that ML had been a boon, not just for consultation, but for facilitating social interaction within their community. The mobile phone has also integrated itself into administrative and management processes within large hospitals. One doctor said: You don’t have walk down somewhere or you send a patient down... it reduces the whole bureaucratic... pushing around of patients. Personal mobile phones, both on the ML program and not, are regularly used to set up diagnostic tests at the laboratory, to confirm insurance claims forms, and also for the doctor on call during the night to discuss a change in a patient’s care with the doctor who admitted that patient. In short, where a U.S. hospital might use infrastructure such as pagers and site-wide internal communications systems, large Ghanaian hospitals improvised solutions using personal mobile phones. C. Discussion In reflecting upon differences between GCN and ML in usage and appropriation, we will focus on an important distinction that we will call ‘specificity’. Two solutions with different ‘specificity’ can be targeted at exactly the same task and be based on an equally nuanced understanding of workflows and use cases; the difference lies in the types and number of layers of technology which make up the solution and the degree to which they are specific to the solution. For example, the GCN system presents not only a robust asynchronous communication medium, but also an in-hospital server as well as an end-user Web application. ML, in its first phase, focuses exclusively on tackling the cost barrier of existing phones using existing networks. As we have seen, these two projects with similar goals at the start ultimately resulted in two very different usage outcomes – particular in the area of adoption. Investing time and resources in needs assessments and design process, the GCN project produced a highly ‘specific’ system to address not only failures in internet infrastructure but also social network gaps. However, adoption has been slowed by the need to introduce the system incrementally into hospitals. The ML project achieved broader adoption over a much shorter period of time, in part by relying on the existing availability and high adoption levels of mobile phones. Similarly, the lower specificity of the ML solution seems to contribute to a greater range of user appropriation behaviors.

While it is tempting to conclude that providing solutions with lower specificity is strictly more desirable – and in many cases it may be more desirable – it must also be remembered that utility comes in many forms. GCN’s higher degree of specificity is due to multiple factors. First, the social networking application is required to meet the GCN goal of enabling isolated doctors to build social capital in an extended geographic network of colleagues. While ML assumed that the doctors that needed to work together already had each other’s mobile phone numbers, GCN had determined that a large portion of rural, immigrant, and junior doctors did not have a strong network of contacts [20]. Second, GCN’s emphasis on overseas consultants implies a need for lowbandwidth, asynchronous communication as opposed to voice calls. Third, GCN operated as a development project, maintaining statistics of usage metrics in order to facilitate evaluation; for now, ML relies entirely on the built-in reporting mechanisms in the OneTouch network. GCN can map consultations made to specific case outcomes, while OneTouch – which does receive very high praise from enrolled doctors who offered their feedback on the system – cannot. Moreover, as previously mentioned, GCN’s goals include technological innovation as well as social impact. GCN assumes connectivity in rural areas will remain an ongoing challenge, while ML relies on the assumption that, with time, OneTouch coverage will be able to reach even the most remote doctors. This question echoes one of the fundamental tensions in ICTD research: as pragmatists, we aspire towards demonstrating the greatest social impact in real communities today, while as researchers, we try to identify what fundamental limitations exist and how these can be tackled in years to come. It is to be expected that untested technology would experience more hurdles in the short run, while its long-term contribution is yet to be seen. In short, then, a project can easily have a number of goals that might be frustrated by a lower-specificity solution. A final note about lower-specificity solutions concerns their potential to be too widely appropriated. In the smallest district hospital (where GCN experienced greater adoption), both doctors tended to switch off or mute their phones while at the hospital in order to minimize distraction. Even in the larger urban hospitals, half of the 20 doctors who discussed their mobile phone usage claimed to keep their phones off while working because of distraction. Hospital administrators have related concerns; at one teaching hospital, a memorandum was circulated to all the doctors banning the use of mobile phones in many locations …to forestall the negative impact of mobile phone frequencies on medical equipments and improve the work ethics of staff. It is also announced… that it is a serious offence to disconnect life-supporting equipment in order to use their power sources (socket) to charge mobile phones. In view of concerns such as these, we suggest that researchers keep in mind both the (immediate) benefits and (eventual) costs of rapid adoption.

127

VII. RELATED WORK ICTDs for Healthcare in the Developing World The research on ICTDs to promote healthcare has a long and rich history [15],[34]. In the context of developing regions, remote medical consultation has been a popular and relatively successful approach [32]. Computers [3], mobile phones [25] and a combination of both [27] have demonstrated utility in a variety of settings. Framing the Problem [21] provides a useful overview and categorization of how ICTs are conceptualized from a variety of different fields. Theory on ‘framing the problem’ can be drawn from socialconstructivism, although we are not advocating here for a change from ICTD’s traditional focus on pragmatism and advocacy/participatory research. Much of the work on qualitative methodology emphasizes the importance of reflexivity and self-awareness in order to minimize such bias [8], but the fact remains there are institutional accountabilities which no amount of methodology can shake. We can learn from the example of anthropologists who have challenged their own role in the colonial apparatus [11],[30]. For example, [17] provides an interesting comparison of how different sectors tackle the management of kiosks in India. Partnerships and Deployment ICTD researchers have an important role to play at the crossroads of business administration and government policy, identifying how both contribute to the goal of development and how their contributions interplay. Many examples in ICTD research and development literature study the role of institutional players, although these papers are typically evaluative rather than action-oriented [18],[23],[24]. The literature on partnership selection and cultivation in ICTD is small but growing rapidly [26],[28],[31]. Specificity In reaction to the variable success of many deployments of general ICT, such as a number of telecenter and ‘hole-in-thewall’ computing initiatives, much of the focus of the HCI ICTD research community has turned towards purpose-built technology [4],[33]. Many of those doing systems research focus instead on providing basic computing or connectivity functionality [13],[29]. [6] presents a useful survey of ICTD literature which mentions a pre-1999 trend towards non-

specific technologies such as connectivity and a post-1999 trend towards specific solutions such as software and VOIP. VIII. CONCLUSION In this paper, we have presented parallel case studies of two ICT projects addressing the need for improved medical consultation among doctors in Ghana. We have examined how the two projects have been shaped by the institutional context and the identity of their researchers. We looked at how the partnerships formed affect the solution outcome, delivery, and adoption; we argued that even the task of assessing technological infrastructure is far from objective; and we noted the implications of the ‘specificity’ of each solution. We conclude with an open question raised by these discussions, summarized in Table 3. In ICTD research, much attention has been given to the socio-cultural, political, and economic contexts of target communities – yet ‘difference’ is a measure between communities; it is only by critically examining our own research community that we can understand the influence and impact of communities on each other. ICTD is linked with economic, social, political, and human development agendas [6]. Regardless of whether we take an integrated approach [10] or focus instead on specific local needs [14], there are still institutional, societal, and individual layers at play in our interventions. Researchers are embedded in contexts of existing friendships, collaborations, expertise and agendas, and we need to be conscious of what kind of consequences our decisions to draw on them have on project outcomes. In this paper we have seen impacts on everything from research agendas to infrastructure assumptions, yet the literature on these avenues of choice remain fragmented across a variety of other communities including CSCW and organizational behavior. ICTD researchers need to increase the exchange of ideas between these communities. For example, in the course of establishing GCN, the task of generating a viable ‘business model’ for the technology providers was not addressed, but a social entrepreneurship community would never allow such an omission. ML has had great adoption success, but technical infrastructure for evaluating its impact was neglected. We need a theoretical framework under which to unite the different fields of research for ICTD practitioners, one surrounding not just technology design, but also technology framing, partnership, assumptions, and deployment.

Table 3 – Impacts of project context on project decisions and outcomes By focusing on connectivity in the rural North, GCN targets a minority of doctors with the greatest need. Framing By focusing on airtime cost, ML impacts the majority of doctors, the bulk of whom are in the urban South. Working through MOH allowed GCN to iterate rapidly with many hospitals early on, but with greater overhead later. GCN`s Partnerships and commitment to the U.S.-based GPSF was a key factor that tied it to a computer-based platform instead of mobile phones. Deployment GMA advertised ML effectively while OneTouch quickly took ownership of project execution and maintenance. ML’s approach is highly dependent on the vagaries of a single national operator while GCN’s is at the mercy of individual hospitals. GCN found the networks to be more challenged than anticipated. Infrastructure ML found mobile phone coverage lower than anticipated. GCN`s higher specificity allowed it to incorporate evaluation indicators and address not only physical communication gaps but also Solution gaps in social capital. Usage limited by access to computers in larger hospitals. ML adoption was much higher given the real-time nature of mobile phones and the fact that they were already widely tested and used. Lower specificity facilitated adoption and appropriation while making it difficult to evaluate the program.

128

REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]

[11] [12] [13] [14] [15]

[16] [17]

Field notes from the GCN investigators’ journals. Phone conversations between the GCN investigators and the ML investigator. K. Brauchli, D. O’Mahony, L. Banach, and M. Oberholzer, “iPath – A Telemedicine Platform to Support Health Providers in Low Resource Settings,” J. IT in Healthcare 3, 4 (2005), 227-235. P. Braund and A. Schwittay. “The Missing Piece: Human-Driven Design and Research in ICT and Development,” Proc. ICTD 2006, IEEE (2006), 2-10. E Brewer, M. Demmer, B.W. Du, M. Ho, M. Kam, S. Nedevschi, J. Pal, R. Patra, S. Surana, and K. Fall. “The Case for Technology for Developing Regions,” IEEE Computer 38, 6 (2005), 25-38. J. Bussell. “International Norms on ICTs for Development: New Data, Initial Findings, and Opportunities for Analysis,” American Political Science Association Annual Meeting, APSA (2005). V. Cerf, S. Burleigh. A. Hooke, L. Torgerson, R. Durst, K. Scott, K. Fall and H. Weiss. Delay-Tolerant Networking Architecture, RFC 4838, IETF, 2007. J.W. Creswell and V.L. Plano Clark. Designing and Conducting Mixed Methods Research. SAGE, Thousand Oaks, CA, 2007. M. Demmer. A Delay Tolerant Networking and System Architecture for Developing Regions. Ph.D. dissertation, Univ. of California, Berkeley, CA, 2008. R. Duncombe and R. Heeks. “Information, ICTs and Small Enterprise: Findings from Botswana,” In H. Katrak and R. Strange (eds.), SmallScale Enterprises in Developing and Transitional Economies. Palgrave, Basingstoke, UK, 285-304. P. Dourish. “Implications for Design,” Proc. CHI 2006, ACM (2006), 541-550. K. Fall. “A Delay-Tolerant Network Architecture for Challenged Internets,” Proc. SIGCOMM 2003, ACM (2003), 27-34. R. Fonseca and J. Pal. “Computing Devices for All: Creating and Selling the Low-Cost Computer,” Proc. ICTD 2006, IEEE (2006), 11-20. R. Gerster and S. Zimmerman. “ICTs for Poverty Reduction,” Information for Development Online, Jul-Aug 2003. http://www.i4donline.net/issue/july03/pdfs/icts_poverty_reduction.pdf W.R. Hersh, D.H. Hickam, S.M. Severance, T.L. Dana, K.P. Krages, and M. Helfand. “Diagnosis, Access and Outcomes: Update of a Systematic Review of Telemedicine Services,” J. Telemed. Telecare 12, S2 (2006), 3-31. D. Kuo, D.R. Gifford, and M.D. Stein. “Curbside Consultation Practices and Attitudes Among Primary Care Physicians and Medical Subspecialists,” JAMA 280, 10 (1998), 905-909. R. Kuriyan and J. Bussell. “Understanding Concepts of ‘Development’ and Linkages to Deployment Strategies in ICT4D in India,” in A. Palackal and W. Shrum (eds.), Information Society and Development: The Kerala Experience. Rawat, New Delhi, 2007.

[18] R. Kuriyan and I. Ray. “Public Private Partnerships and Information Technologies for Development in India,” Proc. ICTD 2007, IEEE/ACM (2007), 310-320. [19] B. Latour. Science in Action. Harvard Univ. Press, Cambridge, MA, 1987. [20] R. Luk, M. Ho and P.M. Aoki. “Asynchronous Remote Medical Consultation for Ghana,” Proc. CHI 2008, ACM (2008), 743-752. [21] W. Orlikowski and C.S. Iacano. “Desperately Seeking “IT” in IT Research – A Call to Theorizing the IT Artifact,” Information Systems Research 12 (2001), 121-134. [22] R. Patra, S. Nedevschi, S. Surana, A. Sheth, L. Subramanian and E. Brewer. “WiLDNet: Design and Implementation of High Performance WiFi Based Long Distance Networks,” Proc. NSDI 2007, USENIX (2007), 87-100. [23] L.R. Peattie, Rethinking Ciudad Guyana. Univ. of Michigan Press, Ann Arbor, MI, 1987. [24] E. Peter. “The Challenges of the Institutional Turn: New Interdisciplinary Opportunities in Development Theory,” in V. Nee and R. Swedberg (eds.), The Economic Sociology of Capitalism. Princeton Univ. Press, Princeton, NJ, 2005. [25] K. Phipps, G. Sangudi, and S. Woolway. What Works: HealthNet Uganda’s Evolution from NGO to Sustainable Enterprise: Portable Healthcare Service Delivery to Uganda’s Rural Areas. World Resources Institute, 2003. [26] D. Ramachandran, M. Kam, J. Chiu, J. Canny, and J.F. Frankel. “Social Dynamics of Early Stage Co-Design in Developing Regions,” Proc. CHI 2007, ACM (2007), 1087-1096. [27] J. Royall. “SatelLife – Linking Information and People: The Last Ten Centimeters,” Development in Practice 8, 1 (1998), 85-90(6). [28] Y. Schwartzman and T. Parikh. “Establishing Relationships for Designing Rural Information Systems,” Extended Abstracts, CHI 2007, ACM (2007), 1845-1850. [29] A. Seth, D. Kroeker, M. Zaharia, S. Guo, and S. Keshav. “Low-Cost Communication for Rural Internet Kiosks Using Mechanical Backhaul,” Proc. Mobicom 2006, ACM (2006), 334-345. [30] L. Suchman. “Practice-Based Design of Information Systems: Notes from the Hyperdeveloped World,” The Information Society 18, 139-144. [31] S. Surana, R. Patra, S. Nedevschi, and E. Brewer. “Deploying a Rural Wireless Telemedicine System: Experiences in Sustainability,” IEEE Computer 41, 6 (2008), 48-56. [32] D.J. Vassallo, P. Swinfen, R. Swinfen, and R. Wootton, “Experience with a Low-Cost Telemedicine System in Three Developing Countries,” J. Telemed. Telecare 7, S1 (2001), 56-58. [33] M. Warschauer. Technology and Social Inclusion. MIT Press, Cambridge, MA, 2003. [34] P. Whitten and B.D. Sypher. “Evolution of Telemedicine from an Applied Communication Perspective in the United States,” Telemedicine and e-Health 12, 5 (2006), 590-600.

129

Improving Child Literacy in Africa: Experiments with an Automated Reading Tutor G. Ayorkor Mills-Tettey, Jack Mostow, M. Bernardine Dias, Tracy Morrison Sweet, Sarah M. Belousov, M. Frederick Dias, Haijun Gong

Abstract—This paper describes a research endeavor aimed at exploring the role that technology can play in improving child literacy in developing communities. An initial pilot study and subsequent four-month-long controlled field study in Ghana investigated the viability and effectiveness of an automated reading tutor in helping urban children enhance their reading skills in English. In addition to quantitative data suggesting that automated tutoring can be useful for some children in this setting, these studies and an additional preliminary pilot study in Zambia yielded useful qualitative observations regarding the feasibility of applying technology solutions to the challenge of enhancing child literacy in developing communities. This paper presents the findings, observations and lessons learned from the field studies. Index Terms—Developing Regions, Education, Educational Technology, Literacy

L

I. INTRODUCTION

is a key part of the global development agenda. It is a complex concept with multiple definitions. An international expert meeting at UNESCO in 2003 proposed the following definition: “Literacy is the ability to identify, understand, interpret, create, communicate and compute, using printed and written materials associated with varying contexts. Literacy involves a continuum of learning in enabling individuals to achieve their goals, to develop their knowledge and potential, and to participate fully in their community and wider society.” [22] The United Nations recognizes literacy as a human right, noting that basic education, of which literacy is the key learning tool, was recognized as a human right over 50 years ago in the Universal Declaration of Human Rights. ITERACY

The research reported here was supported in part by the discretionary gifts to the TechBridgeWorld research group at Carnegie Mellon University, by the Qatar Foundation for Education, Science, and Community Development, by the National Science Foundation under ITR/IERI Grant No. REC-0326153, by the Institute of Education Sciences, U.S. Department of Education through Grant R305A080628 to Carnegie Mellon University, and by the Heinz Endowments. The opinions expressed are those of the authors and do not necessarily represent the views of any of our sponsors. G. A. Mills-Tettey is with the Robotics Institute at Carnegie Mellon University, Pittsburgh, PA 15213 (phone: 1-412-268-8645; fax: 1-412-2686436; email: [email protected]). J. Mostow, M. B. Dias, S. Belousov, and M. F. Dias are also with the Robotics Institute at Carnegie Mellon University, Pittsburgh, PA 15213 (email: [email protected], [email protected], [email protected], [email protected]). T. M Sweet and H. Gong are with the Statistics Department at Carnegie Mellon University, Pittsburgh, PA 15213 (email: [email protected], [email protected])

As a contribution to the discourse on applying information and communication technologies (ICTs) to address development challenges, we describe a study exploring the potential role of computing technology in improving English literacy among Ghanaian and Zambian children who attend school in English but have low reading achievement levels. Low functional literacy among individuals who have completed primary school is not an uncommon problem in developing communities. For example, UNESCO reports that in 2000, more than 1 in 3 adults with a fifth-grade education in Chad and Niger reported that they could not read [21]. In other cases, individuals may finish primary school reading below the expected level. In a representative sample of Ghanaian public schools, reading achievement levels measured by the government-administered Criterion Referenced Test in 2000 indicated that fewer than 10% of the children in grade six were able to read with grade level mastery [9]. Several factors contribute to this problem. For the average child from a rural or low-income urban background in Africa, reading is not part of daily family life, and sometimes parents are not themselves literate. In Ghana and Zambia specifically, most children speak one of a number of local languages at home but attend a school taught in English, the official language for both countries. Typically, under-resourced schools with overcrowded classrooms offer few opportunities for individual attention while developing reading skills. The project described in this paper is a proof-of-concept study to investigate whether an automated computer-based reading tutor that provides guided reading practice can significantly improve the reading proficiency of children in a developing community, even if they have no prior familiarity with computers. We focus our study on children in Accra, Ghana, and Mongu, Zambia. As illustrated in Figure 1, we began by employing the Reading Tutor in a preliminary three-week-long pilot study in Accra, Ghana. The pilot study was used to explore technical and operational feasibility and to motivate partnerships and funding for a longer term study. We followed up with a fourmonth-long controlled study in Accra, in collaboration with the Ghana-India Kofi Annan Centre for Excellence in ICT (AITI-KACE) and with input from Associates for Change, an

130

educational research firm in Accra. The controlled study aimed at quantitatively measuring the educational effectiveness of the tutor with children in Accra. Observations and lessons from these experiences fed into another pilot study in Mongu, Zambia, with the plan of following up with an additional controlled study in Zambia in the future. Exploring the role of technology in improving English Literacy in Africa Existing tool: Automated Reading Tutor for guided oral reading practice Ghana

New tools

Zambia

June 2005

July 2008

Pilot study • Gauge children’s reactions to Reading Tutor • Explore out-of-school usage model • Assess feasibility of test instruments

Pilot study • Gauge children’s & teachers’ reactions to Reading Tutor • Explore in-school usage model • Explore government literacy requirements and tests

May-Sept 2007

Future work

4-month controlled study • Measure reading progress • Compare progress across 3 socio-economic groups • Assess sustainability of outof-school usage model

Year-long controlled study • Measure reading progress • Compare to nontechnological approach to guided reading practice • Assess sustainability of inschool usage model

Reported work

Proposed future work

Figure 1 - Project outline

Section II describes the skills that are involved in reading and introduces the automated tutor we used in this project. Sections III, IV, and V describe the pilot field study in Ghana, the 4-month-long controlled study in Ghana, and the pilot study in Zambia respectively. Section VI distills practical lessons learned in the implementation of these field studies and Section VII concludes with a discussion of future work. II. BACKGROUND As a cognitive proficiency, reading involves several component skills such as phonemic awareness, decoding, fluency, vocabulary, and comprehension [14]. Phonemic awareness is the ability to perceive individual sounds or phonemes in words. Building on this awareness, a child learns the alphabetic principle that spelling generally maps systematically to pronunciation. The child also learns specific letter-sound correspondences and the ability to correctly pronounce written words (decoding). Fluency is the ability to read text accurately, quickly, and expressively and is an important foundation for comprehension. A rich vocabulary is essential for comprehension and effective communication. Children develop these skills through a variety of experiences, including skilled instruction. Research has shown that regular guided oral reading plays an important role in developing

reading skills, particularly fluency and comprehension [14]. Such guided oral reading may happen in small groups in a classroom setting, or with parents at home. Technology may also assist in this process: Scientific Learning’s Reading Assistant [1] and Project LISTEN’s Reading Tutor [11] are examples of computer-based tools that use automated speech recognition to provide a guided reading experience for the user. Project LISTEN’s Reading Tutor has demonstrated success in improving the reading ability both of children whose first language is English [11][12][13] and of children learning English as a second language (ESL) in the United States [16] and in Canada [18]. The project described in this paper used Project LISTEN’s Reading Tutor, developed at Carnegie Mellon University. The Reading Tutor displays stories on a screen and “listens” to a child read aloud. By using speech recognition to analyze the child’s reading, the Reading Tutor is able to give graphical and spoken feedback. It gives help when it detects a long pause, a severely misread word or a skipped word, and also when the reader clicks for help. The tutor may speak the whole word out loud or decompose the word into syllables or phonemes and speak out each part while highlighting it. It may also give a “rhymes with” hint, or read the sentence by playing a fluent human narration to model expressive reading. The Reading Tutor includes a wide variety of stories at different reading levels. It takes turns with the child in selecting a story to read. It monitors the child’s reading progress and selects stories at an appropriate level for the child. For readers at early stages of development, the Reading Tutor also includes word-building exercises to develop knowledge of spelling-to-sound correspondences. Videos of the Reading Tutor in use may be found at Project LISTEN’s website [17]. III. GHANA PILOT STUDY A. Goal The 2005 Ghana pilot study aimed to evaluate the practicality of a technological approach to guided reading practice in Accra, and investigate the feasibility of conducting a longer term controlled study. Specifically, we wished to answer the following questions: 1. Partners and logistics: How feasible is it to engage partners and arrange the logistics for such a study? 2. Learning to use the Reading Tutor: How quickly do children with no prior computer experience learn to operate the Reading Tutor, and what instruction do they need in order to do so? 3. Speech: Does the speech recognition software perform acceptably with Ghanaian accents, and can the students understand the narrated speech? 4. Tutor content: Do the children find the reading material in the tutor engaging? 5. Usage sessions: What is an effective length for a tutoring session?

131

6.

Reading measures: Which test instruments can be used easily and effectively to assess reading proficiency in the Ghanaian setting?

B. Participants and Methodology We chose to focus on the needs of children from lowincome families attending public school because they have a high risk of low achievement in reading. In consultation with Associates for Change, we targeted children in grades two through four since this is a key period for developing reading skills after the initial adjustment to the primary school environment. We restricted our work to an urban environment where computers are more readily available, but where significant literacy challenges still exist. We employed an out-of-school usage model in which practice with the Reading Tutor was supplemental to regular school activities. The pilot study involved qualitative observations of children as they used the Reading Tutor. The study was conducted by the first author who is a native of Ghana. She was assisted by a local volunteer. Two groups of children participated in this study. One group comprised twelve children in grades two through four from an under-resourced public school. They used the Reading Tutor at an internet café near their school for 20-30 minutes each day over a threeweek period. The other group comprised six children from a mixed low- and middle-income neighborhood. They used the tutor on laptops in the home of the researcher, for 20-30 minutes each day, three days a week over the same period. C. Results and Observations Partners and Logistics: The school, internet café, and the parents of all participating children agreed without hesitation to participate in the project, and logistics were arranged very quickly. The internet café donated time on four desktop computers; thus, four children could use the Reading Tutor at a time. We engaged the remaining children in other activities while they waited their turn. This worked fairly well for the pilot project, but would not be a feasible approach for a longer study involving a larger number of children. Learning to use the Reading Tutor: Although one child had previously played a computer game, none of the other children had used a computer before. The children were given an initial 10-15 minute hands-on lesson introducing them to the computer and showing them how to use the mouse, keyboard and software. The children had trouble understanding the Reading Tutor’s built-in automated tutorials on how to operate it and use the keyboard, perhaps due to the unfamiliar narration accent or the use of words such as “roster” that are common in American but not Ghanaian English. This difference in language use is an example of the need to localize tools for a given setting. We explained the tutorials one-on-one to the children during their first two sessions. Some of the questions the children asked during the first few sessions were on keyboard use and on navigating the tutor. By their second or third session, most of the children were able to operate the Reading Tutor without help.

Speech: The speech recognition capability appeared to work adequately with the children’s accents: it accepted a great majority of words that were read correctly. Also, based on the graphical feedback given by the tutor, the children quickly learned to recognize when it did not “hear” them correctly and to repeat themselves when necessary. After the initial difficulty with the built-in tutorials, the children appeared to understand the prompts given by the Reading Tutor. Tutor content: Overall, the children were very enthusiastic about using the Reading Tutor. Most of them appeared to enjoy the world-building exercises and the stories in the tutor. However, we noticed that a couple of children who were older than the norm (11 or 12 years old) but had a kindergarten or first grade reading level were not fully engaged by the content of the simple stories, such as “Sam sat on a mat,” available for their reading level. There was a discrepancy between these children’s maturity and their reading ability. Furthermore, we believe it would be good to incorporate more local content, such as Ghanaian folk tales, into the tutor. Usage sessions: We noticed that the better readers could use the tutor for more than half an hour at a time without getting bored, whereas the more challenged readers would tire after about twenty minutes, but would still look forward to their next turn. We thus decided to limit the length of usage sessions in the controlled study to 20-30 minutes. Reading measures: Although the pilot study did not aim to quantitatively measure reading progress, we identified and tested reading measures that we would use in the controlled study. The instruments we chose to use to assess reading ability were an oral reading fluency test [5] and the Test of Written Spelling (TWS) [8], both of which are hand-scored tests that have been frequently used in previous studies involving the Reading Tutor, and are psychometrically reliable, fast and easy to administer. The fluency test is timed reading exercise in which the child is given a gradeappropriate story to read. It is scored as the number of words read correctly in one minute. Fluency is essential to comprehension and is a sensitive measure of growth in proficiency [19]. The TWS is a dictation exercise in which the child writes down words read aloud by the tester in order of increasing difficulty. The test ends once the child has incorrectly spelled five words in a row, at which point it is assumed that the difficulty of the remaining words exceeds the child’s proficiency. This allows the same test to be used for multiple grade levels. TWS is scored as the number of words spelled correctly. Spelling is like decoding in that it tests printto-sound mappings but it is easier to assess reliably. We tried out the fluency test and the TWS with the pilot study participants, and they did not have trouble with the format of either test. To combat their nervousness about being examined, we avoided using the term “test.” We instead explained that these exercises were to help figure out how the Reading Tutor could assist them and we reassured the children that they would not be given a grade. We modified the passages for the fluency test to use Ghanaian names rather

132

than American names, to make them more recognizable to the students. The children were familiar with the dictation format of the TWS, but again were unfamiliar with some words such as “tardy” which are not regularly used in Ghanaian English.

gender boundaries into two groups, Tutor-1st and Control-1st, as shown in Table I. We used a two-treatment crossover study design: for the first half of the study, the Tutor-1st group used the Reading Tutor while the Control-1st group had no additional reading intervention, and for the second half, the roles were switched: the Control-1st group used the Reading Tutor while the Tutor-1st group had no additional reading intervention. Each half of the study lasted nine weeks (two months) during which there were daily usage sessions of approximately half an hour per child, although attendance and usage varied considerably. This crossover study design, illustrated in Table II, had the advantage of equity in the sense that all participating children had the opportunity to use the reading tutor. It also avoided having to pair similar children for comparison since each child essentially served as his or her own control. The potential disadvantage, which was that carryover treatment effects from the first half may have affected the second half, was mitigated by having half the children participate first in the control, while the others participated first in the treatment group.

Figure 2 - A child in Accra reading a story using the Reading Tutor

IV. GHANA CONTROLLED STUDY A. Goal Following on the successful pilot study, the goal of the controlled study in Ghana was to quantitatively measure the efficacy of the Reading Tutor in helping children improve their reading skills. Specifically, we wished to determine: 1. Does regular use of the Reading Tutor improve oral reading fluency and spelling? 2. Do treatment effects depend on other factors such as school/socio-economic background, gender, or grade level? In addition, we wanted to learn about the operational sustainability of an out-of-school usage model, taking into consideration installation and maintenance of the software, training of staff responsible for the day-to-day running of the project, transportation of the children between the school and the project site, and other logistics. The controlled study was deployed by project staff at AITI-KACE, with remote training and support by our team in Pittsburgh. B. Study Design / Methodology The controlled study involved eighty-nine children from three schools, representing three socio-economic backgrounds. The participating schools, recruited by Associates for Change, were S1: a private school in a middleincome community, S2: a public school in a low-income community, and S3: an informal educational program for highly disadvantaged children who have never attended formal school. The study involved children in grades two through four of S1 and S2, and in the “Intermediate” and “Advanced” levels (roughly corresponding to grades two and three respectively) of S3. The children were split randomly across school, grade and

Table I - Study participants by school, grade and gender Number of Children Tutor-1st Control-1st School Grade Female Male Female Male S1 Grade 2 3 2 2 3 (29 children ) Grade 3 5 2 1 3 Grade 4 1 3 3 1 S2 Grade 2 2 3 3 2 (30 children) Grade 3 3 2 1 4 Grade 4 3 2 3 2 S3 Intermediate 3 4 3 5 (30 children) Advanced 3 5 5 2

A battery of three fluency tests and one TWS were administered to all the children at the beginning of the study (pre-testing), between the two halves (mid-way testing), and at the end (post-testing). The testing was conducted at the children’s schools by AITI-KACE project staff. Three fluency tests were used to better estimate reading ability. The passages used for these tests correspond roughly to first, second, and third grade reading levels respectively, since a test close to the child’s current reading level should give a more sensitive measure of progress. We found that the scores on the three passages were highly correlated, and so for analysis purposes these scores were combined into a single mean fluency score. Table II – Cross-over Study Design Control-1st Group Tutor-1st Group Pre-testing of all children First half of cross-over Reading Tutor Control: no special (1st nine weeks) intervention Mid-way testing of all children Control: no special Reading tutor Second half of cross-over intervention (2nd nine weeks) Post-testing of all children

The study participants used the Reading Tutor in a

133

Experiment 1 (1st nine weeks) Experiment 2 (2nd nine weeks)

Table III – Modified Study Design Control-1st Group Tutor-1st Group Pre-testing of all children School + Control: school only Reading Tutor Mid-way testing of all children Control: No school No school + Reading tutor Post-testing of all children

D. Results and Analysis Figures 3 and 4 illustrate the pre-test scores of children at the three schools, representing their reading proficiency going into the study. It is clear that the S1 children had much higher levels of reading achievement than the S2 children who in turn had higher levels than the S3 children. Pre-testing for our study occurred about two months before the end of the school year. To provide some context for these scores, Table IV compares the fluency pre-test scores with end-of-year norms from schools in the United States [6]. By this standard, the average reading proficiency of the S1 students appears to be at or above the U.S. average whereas that of the S2 and S3 children is lower. Norms from schools in Ghana are not available for comparison. In analyzing each experiment, we focus on gains in fluency and spelling test scores. For experiment 1, the gain is the difference between pre- and mid-test scores; for experiment 2, the gain is the difference between the mid- and post-test scores. A positive gain indicates improvement in the child’s reading proficiency. In each experiment, we use a standard statistical t-test to compare the mean gains of the treatment and the control group. This test yields a p-value indicating how significant the difference is between the means of the two groups. For this analysis we consider p-values of less than 0.016 to indicate statistical significance. This value is smaller and thus more conservative than the commonly used threshold of 0.05 because multiple comparisons (due to the different

Fluency pre-test scores by school and grade 180 160 Mean fluency score

C. Deviation from Study Design In a normal crossover study, experimental conditions are held constant over the course of the study. However, due to logistical challenges that delayed the start of the project, the first half of this study took place while the students were attending school, whereas the second half overlapped with the school vacation. Because of the different conditions in place during the two halves of the study, we analyzed the results as two different experiments rather than as a single crossover study. The first experiment measured the effect of the Reading Tutor while the children attended school (and thus attended English class as usual). The second experiment measured the effect of the tutor while the children did not attend school (and so did not attend English class). This modified study design is illustrated in Table III.

schools) require an adjustment of significance levels [10]. We also compute a measure of effect size, that is, the magnitude of the treatment effect. The measure we use for effect size is Cohen’s d: the difference between group mean gains divided by the within-group pooled standard deviation [3]. The effect size is computed for each school because the treatment effect depends on the school. An effect size of 0.2 is generally considered small, 0.5 medium, and 0.8 large [4].

140 120

Grade 2

100 80

Grade 3 Grade 4

60 40 20 0 S1

S2

S3

School

Figure 3 - Fluency pre-test scores by school and grade TWS pre-test scores by school and grade 40 35 30 TWS score

computer lab at AITI-KACE because it had equipment and technical capability that their schools did not. The computers used for the project had 2.4GHz Pentium IV processors, 256MB of RAM, and 16GB hard drives.

25

Grade 2

20

Grade 3

15

Grade 4

10 5 0 S1

S2

S3

School

Figure 4 - TWS pre-test scores by school and grade Table IV - Fluency pre-test scores compared with U.S. norms Mean pre-test fluency (standard deviation) Year-end U.S. Norm [8] S1 S2 S3 (50th percentile) 2 / Int. 82.4 (44.8) 32.3 (25.9) 11.1 (8.5) 89 3 / Adv 123.3 (31.4) 45.8 (37.4) 20.4 (20.8) 107 4 132.7 (32.7) 59.8 (41.5) -123 Grade / Level

Experiment 1 results by school: Table V shows the results and t-test analysis for experiment 1, when the children were attending school. The reading proficiency of the S2 children who used the Reading Tutor (the “treatment” group) improved significantly more than those who did not (the “control” group), as evidenced by larger gains in both fluency and TWS. The S3 treatment group significantly out-gained the control group in fluency but not in spelling, although there was a positive trend. Finally, there was no significant difference in gains between the S1 treatment and control groups, for either test. Omitted from this analysis are five S1 children who were not present for the midway testing. Due to the small numbers, we did not break down the data to examine

134

gains for each grade level within each school. Table V - Comparison of treatment and control gains for experiment 1

Test

Treatment (Tutor-1st group, School N=41)

Control (Control-1st group, N=42)

Mean Gain (SD)

t-test comparison of treatment & control gains p-value

Effect size

S1

8.6 (20.5)

21.5 (11.7)

0.0833

(# words read S2 correctly per min) S3

61.4 (22.9)

23.4 (21.0)

<0.0001

1.731

13.4 (9.3)

5.1 (4.4)

0.0054

1.144

TWS

S1

-1.9 (4.9)

-0.2 (2.6)

0.2997

-0.457

(# words spelt correctly)

S2

3.9 (2.7)

0.3 (3.4)

0.0039

1.153

S3

2.8 (4.0)

0.6 (1.7)

0.0648

0.719

Fluency

-0.775

Experiment 2 results by school: Table VI shows the results for experiment 2, when the children were not attending school. Interestingly, there were negative fluency gains for the S2 children over the vacation, and these were more dramatic for the control group than for the treatment group. We discuss this observation in the Discussion subsection that follows. For the S2 children, the difference in TWS test score gains between the treatment and the control group was not statistically significant. Finally, there was no significant difference between the treatment and the control group in the other two schools for either mean fluency or TWS. Omitted from the analysis are two S1, two S2 and ten S3 children who were absent for testing. Table VI - Comparison of treatment and control gains for experiment 2

Test

Treatment (Control-1st group, School N=33)

Control (Tutor-1st group, N=37)

Mean Gain (SD)

t-test comparison of treatment & control gains p-value

Effect size

S1

0.3 (13.4)

-3.8 (22.6)

0.613

(# words read S2 correctly per min) S3

-9.1 (17.1)

-39.8 (16.8)

<0.0001

1.816

4.8 (6.0)

17.8 (27.0)

0.1321

-0.665

Fluency

0.222

TWS

S1

2.0 (4.1)

2.9 (5.1)

0.6431

-0.197

(# words spelt correctly)

S2

0.4 (2.0)

0.6 (3.0)

0.858

-0.068

S3

4.8 (6.0)

17.8 (27.0)

0.1321

-0.665

Effect of Gender: In addition to exploring how test score gains varied by school, we examined the data for evidence that test score gains were affected by gender. We focused this analysis on experiment 1 since that data was more expressive. A t-test comparison of the mean gains of females to that of males in each experiment resulted in large p-values (> 0.4), indicating that the mean gains of females and those of males were not significantly different from each other. An Analysis of Variance (ANOVA) also confirmed that gender did not have a significant effect on TWS or fluency gains in experiment 1. Details of these analyses are omitted for brevity.

E. Discussion Effectiveness of the Reading Tutor: The results provide evidence that during the school term, the S2 students (and to a lesser extent the S3 students) who used the tutor gained considerably more than those who did not use the tutor. Thus, the Reading Tutor was helpful for the S2 children. This is a positive outcome since the S2 group most closely represents our target population of children from low-income families attending public school. The proficiency of the S1 children did not appear to be influenced by reading practice with the tutor. A possible explanation for this is that the S1 children might not have had much room to benefit from the tutor since they were fluent readers going into the study. Loss in proficiency over vacation period: Of the statistically significant results highlighted in Tables VI and VII, the negative gains of the S2 children in experiment 2 stand out. A possible explanation for this might be found in studies in the United States which have documented that reading achievement test scores for children from low-income families deteriorate significantly over the summer vacation (a phenomenon referred to as the “summer reading setback”) whereas those for children from middle-income families remain steady or increase slightly [2]. This trend has been attributed to the discrepancy in the reading opportunities and materials available to these two groups over the vacation period. In light of these studies, it is interesting to note that for the S2 children, those who used the Reading Tutor did not deteriorate in reading ability as much as those who did not use it. However, the sessions with the Reading Tutor (totaling, on average, 11.5 hours of reading per child), without English class at school, were not enough to prevent negative gains. Effect of pre-test scores: A complicating factor in the data analysis is the observed unequal average pre-test scores of the two groups, Tutor-1st and Control-1st, despite supposed random assignment of children to the two groups. Table VII shows that this disparity is statistically significant for the S2 children. Higher pre-test scores can be the cause of greater gains [20], and so to determine whether the greater gains of the Tutor-1st group in experiment 1 were due to the Reading Tutor or the higher pre-test scores, we computed the correlation between pre-test scores and gains. We found no significant positive correlation between pre-test scores and gains: the correlation coefficient of mean fluency gains vs. mean fluency pre-test scores was −0.301 (a small negative correlation), and the correlation coefficient of TWS gains to TWS pre-test scores was 0.005 (no correlation). This suggests that the greater gains of the Tutor-1st group in experiment 1 are indeed attributable to the Reading Tutor rather than to their higher pre-test scores.

135 Table VII - Comparison of pre-test scores for the Tutor-1st group and the Control-1st group Tutor-1st Control-1st (N=41) (N=42) Test School Mean preMean pretest score test score S1 122.0 (45.9) 99.2 (33.9) Fluency S2 65.8 (33.9) 26.2 (26.9) (# words read S3 18.6 (13.2) 12.1 (19.1) correctly per min) S1 24.7 (9.7) 24.0 (9.6) TWS S2 11.1 (7.0) 4.3 (3.7) (# words spelt S3 2.1 (2.2) 1.9 (3.1) correctly)

2. 3. p-value 0.137 0.0015 0.308 0.850 0.003 0.840

Tutor usage: Another significant difference between the two experiments is that tutor usage in experiment 2 was much lower than in experiment 1, due to a higher level of absenteeism from the study during the school vacation. As Table VIII shows, this is particularly true for participation of the S3 students whose total usage per student in the second half of the study dropped to under a third of what it was in the first half. This reduced usage may partly explain why there is no statistically significant effect of the Reading Tutor for S3 children in experiment 2. Nineteen children who were absent for testing either at the beginning or end of each experiment are not included in this summary. Table VIII - Tutor usage in each experiment School Experiment 1 Experiment 2 Number of days of tutor S1 29.6 (12.0) 22.2 (9.4) use per student S2 36.9 (2.4) 24.0 (11.7) S3 30.7 (2.8) 12.6 (3.9) Total time spent reading S1 12.4 (5.4) 11.9 (5.2) with tutor per student S2 18.6 (3.1) 11.5 (6.7) (hours) S3 17.3 (2.7) 6.8 (2.5) Average daily time S1 22.9 (7.1) 32.6 (5.3) spent reading with tutor S2 30.1 (3.7) 26.6 (7.0) per student (hours) S3 33.9 (4.6) 31.9 (5.3)

V. ZAMBIA PILOT STUDY The work in Zambia complements the prior studies in Ghana by investigating an in-school usage model and testing the tutor in a different English-speaking African country. ProjectEDUCATE, a non-profit organization supporting some schools in Zambia, introduced us to two under-resourced public schools in Mongu, the capital of Zambia’s Western Province. The schools had received donated computers and were enthusiastic about the possibility of testing the Reading Tutor. However, the Mongu District Education Board Secretary’s office directed us to select only one school for the pilot test. Three of the authors conducted the study in Zambia. A. Goal The goals of the Zambian pilot study were to answer the following questions: 1. Computing infrastructure: What is the state of the school’s computing lab, how is it currently used, and can it feasibly be used for sessions with a computer-based reading tutor?

4. 5.

Training Teachers: How long does it take to train teachers to guide students in the use of the tutor? Children’s response: How do the children respond to the Reading Tutor? Test instruments and Reading Tutor content: What test instruments and Reading Tutor content are appropriate for the Zambian setting? Feasibility of long term study: Would a remote partnership between the school and our research group be a feasible model for a longer-term controlled study?

B. Methodology and Implementation The selection of the school for the pilot test was done after meeting with the headmasters and teachers at both schools and assessing the state of the computer labs and the potential for a successful study. Subsequently, the pilot study consisted of conducting interviews with teachers at the selected school, providing training for the teachers, and making qualitative observations of teachers and students as they used the Reading Tutor. We also explored with the teachers possible details of a longer controlled study. C. Results and Observations Computing infrastructure: Both schools had a computer room with about 20 computers, most of which were 266 Hz Pentium II machines with 64-128MB of RAM and 10GB hard drives. Some teachers from each school had taken a computer skills training course when the computers were initially donated, but had not had the opportunity to reinforce their skills through additional training or guided practice. The schools did not have internet access and the computers were being used primarily for tasks such as typing exam questions, rather than as educational tools for students or for accessing online teaching resources. Given the large class sizes—which ranged from 50 to over 100 students—and strict timetable, the teachers also faced challenges in feasibly using the one lab to teach their students from grades one to nine about computers. Some computers at both schools were not functioning due to broken keyboards, mice, and power strips; some computers were not being protected from dirt during the dusty winter months; power outages were a daily occurrence in Mongu; and for one school, even maintaining electricity for the computer lab was a challenge due to limited financial resources. For the pilot study, we selected the school with better maintained equipment and a higher likelihood of maintaining communication by telephone and email since the school had a telephone and the headmaster had a working email address that he accessed weekly. Training Teachers: We trained three teachers to use the Reading Tutor. The hour-long session covered the educational features of the tutor as well as administrative tasks such as managing users. After basic instruction and some time to practice on their own, the teachers were able to guide students in using the tutor. They would often give feedback to the children as a complement to the tutor when the students had

136

difficulty reading stories. For example, they would instruct the children to click for help when they needed it. The teachers would sometimes correct a mispronounced word that the tutor did not detect. We think this involvement of teachers in the early stages as the child gets used to the tutor could be an important part of longer-term use of the Reading Tutor in this school setting. A similar role was played by project staff in the Ghana studies. The teachers especially appreciated the ability to track their students’ performance using the tutor, including the number of new words the children had read and the number of words they read per minute. Children’s response: We observed one group of eleven children in grades two through four as they read one or two stories from the tutor. The students were selected by the teachers and had varying levels of reading ability, English comprehension, and speaking fluency. As in Ghana, all but one student were completely new to computers. We introduced them as a group to the basic components of the computer and then gave them the opportunity to demonstrate use of the mouse and keyboard to each other. We provided verbal instruction to students individually as they began reading stories from the Reading Tutor. Just as in Ghana, the speech recognition capability appeared to work adequately with the Zambian students’ and teachers’ accents. Some Grade 2 students seemed to have difficulty understanding English. This is because English is introduced as a language of instruction only in Grade 2, as a part of the Zambian Ministry of Education’s initiative to encourage basic literacy by teaching in a familiar language in the first year of school [21]. The students enjoyed using the Reading Tutor; when we returned to the school on another day to meet with the teachers, the group of students was using the word building exercises in the Reading Tutor on their own time.

Figure 3 –Teachers and a child test the Reading Tutor in Mongu

Test instruments and Reading Tutor content: We discussed with the teachers other reading assessment options for a longer term study, in addition to the fluency test and Test of Written Spelling. We learned that through the Zambia Primary Reading Programme (PRP), the students’ literacy levels are measured by reading standardized story books aloud to their teachers. Each color-coded book is associated with a given reading level. Once the teacher determines that the child can successfully read at a given level, he/she moves the child up to a higher level reading group with a different set of books. The teachers were interested in engaging the students with the computers in a way that would support curricular requirements from the Ministry of Education, and this was also emphasized to us by the Ministry of Education officials. Accordingly, we discussed incorporating the standardized PRP reading material into the tutor, which will also have the advantage of providing additional content choices for students. We saw that many of the existing Reading Tutor stories, such as those related to baseball, zoos, or recycling, had little relevance to children’s lives in Mongu. Feasibility of long term study: The teachers suggested that a random subset of the children in grades 2 through 4 could be selected to participate in a controlled study. They explained that students in these grades have 60 minutes of reading class per day and suggested that the intervention group of students could spend 30 minutes of that time working with the Reading Tutor, with teacher supervision, while the control group of students would remain in class. We anticipate several challenges in conducting this project as a remote field study. Although we have email and telephone contact information for the key collaborators at the school, we expect communication to be difficult given their limited internet access and the frequent disruptions in telephone communications due to daily power cuts. Success would depend on the teachers taking ownership of the project as a result of their enthusiasm. We hope that communication with local contacts will help address some of the expected challenges. For example, we plan to involve the Mongu District Education Board Secretary’s office in evaluating the project’s progress at the school. We also engaged a local ProjectEDUCATE technical contact who joined us at our meetings with the schools, received training from a member of our team, and assisted us with installation and use of the Reading Tutor software. We hope that he will stay in touch with us remotely, help incorporate new stories into the tutor, check in periodically with the school, and be available to the teachers to assist with any technical challenges. VI. LESSONS LEARNED From the experiences in Ghana and Zambia, we can glean many useful lessons for our future work as well as for others implementing similar studies. These can be broadly categorized into lessons about building relationships and support, lessons about running the study, and lessons about the viability of automated tutoring.

Figure 4 – Students in Mongu using word-building exercises as a group

137

A. Building Relationships An important requirement for running these studies was the process of developing partnerships at each stage in the project. Significant time must be devoted to building relationships and developing a shared vision. Phone conferences can be a useful tool, but some face-to-face meetings are essential, particularly in the early stages of project planning. A staged implementation, as in a pilot study followed by a controlled study, helps distill key questions to ask, refine design and implementation decisions, and identify potential problems. B. Logistics We learned that it was important for someone involved in the local day-to-day running of the project to have significant decision-making power and to feel ownership of the project. This individual must have the ability, for example, to replace equipment or interface with representatives of the school administration. Otherwise, problems can easily stall progress. It is also essential to have access to some local technical expertise to troubleshoot and repair problems—building this local support base during the pilot studies is crucial. For the controlled study in Ghana, we found that the remote collaboration required regular, sometimes daily, communication among the project partners; email, instantmessaging and voice-over-IP were a cost effective and feasible means of achieving this. Not surprisingly, we learned that a process that requires a significant change in behavior and extra work on the part of parents, such as having the children come to school or participate in a study during a vacation, is hard to sustain and should be avoided if possible. However, it was also clear that unexpected situations are bound to arise for any study that involves cross-continent collaboration and so flexibility and the ability to adjust the study design if necessary is essential. For example, in the controlled study in Ghana, the second half of the study had to be held during the school vacation although this was not the original plan. In running the Ghana studies, transportation of the children was the greatest expense and so although scheduled out-ofschool use in an ICT center was the model chosen for the controlled study, it is not a viable long-term usage model. However, as we learned in Zambia, there are many challenges to be addressed for an in-school usage model for public underresourced schools, even in a school that already has a computer lab. These issues include the limited availability of computers, inadequate experience on the part of the school in maintaining computer labs, scheduling challenges with respect to the school timetable, and also the need to work within the constraints of the existing literacy curriculum. C. Viability of Automated Tutoring A key lesson was that even without prior computer experience, the participating children were quickly able to acquire the skills needed to use the tutor. In general, they were excited about using the computer and about the interactive features of the Reading Tutor. Assistance from the project staff (in Ghana) or from the teachers (in Zambia) helped those children who were initially nervous about computer use or

about reading in English. In both locations, the children’s natural curiosity overcame any initial apprehension of the unfamiliar technology. In observing the children using the Reading Tutor, we noticed that children with a basic foundation in English and not much prior experience with computers, such as the S2 children in Ghana, were easily engaged with the tutor. It was clear that insufficient familiarity with the English language was a challenge for some, particularly the S3 children in Ghana and some of the Grade 2 students in Zambia. For children who did not yet understand English well enough to benefit from an automated tutor that uses only English, a tool that bridges between the local language and English, e.g. by giving prompts or explaining words in the local language, might have been better. Finally, we noticed that the S1 children who were already fluent readers and experienced with computers seemed to get bored and distracted easily when using the Reading Tutor. This might have been because the Reading Tutor displays the story being read one sentence at a time, and there would sometimes be a short delay in loading the next sentence. For a fluent reader, even this short delay was noticeable and sometimes frustrating. There were also instances that indicated that some children got bored of repeated activities and enjoyed variety. For example, many children enjoyed the word-building exercises in the Reading Tutor but would sometimes complain that there were too many of them. Although the Reading Tutor is designed for use by a single user at a time, we noticed that in both pilot studies children would gather around a single computer and try to help each other. This also happened to a lesser extent during the controlled study in Ghana and has been observed in other technology interventions in developing communities [15]. It suggests that we should investigate multi-user scenarios. Finally, there are several usability improvements that can be made to the Reading Tutor to reduce the required technical support, particularly in a developing community setting. These features would apply to any PC-based intervention. • Installation must be easy and straightforward. • Customization must be easy (both for adding local content and to control for the installation footprint for machines without much hard disk capacity). • An easy administrative interface is needed for controlling options such as level of logging, again to deal with limited-capacity machines. VII. CONCLUSIONS AND FUTURE WORK This paper presents our initial experiments in Africa with an automated reading tutor to improve child literacy. It investigates the viability and effectiveness of a computerbased reading tutor in improving the reading ability of children, particularly those attending under-resourced public schools. Although literacy and education for all are at the top of the global development agenda, not much work has been done regarding the role of technology in this process. This

138

work is a useful contribution both because it demonstrates that there is promise for the effectiveness of the approach and because of the practical lessons regarding the implementation of the study. We have presented an initial proof-of-concept investigation. Many additional questions would need to be explored to understand the potential for large-scale application of these technologies. Next steps for our project include trying to set up a year long in-school study in Zambia. The purpose of the study would be to test the feasibility of an in-school usage model in a developing community setting. It would measure the children’s reading gains over an entire school year of using the Reading Tutor and would compare these gains to those obtained with a non-technological approach to guided reading practice. An important goal of the year-long study will be to learn about sustainability and the feasibility of incorporating teachers closely into this work. Additionally, we will continue to develop an understanding of what features would be required of a literacy-enhancing tool developed specifically for use by children in developing communities. What features would be needed, taking into consideration limited resources and equipment, as well as current instructional practices? What tutorial methods are most useful? Could a lighter weight tool be designed for alternative platforms, such as mobile phones, as has been done with the educational games for English-as-a-second language instruction in the MILLEE project [7]? There is significant scope for further research on the role of computing technology in improving child literacy in developing communities.

[3] [4] [5] [6] [7]

[8] [9]

[10] [11]

[12]

[13]

[14]

ACKNOWLEDGMENT We are grateful to Steve Fienberg for his advice on the statistical analysis as part of the Statistical Practice course at Carnegie Mellon University. We express our appreciation to the Ghana-India Kofi Annan Centre for Excellence in ICT (AITI-KACE), Accra, Ghana, for their role in the controlled study, particularly to Dorothy Gordon for her support, Patricia Nyahe for her tireless management of the study, and the UNESCO Ghana Office for sponsoring their work. We also thank Leslie Casely-Hayford of Associates for Change, Ghana, for her advice and contributions to the pilot and controlled studies in Accra. We acknowledge the valuable contributions of Cybercity Internet Café and ProjectEDUCATE to the pilot studies in Ghana and Zambia respectfully. Finally, we are indebted to all the participating children and schools, and to all others who contributed to the project in various ways.

[15]

[16] [17] [18]

[19]

[20] [21]

REFERENCES [1]

[2]

M. J. Adams, "The promise of automatic speech recognition for fostering literacy growth in children and adults,” in M.C. McKenna, L.D. Labbo, R. D. Kieffer, & D. Reinking (Eds.), International Handbook of Literacy and Technology, Volume 2. Mahwah, NJ: Lawrence Erlbaum Associates, 2006. R. L. Allington and A. McGill-Franzen, “The Impact of Summer Setback on the Reading Achievement Gap,” in Phi Delta Kappan 85(1):68-75, September 2003.

[22]

J. Cohen, Statistical power analysis for the behavioral sciences (2nd ed.), Hillsdale, NJ: Lawrence Earlbaum Associates, 1988. J. Cohen, “A power primer,” Psychological Bulletin, 112, 155-159, 1992. S. L. Deno, “Curriculum-based measurement: The emerging alternative,” in Exceptional Children, 52(3):219–232, 1985. J. Hasbrouck and G. A. Tindal, “Oral reading fluency norms: A valuable assessment tool for reading teachers,” The Reading Teacher, 59(7), 636644, 2006. M. Kam, A. Agarwal, A. Kumar, S. Lal, A. Mathur, A. Tewari, and J. Canny, “Designing E-Learning Games for Rural Children in India: A Format for Balancing Learning with Fun,” Proceedings of ACM Conference on Designing Interactive Systems (DIS '08), Cape Town, South Africa, February 25-27, 2008. S. C. Larsen, D. D. Hammill, and L. C. Moats, Test of Written Spelling, Pro-Ed, Austin, Texas, 1999. M. Lipson and K. Wixson, “Evaluation of the BTL and ASTEP Programs in the Northern, Eastern, and Volta Regions of Ghana,” Report prepared by the International Reading Association for The Education Office, USAID/Ghana, August 2004. Available online at http://www.reading.org/resources/issues/reports/ghana.html R. G. Miller, Simultaneous statistical inference, 2nd ed., Springer Verlag, pages 6-8, 1981. J. Mostow and G. Aist, “Evaluating tutors that listen: An overview of Project LISTEN,” in K. D. Forbus & P.J. Feltovich (Eds), Smart machines in education (pp. 169-234), Cambridge, MA: AAAI Press/The MIT Press, 2001. J. Mostow, G. Aist, P. Burkhead, A. Corbett, A. Cuneo, S. Eitelman, C. Huang, B. Junker, M. B. Sklar, and B. Tobin, “Evaluation of an automated Reading Tutor that listens: Comparison to human tutoring and classroom instruction,” Journal of Educational Computing Research, 29(1), 61-117, 2003 J. Mostow, G. Aist, C. Huang, B. Junker, R. Kennedy, H. Lan, D. Latimer, R. O’Connor, R. Tassone, B. Tobin, and A. Wierman, “4Month evaluation of a learner-controlled Reading Tutor that listens,” In V. M. Holland & F. P. Fisher (Eds.), The Path of Speech Technologies in Computer Assisted Language Learning: From Research Toward Practice (pp. 201-219), New York: Routledge, 2008. National Institute of Child Health and Human Development, Report of the National Reading Panel. Teaching children to read: An evidencebased assessment of the scientific research literature on reading and its implications for reading instruction (NIH Publication No. 00-4769), Washington, DC: U.S. Government Printing Office, 2000. U. S. Pawar, J. Pal, K. Toyama, “Multiple Mice for Computers in Education in Developing Countries,” International Conference on Information and Communication Technologies and Development, ICTD 2006, pp.64-71, May 2006. R. Poulsen, P. Hastings and D. Allbritton, “Tutoring Bilingual Students with an Automated Reading Tutor That Listens,” Journal of Educational Computing Research, 36(2), 191-221, 2007. Project LISTEN Videos, available at http://www.cs.cmu.edu/~listen K. Reeder, M. Early, M. Kendrick, J. Shapiro, and J. Wakefield, “The Role of L1 in Young Multilingual Readers’ Success With a ComputerBased Reading Tutor,” Talk at the Fifth International Symposium on Bilingualism, Barcelona, Spain, April 2005. M. R. Shinn, N. Knutson, R. H. Good, W. D. Tilly, and V. L. Collins, “Curriculum-based measurement of oral reading fluency: A confirmatory analysis of its relation to reading,” School Psychology Review, 21:459-479, 1992. K. E. Stanovich, Progress in Understanding Reading: Scientific Foundations and New Frontiers, New York: Guilford Press, 2000. UNESCO, “Education for All Global Monitoring Report 2005 – Education for All: The Quality Imperative,” United Nations Educational, Cultural and Scientific Organization (UNESCO) Publishing, 2005. Available online at http://portal.unesco.org/education/en/ev.phpURL_ID=35939&URL_DO=DO_TOPIC&URL_SECTION=201.html UNESCO, “The Plurality of Literacy and its Implications for Policies and Programmes.” UNESCO Education Sector Position Paper. United Nations Educational, Scientific and Cultural Organization, Paris, 2004. Available online at http://unesdoc.unesco.org/images/0013/001362/136246e.pdf

139

Improving Literacy in Rural India: Cellphone Games in an After-School Program Matthew Kam, Anuj Kumar, Shirley Jain, Akhil Mathur, and John Canny

Abstract—Literacy is one of the great challenges in the developing world. But universal education is an unattainable dream for those children who lack access to quality educational resources such as well-prepared teachers and schools. Worse, many of them do not attend school regularly due to their need to work for the family in the agricultural fields or households. This work commitment puts formal education far out of their reach. On the other hand, educational games on cellphones hold the promise of making learning more accessible and enjoyable. In our project’s 4th year, we reached a stage where we could implement a semester-long pilot on cellphone-based learning. The pilot study took the form of an after-school program in a village in India. This paper reports on this summative learning assessment. While we found learning benefits across the board, it seemed that more of the gains accrued to those children who were better equipped to take advantage of this opportunity. We conclude with future directions for designing educational games that target less well-prepared children in developing regions. Index Terms—cellphone, English as a Second Language (ESL), literacy, mobile game, pilot study

L

I. INTRODUCTION

ITERACY is one of the great challenges in developing regions. Despite huge improvements in recent decades, literacy levels in many poor countries remain shockingly low. Even more challenging is the tension between regional and global “power” languages, that economic opportunities are often closed to those literate only in a regional language. For instance, India is a country with 22 regional and 2 national languages, i.e. Hindi and English. But English, together with computer skills, are the two most requested skills in surveys of poor parents [1]. English is a great economic enabler. It is the Manuscript received September 22, 2008. This work was supported in part by the U.S. National Science Foundation under Grant 0326582, a Qualcomm BREW Wireless Reach award, and sponsorship-in-kind from Sony Creative Software. Matthew Kam is an Assistant Professor in Carnegie Mellon University’s Human-Computer Interaction Institute (phone: +1 412-268-9805; fax: +1 412268-1266; email: [email protected]). He was with the Berkeley Institute of Design and the Department of Electrical Engineering and Computer Sciences at the University of California, Berkeley when this work was carried out. John Canny is the founding director of BID and holds the Paul and Stacy Jacobs Distinguished Professorship of Engineering at UC Berkeley. Shirley Jain participated in the pilot study as a curriculum developer and local supervisor. Anuj Kumar and Akhil Mathur are with the Dhirubhai Ambani Institute of Information and Communication Technology, Gujarat, India.

language of all professions and higher education, but also important for mid-level service jobs: retail, clerical, teaching, law enforcement, etc. that are the most common steps above menial labor. The value of English is widely recognized by ordinary Indians [2], and it is in fact the poorest citizens who are lobbying most strongly to expand English teaching. English is thus the language of power in India associated with the middle and upper classes [3][4]. In other developing regions, it is another language such as Spanish, Mandarin, or French which is not native to most of the population. We believe that many of our lessons will apply to other languages although our focus is on English as a Second Language (ESL). But the public school systems in developing regions face insurmountable difficulties. In India, for example, we were consistently unable to converse in English with those teachers responsible for teaching English in poor schools, where the overwhelming majority of children in the country struggle to learn. More important, public schooling is out of the reach of large numbers of children in rural areas and the urban slums who cannot attend school regularly, due to their need to work for the family in the agricultural fields or households [5]. At the same time, cellphones are increasingly adopted in the developing world, and an increasing fraction of these phones feature multimedia capabilities for gaming and photos. These devices are a promising vehicle for out-of-school learning to complement formal schooling. In particular, we believe that ESL learning games on cellphones present an opportunity to dramatically expand the reach of English learning, by making it possible to acquire ESL in out-of-school settings that can be more convenient than school. Games can make learning more engaging while incorporating good educational principles [6]. More important, a large-scale evaluation with urban slums children in India has shown significant learning benefits from games that target mathematics [7]. We believe that similar outcomes can be replicated with e-learning games that target literacy. The challenge in evaluating any language learning project, however, is that language acquisition is a long-term process on the learner’s part. Worse, with a novel technology solution that has yet to be institutionalized, there were tremendous logistical obstacles in running a pilot study over a non-trivial duration. After 3 years, in which we commenced with needs assessments and feasibility studies, followed by subsequent rounds of field testing interleaved with numerous iterations on our technology designs, we have established the necessary relationships with local partners for such an evaluation. This

140

paper describes the results from a semester-long pilot study – the longest so far in this project – which took place during the project’s 4th year. The study involved 27 rural children who participated in an after-school program that we implemented in their village.

conducting a learning assessment for a novel educational technology over a substantial timeframe when it is not yet integrated into the everyday operations of a formal entity.

III. OVERVIEW OF PILOT STUDY II. RELATED WORK Most work on technology-assisted language learning in the developing world does not explore the convenience that the cellphone’s mobility offers. Banerjee et al. [7] report a largescale evaluation with mathematics learning games on desktop computers, carried out over 2 years with urban slums children in India. Mitra et al. [8] describe a study in India with slums children over 5 months, which involved a “hole-in-the-wall” public computer installed with speech-to-text software. Dias et al. describe a computer-based tutor software for improving reading in Ghana [9]. Kothari’s karaoke-like approach [10] targets native language – not second language – literacy using television. With the cellphone’s increasing ubiquity in Africa, Brown [11] argues that it is timely to envision a future where the cellphone plays a pivotal role in education in Africa. Kam et al. [12] describe how a set of ESL-learning cellphone games that targets children in rural India have undergone numerous iterations, based on successive, short-term formative studies. With the exception of Kam et al., the only education-related projects we know of in the developing world that leverage the cellphone are Islam et al. [13] in Bangladesh, and Librero et al. [14] in Mongolia and the Philippines. Both projects rely on Short Messaging Service and target university students, unlike our approach. To date, Horowitz et al. [15] is the only study we know of that examines the cellphone for promoting literacy. However, this study took place in an industrialized country (USA), even though participants included households below the poverty line. In the study, Sesame Street videos that target the English alphabet were streamed to preschool children over cellphones throughout an 8-week period. Our paper therefore contributes to the literature as the first learning assessment on cellphonebased language instruction in developing regions. He et al. [16] describe a 2-year randomized evaluation of a LeapPad-like device that supports custom software modules for English learning. This interactive system involves a paper book attached to a stylus and supports audio output. It lacks a visual display, unlike a cellphone, but overlaps with our goal of making literacy learning more accessible in the developing world using portable devices. Among the learning technologies for developing countries, one of the most novel devices is the multiple-mice computer described in Pawar et al. [17]. It was intended for collocated learning by a group of children around each computer, with a mouse input device for each child. It has since been extended to distance learning in Moraveji et al. [18]. The evaluations in both papers were short-term, and underscored the difficulty in

The pilot study was carried out in collaboration with a nongovernment organization in North India under the terms of a Memorandum of Understanding. The study took the form of an after-school program, which we held during the afternoons at a private village school affiliated with this NGO. However, our goal was to investigate learning impacts that ESL learning games on cellphones have on lower-income rural children. As such, students who were already enrolled in this school were ineligible to participate in the study. Instead, we invited those parents who could not afford the fees for this private school – and hence sent their children to less expensive schools in the same area – to give consent for their children to participate. In the after-school program, we ran three sessions per week, on average. Each session lasted two hours in the afternoon. Children from neighboring villages attended the after-school sessions after finishing their regular classes in the morning. In the after-school sessions, we loaned cellphones preloaded with ESL learning games to participants. The after-school program took place from late December 2007 to early April 2008, and spanned sessions on 38 days in total.

IV. DATA COLLECTION As our preparation for this pilot study, we made two trips to India, i.e. once in the summer of 2007 to familiarize ourselves with the pilot location and end-user community, and a second time in December 2007 to kick-off the actual pilot. 4 local staff members were hired to run the after-school sessions on an everyday basis. 3 of them were engineering undergraduates in their last semester, while the last member had graduated a few years ago. We spent two weeks training them to run the after-school sessions and perform data collection, and continued to coordinate with them regularly via conference calls and emails after we left India. We interviewed participants on their demographics such as their ages and the grades they were currently enrolled in in school. During the interviews, we also asked other questions, such as the number of cellphones that their households owned, what they currently and/or had previously used cellphones for, their television watching habits and frequency, as well as their parents’ occupations. The questions on media and technology exposure were included because these variables were expected to impact participant ability to learn using cellphone games. To ensure that each participant has the basic numeracy and ESL literacy to benefit from cellphone-based learning in the program, participants were required to pass a qualifying test, i.e. obtain at least 50% of the total score. The test required them to complete one-word blanks using English words about themselves, e.g. name, age, school, grade, etc. They were also

141

asked to fill in the missing letters in the alphabetic sequence, write numbers in the Arabic notation, match words with their pictures, spell the words for everyday objects, and describe a picture of a market scene with short sentences. The qualifying test was designed such that an average child in India with no learning disabilities who has finished 1st grade in a reputable urban school should obtain a perfect score on it. By using the qualifying test as a screener, we ensured that participants were numerate. This was important because we had previously found it difficult to teach children to use the cellphone keypad’s to play e-learning games when they were not familiar with the numbers from 0 to 9 in Arabic notation. Similarly, by ensuring that participants were familiar with the English alphabet, we could target a more advanced syllabus that went beyond the alphabet. We made this decision since Horowitz et al. [15] had already investigated the efficacy of cellphone-based learning for the English letters in the context of preschool children in the USA. Since success in acquiring a second language is correlated with literacy in one’s native language, we administered a test which evaluated the ability of the participants to read in Hindi. Every child was given a short passage that described a diet for promoting dental health. Each child was then asked to read the passage aloud so that we could observe his or her fluency and accuracy. These sessions were videotaped. After that, every participant was asked to write answers to written questions that tested his or her comprehension of the passage, in Hindi. We had designed this test such that an average child who has finished 3rd grade in a reputable urban school in India should obtain a perfect score on it. Our primary method of assessment was to administer preand post-tests which evaluated participants on their ability to spell the common nouns that the curriculum for the pilot study targeted. Although the curriculum targeted other competencies such as listening comprehension and the recognition of written words, our assessment emphasized spelling, which as a recall task was cognitively more difficult than recognition tasks. We maintained attendance records for the participants for every session. We also videotaped each session so as to have contextual data that could potentially account for their test performances. The video recordings captured the classroom proceedings, and individual participants’ interactions with the games. The latter recordings captured participants’ levels of engagement with the games as shown in their facial and body expressions. The pilot staff member who was responsible for videotaping the sessions tried to ensure that every participant was videotaped playing at least one level in the curriculum per day. The recordings were later transcribed and translated from Hindi to English. Finally, for every session, we asked pilot personnel to write a report which summarized what happened in that session, as well as how well each participant interacted with the games. The latter not only covered usability and learning obstacles, but also included pilot staff’s observations on the attitude and persistence that each child demonstrated towards learning.

V. PARTICIPANTS Owing to the strong relationships that our NGO partner had built with the local community over more than a decade, we were able to generate a high level of support among parents in this community. In total, we obtained consent for 47 children to participate in the study. However, we needed to turn 16 of them away; 15 children did not pass the qualifying test while the 16th was attending private tuition for English. The latter represented a confounding variable. Of the 31 children whom we started the pilot with, 4 of them left the program mid-way. Reasons for attrition include time conflicts with private tuition (2 children) and disinterest in attending the sessions (another 2 children). From post-deployment interviews, we understand the latter was due to caste tensions between those 2 children, who belonged to the lower castes, and some upper-caste children in the program. A. Demographics The 27 children who participated in the study until it ended were aged 7 to 14 (mean = 11½ years) and belonged to grades 2 to 9 (mean = 6th grade). There were 11 boys and 16 girls. 5 children came from the upper castes while others belonged to the lower castes. The gender and caste breakdown seemed to mirror the demographics in the community. Every participant attended between 8 and 29 sessions (mean = 20) in the afterschool program, broken down according to the following three functions: • Cellphone training: 0 to 5 sessions (mean = 4) where we taught participants how to use the cellphones, perform alphanumeric input and play mobile games, • ESL learning: 4 to 17 sessions (mean = 10) in which participants played ESL learning games on the cellphones, and • Assessment: 4 to 7 sessions (mean = 6) for administrative tasks and data collection, e.g. demographics interviews and various tests. In India, traditionally, only the upper castes owned land. As such, the upper castes earn their livelihood on the land or run small businesses, while lower castes graze their goats, work as daily-wage laborers or perform menial jobs in the homes of the upper castes. Land-owning and non-land-owning families told us that they earned up to 100,000 (US$2,500) and 50,000 rupees (US$1,250) respectively per year. B. Hindi and English Baseline 26 of the 27 participants were enrolled in the same school, where Hindi is the medium of instruction. The last participant was a school dropout. Assuming regular school attendance, the typical participant would have taken classes on Hindi and English for 5½ and 3½ years respectively prior to the study. We devised a grading rubric to evaluate each participant on the Hindi literacy test and qualifying test. On the former test, participants scored 7.9 out of 18 on average (σ = 4.5, n = 19). 2 participants turned in blank answer sheets. We observed the following problems in the submissions:

142

• Wrong answers due to poor comprehension of the questions (5% of the test-takers) or passage (53%), or responses that simply repeated the questions (32%) • Spelling errors (16%) • Grammatical errors, i.e. using the incorrect form of the verb for the subject’s gender (21%), or the incorrect form of the noun for the subject’s singularity vs. plurality (5%) • Inability to phrase responses in complete sentences (5%) TABLE I BREAKDOWN OF PARTICIPANT PERFORMANCE ON QUALIFYING TEST Poor Fair Good 13% left blanks 32% filled in blanks 55% filled in blanks About empty or filled them with at least 1 with correct myself Alphabet

Word recognition Spelling Picture description

in Hindi

misspelling

5% filled in less than 7 blanks in alphabetic sequence correctly 14% matched up to 2 out of 6 words with correct pictures 41% spelt up to 2 out of 6 words correctly 18% made no attempt to answer this section

13% filled in ~11 blanks in alphabetic sequence correctly 14% matched 4 out of 6 words with correct pictures 18% spelt about 3 out of 6 words correctly 50% wrote answers as individual words, not full sentences

spellings for most questions 82% filled in at least 12 out of 13 blanks in alphabetic sequence correctly 72% matched at least 5 out of 6 words with correct pictures 41% spelt at least 4 out of 6 words correctly 32% wrote intelligible answers in sentences

On the qualifying test, on average, participants scored 44.0 out of 50 (σ = 5.5, n = 22). Our grading rubric indicated how participants should be classified as “poor,” “fair” and “good” on every section of the test. Table I gives the breakdown of how test-takers were distributed across categories for selected sections, and descriptions of the categories. In summary, the average participant had a good knowledge of the alphabet and a fair vocabulary of written words that she could read. On the other hand, she was weak in recalling and spelling everyday nouns, and even weaker in constructing complete sentences with these words. Despite the wide range in the ages of the participants, it appeared from their performance on the above tests that the variation in their English proficiency was much narrower. More specifically, we estimated that the average participant was comparable to an urban child in India who had taken between 1 and 2 years of English classes. Notably, only 10 children (45%) could spell their names correctly in English on the qualifying test papers. We had a chance later to interview the teacher who taught them English in their school. She revealed that her pedagogical approach revolved around having students copy sentences from English textbooks into their notebooks. She felt that it was not worth putting in more effort to teach English since she believed she was underpaid. C. Technology Baseline Among the 27 participants, 25 of them came from families who owned at least one cellphone; 5 participants belonged to families that owned 2 phones each while 2 participants came from families which owned 3 phones each. The cellphone was usually used by the eldest male member in every family, and in fact, 3 boys aged 13-14 possessed their own cellphones. 8

participants came from families that owned cellphones with a color screen, as opposed to monochrome display. Two of the above cellphones – both of which belonged to 2 of the above 3 boys – contained built-in cameras. Nonetheless, cellphone ownership and access were separate issues. Among the 25 children whose families owned at least one phone, 6 of them – 5 girls aged 7-11 and a boy aged 12 – were prohibited by their parents from using the phone, either entirely or most of the time. In general, in poorer families, it seemed that children were allowed to receive (free) incoming calls, but not play mobile games lest they drop the devices. On the other hand, in wealthier families, children were allowed to play on the phones. As such, although most participants were familiar with cellphones, it appeared that rural parents were more willing to entrust these relatively costly devices to their sons (vs. daughters). In total, 15 of the 27 participants (56%) reported that they had played cellphone games before.

VI. CURRICULUM AND GAME DESIGN One of the major challenges with carrying out a pilot study over a non-trivial timeframe was that we needed to develop sufficient digital content that could last throughout its entire duration. We ensured that our syllabus was aligned with local ESL learning needs in India by recruiting a local ESL teacher as our curriculum developer. She had a decade’s experience as an ESL teacher at a prestigious urban school, located in the same geographic region as the after-school program. A. Curriculum Design Given the above attendance rate, the ESL curriculum for the pilot was designed to be comparable to the amount of material that a qualified teacher could reasonably cover in 18 hours with rural children in a classroom. The syllabus was situated within the classroom theme, which participants could readily relate to. Concretely, the syllabus included: • Common nouns that are found in the typical classroom, e.g. chair, table, door. • Verbs that can be performed with the above nouns, e.g. sit, write, open, close. • Sentence structures for constructing sentences out of the above nouns and verbs, e.g. “This is a __.” • Sentence structures for phrasing question-and-answer sequences with the above nouns and verbs, e.g. “What is this?”, “Where is the __?” The curriculum design took participant performance on the qualifying test into account. The curriculum was also based on our attempts to converse informally with participants, during which we learned they did not comprehend simple questions about themselves, did not know the English words for objects around them (e.g. in the classroom), and made grammatical errors. The curriculum therefore targeted the above syllabus in terms of listening comprehension, word recognition (of the written word), sentence construction and spelling.

143

B. Game Design We designed a set of ESL learning games for the cellphone platform that targeted the above curriculum, and piloted them in the after-school program. Our designs drew on 3 resources, namely: 1) recurring patterns in state-of-the-art commercial software applications for language learning, which represented best practices that we reused to avoid reinventing the wheel, 2) traditional village games, which more closely matched the expectations and understandings that rural children have about games, in comparison to contemporary videogames that were largely Westernized, and 3) lessons from several previous rounds of field-testing and iterations with rural children elsewhere in India [12]. In this subsection, we walk the reader through a subset of the screen designs.

Fig. 1 and 2. Figure 1 introduces the English vocabulary for common nouns in the classroom. Word-picture association is a technique employed by many successful commercial language learning software. As the boy moves to each object, the software highlights the corresponding word in a different (green) color and says the word aloud. Figure 2 situates these objects in a classroom scene and builds on the earlier screen by demonstrating how to use the nouns in complete sentences. As the boy moves to each object, the software says the “This is a __” phase aloud for the corresponding object.

In earlier field studies, we observed that rural children did not readily associate a game with learning. It seemed that they viewed a game as an activity to be played purely for pleasure, and did not pay attention to the educational content embedded within game activities. On the other hand, when educational content such as English words and phrases were introduced in non-interactive screens separate from interactive game screens, the rural children appeared to grasp more intuitively that the software was trying to teach them those English words and phrases. Users subsequently paid more attention to the latter. Figures 1 to 4 show some screenshots in which we introduced words and phrases – both written and spoken – to the learners.

Fig. 3 and 4. Figures 3 and 4 introduce additional phrases that the nouns and verbs in the syllabus can be used in. Figure 3 teaches a phrase that associates the verb “sit” with the noun “chair.” Figure 4 shows how to ask questions using the “Where” keyword. Abstract phrases and function words such as “where” are difficult to convey graphically. Hence, when they are taught for the first time, the software explains their meanings orally in Hindi.

The games tested players on their comprehension and recall of the words and phrases. For example, the game shown in Figure 5 says the word aloud for one of the objects displayed on screen. The player needs to identify the correct object and push it onto the area that is blinking blue. At the same time, he needs to avoid the balls thrown by the computer-controlled opponent. This game was an adaptation of Giti Phod, which was one of the traditional games that children play in Indian villages. In Giti Phod, players in a team have to arrange some objects (e.g. rocks) into a given configuration (e.g. a heap), while avoiding being hit by a ball thrown by members in the opposing team. In our experience, we have observed that rural children found it more intuitive to understand videogame rules when the designs of these videogames drew on the rules found in the traditional village games that they play everyday.

Fig. 5. A word-picture matching game which is an adaptation of one of the traditional village games that children in rural Indian play everyday.

Given that television has become a pervasive media among all economic classes in India, it only made sense to draw on popular culture in India to make our designs more appealing to children there. One of these sources is Sesame Street, which is a successful television program for young children that has local co-productions around the world – in both industrialized and developing countries. Its producers in India have found some of its localized characters to be popular with children in India, and we incorporated those characters into our designs

144

for teaching (Figure 4), quizzing (Figure 6) and congratulating (Figure 7) the user.

Fig. 6 and 7. Localized characters from the Indian production of Sesame Street tested the player on his ability to engage in question-and-answer style dialogues, and performed a victory dance for the player upon successful completion of each level in the game.

The activity that targets spelling skills is shown in Figure 8. The player is given an image (e.g. blackboard) and is required to spell the word corresponding to it. Some of the letters in the word are displayed, while blanks are shown for the remaining letters. The player moves between blanks with the arrow keys. Once he has filled in all the blanks and submitted his answer, the correct and wrong letters are displayed in green and red respectively. This feedback constitutes the first level of hints that we have designed to help the learner arrive at the correct spelling. If the player spells the word correctly, he proceeds to the next game. Otherwise, all blanks are cleared after a short pause and he is required to spell the same word again.

Fig. 8. The first level of hints in the spelling activity. After the player has tried to spell the word by filling in the blanks with letters, the correct and incorrect letters are shown in green and red respectively.

If the player is unsuccessful in spelling a word correctly after two attempts, the second level of hints (Figure 9) appears to provide him with additional learning support. Based on the blank that the cursor is currently located at, the game displays a set of possible letters for him to narrow down the choice of candidate letters.

Fig. 9. The second level of hints in the spelling activity. For every blank, a set of possible letters are displayed to provide the player with some assistance, if he was unable to spell the word correctly after two attempts.

The curriculum is broken up into a total of 6 levels in the games. On every screen, the player can access a menu through a shortcut button. Among various options, this menu permits him to move to an earlier level in the curriculum to repeat the material, as well as to move to higher levels in the curriculum. The software was designed so as not to require airtime, which was expensive for most rural families. We implemented the games on Adobe’s Flash Lite and Qualcomm’s BREW (Binary Runtime Environment for Wireless) platforms. We piloted the games on Motorola’s Razr V3m cellphone model, which has a fairly large screen. VII. PILOT SESSIONS In those sessions where participants were taught how to use the cellphone, they were shown how to move their sprites with the arrow buttons. They were also taught how to perform alphanumeric text entry, since most of them did not know this. Sprite movement and text input were essential skills for the games we designed. Pilot staff were therefore asked to write some simple, short sentences on the blackboard, and ensure that each participant demonstrated his ability to enter those sentences via text input. Some other sessions focused on administrative tasks, such as the above tests and demographics interviews. We learned that a few participants had difficulty reading a small subset of the English alphabet despite having passed the qualifying test. We spent two sessions coaching them on those less-frequently encountered letters, so that they would be better prepared for the syllabus targeted in the pilot. Next, at least 8 children had seen the localized Sesame Street characters on television, but did not know their names. To help participants better relate to the characters, so that our games would appeal to them even more, we introduced the characters at the start of the semester. We also screened 3 localized episodes on separate occasions. These episodes were chosen such that they were educational but did not target English learning. Each episode lasted ½ hour, and we observed that participants enjoyed the humorous acts performed by the characters.

145

Fig. 10. In the after-school sessions, each participant was loaned a cellphone preloaded with English language learning games. Participants were taught how to start the games, and were asked to focus on learning English when playing the games on their own.

The remaining, and majority of, sessions focused on ESL learning. A 2-hour session was typically structured as follow: after an exchange of greetings, pilot staff took attendance and briefed participants on the learning objectives for that day. If new games were deployed that day, pilot staff explained and demonstrated how to play them to the participants in small groups. Each participant was then handed a cellphone to play the games on her own (Figure 10), and were told to focus on learning the English syllabus that the games covered. Children who were absent on previous sessions received help from pilot personnel in learning how to play those games that they were unfamiliar with. Pilot staff were limited to providing technical support; and were explicitly instructed not to teach English or communicate with participants in English. There was a short break of 10 to 15 minutes in the middle of each session. At the end of each session, pilot staff took back the phones so that they could charge their batteries overnight and download new games onto them. Each participants received a small packet of biscuits after every session. VIII. QUANTITATIVE RESULTS On the pre- and post-tests, test-takers were awarded 1 point for each common noun in the syllabus that was spelt correctly. A. Post-Test Gains The mean pre-test score was 5.2 out of 18 (σ = 3.3, n = 27) while the mean post-test score was 8.4 out of 18 (σ = 5.5, n = 24). Participants exhibited significant post-test gains on a onetailed t-test (p = 0.007). We present the frequency histograms for both scores in Figure 11. They illustrate that the score distribution had shifted toward the higher end of the spectrum after the deployment.

Fig. 11. Frequency histogram of participant scores on the pre- and post-tests.

The average post-test gains was 3.4 out of 18 (σ = 3.3, n = 24). The gains exhibited a fairly large variation, and ranged from -2 (two participants exhibited negative gains) to 9 out of 18. We present the frequency histogram for post-test gains in Figure 12.

Fig. 12. Frequency histogram of participant post-test gains.

B. High-Gains vs. Low-Gains Learners We sought to understand how participants’ post-test gains were correlated with their demographics and performance on other tests. We also categorized participants into two groups, namely, “high-gains learners” and “low-gains learners,” based on their post-test gains. A learner whose post-test gains exceeded the mean of 3.4 was categorized as a “high-gains learner,” else he was classified as a “low-gains learner.” In all, 9 participants were classified as high-gains learners while 15 participants were categorized as low-gains learners. 3 of the 27 participants could not be classified since they were absent on the day when the post-test was administered. On a normalized scale, when the 27 participants were taken as one group, the average pre-test score was 29% whereas the average post-test score was 47%. The latter score did not seem high in absolute terms, i.e. on average, a participant could not spell over half of the common nouns targeted in the syllabus by the end of the intervention. However, once the participants had been classified, on a normalized scale, high-gains learners scored 41% (80%) on the pre-test (post-test) whereas lowgains learners scored 19% (27%) on the pre-test (post-test), on average. In other words, high-gains learners not only showed larger post-test gains but also appeared to have a higher mean

146

pre-test score at the start of the intervention. More important, post-test gains for both high-gains (p < 0.001) and low-gains learners (p = 0.076) were significant, with effect sizes of 0.54 and 2.24 respectively. That is, both categories of participants exhibited learning gains. (But average post-test gains for lowgains learners were only marginally significant, because of the 2 participants who obtained lower scores on the post-test compared to their pre-test.) TABLE II HIGH-GAINS VS. LOW-GAINS LEARNERS IN TERMS OF DEMOGRAPHICS Age Grade Days Days (Years) Enrolled Spent Spent in School Learning Playing ESL ESL Games Games LowMean 10.7 5th 3.9 10.5 gains 1.8 1.9 1.4 4.2 σ learners Min. 7 2nd 0 5 (n=15) Max. 14 9th 5 16 Highgains learners (n=9)

Mean

σ

Min. Max. Is difference between means significant? Correlation with post-test gains (r)

12.8 1.2

8th 0.9

3.8 0.8

10.2 5.2

11 14 Yes

7th 9th Yes

2 5 No

4 17 No

(p = 0.002)

(p < 0.001)

(p = 0.4)

(p = 0.4)

0.45

0.61

0.11

0.10

In Tables II and III, we examined how high-gains learners may differ from low-gains learners in terms of demographics. On the whole, the high-gains learners did not appear to differ significantly from low-gains learners in terms of the number of days that they spent on learning how to play the cellphonebased games (p = 0.4) or actually playing the games to learn ESL (p = 0.4). Instead, high-gains learners belonged to higher ages (p = 0.002) and were enrolled in more advanced grades in school (p < 0.001). In fact, post-test gains exhibited high correlation with grade levels that participants were enrolled in school (r = 0.61) and medium correlation with age (r = 0.45).

TABLE III HIGH-GAINS VS. LOW-GAINS LEARNERS IN TERMS OF DEMOGRAPHICS Sex Caste Media Exposure Attitude* 67% 77% 73% have played 24%, 38% & 38% Low(33%) (23%) games on cellphones were described as gains were belonged prior to pilot; below average, learners females to lower 60% (40%) watched average and (n=15) (males) (upper) less (equal to or more) above average castes

Highgains learners (n=9)

44% (56%) were females (males)

88% (12%) belonged to lower (upper) castes

than 1 hour of TV per day 56% have played games on cellphones prior to pilot; 40% (60%) watched less (equal to or more) than 1 hour of TV per day

learners respectively 29% and 71% were described as below average and above average learners respectively

*The Attitude column is based on the observations that pilot personnel have on the seriousness and aptitude that participants exhibited as learners throughout the pilot. These qualitative comments were subsequently coded into the “below average”, “average” and “above average” learner categories.

The proportions in Table III were presented for the sake of completeness. We were unable to perform any statistical tests on these proportions due to the small sample size, which for example did not satisfy the standard binomial requirement. We thus caution the reader against drawing firm conclusions from these statistics. However, when examining individual learners to identify surprising cases, we took the demographic variables in Table III into consideration. The analysis is deferred to the following section. TABLE IV HIGH-GAINS VS. LOW-GAINS LEARNERS IN TERMS OF TEST SCORES QualifyQualifyHindi PrePosting Test ing Test, Literacy Test Test (out of Spelling Test (out (out 50) Section (out of of 18) of 18) (out of 6 18) words)* LowMean 42.9 1.2 6.3 3.5 4.8 gains 2.9 1.1 4.2 2.3 2.4 σ learners Min. 37 0 0 0 2 (n=15) Max. 46.5 4 14 10 12 Highgains learners (n=9)

Mean

σ

Min. Max. Is difference significant?

47.1 1.8

3.4 1.6

12.0 1.7

7.4 3.1

14.4 3.6

43.5 49 Yes

2 6 Yes

10.5 14 Yes

2 13 Yes

6 18 Yes

(p < 0.001)

(p = 0.001)

(p < 0.001)

(p = 0.003)

(p < 0.001)

Correlation with 0.57 0.70 0.45 0.46 0.86 post-test gains (r) *In this column, we present the number of words that participants spelt correctly on the spelling section of the qualifying test, out of a total of 6 words.

Table IV compares the high-gains and low-gains learners in terms of test scores. The former outperformed the latter on the Hindi test (p < 0.001). Next, we analyzed the qualifying test results at two levels, namely, the score for the entire test as well as the score on the spelling section. We found that highgains learners outperformed low-gains learners on the entire test (p < 0.001) as well as on the spelling section (p = 0.001). High-gains learners also obtained higher scores on the pre-test (p = 0.003) and post-test (p < 0.001), vis-à-vis low-gains learners. In fact, participants’ post-test gains exhibited a high degree of correlation with their qualifying test scores, for both the entire test (r = 0.57) and spelling section (r = 0.70). On the other hand, post-test gains had a lower correlation with Hindi literacy levels (r = 0.45) and pre-test scores (r = 0.46). IX. QUALITATIVE RESULTS The above quantitative results suggested that current levels of spelling proficiency and grades enrolled in school were the strongest predictors of success in learning how to spell new words through the cellphone-based games which we designed. Higher levels of Hindi literacy and academic preparation were also associated with higher post-test gains. On the other hand, the number of sessions that participants had with the cellphone games – both for learning how to play the ESL learning games and learning ESL through the games

147

– were not associated with post-test achievements. Among the 24 children whom we have post-test gains data on, 5 of them were classified as high-gains learners despite having played the games on only 4 to 7 days (mean = 6 days). Conversely, 7 participants were classified as low-gains learners in spite of having played the e-learning games for 13 to 16 days (mean = 14.7 days). More important, pilot personnel described 6 of these 7 low-gains learners as “hardworking” or “serious” about learning ESL. Similarly, we were curious about how the two students who exhibited negative post-test gains, as well as the school dropout, had interacted with the cellphone-based games. The above quantitative trends raise the following questions: How did some of the high-gains learners played the games such that they benefited despite lower attendance? In the case of some low-gains learners, why did they improve little on the post-test despite spending numerous days with the games and being perceived as diligent? In order to address such questions at the interaction design level, we turned to our video records and daily reports. Our hope was to recommend improvements to the technology designs and/or after-school setting. A. Interaction Patterns with the Technology At first glance, it seemed that participants needed to attend the after-school program for more days. Our video recordings showed that only 3 of the participants reached the last level in the curriculum by the last session in the program. This was a surprise. Given that the curriculum was designed for 18 hours of instruction, we expected an average attendance rate of 10 gameplay sessions to constitute enough time with the games. On examining the video recordings more closely, we saw that at least 8 participants were using the game menu to skip ahead to other levels whenever they were unable to spell the words in the current level correctly after a few attempts. (We note that the menu was not necessarily a negative feature. Among those 8 participants, at least 2 of them used the menu to skip those words that they already knew how to spell.) We needed to understand why learners gave up on retrying the spelling activity for difficult words despite the hints in the spelling activity. On the whole, we observed 4 different levels of behavior associated with the spelling activity in the videos: 1) When students encountered a word that they could spell, they pressed the keypad buttons quickly and with ease to fill in the blanks for the missing letters. 2) When students saw a word that they did not know how to spell, some of them learned to spell it correctly with the help of the first level of hints. 3) Some of those students who failed to learn how to spell a word with the first level of hint eventually learned how to spell it correctly with the help of the second level of hints. 4) Other students never succeeded in learning how to spell certain words despite both levels of hints. In general, we observed that high-gains learners succeeded in learning how to spell words after having seen their written forms displayed on earlier screens (i.e. such learners were able to spell those words correctly – without requiring any hints –

on their first attempt in the spelling activity), or with only the first level of hints. It seemed that they did not require much scaffolding support from the software. In fact, from the video recordings of 9 high-gains learners, we saw that 5 (56%) and 1 (11%) of them depended on the first and second levels of hints respectively. In contrast, 12 (80%) and 8 (53%) out of the 15 low-gains learners who were videotaped relied on the first and second levels of hints respectively. It seemed that the low-gains learners, as compared to the high-gains learners, were less able to rectify their errors in filling in the blanks for the missing letters through the first level of hints, and required the second level of hints to attain the correct spellings. Worse, the inability on the part of the low-gains learners to spell correctly with help from only the first level of hints made some of them visibly unhappy or bored when the second level of hints appeared. The reason for this distress was unclear. The learner could be frustrated that he was spending too much time to learn how to spell the word. Alternatively, on seeing the second level of hints show up, he could be demoralized that he had just been relegated to the ranks of the most inferior learners and needed the second level as a “crutch” in order to succeed. Furthermore, some learners struggled despite both levels of hints. In the videos, two of them turned to their neighbors and asked for the correct letters, and/or to chat. In some cases, participants were embarrassed to ask their neighbors for help again after so soon, and hence used the menu to skip to other levels in the games. More important, we observed that participants – especially among the low-gains learners – may be able to spell the words in the spelling activity, but were not able to spell the same words on the post-test. We offer two plausible explanations. Firstly, some children may have learned to spell the words by their last session in the program, but had forgotten their spellings between the last session and the post-test. Secondly, some participants never learned to spell the words in their entirety, since the spelling activity only involved filling a few blanks and did not require the learner to spell the entire word. Nowhere in the video recordings did we observe any child struggling with usability problems. X. CONCLUSION Our reactions to the results of the learning assessment were mixed. In an underdeveloped region where rural children did not have access to quality English instruction in their regular school or elsewhere, we were excited to see the participants -both high-gains and low-gains learners – in the after-school program exhibit statistically significant post-test gains that could be reasonably attributed to our cellphone-based English learning games. On the other hand, the learning benefits were uneven among participants. This could be a cause for concern. To begin with, high-gains learners outperformed low-gains participants on the pre-test, qualifying test and Hindi literacy test. In fact, participants’ post-test gains appeared to be highly correlated with their existing levels of spelling proficiency (as

148

measured by their performance on the spelling section of the qualifying test) and the grades in school that they are currently enrolled in. This observation suggests that those rural children with a stronger academic foundation are the same children who are most well positioned to take advantage of the benefits that cellphone-based learning confers. Our results are consistent with the outcomes of a study with rural and urban low-income children in India described in He et al. [16]. This study showed that weaker students benefited more from a teacher-directed pedagogical intervention, while stronger students benefited more from a self-paced, machinebased approach to English learning. These results should not, however, be interpreted to mean that we rule out technologyaugmented learning completely in the context of low-income children. Horowitz et al [15] reported a study on videos for learning the English alphabet streamed over cellphones. In this study, a greater proportion of lower-income parents, vis-àvis their higher-income counterparts, perceived the videos to have improved their children’s knowledge of the alphabet. In the face of the above overwhelming odds, what can we do to promote more equitable educational opportunities in the developing world? One possible – and perhaps cautiously optimistic – interpretation of the above results is that future research needs to be directed at understanding how e-learning software can provide more scaffolding support for those rural children who have less academic preparation. As an example, the spelling activity needs to be redesigned such that the learner is guided to spell the entire word eventually. With this redesign, however, gameplay becomes prolonged and can potentially increase player frustration, as we have witnessed above. One remedy is to have e-learning games track learner performance, so that the software can be adaptive in skipping stages that are similar to those that the player has previously performed well in. Another implication for instructional design, which calls for additional investigation, is scaffolds such as hints that are less conspicuous, so that their appearing on screen does not diminish the learner’s sense of self-esteem or achievement. Unfortunately, adaptive educational applications require the application state to be stored and retrieved on the same mobile device. From the logistics standpoint, this requirement is more difficult to implement in developing regions since it is harder to ensure that the same learner uses the same phone – which stores his performance from prior session(s) – across sessions. For instance, in an after-school program where attendance fluctuates from session to session, it would be prudent to keep a shared pool of cellphones, such that children who show up for the day’s session can draw from. In these circumstances in which it is not possible to reserve a cellphone for each child, a wireless networked mechanism for synchronizing application state across all cellphones may be necessary. This, and other issues that we have raised above, require further investigation for cellphone-based literacy learning to be more effective in targeting less academically advanced rural children. As we think more widely beyond the cellphone to consider it as a component in the broader learning environment, since

the after-school program is a model that is readily replicable, we encourage the reader to adopt and experiment with the lessons from this paper. Our results suggest that the cellphone – which remains a relatively scarce resource in the developing world – is most effectively utilized in an after-school program that targets more advanced children. This restriction may be a necessarily evil until we gain a deeper understanding of how to design instructional scaffolds for less well-prepared rural children. Next, children’s tendency to seek help from their neighbors can be channeled productively if the latter are taught to offer help appropriately (e.g. instead of only telling their neighbors the correct spelling, help them to associate and remember the correct spelling). Such peer coaching strategies are especially crucial since cooperative group learning is unfamiliar to many rural children, whose schools (if they attend one) are more likely to implement rote learning. Alternatively, such an afterschool program can hire facilitators to provide academically less prepared learners with similar coaching. ACKNOWLEDGMENT We thank Suraksha (Urvashi Sahni and Shalini Mathur) and Sesame Workshop India for collaborating with us on this pilot study. Lauren Bailey and Anuj Tewari provided much-needed administrative assistance, while Aman Anand and Siddhartha Lal provided technical assistance. We are especially indebted to Mehnaaz Abidi, Jatin Chaudhary, Neelima Purwar, Gautam Singh and Kavish Sinha for providing ground support. Lastly, we thank the parents and their children for their participation. REFERENCES [1]

J. Pal, M.Lakshmanan, and K. Toyama, “’My child will be respected:’ Parental perspectives on computers in rural India,” in Proc. of 2nd IEEE/ACM International Conference on Information and Communication Technologies and Development, Bangalore, India, Dec. 2007.

[2]

S. Shukla, “From pre-colonial to post-colonial: Educational transitions in southern Asia.” in Economic and Political Weekly, 31(22), 1996, pp. 1344-49.

[3]

D. Faust, and R. Nagar, “Politics of development in postcolonial India: English-medium education and social fracturing,” in Economic and Political Weekly, India, July 28, 2001.

[4]

M. P. Kishwar, “Deprivations’s real language,” in The Indian Express, September 15, 2005. Available: http://www.indianexpress.com/printerFriendly/12662.html.

[5]

Azim Premji Foundation, “The social context of elementary education in rural India,” 2004. Available: http://www.azimpremjifoundation.org/ downloads/TheSocialContextofElementaryEductaioninRuralIndia.pdf

[6]

J. P. Gee, What Video Games Have to Teach Us About Learning and Literacy, Palgrave Macmillan, 2004.

[7]

A. Banerjee, S. Cole, E. Duflo, and L. Lindon, “Remedying education: Evidence from two randomized experiments in India,” NBER Working Paper No. 11904, December 2005.

[8]

S. Mitra, J. Tooley, P. Inamdar, and P. Dixon, “Improving English pronunciation: an automated instructional approach,” in Journal of Information Technologies and International Development, Vol. 1, No. 1, Fall 2003, pp. 75-84.

[9]

M. B. Dias, G. A. Mills-Tettey, and J. Mertz, “The TechBridgeWorld initiative: Broadening perspectives in computing technology education

149 and research,” in Proc. of the international symposium on Women and ICT: Creating Global Transformation, ACM Press, 2005. [10] B. Kothari, A. Pandey, and A. R. Chudgar, “Reading out of the ‘idiot box’: Same-language subtitling on television in India,” in Journal of Information Technologies and International Development, Vol. 2, No. 1, Fall 2004, pp. 23-44. [11] T. H. Brown, “The role of m-learning in the future of e-learning in Africa?,” in Proc. of 21st ICDE world conference, Jun. 2003. [12] M. Kam, A. Agarwal, A. Kumar, S. Lal, A. Mathur, A. Tewari, and J. Canny, “Designing e-learning games for rural children in India: A format for balancing learning with fun,” in Proc. of ACM international conference on Designing Interactive Systems, Cape Town, South Africa, Feb. 2008. [13] Y. M. Islam, Z. Rahman, S. S.Razzaq, M. A. Sayed, and S. Zaman, “Effect of feedback during lecture style delivery both in a face-to-face classroom and during a distance education television session in a developing country like Bangladesh without the use of Internet,” in Proc. of 6th IEEE international conference on Advanced Learning Technologies, pp. 469-471, 2006. [14] F. Librero, A. J. Ramos, A. I. Ranga, J. Triñona, and D. Lambert, “Uses of the cell phone for education in the Philippines and Mongolia,” in Distance Education, vol. 28, no. 2, pp. 231-244, Aug. 2007. [15] J. E. Horowitz, L. D. Sosenko, J. L. S. Hoffman, J. Ziobrowski, A. Tafoya, A. Haagenson, and S. Hahn, “Evaluation of the PBS Ready to Learn cell phone study: Learning letters with Elmo,” reported prepared by WestEd, Sep. 2006. [16] F. He, L. L. Linden, and M. MacLeod, “How to teach English in India: Testing the relative productivity of instruction methods within the Pratham English language education program,” working paper, Jul. 1, 2008. [17] U.S. Pawar, J. Pal, R. Gupta, and K. Toyama, “Multiple mice for retention tasks in disadvantaged schools,” in Proc. of ACM conference on Human Factors in Computing Systems, San Jose, CA, Apr. 2007. [18] N. Moraveji, T. Kim, J. Ge, U. S. Pawar, K. Inkpen, and K. Mulcahy, “Mischief: Supporting remote teaching in developing regions,” Systems, pp. 753-762, Florence, Italy, Apr. 2008.

150

Kelsa+: Digital Literacy for Low-Income Office Workers Aishwarya Lakshmi Ratan1, Sambit Satpathy2, Lilian Zia3, Kentaro Toyama1, Sean Blagsvedt4, Udai Singh Pawar5, Thanuja Subramaniam6

Abstract—Almost all formal organizations employ service staff for tasks such as housekeeping, security, maintenance, and transport at their office facility. Many of these workers earn wages in line with menial-labor salaries in their respective countries. They have few onthe-job opportunities to upgrade their skills or learn new ones. Kelsa+ is an initiative through which organizations in developing countries can increase digital literacy and skill development among such low-income workers, through the provision of an Internetconnected PC for the service staff’s free, unrestricted use when off duty. We study a Kelsa+ pilot implementation in Bangalore, India, involving an office facility with 35 service staff. In a preliminary exploration over 18 months, we find that at a cost that is negligible for the organization, workers’ use of the Kelsa+ PC is high and can deliver benefits both to themselves and to the office. For workers, broad gains were seen in confidence, self-esteem, and basic digital literacy, while a few individuals experienced improvements in second-language (English) proficiency and career opportunities. These early results point in the direction of a cost-effective ICT4D initiative that could be run in the developing-country offices of the very organizations promoting development off-site. Index Terms—ICTD, digital literacy, service staff, low-income workers, urban poor

I. INTRODUCTION Those of us working in “information and communication technologies for development” (ICTD) often run projects in remote rural areas or urban slums to work with low-income communities. Meanwhile, we often neglect a low-income group right under our very noses: workers who clean our offices, provide security, maintain facilities, etc. Most offices involve a sizeable group of service staff, who take care of the housekeeping, security, transport, maintenance and so forth of the facility. In developing countries, service staff tends to be employed in large numbers, with many of them earning wages equivalent to those for Manuscript received September 22, 2008. Revised February 20, 2009. 1 Aishwarya Lakshmi Ratan and Kentaro Toyama are with Microsoft Research India, “Scientia”, 196/36, 2nd main road, Sadashivnagar, Bangalore – 560080, INDIA. (Corresponding author: Aishwarya Lakshmi Ratan - phone: +91-80-66586000; fax: +91-80- 23614657; email: [email protected]) 2 Sambit Satpathy is with CISCO 3 Lilian Zia is with INTEL 4 Sean Blagsvedt is with Babajob.com 5 Udai Singh Pawar was with Microsoft Research India 6 Thanuja Subramaniam is with Microsoft, Sri Lanka

menial labor in the same geography. In India, for example, the ratio of support staff to direct employees can range from 10% to as high as 60% – 30 support staff for an employee base of 100 is not unusual. Given that the corporate IT and ITES sector alone employs more than 1.6 million professionals in urban India [1], the support staff at these facilities would involve a population of at least 160,000 and more likely 320,000 urban low-income workers. Such workers typically have limited or low-quality education and earn between $50 and $200 per month. They have few on-the-job opportunities to upgrade their skills or learn new ones, and therefore remain caught in a vicious cycle of low-income work. We propose Kelsa+, a program that offers Internetconnected PCs for free, unrestricted use to the low-income workers in modern offices in developing countries. In our 18month pilot, we tried several things to encourage usage and improve development impact, and recorded a variety of positive, if limited, results. Kelsa+ could thus be a worthwhile project to spread to other offices in developing countries, if these benefits could be delivered more systematically. In this paper, we report on our pilot experiment with Kelsa+ in Bangalore, India. We discuss our design decisions, describe usage patterns, and identify explicit development-focused opportunities and outcomes. We also discuss some of the potential challenges to adoption by other organizations. Although PCs are likely to have been made available to service staff elsewhere, perhaps on an informal basis, to our knowledge, a deliberate exploration of the design and value of such a project has never been conducted before. II. RELATED WORK There are three threads of research that are relevant to the Kelsa+ study: the Hole-in-the-Wall experiments for children, computer-aided learning for adults, and computer kiosk or telecentre initiatives aimed at promoting development in poor communities. The ‘Hole in the Wall’ (HitW) experiments The National Institute for Information Technology (NIIT) [2] ran a series of computer-based education experiments with children from disadvantaged communities in New Delhi, India, in the late 90s. Effectively, they bore a hole in the wall that separated NIIT from the neighboring slum settlement and had an Internet-connected PC set up facing the settlement, with a touchpad built into the wall for navigation. They found that children from the slum communities in the vicinity

151

(mostly 6 to 14 years old) explored PC usage on their own using the device, resulting in significant learning benefits without any formal instruction. HitW interventions for children have been conducted at various rural and urban sites, and the pattern of improved digital literacy from this Minimally-Invasive Education (MIE) intervention has been consistently recorded [3]. Critiques of the project have come from those who see the “digital divide” not as “gaps to be overcome by providing equipment,” but as “social-development challenges to be addressed through the effective integration of technology into communities, institutions, and societies” [4]. Such critics find the HitW intervention to be “technologically deterministic” and lacking in community and parental involvement. Further, they question whether simply having physical access to an IT device and learning a set of basic IT skills would translate into any systematic improvement in the lives of these children. Kelsa+ was initially inspired by the ‘Hole in the Wall’ project. It examines unrestricted PC access not for children, but for adults, specifically low-income urban workers, though within an institutional and social context. Computer-Aided Learning for Adults Computer-Aided Learning (CAL) has existed in various forms for a number of decades as an educational tool in schools [5], universities [6], healthcare institutions, and is becoming an important area for ICTD investigation in developing country contexts [7]. Early experiments in basic computer-assisted instruction, comprehensive historical overviews of which have been done [8], has led to more sophisticated recent work around using robotics for science education [9], facilitated distance learning [10], and so on. However, there have been relatively few investigations around uptake and impact in the context of CAL for adult learning. One study identifies the importance of CAL in maintaining the continuity of learning history of adults, arguing that such use of computers in learning will be able to shift the focus of cognitive energy to cognitive creativity, allowing adults more space for absorption of knowledge and creativity [11]. Kelsa+ examines a subset of questions in the domain of CAL for low-income urban workers in developing countries. Computer kiosk and telecentre ICTD initiatives Most of the adult digital inclusion projects in the developing world have been undertaken as computer telecentre initiatives [12]. These typically, though not always, involve a small number of PCs set up in a rural area, offering a variety of PC-related services and run by an entrepreneur, salaried operator, or a community-based organization. The expectation in these projects is often that once ICTs such as the PC are made physically available in communities where they did not exist before, there will be usage by members of the community and socio-economic development will follow. This approach is evident in the project concept notes and promotional material of several telecentre initiatives: “by providing information about employment, better farming techniques and health we hoped for new sustainable job

opportunities, improved farming knowledge and healthier life.” [13] However, many of these projects have found it difficult to sustain the technology and to establish clear links between ICT access and development impact. In many instances, usage of the facility (and therefore revenues in for-profit kiosks) falls over time, irregular connectivity in remote areas causes periodic interruptions in service provision (and further decline in usage), device maintenance accrues unanticipated costs and where high usage persists, it is often for the usage of applications with limited welfare impact and by community members who are not particularly disadvantaged (making it hard to justify public expenditure) [14],[15]. Many programs face a tension between pursuing financial sustainability and maximizing social outreach [16]. Those kiosks that are able to sustain usage often involve an exceptional kiosk operator or a committed organization that is able to effectively coordinate between the end-users and the desired application [17]. In contrast, Kelsa+ operates in a modern office environment, where exactly the elements for sustainability are already present – IT maintenance and support, good physical, electrical, and connectivity infrastructure, caring staff, etc. A few kiosk projects, such as the Akshaya project in Kerala, have explicitly pursued adult digital literacy as an objective [16], [18]. Despite the project’s impressive scale, however, field studies suggest that these programs achieved neither the reach nor the depth of digital literacy that was sought [28]. We hypothesize that one reason for this was that digital literacy courses were conducted over very short time periods (10 sessions of 90 minutes each), which did not permit learners enough time to familiarize themselves with the technology. In Kelsa+, interaction with the PC is voluntary and continuous in a process we call “digital habituation” [19]. This allows the incremental build-up of digital literacy skills at a user-determined pace, and through user-determined content and applications. III. BASELINE INVESTIGATION Low-income urban workers largely live in the city’s less developed residential settlements with many working as part of the informal economy. A large share of these workers are young and recent urban migrants. Given low levels of education or low-quality education, they are mostly employed in low-paying service sector jobs or work as small entrepreneurs. Among this group, some workers find jobs at formal office facilities, often associated with slightly higher pay, additional perks, and more prestige. In Indian cities, these opportunities have mushroomed since the early 1990s, as both domestic and multinational corporations expanded operations and facilities. Unfortunately, these workers have limited opportunities to upgrade skills or learn new ones, especially within their workplaces. The training institutes that exist, for spoken English, typing, or IT skills, place heavy demands on workers’ time and finances, as they require attendance at external training centres. As a result, even after decades of labor,

152

workers earn only marginally better incomes than when they started. Our pilot focused on the service staff at a single urban corporate facility in India’s IT capital, Bangalore. The facility employs around 55 full-time employees and 35 support staff at any given point in time. The latter group consists of housekeeping or cleaning staff, drivers for the office cabs, security guards stationed at the facility’s entry and exit points, and building maintenance staff to monitor the facility’s electricity, connectivity and other infrastructure. Prior to the pilot, we conducted detailed structured interviews with nearly all of the facility’s support staff (a sample of 30 respondents out of a total group of 35 workers), to understand their baseline socio-economic characteristics, occupational history, current job requirements, use of technology and aspirations. Each interview lasted for 60-90 minutes. This was complemented with participatory observation of work routines. Of the many insights gained from this qualitative investigation, a few are described below. The average age of the support staff is 26 years, and the average worker had ~12 years of formal schooling (class 10). All except two are men: many have migrated to the city alone in search of better work opportunities, even as their families continue to live in the village. The average worker earns ~$100 in income per month [20]. Depending on their job and the week in question, workers’ shifts varied: security guards and maintenance staff rotated for a week each between a 9 pm – 7 am shift, 7 am – 2 pm shift, and a 2 pm – 9 pm shift; housekeeping staff rotated between 9 am – 4 pm and 12 pm – 9 pm shifts; and drivers worked 12 hour shifts from 10 am 10 pm and 10 pm - 10 am. Workers had changed numerous jobs, despite having entered the labor force recently. One worker for instance had started out in the village doing casual wage-labor work, then worked as a private tailor, moved on to work at an export garments factory, then changed jobs to work at a finance company, after which he moved to corporate housekeeping. In the course of their daily work at the office, workers regularly saw PCs but did not touch or interact with them, except to clean them. Their interactions were exclusively with the specific ‘tools of their trade’, involving coffee machines, vacuum cleaners, fax and photocopying machines, radios (for the drivers), and phones. A few workers had occasional interactions with a PC at a cyber café, or at the office for a specific application, e.g., managing the Building Management System or the Security Camera Tracking application. Such workers’ general digital literacy skills were very low though – a result of their repeated restricted interaction with a single niche application. Likely because they worked in an office full of PCs (a software development and research centre), PCs were dominant in the discourse of the workers’ aspirations for themselves and their children. On a four-point Likert scale (‘Not At All Important’ to ‘Very Important’), all except two respondents rated the computer as being either Important or Very Important for their own upward mobility, and all of them felt this way about the importance of computers for their children to get ahead in life. As one respondent claimed,

“Even if you are poor, if you learn computers and try and get used to it - you can improve” [21]. Yet, though considered critical at a conceptual level, functional understanding of the PC was limited and based on workers’ observations from mass media and their environment. One worker’s comments on the PC as a learning device aptly captures this lack of a clear functional understanding, “Students now learn Windows - how to open it and use it, what’s inside it, games, etc. - and I can’t say why exactly that is useful, but it is. I know there is something in the computer that is important for students” [21]. A few workers who had learned to use particular applications meaningfully on a PC had done so through instruction from their peers or seniors at a past workplace. They described having picked up these skills through a combination of observation and ‘learning by doing’. As one respondent described, “I learned to use Outlook from my boss here. I had learned to use Excel at my previous job at a travel agency – a lady colleague taught me there. I learned to browse the internet after observing how my friends did it. I learned to do personal email on my own.” IV. SOLUTION: KELSA+ In response to the expressed desires in our baseline qualitative study, we introduced an Internet-connected PC at the workplace for the exclusive and free use of the facility’s support staff, to be used during workers’ off-duty hours. The project was named Kelsa+, with ‘kelsa’ being the local language (Kannada) word for ‘work’. Kelsa+ was, therefore, meant to signify ‘after-work’, ‘beyond-work’ and ‘improvingwork’. The intervention involved three overlapping phases of activity: (1) a ‘Hole-in-the-Wall’ phase inspired by the NIIT studies, when we were just observing what workers did on the PC when left on their own, (2) a Learning Modules phase, which involved trying different things to improve productive value of the PC for the workers, and (3) a Pre-Expansion phase, when we conducted interviews with management at other firms to understand what issues we’d need to address to make Kelsa+ work in different office locations. We describe each of these below. A) Phase One: An Office ‘Hole in the Wall’ The first phase involved establishing the four basic components of the project. First, a worker-dedicated PC was set up and integrated with existing company structures to have full infrastructural and institutional support. The PC had a basic Windows XP Operating System and the Office 2007 suite of applications installed. It had a dedicated broadband internet connection, as well as peripherals such as speakers, headphones, a printer and a webcam. User log-in on this PC was disabled to minimize barriers to entry, so that a worker could begin interacting with the PC as soon as s/he sat in front of it. That the maintenance of this PC would be handled by the company’s regular IT staff was also established. Second, the Kelsa+ PC was set up in the office’s basement,–a space that housed the maintenance office, the

153

workers’ changing rooms, the office cars, etc, and so was shared by all the service staff. The fact that it was placed in a location that was primarily “owned” by the workers and in which they felt comfortable was deliberate. Had the PC been placed in the cafeteria or in the lobby of the building, workers might have used the facility less, in deference to the office’s full-time staff. In some ways, this placement choice of using a space that the target group felt comfortable accessing was probably just as important as the ‘Hole in the Wall’ project's decision to place their kiosks near schools and outside of buildings. As a worker later affirmed, “When the computers are inside, it is difficult for us to feel comfortable touching them. Now, this is like a computer for us, outside [in the basement]” [20]. Third, in a set of initiation activities, the PC was introduced to the workers as a facility that they could use for any purpose whenever they were off duty. Workers were informed that all activities on and around the PC would be recorded, both for research and security purposes, specifying that none of this material would ever be used to restrict usage. Signage in the local language and English indicating this logging was also placed at the Kelsa+ PC location. For any questions or concerns, two of the authors were pointed out to the workers as the people to contact. Finally, the determination of who would use the PC when and how was left entirely to informal processes of negotiation between workers. No schedules were drawn up. No instructions were given. That usage and learning was entirely voluntary and self-paced was established. No restrictions were laid on how many users could simultaneously work on the PC. The intent was to provide a wide open area, conducive to groups of peers interacting with the PC together. B) Phase Two: Learning Modules After a year of allowing unrestricted usage of the PC, we collected feedback from the workers in a series of focus group discussions. Strong interest was displayed in achieving particular kinds of learning using the PC, including learning English, office productivity software, and accounting software (Tally). To further this goal, we started by introducing English as a Second Language (ESL) content, both as CDs and compiled weblinks, to the workers. No formal instruction was given or classes were taken. The voluntary and self-paced learning nature of phase one continued. We only acted as initiators of introducing new content at the Kelsa+ PC towards a specific productive goal, and not as evaluators or routine instructors. C) Phase Three: Management Discussions Kelsa+ is housed within an institutional context, and we made an effort to understand the effect that the intervention was having on its social and institutional environment, to avoid some of the limitations that the Hole-in-the-Wall study had faced. As we continued with other explorations on Kelsa+, we collected feedback from management to understand their reactions to the Kelsa+ facility for workers. We also began a set of investigations to understand what it would take to expand the project from a small-scale pilot at

one office location, to multiple sustained deployments across various firms. For this, the project was consolidated as a ‘Corporate Social Responsibility’ initiative, and proposed to a number of major corporations in Bangalore. Senior management at these firms were interviewed in relation to their own service staff populations. This data was used to compile a variety of solutions to the expressed concerns, so that the project could possibly fit into other institutional contexts. V. STUDY METHODOLOGY We employed a mixture of research methods in studying Kelsa+ through its three phases of activity, which we describe below. Collective usage: quantitative metrics The Kelsa+ PC had a logging application installed, which tracked all events initiated on the PC [22]. This included launches of all applications, as well as URLs visited. The logging tool also allowed an examination of how much time users had spent on the Kelsa+ PC on any given day. These three metrics (collective time spent using the machine, applications used, websites visited) were recorded over the 18-month period (Jan 11, 2007 – June 2 2008), with the analysis conducted on consecutive fortnights of activity. Application and internet usage was measured using number of launches as well as active time spent on each application/website. There were two breaks in logging, one when the PC’s connectivity was interrupted in mid-November 2007 and the other in late-March 2008 when the PC’s OS was re-installed, both in response to virus attacks. The categorization of the log data into meaningful groups, including application categories and website categories, was performed based on the researchers’ examination of keywords that tagged the data optimally despite the heterogeneity of usage. These are available on request from the authors. Collective usage: qualitative measures The activities on the Kelsa+ PC were also recorded using a screenshot logger, with a screenshot of the PC taken every minute. In addition, the activities around the PC were recorded using a motion-detecting web camera. These screenshot and video logs were processed manually, and hence, selectively, to answer specific questions around usability, group usage/sharing dynamics and number of distinct users. For the latter, 14 days of video logs were randomly picked from the 18-month study period to estimate the average number of distinct users per day. Individual impact: quantitative metrics of change Given the high job turnover rate among workers in this segment (described in the baseline study’s results), longitudinal tracking, i.e. measurements for the same individual over time, was very challenging. However, for a limited subset of workers, we obtained before and after results for particular behavioral or proficiency tests and these cases

154

were used to understand possible patterns of change. These included the following tests: • A Self-Esteem questionnaire: A local languagetranslation of the Rosenberg Self-Esteem questionnaire [23] was administered to a sub-set of workers prior to the Kelsa+ intervention (17 workers), and then five weeks after it was introduced (27 workers). Eleven respondents overlapped between the two administrations (0.65 of pre-test group). The differential in scores obtained for these workers was calculated. • A Digital Literacy test: A brief digital literacy test covering usage of basic applications on the PC was administered prior to the introduction of the Kelsa+ PC (30 workers). The time taken to complete a task, as well as the number of prompts needed to complete the task, were recorded as metrics of proficiency. The same test was administered to three workers, 18 months after the intervention, of whom two had taken the pre-test as well. The differential in scores obtained for these workers was calculated. Normal staff turnover at the rate of ~2 workers per month had disallowed a pre- and post-test comparison for more workers over this extended time period. • An English proficiency test: An adapted version of Cambridge University’s ‘Key English Test’ for Beginners [24] was administered to a sub-set of workers first in March 2008 (20 workers), prior to the introduction of the ESL digital content. A second test in the same format was administered in August 2008 (17 workers), after over three months of the ESL material having been available for workers’ usage. The questions focused on testing basic English vocabulary, grammar, and reading comprehension. Seven respondents overlapped between the two administrations (0.35 share of pre-test group). The differential in scores obtained for these workers was calculated. The post-test also asked a few questions on workers’ Kelsa+ usage or non-usage over the week preceding the test.

To assess the relevance of Kelsa+ in other institutional contexts, over 15 corporations were contacted from a range of sectors (biotechnology, energy, IT) with the majority being IT firms. They were first briefed about the Kelsa+ concept. Structured interviews were then conducted with them in which they shared details on the size and composition of service staff at their facilities, their evaluation of the possible benefits from, as well as concerns around, implementing Kelsa+ as a CSR initiative for the service staff at their facility. VI. STUDY RESULTS Given the multiple research methods that were used, in this section we present our results in correspondence to thehighlevel categories described in the methods section: collective usage,individual impact and management feedback. VI.I Collective Usage Time used After the initial fortnights, the Kelsa+ PC saw high and sustained adoption by the support staff (Figure 1), recording average collective usage of 10.13 hours per day. Usage peaked at 17.83 hours per day in fortnight 29. Further, usage was not restricted to a small minority of persistent users, but was spread across a broad base of workers. From the video log analysis, we found an average of 13 workers to be primary users of the PC each day, with all staff workgroups represented in the user base. 40 unique workers were observed using the Kelsa+ PC at least once (despite a staff size of only 35 at any given point in time – this occurred due to staff turnover) during the sampled period. A quarter of the workers were identified as high-frequency users, seen using the PC on half or more of the observed days [19].

Individual impact: qualitative narratives of change A subset of the staff from across workgroups (6 workers, ~0.17 share) was interviewed in detail 18 months after the intervention was launched. These workers discussed their background, prior exposure to a PC or not, their usage of the Kelsa+ facility, and their present job, following a structured interview protocol. Their perceptions of change since the introduction of the Kelsa+ PC were also recorded. The results from this analysis describe socio-economic mobility pathways associated with usage of the Kelsa+ PC. Management interviews We conducted structured interviews with five facility managers at the pilot location (across transport, security, housekeeping, etc.) and recorded their observations around the Kelsa+ intervention including: the usage patterns of workers, any effects on worker productivity (positive or negative), any other changes in service staff behavior or work since PC access was given, etc.

Figure 1: Hours of active usage of the Kelsa+ PC per day (average over each fortnight) Using a different measurement instrument, the responses to the questions on Kelsa+ usage or non-usage included in the second English proficiency test revealed that all but two workers in the sample (0.88 share) were active users of the

155

Kelsa+ PC. They reported using it on an average of 4 days in the week preceding the test, and for an average of 60 minutes per sitting.

Usage dynamics The applications and content accessed on the Kelsa+ PC changed over the study period. Figures 6 and 7 (on pp. 13) showcase the changing distribution of application and internet usage over 31 fortnights. Initial application usage was spread evenly across the gamut of available software on the PC, indicating heavy exploratory usage. However, starting in the third fortnight, Internet Explorer becomes the dominant application launched, followed closely by multimedia applications. There is a shift away from basic interactive applications that are very popular initially (Microsoft Paint and offline games, whose usage fell from 15% to 0), to increased usage of the Internet with its dynamic content and more sophisticated applications. As Figure 7 shows, workers spent many months primarily using the internet for entertainment – viewing music videos or films on YouTube or Stage6. However, we see a rise in using the Internet for communication, through email and social networking sites like Orkut, eight months after the project was started (correlated with 6 workers simultaneously creating email accounts at this time). There is a strong motivation to use the computers for selfexpression, as Figure 3 shows. The ability to create a personal digital presence, both through images and later through email, was a source of great pride among workers, and seemingly altered the way they perceived themselves. Learning Workers followed individual and group learning paths. Several workers simply observed their peers using the PC for weeks before attempting to touch the PC themselves. As one worker commented, “For the first one and half months, I just watched how other people used the computer.” We asked why he did this without trying to use the PC himself, to which he responded that “what if something happened when I used it?” [19] Observation and individual exploration established ‘learning by doing’ routines, which were instrumental to meaningful PC literacy gains.

Figure 2: Combination of workers observing and being primary users themselves

Group usage of the Kelsa+ PC was very popular; in fact, some users actively sought out colleagues with whom to use the PC. Peer learning sometimes took the form of an informal teacher-student relationship for a particular task or application. For instance, as [19] describe, “many workers had gone through the email account creation process with a peer who was an existing email user. During the registration process, in the space where a second email address was required (usually of the person creating the account), the experienced email user friend would enter his email address, since for most workers, this was their first email account.” For other workers, group usage involved a symbiotic learning relationship: “I use it with a friend generally, so that we can learn from one another. What he doesn’t know I tell him, and vice versa.”

Figure 3: Workers creating personalized desktop backgrounds using the Kelsa+ PC’s webcam Usability preferences Given their educational and linguistic background, workers’ revealed a number of adaptations in input techniques to achieve functional use of the PC. Navigation using a mouse dominates since typing is problematic. Browser history is repeatedly used as an easy way to access content of choice, indicating implicit peer learning [Figure 4]. Search queries (with the “.com” tag) are repeatedly used to access online content within and across sites, given that it involves only memorizing and typing in one keyword or a short phrase, as opposed to an entire url. Email forwarding is the dominant method of online communication, with one user forwarding up to 15 emails per day to 10 or so people (mostly images).

156

Figure 4: Browser history is repeatedly used as a way to access content without typing out URLs

VI.II Individual Impact A) Self-esteem and confidence gains The 11 workers who took the self-esteem test just before the Kelsa+ intervention and five weeks after its installation, recorded a 5% jump in their self-esteem score. This change clearly cannot be attributed just to the PC installation without a control group. However, qualitative interview responses from workers appeared to reiterate the effect that the Kelsa+ PC had had on individuals’ sense of hope, confidence and self-esteem. One driver commented, “I see some changes in my life – I couldn’t speak English before, now I can speak a little… I haven’t yet changed my job or used English outside, but I now have some hope to learn… I have that courage.” Further, use of the Kelsa+ PC was seen as a first step towards further possibilities: “To do any computer course outside, I think I should know some basics.... having picked up some basics today I have the confidence that when I take up a course... I will be able to catch up ... and I am very inspired by using the PC here to know more and learn.” Given that the Kelsa+ PC was in many cases the first time a worker had touched a PC, it elicited responses such as: “I felt so happy that day when we had the interview. For the first time I touched a computer and did so many things without a mistake….. I don’t have an email account. So now maybe I can make one on this computer.” [20] The continuous access to a PC at a location that workers visited everyday lowered the barrier to usage considerably, “Since, the computer is here, we get awareness! Also because we can see the computer daily ....my desire to learn and use it has increased very much.” And all this of course translated into changes in how workers viewed their workplace: “In all my service, this is one of the best workplaces I have seen.” B) Digital literacy gains Two workers, a housekeeping worker and a driver, with minimal prior PC exposure and varied demographic characteristics, were administered PC literacy tests before the

Kelsa+ intervention and 18 months after it was initiated. Churn in workers and the small sample size, curbed such a pre- and post-test for a larger share of workers. Such testing would be essential to establishing average impact across workers. However, the results for these two respondents are encouraging, given their low educational attainment and given that in these cases we can clearly trace the digital literacy skill gains to workers’ Kelsa+ PC usage alone. Both workers had experienced gains in a core set of PC literacy skills (see Table 1). While they were unable to perform any of the tasks in the pre-test (save for turning on the PC), both were now able to open a web browser window, launch a search application, navigate through the results, and open and close a document-processing application. Similar to the Hole-in-the-Wall gains for children, these measured improvements in digital literacy for two such demographically and occupationally distinct workers, from entirely voluntary and informal learning processes, are promising. Neither of these workers had interacted with a PC outside of the office during this time. Neither of them had been schooled in English. One of them had left school after class 7, the other after class 4. This only goes to show how despite the average educational attainment among the pilot location’s service staff group being higher at Class 10, even those who had much lower levels of education in a local language saw gains in basic digital literacy from Kelsa+ usage. Table 1: PC Literacy Test Results A R Occupation/ Age Driver/ 55 Housekeeping/ 23 Education/ Class 4/ Tamil Class 7 / Kannada Medium of instruction Annual Income $1600 $1000 (US$) Previous PC Never touched a Had touched a PC Exposure PC before, but never used one Reported usage 30-45 min each 30 min every day frequency session; 3-4 sessions a week Key applications Local-language Local-language used news portals, music and films, games, localemail, games language music and films PC literacy scores PrePost-18 PrePost-18 mths mths 9 9 9 Turn on the PC ? 9 9 Play Windows X X (offline) Games 9 9 Open Internet X X Browser and go to a search engine 9 Enter a search X ? X query 9 9 Open the best X X

157

search result and go back to results list 9 Sign-in to E-mail X X X client 9 Start MS-Word X X ? application Type in MS-Word X X X X 9 Save document X X X 9 Print document X X X 9 9 Close Word X X application 9 9 Shut down PC X X 9 : Successfully completed the task ? : Partially fulfilled the task/ completed the task with extensive prompting X : Does not know/ did not attempt Source: [19] C) Improvements in English proficiency Over a 3-month period of Kelsa+ usage, the English proficiency scores for the seven workers who took both preand post-tests improved from 32% to 41%. Looking at their individual trajectories, we find that of the seven, four saw increases in their English proficiency while the other three saw declines. Yet, the increments upward (average movement of +19.4%, p=0.04) appear to be significant while the movements downward are not distinguishable from measurement error of the same proficiency level (average movement of -3.8%, p=0.13). This possibly indicates an interaction between individual worker motivation or initiative and the availability of the PC-based learning material, to produce differential skill gain outcomes. A key issue is, therefore, the need to understand motivational tools that might make available skill-development tools more widely used across all workers.

Table 2: English Proficiency Test Results Overall Test 1 Test 2 Average score (%) 45% 53% +8% Sample size (n) 20 17 Pre- and post-test cases (7 workers) Average score (%) 31.7%

41.2%

+9.5% (p=0.14)

D) Socio-economic mobility pathways While the measured estimates may indicate gains in particular skills for select individuals, the true impact of such facilities would need to manifest as improved socio-economic outcomes for workers. Three cases illustrate the variety of ways in which these skill gains have translated into individual workers’ ‘development’. These workers displayed particularly

high motivation levels, and their experiences are anecdotal. At best, their outcomes indicate the extent of welfare impact that is possible when worker initiative and circumstance interact with access to the Kelsa+ facility to produce welfareimproving outcomes. Upward mobility within the same office A housekeeping worker with a Class 7, local-language, rural government school education, and no prior exposure to a PC, began using the Kelsa+ PC for 60-90 minutes each day after his work shift. He used it for games, internet video, music, etc. A few months later, he created an email ID with the help of a driver colleague, began using the Learn-English CDs available at the Kelsa+ PC, and various Microsoft Office applications. He was given an initial typing job on MS Excel by the IT staff in the office, which he completed and submitted satisfactorily. He was then given additional inventory data entry jobs (on Excel), before being hired as a dedicated worker for the IT staff. English skill acquisition A driver with no prior exposure to a PC and a Kannadamedium government school Class 10 education, took an interest in using the PC for music, movies, as well as learning English, spending 2-8 hours of off-duty time per workshift on the Kelsa+ PC. He specifically requested an extension to stay in this job for a few more months (a number of drivers left the facility due to inconvenient shift changes at this time; contract workers are often moved from one location to another), since he had begun using the Learn-English CDs at the Kelsa+ PC and wanted to acquire a certain level of competence before having to leave. Shift to a new career path A security guard with a Class 12, local-language, rural government school education, sat at the Kelsa+ PC for 60-120 minutes before or after a shift. He used the Kelsa+ PC to practice typing as he underwent a data-entry training course outside the office. The practice afforded by access to the PC at the workplace was key, in his opinion, to his successful job interview for a data-entry job. He has since moved from his security guard job to a data entry position at another firm. His pride in his new job was reflected in his comments, “Today I can stand up in front of my father and friends and say that I am no more a watchman, but I am doing a computer job."

VI.III Management feedback A) Managers at pilot location The response from the management at the pilot location was strongly positive. They felt that such exposure to technology made their workers confident and knowledgeable, while also being a source of recreation. Moreover, they commented that providing a PC access facility for the service staff at the workplace had a special connotation for workers: “sitting in a cyber [café] and learning and sitting in their office and learning is something different.”

158

Certain improvements in the workers' ability to communicate in English were described. In addition, some workers had demonstrated initiative in integrating PC usage into their own everyday workflows. A worker in the housekeeping division, for instance, had begun to type up the weekly stock order list for the office pantries on MS Excel and was emailing these to his supervisor, instead of writing them out by hand (Figure 5).

location for the PCs due to safety measures in some companies, risk of downloading or sending malicious or offensive content from the PCs, legal concerns regarding company liability, the risk of computers being used for entertainment purposes only, and the associated effort for regulation and supervision of usage being too huge for companies to undertake. [25] The fact that most of these workers are contracted and are not direct employees of these companies also contributed to reluctance by some corporations: “These workers are not our employees so why should we educate them. Their employers may not like it.” In general cost was not perceived to be an obstacle in proceeding with Kelsa+ amongst most of the companies interviewed. Connectivity charges were considered the biggest cost component, but procuring PCs for the initiative was not perceived to be an issue. VII

DISCUSSION

In proposing Kelsa+ as a potentially effective ICT4D intervention for low-income office service staff, we return to the related work in this field and discuss this intervention’s comparative strengths and limitations. (1) Extending ‘Hole-in-the-Wall’ possibilities Similar to the HitW intervention, Kelsa+ highlights how acquiring basic digital literacy skills has little to do with socio-economic background or educational quality, but much more to do with immersion, practice, and learning by doing. Persistent and minimally-invasive access to the PC for workers’ continuous exploration in the office environment through Kelsa+ has allowed for the following primary benefits:

Figure 5 a and b: Housekeeping workers maintaining task lists on MS Excel at the Kelsa+ PC (even as they multitask with playing games) B) Managers at potential implementation locations The feedback from the fifteen corporations contacted as potential implementers of Kelsa+ facilities fell into two camps. One set of corporations saw Kelsa+ as a tool for their workers to learn English, which they considered a major factor in increasing workers’ confidence and employability. The intervention would bring more equity and awareness to their service staff, and change the way they viewed technology and the world. As one manager commented, Kelsa+ is "not just a little feel good project, but something that can have an impact if done right.” [25] On the flip side, the managers at a number of other corporate facilities were very concerned about the potential risks of this intervention. Security of both physical equipment and content was perceived as the number one concern. Other perceived risks included difficulty in finding a suitable

(a) Improvements in basic digital literacy, including the ability to turn a PC on, operate an input device effectively, identify various applications, and navigate through the web to preferred content. (b) Improvements in English proficiency, driven not only by the use of dedicated ESL content, but also by the workers’ repeated interaction with English language content on the web, and sporadic use of email and office productivity tools. (c) Improvements in hope, confidence and self-esteem. Kelsa+’s achievement lies in its demonstration of how easy PC usage can be, and its encouragement to workers to believe that they can learn more as they work and at no additional cost. Should this hope translate systematically into real outcomes manifested in test scores, or new jobs, workers will be further inspired to experiment and learn, thereby establishing a virtuous feedback loop. In a departure from the HitW studies, Kelsa+ also explores the effects of the intervention on the institutional context that houses such an intervention. The pilot office facility saw greater staff morale and attachment to the workplace from having such a recreation and learning facility at the

159

office. Worker initiative to incorporate the Kelsa+ PC into their work routines was welcomed. Managers responded positively to improvements in worker confidence and knowledge, and to enhanced worker capacity through improved digital literacy and communication skills. At the same time, it is unclear whether these dynamics of security, productivity, and open exchange seen in the pilot will repeat themselves in other Kelsa+ locations. For instance, restrictions in the timings that workers can stay at the office facility, before or after their workshifts, will indeed influence the extent to which they can experiment with the PC and gain from access. Resolving such issues will involve political bargaining between stakeholders, a process inherent to the effective design and implementation of any sustainable ICTD intervention [26]. (2) Slow and sustained wins the race What makes the Kelsa+ installation any different from other telecentre/ kiosk or digital literacy initiatives in developing countries? There are three key aspects of the Kelsa+ system’s design that distinguish the user experience it offers lowincome service workers from other similar ICTD initiatives: (a) Availability and support: The office environment provides 24x7 availability and all the necessary infrastructural, technical, social and financial support required to sustain the health of the PC for the workers. Skilled maintenance is always at hand, more experienced users are all around to ask for advice, and the institution is already an integral part of workers’ daily routines. As a result, the barrier for workers to begin experimenting with the PC is considerably lowered. (b) Digital habituation: Interaction with the Kelsa+ PC has no fee and is not limited in any sense. This allows workers to respond to the PC spontaneously and evolve behaviors incrementally as they build skill. Workers engage in a slow process of familiarization with the technology as they learn basic navigation techniques and understand what the technology can be used to achieve (or not) and how; a process [19] term ‘digital habituation’. Such habituation “constitutes a critical intermediate step between providing PC access to a disadvantaged community and achieving sustained development impact.” [19] (c) Learning by doing / learning through peers: An advantage offered by the Kelsa+ PC and usually not available at regular telecentres, formal IT training institutes, or libraries, is the ability to learn passively and informally from peers, in addition to learning actively through doing. As a security guard who had done two years of computer courses at a government college noted: “They used to teach us basics...but I didn’t pick up much from the class… I generally learnt things on my own after coming here...once I see some people using [certain applications or features] ....and then next time I generally go about following the same.”

(3) Low marginal cost for well-endowed providers In the PC kiosk franchise arrangement, there is often an entire livelihood, that of the kiosk operator, depending on the success or failure of the PC kiosk business. In contrast, one of the major advantages of Kelsa+ is that it is an intervention that costs marginally nothing to those financing and running it, i.e. large private corporations. The providers in this case already manage large IT budgets and maintaining an additional set of PCs has a negligible cost. The Kelsa+ intervention involves a set of basic components, with the major costs being the upfront capital expense for the devices and software (operating system, internet browser, office productivity suite, anti-virus application). Additionally, there are monthly connectivity charges. There exist several procurement and connectivity options that determine the final cost of a given Kelsa+ facility. For a Kelsa+ facility with 40 workers, with capital costs amortized over a 3 year period, the monthly cost can range from $3.4 per worker (when a new PC is purchased, Microsoft Office is installed, and a 512 Kbps Unlimited Data Transfer connection is used) to $1.3 per worker (when a refurbished PC with Open Office is used, over a 256 Kbps Unlimited Data Transfer connection) [25]. (4) Development through ‘agency’ enhancement A key aspect of the Kelsa+ intervention is the freedom that accompanies exploration on the PC. The opportunity cost for workers to invest in their own learning is minimized through free access and convenient placement of the PC within their workplace. More importantly, by refraining from any restrictions on workers’ access to particular applications versus others, the agency of the worker is respected and encouraged in determining particular usages of the PC that meet his/her need [27]. Kelsa+ encourages a peer-learning model, through which workers with heterogeneous exposures and skills are allowed an opportunity to share their knowledge at a location where they congregate organically, i.e., the work place. Learning is driven by the user’s demands, and when not met by peers, workers are able to consult employees with more knowledge or step up and draw from training courses offered externally. In all this, the worker is centre-stage and his/her decisions dictate usage and impact, which differs from more paternalistic interventions in the ICTD space where certain ‘developmental’ results are expected and deviations from those results are treated as shortcomings. VIII LIMITATIONS AND ONGOING WORK Though based on a simple premise, Kelsa+ is a powerful idea because it takes the ICTD discussion and integrates it with the daily workings of mainstream institutions in developing countries. Many populations in need operate in or

160

close to existing IT infrastructures. As Kelsa+ shows, the innovations needed to have technology be useful to such groups are, therefore, mostly of political will, design, and process. There are many shortcomings in the current study. The evidence presented is based on a handful of cases, and such anecdotal positive effects cannot be mistaken for large-scale systematic impact. It may well be that the workers whose results we have recorded are exceptionally motivated and do not form the norm in this segment. The facility where the pilot was implemented was in fact a small software development and research centre, which may have presented many unique conducive factors for Kelsa+ to take off, including caring staff, an environment of trust, ubiquitous IT presence, strong IT maintenance support, 24x7 access, and open policies regarding employee PC usage. The interview responses from the participants may also have been influenced by the peer relationship they share with the researchers, which would prevent the sharing of negative feedback. It is clear that objectively verifying the value of Kelsa+ as a general ICTD intervention involves exploring it in a variety of institutional contexts. This is our ongoing effort. In extending the scope of Kelsa+ to become a systematic tool for socioeconomic mobility among urban low-income office service staff, our current efforts involve emphasizing longitudinal tracking of workers to map welfare impact, deploying Kelsa+ in non-IT corporate facilities, introducing some structured learning components to test their effect on learning outcomes and worker welfare, introducing certification options to prove skill acquisition externally during job-search, offering supplementary income generation possibilities for simple mini-tasks completed at the Kelsa+ PC, understanding the specific constraints that prevent women workers from using the Kelsa+ facility, and testing motivational tools that might influence the usage of the PC for explicit skill-building across domains and workers.

ACKNOWLEDGMENT To the workers involved in this project, we owe our most sincere thanks, for allowing us to examine and learn from their experience. Many thanks to the facilities management and IT staff at the pilot location for their support. Thanks to Itamar Kimchi for his help with refining the log analysis results. REFERENCES [1] [2] [3]

[4] [5]

[6] [7] [8] [9] [10] [11]

[12] [13]

This study draws the ICTD community’s attention to a subset of the poor who spend the bulk of their time around sophisticated IT infrastructures, but so far, do not gain from such proximity. We have described a pilot implementation of the Kelsa+ project in an office facility with 35 service staff, among whom the average worker earns $100 a month and has studied till class 10 in the local language. Over a period of 18 months, we saw broad improvements in workers’ confidence, self-esteem, and basic digital literacy, while a few individuals experienced increases in second-language (English) proficiency and career opportunities. Verifying these results through larger-scale and wider-scoped Kelsa+ implementations and measurements is now necessary to build on the promise of this simple, cost-effective, yet powerful ICTD intervention.

[14]

[15]

[16]

[17]

NASSCOM, ‘Indian IT Sector Score 10-in-10,” February 2007, URL: http://www.nasscom.in/Nasscom/templates/NormalPage.aspx?id=50833 NIIT: www.niit.com R. Dangwal, S. Jha, S. Chatterjee, and S. Mitra, “A Model of How Children Acquire Computing Skills from Hole-in-the-Wall Computers in Public Places,” Information Technologies and International Development, 2(4), 2005, pp. 41–60. M. Warschauer, Technology and social inclusion: Rethinking the digital divide. Cambridge, MA: MIT Press, 2003. L. Linden, “Complement or Substitute? The Effect of Technology on Student Achievement in India,” unpublished manuscript, Columbia university, June 2008. URL:http://www.columbia.edu/~ll2240/Gyan_Shala_CAL_2008-0603.pdf . Last accessed September 22, 2008. A. Lane, and M. Porch. “Computer Aided Learning and its impact on the performance of non-specialist accounting undergraduates,” Accounting Education, 11 (3), 2002, pp. 217-233. The One Laptop Per Child (OLPC) initiative: www.laptop.org ; Multipoint for education: http://www.microsoft.com/unlimitedpotential/programs/multipoint.mspx T. Oppenheimer, The Flickering Mind: Saving education from the false promise of technology. Random House Publishers, 2003. A. Mukherjee, “Build Robots Create Science – A Constructivist Education Initiative for Indian Schools”, Proceedings of Development by Design (dyd 02), Bangalore, India, 2002. R. Wang, et al. “Distance Learning Technologies for Basic Education in Disadvantaged Areas” Proceedings of the 8th Global Chinese Conference on Computers in Education (GCCCE 04), 2004. W. Schinagl, “New learning of adults in information and knowledge society”, Journal of Universal Computer Science, 7 (7), 2001, pp. 623628. URL:http://www.jucs.org/jucs_7_7/new_learning_of_adults/Schinagl_ W.pdf R. Heeks, “ICT4D 2.0: The Next Phase of Applying ICT for International Development,” IEEE Computer magazine, June 2008, pp. 26-33. S. Bailur, “The Complexities of Community Participation in Rural Information Systems Projects: The Case of ‘Our Voices’”, Proceedings of the 9th International Conference on Social Implications of Computers in Developing Countries, São Paulo, Brazil, May 2007. K. Kiri, and D. Menon, "For Profit Rural Kiosks in India: Achievements and Challenges," i4d magazine, June 2006. URL: http://www.i4donline.net/articles/currentarticle.asp?articleid=700&typ=Features R. Veeraraghavan, N. Yasodhar, and K. Toyama, “Warana Unwired: Replacing PCs with Mobile Phones in a Rural Sugarcane Cooperative,” Proceedings of the 2nd IEEE/ACM International Conference on Information and Communication Technologies and Development, December 15-16, 2007, Bangalore, India, pp. 89-98. R. Kuriyan, I. Ray, and K. Toyama, "Integrating Social Development and Financial Sustainability: The Challenges of Rural Kiosks in Kerala," Proceedings of the 1st IEEE/ACM International Conference on Information and Communication Technologies and Development, May 2006, Berkeley, USA. N. Rangaswamy, “Social Entrepreneurship as Critical Agency: A study of Rural Internet kiosks,” Proceedings of the 1st IEEE/ACM International Conference on Information and Communication Technologies and Development, May 2006, Berkeley, USA.

161 [18] J. Pal, “Examining e-literacy Using Telecenters as Public Spending: The Case of Akshaya,” Proceedings of the 2nd IEEE/ACM International Conference on Information and Communication Technologies and Development, December 15-16, 2007, Bangalore, India, pp. 59-67. [19] A. L. Ratan, and S. Satpathy, “Digital Habituation as a Basis for Digital Inclusion,” Microsoft Research Technical Report, June 2008. [20] U.S. Pawar, A.L. Ratan and S. Blagsvedt, “An ‘Office Hole-in-the-Wall’ Exploration,” Microsoft Research Technical Report, January 2008. [21] A.L. Ratan, “Lessons from Low-income Workers in Bangalore on the Value of Information Technology.” Paper presented at the Conference on Living the Information Society: The Impact of ICT on People, Work, and Communities in Asia, Manila, Philippines, April 23-24, 2007. [22] R. Veeraraghavan, G. Singh, K. Toyama, and D. Menon. “Kiosk Usage Measurement using a Software Logging Tool,” Proceedings of the 1st IEEE/ACM International Conference on Information and Communication Technologies and Development, May 2006, Berkeley, USA, pp. 317-324.

[23] The Rosenberg Self-Esteem Index: http://chipts.ucla.edu/assessment/Assessment_Instruments/Assessment_f iles_new/assess_rse.htm [24] Key English Test (KET) certification preparation, Cambridge University. URL: http://www.cambridgeesol.org/exams/generalenglish/ket.html [25] L. Zia, “Internship report: Digital Literacy for Low-Income Workers”, unpublished, Cambridge University, 2008. [26] R. De, and A.L. Ratan, “'Whose Gain is it Anyway?' Structurational perspectives on deploying ICTs for development in India's microfinance sector,” Information Technology for Development, forthcoming. [27] A. L. Ratan, and S. Bailur, “Welfare, Agency and ‘ICT for Development’,” Proceedings of the 2nd IEEE/ACM International Conference on Information and Communication Technologies and Development, December 15-16, 2007, Bangalore, India, pp. 119-130.

Figure 6: Distribution of applications used at the Kelsa+ PC over 18 months. Dominance of internet usage can be clearly seen. Decline in use of applications like offline games also seen.

162

Figure 7: Distribution of internet usage at the Kelsa+ PC per fortnight over 18 months. Increase in Internet video usage is clear. Steady and high usage of email is seen after fortnight 15. Social networking websites see increased usage starting in fortnight 21.

163

Mapping the Dynamics of Social Enterprises and ICTD in Cambodia Kelly Hutchinson, Alemayehu Molla

Abstract— As Information and Communication Technology (ICT) and social enterprises become drivers of economic growth, the nexus provides opportunities for new models of business to bring benefits to communities in developing countries. Recognising the complex dynamics and range of actors in this diverse and emerging sector, this study chooses to document the external influences, use and impact of ICT on social enterprises (SEs) -enterprises that have both a business and social development goal. The problem investigated is the potential gap between the rhetoric of the ‘promise of e-business’ versus the dynamics of enactment and impacts of ICT in practice in the social enterprise sector. The main contribution of the research is to identify the real development impact of ICT use by social enterprises by assessing the role of ICTs in achieving their goals. Its major finding is that SEs’ social and business missions are inextricably linked .and their use of ICTs and benefits from it can be used as a proxy to assess ICTD. Thus SEs provide a new paradigm in the study of ICTD. Index Terms—ICTD, ICT, social enterprise, international development, community informatics, Cambodia, CICs.

T

I. INTRODUCTION

he usefulness of ICTs to achieve development goals are increasingly accepted by international development agencies such as the United Nations (UN) and the World Bank. As ICTs are relatively new tools within the development arena, their impact is emerging through the work of a number of researchers within the new discipline, ICT for Development (ICTD). Whilst the position of most of the multilateral agencies is grossly optimistic about the potential of ICT for addressing development needs of enterprises in developing countries, there is a need for empirical evidence as to the real impact of ICTs such as e-business in developing countries This paper focuses on the intersection of social enterprises (SEs) as dual social-business mission organisations, and the application of ICTs by those organisations in an international development context. Social enterprises generate revenue to support their social mission, as well as providing employment for often marginalised people [8], [9]. Social enterprises do business with an explicit social mission–helping poor and marginalized people to participate in sustainable business Manuscript submitted September 22, 2008.

opportunities. For these groups, social enterprises can be an important source of jobs, income, training, business mentoring, and technical input [2], [3], [35]. As ICTs and social enterprises become drivers of economic growth, it is important to investigate this nexus to see what opportunities these new models of business can bring to communities in developing countries. The study takes ebusiness as a typical instance of ICTs and argues that social enterprises are well positioned to benefit from the application and use of ICTs and to deliver the promise of e-business benefits for development. The paper explores how social enterprises are using ebusiness for their comparative advantage, and the outcomes (benefits) of e-business to social enterprises and the impact (attributable to e-business) of social enterprises in the target community. The study is situated in Cambodia where there are many social enterprises active in IT services, handicraft, hospitality and processing/production., Cambodia, therefore provides a space where a vibrant social entrepreneurial spirit presented a unique perspective on the perception, use and benefit of ICT in an international development context. Through mapping organisations in the Cambodian social enterprise sector, this paper aims to fill the knowledge gap, i.e., empirical evidence on the impact of ITCD, by reviewing donors’, associations’ and NGOs’ experience in order for others to learn how to use ICT to benefit their communities. Exploring how ICT is enacted in practice by social enterprises, although descriptive in nature, represents an original contribution to the body of knowledge and practice of ICTD. II. LITERATURE REVIEW ICT was promoted as offering businesses in developing countries a potential for creating new exchange mechanisms to enable them to compete on a more equal basis in world markets [38]-[41], [37], [45]-[47]. Apart from donor and multilateral agencies, there is a growing body of academic research such as SMEs’ uptake of ICTs in Botswana [12], the contextual study into ICTs in international development agenda in Vietnam [4] and the issues facing sustainability of community ICTs [15]. Following on from this nexus between ICTs and development is more applied research which builds on the agenda of donor agencies regarding the uptake of ICTs by SMEs and the role of change agents in facilitating this process [13].

164

Whilst the position of most of the multilateral agencies is grossly optimistic about the potential of ICT for addressing development needs of enterprises in developing countries in general and that of SMEs in particular, there is an acknowledgment that tailored initiatives (such as e-commerce) are best suited for ICT to achieve development goals [20][21]. However, E-commerce has found it challenging to deliver the purported benefits espoused in the early studies of UNCTAD [38] and the OECD [30]. For instance, the DFID Globalisation and Poverty study explored the benefits of ebusiness for developing countries found that such benefits are largely missing and can only be realised if developing countries’ enterprises build their ability to develop tailored marketplaces [22] , [31]. More specific research [27]-[28] into the reality of e-commerce benefits among businesses in developing countries also cautioned the over-optimistic expectations of e-commerce benefits. Overall, current evidence appears to suggest that for e-commerce to become more widespread in a way that benefits producer firms in developing countries, greater attention will need to be given to how firms relate to each other within global value chains [25], as well as the specific types of transactions they are involved in. Core to this debate is the call for empirical evidence as to the real impact of ICTs such as e-business in developing countries– ‘whose development does e-commerce effect?–as raised by [30]. Finally the call for ‘bottom-up’ approaches in developing countries that are based on realistic assessments of ICTs and e-business opportunities and obstacles, and region and value chain specific solutions provide impetus for this current research. This study argues that social enterprises are one form of enterprise that are likely to benefit from the application and use of ICT and can then deliver the promise of e-business benefits for development. Emerson & Twersky defined a social enterprise as a ‘generic term for a non-profit enterprise, social-purpose business or revenue-generating venture founded to support or create economic opportunities for poor and disadvantaged populations while simultaneously operating with reference to the financial bottom line’ [14]. Davis [8] sees social entrepreneurship as extending the definition of entrepreneurship by its emphasis on ethical integrity and maximizing social value rather than private value or profit. The diversity of the emerging field prompted Alter to write the Social Enterprises Typology [1]. Alter’s definition captures purpose, approach and practical applications: A social enterprise is any non-profit-owned revenue-generating venture created for the purpose of contributing to a social cause while operating with the discipline, innovation and determination of a for-profit business [1]. Key to the understanding of social enterprises is their focus on sustainability and their dual social and business mission. Sustainability is a core principle of social enterprises and refers to a business venture’s simultaneous pursuit of beneficial outcomes along three dimensions: economic, social and environmental. As a way forward in building sustainable

communities, social enterprises generate revenue to support their social mission, as well as providing employment for often marginalised people [8], [9]. A social mission is the clearly stated goal of providing benefit to some section of society. This requires identifying the target group the enterprise wishes to support. It also requires clear statements of intent to make sure that it is understood that the purpose of the organisation is to meet the social, as well as the business mission. The growing relevance of social entrepreneurship for development builds on the success of SMEs in reducing poverty and improving livelihoods The intersection of ICT and social entrepreneurship therefore provides a way forward in building sustainable communities. As a framework Alter’s [1] social enterprise typology illustrates how social enterprises have come up with ownership models, income and capitalisation strategies, and the unique management and service systems designed to maximize social value. The international development focus has shifted in recent years to building sustainable enterprises and encouraging the emergent SME sector [10], [42]. In acknowledging this shift and increased focus on economic and social sustainability we find that social enterprises are playing an increasing role with the support of donors, governments and communities. III. ASSESSMENT FRAMEWORKS In order to assess the outcome and impact that ICT may deliver to social enterprises this study considers how social enterprises’ use ICT to achieve their dual mission and how they in turn measure success. When considering how to measure impact, the line of investigation reviewed a range of areas including livelihood and capability frameworks, and current work in the field of ICTD. Impact can be measured using livelihood and capability frameworks [28]. In particular DFID’s Sustainable Livelihoods Framework (SLF) is an effective tool for uncovering and highlighting the relationships between the livelihoods of vulnerable people, businesses and the institutional context in which business transactions take place. The challenges facing the social enterprise sector are similar to those of other non-profit organisations where the social mission has driven programs yet there has been little thought how to measure the achievement of their social goals. Some effort has been put into developing tools to assess the impact of social enterprises in what is commonly known as ‘social accounting’. Payne clearly states that social accounting is not a social impact assessment as the official title implies, ‘Social and Ethical Accounting, Auditing and Reporting’ (SEAAR) [33]. Built around stakeholders, AA1000 (http://www.accountability.org.uk) seeks to link the defining and embedding of an organisation's values to the development of performance targets, thus tying social and ethical issues into the organisation's strategic management’ [33, p14]. This research assumes that social enterprise impact is predicated on the organisation's mission, the social objectives it intends to

165

achieve, and what impacts can be measured. Figure 1 based on Alter [1] provides an overview of the research framework and Table 1 details some of the key concepts to be explored.

Fig. 1. Research Map

research [6], [22], [24]. A case study may capture reality in greater detail, analysing a greater number of variables than is possible with other research methods [15]. Yin [47] recommended the use of case-study protocol using a triangulation of data sources. This study looks at the key actors in the Cambodian social enterprise environment in an effort to answer the research question. As this study addresses a new area of academic inquiry there was little primary data regarding the topic, it requires a range of research methods and active dialogue with organisations involved in the sector. Triangulation for this study was created through interviews, surveys and document analysis. TABLE 2 SAMPLE OF ORGANISATIONS

TABLE I

KEY CONSTRUCTS OF INTEREST Construct

Definitions

Examples

Sub-sectors Handicraft

Beneficiaries

Target community who benefit from SEs activities

Disabled, Women, Street Children, Orphans, People Living With HIV/AIDS, Youth, Unemployed, Trafficking Survivors, Villagers, Farmers

Hospitality IT Services

Ways in which social enterprises provide support,

Employment, Training/ Skills Development, As Consumer To Local Producers (Creating Market), As Marketer Of Local Producer’s Products (Intermediary), Supplier to Local Producers, n, Assistance In Business Development, Product Design/ Service Development

Forms of Support

Social accounting and social auditing7

Methods of measuring and reporting an organisational social and ethical performance.

An organisation conducts a social audit makes itself accountable to its stakeholders and commits itself to following the audit's recommendations.

Social Mission Indicators

A set of measures of determining if the social mission is being met

Evaluate Performance against Social Indicators, Ability to Employ More People, Recommendations from Customers, Recommendations from Beneficiaries, Awards/ External Recognition from Peers, Specify, Awards/ Recognition from Development Agencies/Donors

IV. METHODOLOGY In order to bridge the gap between theory and practice a qualitative approach was chosen as applied research leans towards involving practitioners, users and other stakeholders, in order to provide practical outcomes. The research took a holistic–multi-disciplinary, multi-method–approach to triangulate data collection and is detailed as to sample and selection, cross-sectional survey, semi-structured interviews, information observation, document analysis and artefact review as utilised in the case study. Examining a case of one sector’s relationship to ICT and development allows an exploration of particularity and complexity of a single case to understand its activity within important circumstances (Stake 2000) utilising the case study method of exploratory IS

Associations Artisans Association of Cambodia Craft Network Cambodian Silk Forum ICT:CAM

Processing Multi-sector

Social Enterprises Hagar Design Handmade Textile Association Colours of Cambodia Shinta Mani CIST Community Info Centres Digital Divide Data Yejj Info Senteurs D’Angkor

WASMB

Participants were both industry protagonists and those that support them through associations and donor agencies. Supplementary data was collected from a review of documents and artefacts such as brochures, marketing material and websites. Donor and media reports were sought regarding the activities of these social enterprises in order to ensure as complete a picture of the social enterprise sector as possible. Of the 50 organisations identified in the study, 18 were interviewed and 25 were surveyed. 1) Defining Beneficiaries The next step is to identify the target beneficiaries of social enterprises’ blended mission. A useful tool from the UK is the DETR 2000 Index of Deprivation – a comprehensive index for comparing levels of deficiency across a range of issues. The index covers seven aspects of deprivation – income, employment, poor health and disability, education skills and training, housing, geographical access to services and child poverty. The framework was used in this study to assist interviewees in identifying target beneficiaries. The resultant categories were then used in the survey and are presented in Figure 3. below.

166

Fig. 3. Target Beneficiaries

The categories in descending order represent lesser degrees of importance. The findings showed a high incidence of support for beneficiaries with disabilities; however women and youth are also a major focus of social enterprises in Cambodia. As this data is aggregated across different sub-sectors, another picture emerges. The handicraft sector supports beneficiaries from all categories, including street children and people living with HIV/AIDS, which neither IT services nor social enterprises from production sub-sectors supported. Although the survey provides just one set of data, the interviews and document analysis also identified that some other social enterprises supported these disadvantaged people in Cambodia. Mith Samlanh works with street children through their restaurants Friends and Romdeng, and Khana is an association working with NGOs in support of people living with HIV/AIDS. Both recognize the potential that income generation activities can have in creating sustainable livelihoods for their target beneficiaries. Now we turn to social enterprises themselves to measure their impact on their target community and see how ICT does or does not play a role in achieving their blended mission. The analysis will address the blended value proposition in conjunction with the social enterprise typology to determine the outcomes and impact of ICT on social enterprises. V. KEY FINDINGS Outcome and impact of ICT – achieving social enterprises’ mission The paper now turns to assess if ICT does deliver benefit to social enterprises by considering how ICT helps them to achieve their blended mission which combines commercial and community goals. It is hoped that the evidence presented here will contribute much needed insight into the impact of ICTs on development more generally. A. Blending the dual mission In addressing how the dual mission of social enterprises is enacted in practice this paper asks in turn, how ICT interacts in this context. The new organisational model of social enterprises embodies a shift in the pursuit of sustainable development on all levels, financial, social and environmental. This move to a ‘blended value proposition’ [14] reflects the integration of the dual mission, which provides the context as

to the contention of this paper – that it is due to the very nature of social enterprises that ICT does bring benefit directly as it assists the overall mission and the ‘total value creation’ that social enterprises bring their beneficiaries. Initially it was thought that the benefit that social enterprises bring communities could be divided into two sides of the dual mission, business and social as illustrated in Fig. 1. However, as the research process unfolded, the realisation that the blended mission drives social enterprises had a profound influence on the resulting framework for analysing how ICT delivers benefit to target communities. To understand how this is manifested in practice, Table 3 provides some examples of organisations’ statements of their mission from the survey results. TABLE 3 EXAMPLES OF SOCIAL ENTERPRISE MISSIONS CIST Colours of Cambodia

Bringing digital opportunities to underprivileged young people, that is developing an IT training centre Provide fair trade channel for small producers and NGO products

Hagar Design

Our social mission is to train and employ highly disadvantaged Cambodian women while using our profits to financially support Hagar.

Khmer Mekong Food

Great jobs for young people and students Buying raw agriculture from farmer to push them grow more fruit and vegetable Provide nutrition and quality Safe food to consumers

Rajana Association Senteurs D’Angkor SotheaKhmer Silver Crafts

Find Markets for small producers, give them the employment, send them to school Good working conditions, share benefits with staff through a bonus system. Create employment for local villagers Social welfare for people with disability and vulnerable poor people. Work and wage for poverty

What is of interest is the intertwining of both business and social goals in many of the respondents answers. This suggests that the mission of these organisations is a blended proposition providing evidence of this new ‘business’ segment operating with specific commercial and community development goals. 1) Reframing to reflect the blended mission As a result of this investigation, there has been a slight departure from the originally perceived dual mission to the more blended version which is reflected in Figure 2. the Social Enterprise Blended Mission, which highlights the potential outcomes of ICT in social enterprises. As this is a qualitative study, following naturalistic inquiry, it is usual for the research questions to evolve over the period of investigation. What was the original understanding, that the dual mission was two separate yet complementary missions, was due to the perspective as an outsider. Through examining the data it became evident that this distinction is more blurred than originally expected.

167 TABLE 6

INDICATORS OF IMPROVED LIVELIHOODS Indicator

Fig. 2. Social Enterprise Blended Mission

As Figure 2. demonstrates, it is difficult to differentiate between the business and social drivers within the blended value proposition. Improved communication can be both with donors regarding monitoring of a social program such as DDD and UNDP, or to customers purchasing silk handbags for resale in Australia, in the case of Hagar Design through Craft Network. The findings indicate that there is not a clear distinction between an NGO that earns income from selling craft, such as Tabitha Cambodia, to an IT company which has a policy of employing disadvantaged young people, such as Yejj Info. In fact, Yejj Info provides one of the best examples of the blended value proposition with their comprehensive mission statements which combines commercially, socially, ethically and environmentally sustainable 2) Mapping the blended mission in action Social enterprise impact is predicated on the organisation's mission, the social objectives it intends to achieve, and what impacts can be measured as a result of the business initiatives. A defining ambition of social enterprises is that they support those disadvantaged in the community. Leading by example, social enterprises employ disabled computer operators at DDD, market landmine survivor’s art work at Colours of Cambodia, and encourage local communities to invest for their future through a community savings program at Tabitha Cambodia. Exactly how this complex interweaving of social and financial sustainability is achieved is what this section aims to address. This manifestation of dual sustainability illustrates the true outcomes of the blended mission. What this study acknowledges is that these are inextricably linked, so to measure one in isolation of the other denies the new paradigm that social enterprises present. B. Measuring the social impact Social impact indicates a positive effect on the target population as a result of an intervention and can be measured. Social enterprises, like all development programs, have direct as well as indirect impacts. Alter [1] provides some examples of social enterprise impact measurements and corresponding indicators. These were adapted to identify the outcomes (benefits) of ICT to social enterprises and in particular to assess the impact (attributable to ICT) of social enterprises in the target community. Survey respondents were asked ‘In your opinion how has employment with your organisation improved the livelihoods of your staff?’ and their response is shown as percentages of those who either strongly agreed or agreed with the indicators listed in Table 6 below.

% agreement

Earning a livable wage

73%

Provided new career opportunities

63%

Improved food security

71%

Improved family stability

71%

Improved health

57%

Improved access to education

57%

Increased purchasing power

71%

The nature of social development lends itself to monitoring and evaluation [11], and often the indicators for these outcomes are clearly set during the start up of the venture. In the case of donor initiated enterprises there is often a logical framework or ‘logframe’ that sets out indicators for monitoring and evaluation. In the case of social enterprises the need to measure the triple bottom line means a range of indicators from both social and business areas are required. In evaluating the impact of the current sample, a number of indicators were presented to social enterprises in the survey. The 44% incidence of evaluating performance against internally structured social indicators is promising as it suggests that these organisations are very clear of the need to meet their dual mission. This potentially could mean they will be more successful than an organisation that is not as clear on their strategic mission. Another category for measuring the success of achieving the social mission is through recognition from stakeholders. This included acknowledgement from peers, awards, and notes of appreciation from the most important people, beneficiaries and customers which accounted for the highest response rate. Although the survey data had a low incidence of recognition from development agencies/donors through awards, the other data collection discovered a number of awards and pride in showcasing social enterprises success in Cambodia is further explored below. C. Identified benefits of ICT It is through the case of the Cambodian social enterprise sector that this paper demonstrates the opportunity to address social issues through business initiatives. Benefits of using ICT, either directly as in the case of IT services with skilled labour, or indirectly by selling via an ICT-enabled intermediary. If one of the promised benefits of ICT is increased efficiency of both internal and external operations then it is important to measure how ICT impacts on social enterprise operations. In measuring the extent of ICT utilisation in the organisations surveyed, the perception on impact on operations was deemed an appropriate measure. The most common way ICT helped improve operations was through enabling external communications with clients and customers. This was further supported by the interview results, with most social enterprises seeing ICT as a core tool for customer relationship management. Simple ICTs such as phone and email were the dominant forms.

168

a)

Access to Markets

Does ICT enable social enterprises to link into the global trade of goods and services? Through the application of ICT are social enterprises able to access new markets that they previously would have been excluded from? The fair trade movement is a response by civil society to the potentially negative impact of globalisation [33]. It is important to acknowledge that the networked economy can potentially further marginalise producers in developing countries. Those wishing to redress this imbalance see the benefits that ICT can bring as a tool for trade [5]. The majority of enterprises surveyed felt that ICT provided benefit in key areas of communication with customers (80%), and sales and marketing (64%), suggesting that social enterprises are using ICT to connect to markets and facilitate trade in ways previously not imagined. b)

Income Generation

One of the defining characteristics of a social enterprise is its income generation activities. This is in fact what differentiates them from others in the non-profit sector as the goal is clearly to earn income to support the social and business development of the organisation. What is of interest to this study is the role that ICTs have played in achieving this goal. Respondents to the survey were asked to respond to a number of statements which variously reflected their perception of the impact of ICTs on achieving increased sales or income. These statements were developed from the pilot assessment of DDD and some preliminary interviews with other social enterprises to identify the various ways impact is measured. The level of agreement is a combination of agreed and strongly agreed and suggests how these social enterprises see the relationship between income generation, employment and ICTs. TABLE 7 ICT ROLE IN ACHIEVING GOALS

How do these statements apply to your organisation?

% agreement

Email has increased our sales

50%

ICT has increased our profits We have employed more staff as ICT makes it easier for us to do more work for our clients Due to increased sales from using ICTs (website/ email) we have employed more staff Due to increased income from ICT related we have been able to offer more social programs to staff

33% 20% 40% 40%

As the blended value proposition of the social enterprise is an interweaving of social and business indicators, it is not surprising that ICT helps to achieve both these simultaneously. Hence separating the business versus social is not that useful, for it is this unique blended characteristic that perhaps makes the impact of ICT more successful than if it was separated into a silo approach for development. One concern with income generation is that the proceeds go to support the beneficiaries and are not inappropriately used for administration or unrelated costs. Of course this is entirely dependent on the structure of the social enterprise; however

when direct sales are for goods produced by the target beneficiaries, customers like to know their purchase actually goes to help those who created them. This is where ICT can assist with clear communication and monitoring. An example of how this is enacted in practice can be seen on the Khmer Silk Processing Association (KSPA) website (See http://www.phaly-craft.org) where they state clearly the promise that the income goes to those who need it. All income from the sale of these products are partly used to support Future Light Orphanage (FLO) activities and children whose numbers increase everyday. c)

Employment Opportunities

Choosing an appropriate target group requires prior knowledge and theoretical understanding of the social environment of the specific country. DDD identified their target group as people who fit one of these criteria: disabled, orphans (due to war or HIV/AIDS), rural poor and trafficking survivors. They also focus on youth providing entry-level positions and training in order to develop their career opportunities. DDD found that it is important when dealing with disabled people to give them the opportunity to prove their skills. On first appearance a young woman with no hands below her wrist may seem to have limited potential as a data entry operator; however once given the chance she has become a valued employee who upholds the required productivity levels without having any fingers. This is a prime example of how DDD has taken on the challenge of providing employment opportunities for disabled people and met both individual staff goals and general business needs. d)

Capacity Building

Some social enterprises’ social mission encompasses capacity building, where target beneficiaries are provided training and work experience in using ICTs directly. In this case the social mission drives the adoption of ICTs. In the hospitality industry students are trained on hotel booking systems such as at Shinta Mani and are able to be prepared for real life experience not just the theory that class based learning alone provides. By embracing the experiential learning approach, ICTs can be used to simulate real life experiences and provide an advantage for skill development. One of the major concerns of the digital divide is not only physical connectivity the ‘second term level digital divide’ [18]. Skills to effectively use ICTs highlight that connectivity alone is not enough to bridge the gap. It is with this in mind that some social enterprises, particularly in the IT services sector, are providing direct benefit to their target communities. DDD, Yejj Info and CIST all have training and skills development as a core program of assistance. Another social enterprise in the IT sub-sector, Yejj Technology, operates a Cisco Network Academy which provides formal training programs with industry recognised qualifications so that their students from disadvantaged backgrounds can engage in the IT services sector.

169

e)

Strengthening Sponsor Relations

One unique benefit of sponsorship in the highly competitive non-profit sector is that a personal relationship is built between a sponsor and a community. It is this connection that keeps them returning or undertaking fundraising activities in their home country continues sponsorship of either an individual student as in the case of Shinta Mani, or new facilities for the Future Light Orphanage, which is run from the proceeds from the KSPA. Communication via email is a key tool for this, enabling all phases of the relationship, prior to visit and post visit, to continue seamlessly. KSPA has also branched out to provide some guest accommodation at the orphanage on the outskirts of Phnom Penh. Enabling those who wish to make their contribution to the NGO through not only donations, or purchasing goods or services, guests can volunteer to work with the children teaching English or computers providing a true link to the community. The opportunity to personally interact with the children is actively marketed on the KSPA website and also has an online booking form to ease the process. The form is then submitted and emailed to KSPA and then they communicate with the customer via email and finalise the booking and program for their visit. It is the seamless ability to engage and interact with KSPA even via a simple online form that shows how ICT facilitates this relationship. Cambodia unfortunately attracts paedophiles who visit Cambodia to take advantage of poor and disadvantaged children and women. The management of FLO take the care of their children very seriously and screen visitors to ensure they are not placed at risk but benefit from interacting with only genuinely interested volunteers who match KSPA's concern for the welfare of the children as a priority. f)

Social Outsourcing

Focusing on the philanthropic benefit of outsourcing to disadvantaged communities has been a unique strategy to counteract the backlash that US firms face as they outsource jobs overseas. Maximising this market niche, DDD shows that business can benefit communities in developing countries in a viable business model that does not exploit workers. DDD in fact targets beneficiaries from marginalised backgrounds and provides employment where others reject them. Through their social mission they provide educational scholarships that support their workers career development to leave DDD and take up new job opportunities. This then provides an opening for other disadvantaged Cambodians to join the workforce. Focusing on the staff’s professional development and achieving the business goal provides much more of an impact for local communities than charity. Businesses who value social contribution are happy to partner with social enterprises such as DDD as they can satisfy their needs to reduce costs through accessing an ethical supply chain. All of which is enabled by ICT as digitising content is the mainstay of DDD’s services. DDD also understand the benefit of social marketing and the role it can play in bringing a comparative advantage to

competitors. They have actively sought media coverage and see it as a way to promote the ideals of their social enterprise to their potential market. With the backlash about outsourcing jobs overseas a public awareness campaign in the US is very important for an enterprise such as DDD. On their website they have a News section which lists articles and stories which highlight the experience of the organisation from an external perspective. As the article on their website illustrates, DDD knows the benefit their business brings young Cambodians like Nut Pove, the opportunity they would otherwise have missed. But it is sadly not enough to provide jobs alone, this is where ICT can help spread the word of the work being done in Cambodia and it is examples such as DDD which provide evidence of how this is being done. Although social outsourcing is more commonly connected with the IT services sector, the model is also used in the manufacturing sector. In the case of handicraft sector in Cambodia, many social enterprises work with designers from international markets to produce their products. Small boutique labels make connection with suppliers in Cambodia through agents such as AAC and Craft Network to arrange production. For example the designer in the US or Australia send design, colour specifications via email and the production is made to order. As Cambodian enterprises can handle small quantities and offer individual attention to design houses they are increasingly popular. Bronwyn Blue a Small Business Development Consultant, specialising in the craft sector, works with social enterprises to improve their quality control and designs to meet the needs of international market: The core finding presented in this paper is how the blended mission is enacted in practice and in turn, how ICT interacts in this context. This study argues that it is the integration of the dual mission, which is the very nature of social enterprises, that ICT brings benefit directly as it assists the overall blended mission and the ‘total value creation’ that social enterprises bring their beneficiaries.

Fig. 4. Map Total Value Creation

VI.

DISCUSSION

This study found that the major characteristics of social enterprises are that their development objectives are to achieve a blended mission meeting both social and economic goals. From this basis the assumption would be that ICTs should be utilised to realize these goals [19], [7]. It is with this

170

in mind that this study aimed to find a realistic view of the role ICTs play in supporting social enterprises to meet their dual mission. What this study acknowledges is that these are inextricably linked, so to measure one in isolation of the other denies the new paradigm that social enterprises present. Social enterprise impact is predicated on the organisation's mission, the social objectives it intends to achieve, and what impacts can be measured. Social enterprises, like all social programs, have direct as well as indirect impacts. Alter [1] provided some examples of social enterprise impact measurements and corresponding indicators. These were adapted to identify the outcomes (benefits) of ICT to social enterprises and in particular to assess the impact of social enterprises in the target community. In evaluating the impact a number of indicators were presented to social enterprises in the survey and the results indicated that evaluation of performance was against internally structured social indicators such as improved livelihoods, increased income, access to affordable health care and access to education. This is promising as it suggests that these organisations are very clear of the need to meet their dual mission. This potentially could mean social enterprises will be more successful than an organisation that is not as clear on their strategic mission. The goal for development of disadvantaged communities can take many forms and in the case of social enterprises there appears to be a number of models as outlined in Alters Social DDD Hagar Design

Yejj Info Shinta Mani

establishing the social enterprise and of those surveyed three were still registered as an NGO whilst eight others were set up as social enterprises from the outset. The summary of results in Figure 8.e shows a number of social enterprises as classified according to the SE Typology. The level of ICT uptake is not directly linked to the level of integration however it could be claimed that there is some correlation with full integration and management vision with more successful ICT assimilation. This study argues that it is the integration of the dual mission, which is the very nature of social enterprises, where ICT brings benefit directly as it assists the overall mission and the ‘total value creation’ that social enterprises bring their beneficiaries. This is promising as it suggests that these organisations are very clear of the need to meet their dual mission. This potentially could mean social enterprises will be more successful than an organisation that is not as clear on their strategic mission. This bears direct significance on the framing and uptake of ICTs. The institutional form that the social enterprise takes itself is another potential influence as to the uptake of ICT. Management awareness, capacity of staff and commitment to use ICT to achieve the blended mission requires both sides to be successful where meeting economic needs is equally as important and social goals.

KSPA Tabitha

Enterprises Typology [1]. If we consider the social enterprises sector as a norm due to its accepted social constructs of philanthropy and social good, then this in itself could act as an influence on the enterprises uptake of ICT. So it could be claimed that the type of social enterprises may also impact on the uptake of ICT. Perhaps it is the ways the social enterprise enact their social mission that predicates how they will or will not utilise ICTs as the following examples illustrate. The form that a social enterprise takes is an interesting entry point for analysis as represented in Figure 5. Using Alter’s social enterprise typology to analyse the survey data, regarding how social enterprise perceive themselves, can provide some insight into the level of maturity and development of the organisation. One of the approaches to assessing social enterprises is to classify them based on the level of integration between social programs and business activities and this framework helped develop the research design for this current study. This is where the social enterprise typology can assist in analysis of the experience of this sector in Cambodia. Interview results indicated that two social enterprises that were established by NGOs as income generating initiatives have made the transition to being an independent social enterprise. The question this raises is to what extent does the relationship with the founding NGO continue once the transition has occurred? Interview data suggests that NGOs are instrumental in

Fig.5. SE Typology - Business/Program Integration [1]

The final aim of the research was to measure the role of social enterprises’ use of ICTs in achieving their goals. This study’s major finding was that the blended mission requires both sides to be successful, thus meeting economic and social goals as equally as important. In identifying at what point ICT intervention made an impact, the areas that stand out include income generation, access to markets, improved communication with customers, and strengthening of sponsor relations. Whilst the results indicate a positive acceptance of ICT, most social enterprises are still at the early stages of adoption. This means that there is a continued role for external institutions to support further development, with a suggested focus on meeting the dual mission in a sustainable manner, and therefore meeting the requisite standards for capacity and use. Most importantly this study acknowledges that the social and business missions are inextricably linked, so to measure one in isolation of the other denies the new paradigm that social enterprises present. Perhaps the most active sub-sector that best reflects this unique value proposition is the emerging IT sector. Social outsourcing represents a new model and the potential of the IT services sub-sector came to light during this research. In summary the case of the social enterprise sector in Cambodia provided a rich tapestry of examples with a number

171

of organisations receiving external recognition of Cambodia’s social entrepreneurs. The outlook for the other sub-sectors handicraft and processing have yet to fully engage and use ICT to their maximum benefit. As such there is a need for further support to best harness the potential that ICT can bring developing countries. VII. RECOMMENDATIONS If the aim of social enterprises is to have a positive outcome and impact for their target beneficiaries, then they should be open to develop standards that ensure no-one is exploited. Responding to the call to address issues in the regulation of global supply chains, and the potential that ICT can play in the self-regulation required in ethical trade provides another perspective to this study [22]. Through adhering to the standards of fair or ethical trade, Cambodia can build on its reputation that workers are safe and purchases make a real difference. This is an issue for workers in the home-based handicraft business model and increasingly will also become an issue for the IT outsourcing sector. As such one recommendation is to develop a fair trade charter for the social enterprise sector in order to ensure that those who are meant to be benefiting from the enterprise are in fact receiving a real benefit. The development of fair trade monitoring along the lines of the ILO Better Factories (See http://www.betterfactories.org) program is a model that could address this. This is a unique initiative that uses ICT to assist the process of transparent reporting in the garment sector in Cambodia. Using an interactive website, this project enables buyers to log in from overseas to check the status of a particular factory to see whether it has met Cambodian labour standards. One issue is that this model is also seeking to make the transition from a donor initiated program to a selfsustaining model. In fact the most active representative body in the social enterprise sector is the Artisans Association of Cambodia, which already has a working relationship with Garment Manufacturing Association of Cambodia via the Government-Private Sector Forum, so it is feasible that some mutually beneficial arrangements can be made. As there are some correlations between the manufacturing sectors and the sub-sectors of the social enterprise sector, there might be a possibility for a new holistic model to monitor social outsourcing, using ICT as a key tool to promote transparency. Another issue that arose from this research is the lack of regulation of the social enterprise sector in Cambodia, which the above recommendation would address. In the case of the NGOs making handicrafts, they are officially registered as an NGO with the Ministry of Interior, which has no reference for income generation. As such, NGOs do not pay tax, and the revenues from sales overseas are not being reinvested back into the broader economic development of Cambodia. Registration with Ministry of Commerce is required for any entity selling goods or services that may require a certificate of origin for export. These are areas that donors and other institutions working with social enterprises can assist and that

ICT can play a role in facilitating a more streamlined process for all actors. Finally, an area that poses potential is to review the long term impact of ICT in social enterprises in developing countries in terms of its contribution to poverty reduction and improved livelihoods. This study raises the issue of measuring the impact of ICTD-SE and it is recommended that further research be conducted to refine the framework. This would enable better assessment of the inherent challenges and potential impact that the intersection of ICT, development and social enterprises presents producers and communities in developing countries. In conclusion this study contends that the foundations are there for social enterprises to progress along the continuum from simple ICT use to integrated ICT and e-business. This combined with the increased level of awareness of the benefits and the supportive donor environment augurs well for Cambodian social enterprises to engage in broader use and therefore bring benefits of ICT in their development and positive impact in the lives of the target beneficiaries. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]

[12]

[13]

[14] [15]

Alter, S. K. 2004) Social Enterprise Typology [online]. Virtue Ventures LLC. 2004. Badiali, N. (2005) Tall Poppies - Enterprising the Social Economy. The Scottish Council of Voluntary Organisations (SCVO) and Kilmarnock College. Bornstein, D. (2004) How to Change the World: social entrepreneurs and the power of new ideas, Oxford University Press. Boyle, G. (2002) Putting Context into ICTs in International Development: An Institutional Networking Project in Vietnam, Journal of International Development, Vol.14, No.1, 101-112. Caserta, A. (2001) Creating a Fair Trade Partnership through a Fair Trade Global Communication System. Cash, J. I. & Lawrence, P. (Eds.) (1989) The Information Systems Research Challenge: Qualitative Research Methods, Boston, Harvard Business School Research Colloquium. Curtain, R. (2004) Information and Communications Technologies and Development: Help or Hindrance?, Curtain Consulting for AusAID. Davis, S. (2002) Social Entrepreneurship: Towards an Entrepreneurial Culture for Social and Economic Development. Youth Employment Summit. Dees, J. G. (2003) New Definitions of Social Entrepreneurship: Free Eye Exams and Wheelchair Drivers. Knowledge @ Wharton, Wharton School, University of Pensylvania. DFID (2005) Working with the Private Sector to Eliminate Poverty. DFID & the Private Sector. London. DiMaggio, P. (2001) Measuring the Impact of the Non-profit Sector on Society is Probably Impossible but Possibly Useful: A Sociological Perspective. Measuring the Impact of the Non-profit Sector. P. Flynn and V. A. Hodgkinson. New York, Plenum: pp.249-272. Duncombe, R. and R. Heeks (2001) Enterprise Development and Information and Communication Technologies in Developing Countries: Supporting “ICT-Flyers”. Manchester, UK, IDPM, University of Manchester. Duncombe, R. and A. Molla (2005) An Evidence Based Framework for Assessing the Role of Change Agents in the Promotion of E-Commerce for SMEs in Developing Countries. Connecting people and places: challenges and opportunities for development, Milton Keynes, UK, UK Development Studies Association. Emerson, J. (2003) Blended Value Proposition: Integrating Social and Financial Returns. California Management Review, 45. Emerson, J. and F. Twersky. (1996) New Social Entrepreneurs: The Success, Challenge and Lessons of Non-profit Enterprise Creation. San Francisco, Roberts Foundation.

172 [16] Galliers, R. (Ed.) (1992) Information systems research : Issues, methods, and practical guidelines, Oxford, Blackwell Scientific Publications. [17] Gurstein, M. (2005) "Sustainability of Community ICTs and its Future." The Journal of Community Informatics Vol.1 (2). [18] Hargittai, E. (2002)"Second-level digital divide: differences in people's online skills." First Monday Vol.7 (4). [19] Harris, R. (2004) ICT for Poverty Alleviation. e-Primer. Kuala Lumpur, APDIP. UNDP [20] Harris, R. and D. Vogel (n.d.) E-Commerce for Community Based Tourism (Draft) Hong Kong, Harris and Associates. [21] Heeks, R. (2002) Failure, Success and Improvisation of Information Systems Projects in Developing Countries. Development Informatics Group, IDPM, University of Manchester. Working Paper 11. [22] Heeks, R. & Duncombe, R. (2003) Ethical Trade: Issues in the Regulation of Global Supply Chains. Centre on Regulation and Competition. University of Manchester. [23] Humphrey, J., M. Robin, et al. (2003) The Reality of E-commerce with Developing Countries. Globalisation and Poverty, DFID Media @LSE and IDS. [24] Kaplan, B., Truex, D. P., Wastell, D., Wood-Harper, A. T. & Degross, J. (Eds.) (2004) Information Systems Research: Relevant Theory and Informed Practice, Norwell, MA, Kluwer Academic Publishers. [25] Kaplinksy, R. and J. Readman (2001) Integrating SMEs in Global Value Chains: Towards Partnership for Development, UNIDO. [26] Klein, H. M. & Myers, M. D. (1999) A Set of Principles for Conducting and Evaluating Interpretive Field Studies in Information Systems. MIS Qtly.Viewed: 10 November 2006. [27] Molla, A. (2004) The Impact of eReadiness on eCommerce Success in Developing Countries: Firm-Level Evidence. Development Informatics Working Paper. Viewed: 20 November 2006. [28] Molla, A. (2005) Institutions and ICTs in the Makerere university: Bridging digital divide or spreading donor dependence,. IFIP Conference. Abuja, Nigeria. [29] Murray, C. (2001) Livelihoods research: some conceptual and methodological issues. Development Studies Association Annual Conference-Panel on Livelihoods Frameworks and Poverty Analysis, University of Manchester. [30] Odedra-Straub, M. (2003) "E-commerce and development: Whose development?" Electronic Journal of Information Systems in Developing Countries Vol.11 (2): pp.1-5. [31] OECD (2004) Accelerating Pro-Poor Growth Through Support for Private Sector Development, OECD. [32] Paré, D. J. (2001) Does This Site Deliver? B2B E-commerce Services for Developing Countries. B2B E-commerce and Developing Countries. Media@LSE. London, The London School of Economics & Political Science. DFID funded project on e-commerce for developing countries.UK [33] Payne, J. E. (2002) E-Commerce Readiness for SMEs in Developing Countries: A Guide for Development Professionals. Washington, LearnLink. [34] Redfern, A. & Snedker, P. (2002) Creating Market Opportunities for Small Enterprises: Experiences of the Fair Trade Movement. SEED Working Paper 30. Geneva. [35] Spinali, L. and H. Mortimer (2001) A Scan of the Not-For-Profit Entrepreneurship: Status of the Field and Recommendations for Action, Kauffman Center of Entrepreneurial Leadership. [36] Stake, R. (2000)Case Studies. Handbook of Qualitative Research. N. K. Denzin and Y. S. Lincoln. Thousand Oaks, CA, Sage Publications. [37] UN (2003) Tools for Development Using Information and Communications Technology to Achieve the Millennium Development Goals. Working Paper, United Nations ICT Task Force. [38] UNCTAD (2001) The Digital Economy: Integrating the LDCs into the Digital Economy. E-Commerce and Development Report 2001. Brussels, UNCTAD, Electronic Commerce Branch. [39] UNCTAD (2002a) World Telecommunication Development Report: Reinventing Telecoms, UNCTAD. [40] UNCTAD (2002b) E-Commerce and Development Report 2002. Geneva, UNCTAD. [41] UNCTAD (2004a) Partnership for Development: Information and Knowledge for Development, São Paulo. UNCTAD [42] UNCTAD (2004b) E-Commerce and Development Report 2004. [43] UNDP (2006) Global partnership for development. Annual Report. New York, UNDP.

[44] WorldBank (1999) Knowledge for Development. World Development Report. Oxford University Press, World Bank. [45] WorldBank (2001) Attacking Poverty. World Development Report. Oxford University Press, World Bank. [46] WorldBank (2002a) Information and Communication Technologies: A World Bank Group Strategy. Washington, World Bank. [47] WorldBank (2002b) Information Communication Technologies. A World Bank Group Strategy. Washington DC, World Bank. [48] Yin, R. K. (1993) Case Study Research, Design and Methods. California, Sage Publications.

173

Political Incentives and Policy Outcomes: Who Benefits from Technology-Enabled Service Centers? Jennifer Bussell

Abstract—This study investigates the causes of variation in government policies to use information and communication technologies to improve service delivery to citizens. I ask why state governments in India vary in the number and type of services they offer to citizens through technology-enabled citizen service centers. I argue that politicians estimate the expected electoral benefits from providing improved services to citizens and weigh these benefits against the costs of increased government transparency and associated reductions in corrupt income. Politicians then design service center policies to maximize their chances of retaining power. Because levels of corruption and the characteristics of electoral competition vary across the Indian states, we see related variations in technology policies. These variations in policy, and in particular the services made available to citizens, have important effects on who benefits from citizen service centers. I use evidence from sixteen Indian states to test these arguments, and show that the character of the ruling government and the level of state corruption are robust predictors of variation in state-level technology policies. Index Terms—Information and communication technology, development, India, corruption, politics.

I. INTRODUCTION

W

hy do some governments invest in new technologies

while others do not? Why do certain governments implement full-scale reforms of public service delivery, while others do so in a superficial manner? The emergence of new information and communication technologies in the 1990s raised these questions in a stark way. Low-cost, digital technologies offered prospects for increasing the effectiveness of government, particularly through improved service delivery to citizens. International observers expected developing country governments, often criticized for failing to deliver services in an effective and transparent manner [1], to be the prime beneficiaries of new digital technologies [2], [3]. Developing country citizens were expected to benefit from improved Manuscript received September 17, 2008, accepted January 20, 2009. This work was supported in part by the National Science Foundation under Grant No. 0326582. Jennifer Bussell is a Ph.D. Candidate in the Department of Political Science at the University of California, Berkeley, Berkeley, CA 94720-1950 (email: [email protected]).

delivery of government services such as identity cards, birth and death certificates, and licenses; the supply of welfare and redistributive goods; and general government-citizen communication (see, inter alia, [4]-[6]). Developing country governments responded to this opportunity and began to develop new systems for “eGovernance” and digital technology-based service delivery (see, inter alia, [7]-[12]). Yet, after more than a decade, the question remains whether governments have utilized new technologies to enable better service delivery to citizens. In India, the main subject of this study, state governments in nearly all of the twenty major states had implemented policies to provide technology-enabled services to citizens by 2006.1 These “technology-enabled service center” policies deliver public and private services2 to individuals through the use of information technologies, in particular computers and the Internet, at dedicated local centers. These centers provide a “one-stop shop” environment for multiple government departments, thus streamlining the process by which citizens access an array of services.3 Yet the outcomes of efforts in India, and thus the benefits new technologies provide to citizens, differ tremendously across the states, in terms of both the extent and character of services provided. In the low-income states of Chhattisgarh and Orissa, the Chhattisgarh government provides nearly forty services, while Orissa offers fewer than ten. The types of services also vary, with states such as Andhra Pradesh providing a wide range of high-demand services, while others, including Uttarakhand, provide only a few low-demand services. I argue that the observed variation in technology adoption, and the related benefits derived by citizens, result from the strategic calculations of political elites. Drawing on extensive sub-national analysis of technology policies in the Indian 1 The constitutional allocation of responsibilities in the Indian federalist system makes sub-national state governments responsible for the provision of a majority of government services to citizens, rather than the national government. As a result, states are institutionally the most appropriate level for implementing service centers. 2 Service centers often provide a mix of government services and services from the private sector, such as Internet access, digital photography, or telemedicine services. 3 While there may be related policies regarding service delivery via mobile centers or mobile phones, the policies studied here all involve service delivery at an immovable center through the use of information and communication technologies (ICTs).

174

states, I posit that politicians attempt to utilize technology initiatives to maximize their chances of retaining power, but that these policies have both potential benefits and costs. Politicians evaluate the tradeoffs between expected effects on their political interests, especially their potential for reelection, of implementing new policies. When the balance of politicians’ calculations differs across states, we observe variation in the policies these states adopt, and in the associated benefits for citizens. This paper looks specifically at the ways in which electoral politics affect the services offered in Indian state service center initiatives. I argue that electoral incentives, as shaped by the combined effects of preexisting institutions of corruption and party competition, affect the number and type of services made available to citizens. I then test this argument using data on services in the sixteen Indian states4 that implemented technology-enabled citizen service center policies during the period from 1999 to 2006.5 The analysis is based on a new dataset of state service center policies that I developed during fieldwork in sixteen states, in addition to supplementary data collection. I utilize both quantitative and qualitative analytic techniques to evaluate the factors associated with variations in technology-enabled service provision across the states. After considering the details of my argument, I address trends in the number and type of services6 made available. II. ELECTORAL INCENTIVES AND TECHNOLOGY POLICY I argue that politicians will implement technology-enabled service centers, and specific services within those centers, when the expected electoral benefits from doing so outweigh the electoral costs. The benefits to politicians are not likely to differ dramatically across states. Politicians and parties can amplify the electoral benefits of technology policies by targeting services to their most important constituencies, yet because every incumbent should be able to benefit from such a strategy, the overall size of the electoral benefit of services reform should be similar for political incumbents across states. The costs to politicians from service centers come primarily in the form of foregone corrupt income. In the preexisting process of service delivery, politicians use the machinery of the state to extract rents [14]. Yet in the Indian context they do so not simply for personal pecuniary gain but rather to enhance their hold on political power. In a cycle of 4

I consider only those states in which the state government implemented a service center policy, thereby excluding the Northeast states and Jammu & Kashmir, where the central government initiated a service center initiative, and the four remaining states in which no service center policy was implemented (Bihar, Goa, Jharkhand, and Madhya Pradesh). 5 In some states more than one service center initiative was implemented during the period under consideration, for example one in urban areas and a second in rural areas. In these cases I consider the overall services offered across all projects, omitting any overlapping services. 6 Evaluation of services requires an analysis of the services that are stated to be available versus the services that are actually available in the centers. While this is difficult to assess without visiting thousands of individual centers, a reasonable measure of available services can be made from a combination of site visits, interviews with project representatives, and reviews of associated websites.

funds noted first in the literature by Wade [14], citizens pay bribes to bureaucrats, who then channel some portion of these funds to politicians, either in hopes of a job transfer or simply to satisfy a political boss’ demands (see also [15]). Politicians then use a portion of these funds to finance future elections. As a result, any threat to this flow of funds is a threat to politicians’ ability to secure reelection.7 I contend that the size of these costs depends primarily on variation in two key characteristics of the Indian states: the level of corruption and the cohesion of the ruling government. The level of corruption affects incumbent politicians’ dependence on bribes as a source of campaign finance—the greater the share of bribes in overall campaign finance and thus the more dependent politicians are on bribes, the less supportive they will be of policies to increase transparency in service delivery. There is substantial variation in the level of corruption across the Indian states, with fewer than 20% of citizens in states such as Kerala and Himachal Pradesh encountering demands for bribes when interacting with government officials, while more than 60% of citizens Bihar or Karnataka have direct experience with bribing in multiple government departments [16]. The cohesion of the ruling government, by which I mean whether a single party or a coalition of parties rules in the state, can also affect costs. Coalition versus single-party rule is relevant because of the role that supporting minority parties play in the stability of coalition governments. When a large party requires the support of smaller parties to form a government, the lead party often allocates ministerial posts to its coalition partners in order to reinforce their support. In most cases, the incentive for holding a ministerial post is not control over policy, but access to the associated departmental rents. This is because ministers can demand a portion of the bribes collected by bureaucrats in their department during service delivery. Such rents provide a concentrated economic and political benefit to ministers, who can use these funds to finance re-election campaigns. When politicians consider the potential costs of new technology policies, ministers who control departments delivering a high volume of services to citizens—for example, Transportation—may then expect major costs from more efficient service provision. From the perspective of a coalition member, a threat to this income is also a threat to the expected benefits of participating in the coalition. Moreover, minority partners in a coalition government typically reap only a limited, often disproportionately small, electoral benefit, as a party, from service center projects; credit instead tends to accrue to the larger majority party. Yet because minority partners are crucial to the survival of coalition governments, and because decision-making power is decentralized across ministers from all coalition parties, key supporting party ministers often have power to resist new technologies for delivering services. As a result, supporting party ministers have both potential 7 My argument also extends discussions of India as a “patronage democracy” (Chandra, 2007; 2004) by focusing on how patronage and other funds come from corrupt, rent-generating activities by bureaucrats and politicians.

175

incentives to resist technology policies and important decision-making power within the coalition that they can use to influence policy outcomes. In single-party ruled states, the concentrated costs for individual ministers are often outweighed by the overall electoral benefits of technology policy to the party. Singleparty governments also place decision-making authority with a small number of individuals, in particular, in the government’s Chief Minister. As leader of the party, the Chief Minister “internalizes” a large portion of the electoral benefit to the party of implementing service center projects. In single-party governments, the electoral benefits of implementing service center projects are concentrated, particularly for the Chief Minister, while the costs are in part diffused among Ministers and other individual Members of the Legislative Assembly (MLAs) who have limited capacity to resist the directives of the Chief Minister, due largely to party discipline. In contrast to single-party governments, then, in coalition governments the political costs of service center implementation are concentrated in key decision-makers with the capacity to resist—or alter the character of—project implementation, while the electoral benefits are, from the perspective of these key decision-makers, diffused outwards. This logic suggests that the type of ruling government— namely, whether single-party or coalition—should then affect the character of new technology policies. There is qualitative evidence for the importance of both corruption and ruling government cohesion in the choice of services to include in service center initiatives. In interviews with state level bureaucrats, representatives of states with low or average levels of corruption were more optimistic that the introduction of technology would be politically feasible. A bureaucrat from Orissa, a state with just below average corruption, when asked about resistance to service centers within government, noted that “There has been resistance, but we are slowly getting rid of it…People understand now that the story has gone past where they can stop it” [17]. Whereas an official from Haryana, a state with relatively high corruption, stated that, “many departments are not moving forward…with the implementation of front-to-back eGovernment services” [18]. He felt that this was largely due to the threat of increased transparency. For example, in “the Transport Department people in the department do not want to shift to a new system because they have established ways by which they are able to skim money off the top and they don’t want to lose these sources of income” [18]. Analysts have also noted the importance of coalition governments for service center outcomes in individual state cases. Kiran [19] comments that a service center project in the state of Kerala “was opposed by the participating departments on account of the fear that they would lose their existing authority and power. This was particularly evident in Kerala, which is ruled by a coalition government, with different political parties in charge of different departments.” Thus the policy-making incentives for party leaders come from the potential electoral benefits of implementing “good governance” technology policies and the contrasting threats

from decreased corrupt income. I now consider the effects of these political incentives on the character of technologyenabled services. III. CORRUPTION, COALITIONS, AND TECHNOLOGY-ENABLED SERVICE DELIVERY A key question for an analysis of service center policies is which citizens actually benefit from introduction of these technologies. Because state governments are responsible for providing services to their entire population, if they offer technology-enabled services that benefit only a portion of the population, then they are failing to offer the same benefits to other citizens. Fig. 1 shows the variation across Indian states in the number of services made available to citizens.

Fig. 1 – Available Services in the Indian States (2006) My argument offers specific predictions for the ways in which politicians’ calculations over the likely costs and benefits of technology policies may affect who benefits from these policies and to what degree. In cases where the expected costs from technology adoption are high, such as in states with high levels of corruption and in states ruled by a coalition government, governments are likely to implement policies in ways that serve fewer citizens overall. A. The Quantity of Services I evaluate the relationship between level of corruption and policy outcomes using a novel measure of state-level bureaucratic corruption, which is uniquely suited to testing my theoretical claims. This measure draws on a Transparency International survey of corruption in India [16]. The survey asked citizens about both their experience with corruption in acquiring services from government and their perception of corruption in government. Transparency International provides an indexed corruption score by state, based on eleven departments, including the police, municipal services, electricity, and the judiciary. I use this indexed score to develop my measure of bureaucratic corruption across the Indian states. The survey is particularly appropriate for the purposes of this analysis because it focuses explicitly on bureaucratic corruption in low-level service delivery, the area

176

targeted by service center policies, rather than the high-level corruption more often engaged in by top officials. In states with high levels of corruption, the large anticipated costs from more transparent service delivery should discourage service provision overall, leading to fewer services than in lower corruption states. The cohesion of the ruling government should also affect service selection, as coalition governments are likely to resist the inclusion of supporting-party controlled services, leading to fewer services overall. In this section, I discuss descriptive statistics, differenceof-means tests, and OLS models I use to evaluate these hypotheses. It is worth bearing in mind that the sample size is small: with only sixteen state cases, statistical power may be low. If I am able to uncover statistically significant relationships, it is because these relationships are particularly strong.8 Corruption First, the level of corruption in a state is a strong predictor of the number of government services offered in computerenabled service centers. The major findings of this analysis are presented in Table I. Among the states with below average bureaucratic corruption levels, the mean number of services offered is 20.1. This noticeably contrasts with the mean number of services offered in states with above average levels of corruption, which is 10.6. This difference is statistically significant at the .05 level. As noted above, any significant finding with only sixteen cases implies a particularly strong relationship between the variables under consideration. A univariate regression model offers similar findings. The relationship between quantity of services and corruption, when corruption is measured on a scale from zero to one, is not statistically significant, but is in the predicted direction. When corruption is measured dichotomously, the relationship is significant at the .01 level (t-statistic of -3.22) (See Model 1 in Table III below). Table I Indian States, Corruption, and Quantity of Services Corruption Below Average Above Average Absolute score: 240-478 Absolute score: 479-695 Scaled score: 0-.523 Scaled score: .524-1 Andhra Pradesh (41) Delhi (6) Chhattisgarh (37) Haryana (10) Gujarat (19) Karnataka (22) Himachal Pradesh (24) Rajasthan (11) Kerala (11) Tamil Nadu (10) Maharashtra (16) Uttarakhand (2) Punjab (16) Uttar Pradesh (13) Orissa (7) West Bengal (10) Mean services: 20.11 Mean services: 10.57 Difference in means: 9.54 (t-statistic = 2.07) 8

Statistical power, of course, is a function not just of the sample size but also the effect size.

Ruling Party Cohesion The next relationship to consider is between the type of government and the quantity of services. Are single party governments associated with higher numbers of services? Here, the evidence is not as clear-cut. On average, states ruled by single parties do offer more services than coalition states; with a mean of 18.0 services while coalition states have a mean of 11.4 services. The difference between these two means however is statistically significant only at the .1 level. Given the small number of cases and the fact that the services do trend in the direction predicted by my argument, however, it is reasonable to believe that there is a meaningful relationship in the data between the quantity of services and the type of government. In the case studies below I discuss additional evidence for a relationship between government cohesion and services outcomes. Corruption and Government Cohesion Perhaps the more relevant test of my argument is an analysis of the relationship between corruption and government cohesion. The effect of corruption on the number of services is most evident in low corruption, single-party states. These states provide, on average, a larger number of services than any other states, and this difference is statistically significant. The effects of ruling government cohesion and high levels of corruption are somewhat more difficult to parse, as the average number of service provided in high-corruption, singleparty states is not statistically different from that provided in coalition-ruled states. Coalition-governed states provide fewer services on average than single-party states, regardless of the level of corruption in the state. Thus, based on this data, it is only possible to say that either coalition states will provide fewer services regardless of the level of corruption, or that high corruption states will provide fewer services than low corruption states regardless of ruling government cohesion. In order to provide greater clarity on the role of ruling government cohesion in service selection, I evaluate the specific services offered by coalition governments in detail below.

177

Table II Mean Number of Services by Corruption and Ruling Government Cohesion Ruling Difference Government in Means, Cohesion Single Single Coalition Party Coalition Party (t-statistic) Corruption

Difference in Means, Low - High Corruption (t-statistic)

Low

27.4

11.3

High

10.2

11.5

17.2 (3.04)

-0.2 (-0.06)

16.1 (2.87) -1.3 (-0.43)

Alternative Explanations The length of time elapsed since a service center initiative was implemented might reasonably be expected to influence the number of available services. As governments in states that implemented centers early will have had more time to introduce additional services. The level of economic development might also be associated with provision of services, as states with higher incomes might have more funds available to invest in government reforms. In order to test these alternative explanations, I used multivariate regression models to evaluate the relationship of level of corruption, ruling government cohesion, time elapsed since implementation, and state economic development with the quantity of services available in the states. As shown in Table III, the level of corruption shows a clear and robust relationship with the quantity of services, even when taking into consideration these alternative explanations. However the length of time since centers were initiated and the level of economic development in a state show no relationship with the quantity of services. Ruling government cohesion also does not show a statistically significant relationship with quantity of services, but as noted above this may in part be due to the small number of coalition states in the dataset.

Table III Multivariate Analysis of Candidate Explanatory Variables Variable Model 1 Model 2 Model 3 Model 4 Constant 24.7 25.8 24.5 24.9 Above Average Corruption Ruling Party Cohesion Months Since Initiation State Domestic Product per Cap

(7.19) -14.0**

(6.85) -13.7**

(2.29) -13.3*

(2.30) -12.4

(-3.22)

(-3.12) -3.5

(-2.39) -3.4

(-2.17) -4.2

(-0.80)

(-0.75) .02

(-0.88) .1

(0.14)

(0.38) -9.0

(-0.86) N 16 16 16 16 r2adj .38 .37 .32 .30 Entries are unstandardized regression coefficients with t-ratios in parentheses. *p < .05 **p <.01 B. Selection of Services The benefits of computerized service centers to citizens depend not just on the quantity of services, but also on the character of the specific services made available. In this section I consider first whether high demand and high corruption services are made available in services centers across all states. I then consider the specific types of services made available in coalition-government led states. The government of India, and in particular state governments in India, provides hundreds of services to citizens on a regular basis. Table IV lists the fifteen most commonly provided services in computerized centers across the states. Out of the 73 government services offered by at least one service center initiative, only these services are offered by more than 25% of the projects (shown in Table IV).9 I also include one service, ration cards, that is offered in 24% of initiatives, but is interesting to include in the analysis because it is provided by a different department, Food and Civil Supplies, than any of the other services. Are these services in high demand by citizens? Based on Transparency International India’s survey of corruption in public service delivery [16], only a small number of those services most needed by citizens are provided in computerized service centers. Of the top twenty-five services required by citizens, thirteen are offered in at least one state. However only six of these are among the services offered in at least 25% of states (highlighted in bold in Table IV). Thus, in many cases there is a disconnect between the services that are provided by states and those that are in high demand by citizens. This disconnect is most obvious in states with above average levels of corruption. High corruption states provide, 9

Birth and death certificates, which are offered by 62% of initiatives, is the only service provided in more than 50% of states.

178

on average, 2.9 of these thirteen services, while below average corruption states provide 5.2, a difference of means that is statistically significant at the .05 level. When the number of high demand services is regressed on the level of corruption, the relationship is in the predicted direction and significant at the .05 level (t-statistic of -2.33). Table IV Most Common Services offered in Indian Computer-Enabled Service Centers Department Service Municipal Corp/Rural Dev

Food & Civil Supplies Revenue

Grievances Birth & Death certificates Caste/tribe certificates Property tax payment Ration cards

Electricity

Land records Income Certificates Residency certificates Electricity bill payment

Telephone Company

Telephone bill payment

Water Transport

Water bill payment Driving licenses Vehicle registration Home/Police Arms license Bold indicates that this service is one of the top 25 most demanded services by citizens, according to [16]. An alternative measure of citizen demand for service computerization is the estimated level of corruption in a particular service. Because bureaucrats are more likely to demand a bribe for some services than others, service reforms that have the potential to reduce corruption should provide greater benefits to citizens when introduced for services with a high likelihood of corruption. However reform of these services will also provide the highest costs to bureaucrats and politicians in terms of lost income, and so computerization in high-corruption potential services should be least likely in high corruption states. When we consider high corruption services, separate from their overall demand by citizens relative to all services, states with higher levels of corruption are less likely to provide these services through computerized service centers. As seen in Table V, below average corruption states are much more likely to provide services associated with bribe payments than states with above average corruption levels. Out of the six most corrupt services, only one, land records, is available in any of the above average corruption states.

Table V Availability of High Corruption Services in Indian ComputerEnabled Service Centers Reason for Est. % of < Average > Average Bribe Annual States Corruption Corruption Bribes Offering States States ($ ‘000) Police First 458,000 13% 2 0 Info. Report School Fees 340,000 0% 0 0 (exemption) Rural 313,000 0% 0 0 Financial Loan Register 307,000 50% 5 3 Property School 288,000 0% 0 0 Certificate Electricity 166,000 13% 2 0 Connection Estimated annual bribe value based on Transparency International India data [16] and author’s calculations. In the case of coalition-led states, I also expect to see effects of electoral considerations on the specific services that are chosen for inclusion in centers. Services controlled by supporting members of a coalition government should be less likely to be included than those services controlled by the majority party. Figure 2 provides an overview of service provision in coalition-led states. As shown in this graph, politicians from the smaller, supporting members of a coalition are more likely, on average, to acquire ministerial posts that involve control over services commonly offered in service centers. This is in line with expectations that supporting coalition members will demand ministerial appointments in departments with high levels of government-citizen interaction. However, when supporting coalition members control these services, it is less likely that they will be included in service centers, as seen from the right side of the graph. In the case studies below I consider the specific ways in which allocation of ministerial posts is linked to these lower levels of service provision.

179

Figure 2 – Average Service Provision in Coalition States – Lead vs. Supporting Parties Kerala I begin the case discussion with Kerala. This state is viewed by many as an innovator in both development in general [20] and ICT-enabled development in particular [21], [22]. Despite below average income per capita, Kerala has the highest literacy rate in the country and the highest overall human development index [23], [24]. Kerala also applied its development strategies to the use of technology and was one of the first states to use IT to reform service delivery (Kurian, 2003). Yet the services offered in Kerala’s two main service center initiatives, FRIENDS and Akshaya, pale in comparison to many of their counterparts in other states. Through FRIENDS centers, only ten government services are offered, with just three offered in Akshaya centers. Could coalition politics be at least partly to blame for the low quantity of services in Kerala? As noted above, analysts have pointed to the importance of coalition dynamics in shaping Kerala’s service center initiatives [19]. Once the first service center initiative was approved, the FRIENDS initiative, the ruling Left Democratic Front (LDF) continued to wield their influence over the shape of the centers. This coalition was led by the Communist Party of India (Marxist) (CPI(M)), with its coalition partner the Communist Party of India (CPI) holding the next largest number of seats in the state assembly during the period from 1996 to 2001. The coalition in total held 76 seats, five more than the necessary majority of 71. Ministerial posts were allocated across coalition members, with the largest number of ministries predictably allocated to the CPI(M). Six of the fourteen most common services were made available through the FRIENDS centers: electricity bill, telephone bill, water bill, property tax payment, driving licenses, and vehicle registration. Of these services, two of the relevant departments were overseen by supporting coalition members, the Water Department and Transport, led by the Revolutionary Socialist Party (RSP) and an Independent member of the legislature, respectively. Because the RSP held only five assembly seats and the independent, by default, held

only one, neither of the two ministers from these parties had the power to bring down the coalition government on its own: even defection by all five RSP ministers would leave the coalition with 71 seats, sufficient to maintain its majority. The remaining services offered through FRIENDS centers were delivered by departments overseen by CPI(M) ministers. Among the common services not offered through these centers, two key departments, Revenue and Food & Civil Supplies, were overseen by the CPI, the only party with enough assembly seats to threaten the stability of the ruling coalition. The other relevant departments, Local Administration and Rural Development, were both led by representatives of the CPI(M). The lack of services offered by departments under the CPI(M)’s most important coalition partner aligns clearly with predicted outcomes. The Akshaya project, a second computer-enabled center initiative in Kerala, offers a similar story of coalition politics, with an interesting twist. The Akshaya initiative was launched under the United Democratic Front, a Congress Party-led coalition that came to power in 2001. In this coalition the Congress held the largest number of seats, at 62, but needed the support of one or the other of its two main coalition partners, the Muslim League of Kerala or the Kerala Congress (M), to reach the magic number of 71 for a majority. Akshaya was launched in 2003 through a pilot project in the state's Malappurum district. The stated goal of the initiative was to increase access to technology for the citizens of Kerala, and a key part of the initiative was computer literacy, or “e-literacy,” for the head of every household, a goal that mimicked Kerala’s well-publicized success in boosting traditional literacy rates (www.akshaya.net).10 But these centers provided an additional opportunity for the delivery of government services, as noted by those affiliated with the project [25] Overall, however, the Akshaya project has implemented next to no services through its network of what in 2008 is more than 1,000 rural centers. Only three common government services are on offer: electricity bill payment, telephone bill payment, and grievance redressal. The Electricity Department is overseen by a Congress minister, and the Department of Rural Development, which would oversee grievances in the rural areas where Akshaya centers are located, is headed by a representative of the Kerala Congress (M). While it is surprising that the KC(M) would have one of their services offered through the Akshaya centers, it is perhaps more telling to note the number of services overseen by the Kerala Congress (M) that are not offered through these centers. During the UDF government, the KC(M) was in control of both the Revenue Department and Rural Development, which together account for six of the remaining commonly offered services, none of which have been provided through Akshaya. Other coalition parties controlled the departments overseeing nearly all other common services, 10

At nearly 91%, as measured in the 2001 Census, Kerala has the highest literacy rate of any Indian state or Union Territory.

180

with the Kerala Congress (Jacob) and the Kerala Congress (B) overseeing Water and Transport, respectively. Congress ministers oversaw only two common services that were not offered via Akshaya, arms licenses and ration cards. For a portion of these services, the lack of provision through Akshaya is even more surprising, as these services had already been implemented at the district level in FRIENDS centers. Thus even though the basic infrastructure was in place to extend water bill payments, property tax payments, driving licenses, and vehicle registrations through Akshaya outlets, this was not done under the Congress administration. This means that citizens must either go to the district FRIENDS office or avail of these services through traditional department-based offices, which are still be prone to any corruption in the system. The twist in the Akshaya case is the way in which the policy itself was used as a tool to target the core voters of a key coalition member party. The Malappurum district where the pilot project was initiated has a largely Muslim population, and the Muslim League is the dominant party in the region [26]. When the UDF coalition regained power in 2001, a prominent member of the Muslim League, Shri. P. K. Kunhalikutty, became Minister for the Information Technology Department. According to analyst reports, the Minister subsequently consulted with representatives of Malappurum and determined that information technology could be a “real enabler of the local economy” [26]. In order to facilitate IT use, a plan was developed by the local government to provide for computer literacy in the district. This plan was subsequently modified in coordination with the Kerala State IT Mission, under the oversight of the IT Minister, and realized as a “telecenter” initiative for the district [26]. If successful, this pilot would then be extended to other districts in the state. So while the stated intention was to develop an initiative that could eventually benefit the entire state, the initial benefits would clearly be centered on the core constituency of the standing IT Minister, representing a key ruling government coalition member. Akshaya, then, provides an important example of the combined effects of coalition rule. First, a second computer center initiative in the state was launched for the apparent purpose of rewarding key Muslim League constituents, a clear boon for the holder of the IT Ministry post. Second, the lack of services offered through Akshaya reflects the prevalent use of ministerial posts by the UDF government to reward coalition members. Even those services that should be relatively easy to implement because of their inclusion in the earlier FRIENDS initiative have, for the most part, not been made available to Akshaya patrons. Karnataka The next state, Karnataka, has a much higher level of corruption than Kerala, which would, on its own, imply a stronger emphasis among coalition members on the economic side-benefits of ministerial posts. But Karnataka is also a “hard” case, because it is one of the most technically advanced

states in the country, with the IT hub of Bangalore as its capital. On the face of things, Karnataka has also been at the forefront of using ICTs in service delivery. The Bhoomi land records initiative received more media attention than most initiatives in the country, and the former Secretary for eGovernance, Rajiv Chawla, is widely recognized as a father of the Indian movement to incorporate information technologies into Indian government processes. Yet, as we will see, even the presence of early initiatives and an important evangelist do not guarantee de-politicized service delivery in the state. A major service center initiative, Bangalore One, was initiated under the coalition government of the Congress Party and Janata Dal (Secular) (JD(S)) in 2005. This particular government was formed despite the fact that the BJP held the largest number of seats in the assembly, because the BJP was short of a majority and could not agree to a coalition with other parties to acquire a majority [27]. So while the Congress has the greater number of seats in the coalition, it is highly dependent on the support of its JD(S) partner.11 Bangalore One was launched in partnership with the municipal government in Bangalore and was intended to provide a similar one-stop government services environment as offered by Andhra Pradesh’s eSeva initiative [28]. In the case of Bangalore One, however, despite multiple years of experience with technology-enabled service delivery, many of the most commonly offered services are not available. Of the three services that fall under the domain of the Revenue department—residency certificates, income certificates, and land records—none are offered in the Bangalore centers. The exclusion of land records is especially surprising, given that the Bhoomi land records initiative was viewed as such a ‘success’ in the state. Why are these common and highly valued services not offered in Karnataka? The most plausible answer again lies in coalition politics. The minister in charge of Revenue was a politician from the supporting JD(S), who also held the post of Deputy Chief Minister, and so was clearly an important player in maintaining the coalition. Ration cards are also unavailable in the centers, and it was a minister from the JD(S) who oversaw the responsible Food & Civil Supplies department. Of those common services that were offered, the JD(S) was responsible for only one, electricity bill payment, while the Congress was responsible for the remaining services, via Municipal Administration, the department that was also involved in the implementation of the initiative itself. So even in a state with a strong emphasis on technology and history of government IT initiatives, the dynamics of corruption and coalition politics seem closely linked to the selection of specific services made available in the state’s one-stop centers. Rajasthan 11 The coalition eventually collapsed in early 2006, when a rebel JD(S) leader, H.D. Kumaraswamy, pulled out of the government with a group of other JD(S) MLAs in order to form an alternative coalition with the BJP (Rediff, 2006).

181

In Rajasthan, the LokMitra (urban) and JanMitra (rural) centers were opened in early 2002 under the leadership of the Indian National Congress party. During this period the Congress party held a majority of the seats in the state assembly and so did not require the support of any other parties in a coalition. The initiation of service centers was largely an enterprise of the Chief Minister Ashok Gehlot, who sent representatives of the state to Andhra Pradesh in 2001 to evaluate the eSeva initiative there and determine how something similar could be used to improve service offerings to citizens in Rajasthan [29]. According to one of the bureaucrats who participated in that mission, a decision was made to improve on the Andhra model by offering services in both urban and rural areas, through two separate initiatives, at least at the beginning.12 The choice of services in each set of centers was then determined based more on urban versus rural needs, with the LokMitra centers focused on bill payments and JanMitra centers emphasizing grievances and non-government services such as agricultural prices. However, in rural areas only four government services were offered, while in urban areas just seven services were made available. The availability of specific services, particularly in urban areas, seems directly linked to the interests of the Chief Minister. When bureaucrats in charge of implementing LokMitra services attempted to convince various departments to provide their services in the centers, they faced significant resistance. The state telephone company did not want to allow outsourcing of its bill payments through the computerized centers, and the bureaucratic officers found it necessary to request the intervention of the Chief Minister. In this case the Chief Minister did intervene and the Telephone company was forced to provide services for bill payment through the LokMitra centers. On other occasions, however, such as in the case of income tax payments, the Chief Minister was unwilling to take similar initiative and income tax payments were not included in the services offered by the centers [29]. The mixed response of the Chief Minister in the case of Rajasthan may be closely tied to the high level of corruption in public service delivery in the state. As one bureaucrat involved in the initiatives noted, politicians “want to provide the maximum services to their vote bank,…[but] they also want money for the next election” [29]. Thus from the perspective of the Chief Minister, by providing some services, especially those that do not typically involve high levels of corruption, such as bill payment, there is an opportunity to reap some political benefits from the centers. At the same time, those services that may more often involve the payment of bribes, such as income tax, can be restricted from inclusion in the centers in order to maintain the rents from those services. In this way we can see how high levels of corruption may affect service delivery in single party states. Chhattisgarh The final case to consider is Chhattisgarh, a single-party-led state that has below average levels of corruption in public

service delivery. The state of Chhattisgarh was formed in the year 2000 when it was carved out of Madhya Pradesh. Despite below average levels of economic development and a persistent and violent separatist movement in part of the state, the Chhattisgarh government has implemented a substantial eGovernment program through its Choice service centers. These centers, which to date are largely located in urban centers, provide one of the largest sets of services of any state in India. Why is it that Chhattisgarh has implemented such a robust service center initiative? Here again we see the important role played by Chief Ministers in the character of public service reforms. Choice service centers were initially implemented under the first Chief Minister of the state, Ajit Jogi of the Indian National Congress. However the growth of the project from a pilot stage has occurred under BJP Chief Minister Raman Singh. According to government officials associated with the Choice initiative, Chief Minister Singh encouraged an inclusive decision-making process in which the bureaucrats in charge of major citizen-facing departments would participate in an “empowered committee” that was tasked with making all key decisions about the project moving forward. The goal of this model was to ensure consensus on decisions so that no participants would later attempt to block implementations of the centers or any specific services [30]. In addition, the Chief Minister took over leadership a “governing council” made up of government ministers, which is responsible for overseeing the implementation of the project. This meant that the implementation team could go directly to the Chief Minister in case of any problems with implementation. In no other state, other than perhaps Andhra Pradesh, has the Chief Minister taken on this type of direct oversight role in the development and implementation of service centers. What factors might have influenced the Chief Minister to take such a strong position on eGovernment in the state? Corruption does exist in Chhattisgarh, but government representatives argued that the expectation of the Chief Minister was that improvements in service delivery would provide a greater potential electoral boost to the ruling government that any threat from reduced access to rents. “The Chief Minister sees the benefits as greater than the costs. He has been the brainchild behind all of these frameworks. Indian politicians…have very sharp political minds. The Chief Minister in this term is focused on good governance, on accountability, transparency, and responsiveness of the government” [31]. Given the difficulties that the government otherwise faces in areas of development and social stability, it seems that the Chief Minister has adopted service reforms through the use of technology as an important platform for delivering valued goods to citizens. This is possible, at least in part, because he does not likely feel the same threat to campaign resources from increased government transparency that exists in higher corruption states. IV. CONCLUSION

12

JanMitra and LokMitra were merged into a single initiative, eMitra, in 2004. Both Jan Mitra and Lok Mitra are Hindi variations on the term “people’s friend”.

The evidence presented here provides strong support for an argument about the role of corruption and coalition dynamics

182

in affecting which citizens benefit from technology-enabled service delivery. Citizens in more corrupt states are less likely to benefit from new technologies. This is because even if service centers are implemented in their state, these centers are likely to provide fewer and lower demand services than those centers in other states. Citizens may also be at a disadvantage depending on the characteristics of their ruling government. In coalition-led states, state governments will chose services that ensure the stability of the government before considering what services might benefit voters. When MLAs from supporting parties serve as ministers for departments, the services of those departments are less likely to be implemented, particularly if that party holds enough seats to disrupt the majority of the coalition. Who benefits from technology-enabled service centers is thus highly dependent on the political characteristics of a state and in particular the extent of corruption and cohesion in the ruling government. While there may be great potential to improve service delivery through one-stop service centers it is clear that, at least to date, the actual benefits to citizens are politically driven and thus, in the case of the Indian states, highly varied. REFERENCES [1] [2] [3] [4] [5] [6] [7]

[8]

[9] [10] [11] [12] [13] [14] [15] [16] [17]

World Bank, World Development Report: Making Services Work for the Poor. Washington, D.C.: The World Bank and Oxford University Press, 2004. World Bank, ICTs and MDGs: A World Bank Group Perspective. Washington, DC: The World Bank Group, 2003. Accenture, Markle Foundation, UNDP, Creating a Development Dynamic: Final Report of the Digital Opportunity Initiative. 2001. R. M. Davison, C. Wagner, and L. C. K. Ma, “From government to egovernment: a transition model,” Information Technology and People, Vol. 18(3), pp. 280-299, 2005. United Nations, World Public Sector Report: E-Government at the Crossroads. New York: United Nations Publications, 2003. D. West, Global E-Government 2005. 2005. Available: http://www.insidepolitics.org/egovt05int.pdf J. Bussell, “Electoral Competition and Digital Development in India and South Africa,” Proceedings of the IEEE/ACM International Conference on Information and Communication Technologies for Development 2007 (ICTD2007), 2007. K. De Tolly, B. Maumbe, and H. Alexander. “Rethinking E-Government for Development: Issues, Lessons and Future Prospects for the Cape Gateway Portal in South Africa,” presented at the IST-Africa Conference, Pretoria, South Africa, May, 2006. S. Madon, “Evaluating the Developmental Impact of E-Governance Initiatives: An Exploratory Framework,” The Electronic Journal on Information Systems in Developing Countries, 20(5), pp. 1-13, 2004. J. Satyanarayana, EGovernment…the science of the possible. New Delhi: Prentice-Hall, 2004. R. Heeks, “Causes of eGovernment Success and Failure: Factor Model,” 2003. Available: http://www.egov4dev.org/causefactor.htm S. Bhatnagar, “Lessons from eGovernment in Developing Countries” Regional Development Dialogue, No. 24, 2002. Government of West Bengal, IT Department official, January 18, 2008. R. Wade, “The Market for Public Office: Why the Indian State is not Better at Development,” World Development, Vol. 13(4), pp. 467-497, 1985. F. De Zwart, The Bureaucratic Merry-go-round: Manipulating the Transfer of Indian Civil Servants. Amsterdam: Amsterdam University Press, 1994. Transparency International India, India Corruption Study 2005. New Delhi: Transparency International India, 2005. Government of Orissa, IT department official, January 29, 2008.

[18] Government of Haryana, IT Department official, February 4, 2008. [19] G.R. Kiran, “Front-End First: Citizen Payment at FRIENDS Centres in Kerala,” UNPAN eGovernment for Development Success/Failure Case Studies, Number 17, 2002. Available: http://www.egov4dev.org/friends.htm [20] P. Heller, The Labor of Development. Ithaca: Cornell University Press, 1999. [21] S. Madon, “Governance lessons from the experience of telecentres in Kerala,” European Journal of Information Systems, Vol. 14, pp. 401416, 2005. [22] S. Madon, and G.R. Kiran. Information technology for citizengovernment interface: a study of FRIENDS project in Kerala. World Bank Global Knowledge Sharing Program (GKSP), 2002. [23] Reserve Bank of India; GOI, 2001 [24] Reserve Bank of India; GOI, 2005. [25] Kerala Government, IT Department official, June 22, 2004. [26] G.R. Kiran, “Akshaya, Malappuram, Kerala” in Information and Communication Technologies for Development: A Comparative Analysis of Impacts and Costs from India,” Bangalore: International Institute of Information Technology, 2005. [27] Rediff.com, “JD-S, Congress Team up in Karnataka,” May 16, 2004. Available: http://ia.rediff.com/election/2004/may/16karna.htm [28] Karnataka Government, eGovernance Department official, February 22, 2006. [29] Government of Rajasthan, former IT Department official, May 7, 2007. [30] Government of Chhattisgarh, IT agency official, January 23, 2009. [31] Government of Chhattisgarh, IT Department official, January 23, 2009.

183

Results from a Study of Impact of Egovernment Projects in India Subhash C. Bhatnagar and Nupur Singh

Abstract—The paper presents the results from an assessment study of eight e-government projects from India. The assessment framework measured the total value delivered by a project to various stakeholders on three dimensions: (a) cost to clients for accessing services; (b) perception of quality of service and governance; and (c) agency cost and revenue. Data was collected from randomly selected users encapsulating their experience of using the computerized and manual systems. Impact was estimated as the difference between the rating of the computerized and the manual systems. Citizens indicated an overwhelming preference for computerized service delivery. The number of trips required to be made to the concerned offices reduced significantly and waiting time came down by nearly fifty percent. Overall impact showed wide variations across projects. Of the seven agencies where corruption was reported in the manual system, five services saw significant reduction through computerization but the impact was marginal in three projects. Government of India has adopted the framework used in this study to assess the impact of 40 projects implemented at the national, state and local levels. Paper discusses the implications of the results for further development of e-government projects in India and identifies the limitations of the study. Index Terms—E-government, governance, impact assessment, India

are making ICT investments in the public sector. On the other hand, evidence of failed projects has drawn attention to the level of risk involved in implementation. A failure rate of more than 50 percent is widely cited in this context [1]. A study commissioned by DFID [2] researched firm-level impact of ICT in developing countries, 2 but similar studies have not been undertaken for the public sector. A report by the United Nations (2003) laments the fact that documented research on the social or economic impact of e-government development is virtually non-existent [3]. Multilateral organizations such as the World Bank fund nearly seven billion dollars of ICT investment as part of their lending programs and grant assistance to various client countries. A World Bank report noted that the largest yet the least monitored investments are IT components of projects in different sectors, highlighting the relevance of systematic assessment of the impact of these applications [4]. This paper presents an early effort at evolving a systematic framework and methodology for assessing the impact of egovernment projects based on a review of past efforts at assessment of e-government projects. Sections II and III discuss the review and present the key features of the framework. The paper goes on to report the findings from a study that used the framework to ascertain the impact of a selection of eight e-government projects from India. II. EFFORTS TO ASSESS PUBLIC SECTOR IT PROJECTS

I

I. BACKGROUND

N recent years a large number of countries have launched “e-government” programs, and several development agencies and governments have identified e-government implementation as a key policy priority. 1 Driven by the success of a few projects in improving delivery of services to citizens and businesses, an increasing number of governments Manuscript received February 5, 2009. S. C. Bhatnagar is with the Indian Institute of Management, Ahmedabad, India (phone: 91-79-6632-4802; fax: 91-79-6632-6896; e-mail: [email protected]). N. Singh is with the Centre for E-Governance, Indian Institute of Management, Ahmedabad, India (e-mail: [email protected]). 1 Examples of e-government initiatives launched by international agencies can be found on the following links: World Bank (http://www.worldbank.org/egov); ITU (http://itu.int/itudoc/itu-t/workshop/egov); and DFID (http://www.publictechnology.net/modules.php? op=modload&name=News&file=article&sid=1533)

While significant amount of academic and policy research has focused on evaluation of public sector ICT projects, a systematic framework for evaluation has not emerged. One group of studies provides macro-level estimations of egovernment activity using appraisal indices focusing on supply-side, quantifiable measures such as web presence of government, network coverage, institutional and regulatory support and human capital provision. 3 Such factor-based assessments of e-readiness do not have immediately obvious and tractable policy implications. They tend to focus almost 2 The study sponsored by DFID at the London Business Schools collected data from firms in India and Brazil to establish a relationship between ICT investments, profitability, growth and productivity at the firm level. The degree of collateral organizational change was studied as a mediating variable. 3 Examples of such frameworks are (1) UNPAN (2004) E-Government Readiness Report www.unpan.org and (2) Brown University (2004). Global E-Government Report, available on www.insidepolitics.org/egovt05int.pdf

184

exclusively on measurement of physical access to certain types of ICT without incorporating issues such as affordability, appropriateness, ICT capacity and training, and the regulatory and macroeconomic environment [5]. The second group of studies, largely anecdotal, done in a piecemeal fashion, provide project-level evaluations with little prospect for synthesis from past approaches. Evaluative studies have been done to serve a variety of purposes. Some studies looked at implementation success in terms of whether the systems were functioning as they were designed to, or the degree to which the intended outcomes were achieved. Some studies looked at long term sustainability and scope for replication, 4 while some measured the benefits that were delivered to agencies. 5 A few focused on benefits to the clients. 6 There was hardly any comprehensive study that assessed the impact on all the stakeholders and covered both short-term and long-term direct and indirect impacts. A few studies had carried out a cost-benefit analysis. Often, evaluation studies had been done by agencies that were likely to have an interest in showing a positive outcome. A variety of approaches have been used for evaluation. These included surveys, expert opinion, ethnographic studies and internal assessments produced by lending agencies. The utility of such evaluations has been limited because: ▪

Different studies of the same project showed very different outcomes, thus indicating a lack of credibility of results. 7 Part of the reason for different outcomes was the use of very small samples and a lack of rigor in sampling and collecting data from clients of the systems. The results could therefore not be easily generalized over the entire population of clients.

▪

The studies evaluated the functioning of the computerized system but were not able to assess the difference made by ICT use, as the need for counterfactuals (evaluation of systems as they worked before computerization) was ignored. Often, the impact of ICT use was not separated from other interventions that were made simultaneously with the computerization effort.

▪

Finally, since different studies did not use a standard methodology, it was difficult to compare the outcome of one project with that of others.

Work done in EU countries has focused on understanding the ‘processes’ that deliver a return on investment and generate value [6]. Common best practice factors for value 4 Government of India, Ministry of IT has commissioned quick assessment of 29 projects. 5 Korea’s eProcurement agency has evaluated the impact on different government agencies using the system. 6 Global Knowledge Sharing Program got 4 Indian projects evaluated where clients were surveyed. 7 For example the Bhoomi project of issuing copies of land title has been evaluated by Public Affairs council reporting a significant positive outcomes including reduction in bribes. Recent studies by a team from MIT and IIIT Bangalore found that corruption had not declined and major benefits were derived by land sharks.

creation in e-government projects have been documented in reports by governments and external consultants [6] – [8]. An example is the study conducted jointly by the Danish Ministry of Finance and Accenture which identifies two primary dimensions of value: (i) value to investors in terms of tangible financial benefits, cost savings, cost avoidance, and increased revenue; and (ii) value to users in terms of improved services, reduced cost and/or time savings to citizens, and reduced administrative burden to businesses. Investment in government employees and technological infrastructure are identified as key enablers for creating value through egovernment. The focus has been on using business case methodologies from the private sector to demonstrate the economic value of e-government projects. The calculation of time and money spent in finding and using public information are the most direct and measurable benefit of e-government applications to users, as shown in recent OECD and EU studies [9],[10]. These reports identify improved revenue collection, lower costs due to efficient processing of transactions, and a reduction of administrative burdens due to simplification or elimination of procedures as some direct impacts that can be measured. Financial savings to users in terms of time and money spent in finding and using public information are the most direct and measurable benefits of egovernment applications for clients. A significant amount of work has also been done on studying users’ perceptions of quality in terms of attributes such as accessibility, attitude of staff, cost of service, provision of information, procedural fairness and convenience. Calculation of net economic benefit of an IT project has proven to be challenging as outcomes are multi-dimensional and composed of both quantitative and qualitative indicators. Further attention is required to be given to the linkages amongst issues of quality of service, governance, wider impacts on society, and ICT investment. Newer frameworks are evolving around a notion that an exclusive focus on financial costs and benefits for the government ignores many important non-economic benefits [9]. One such benefit relates to improvements in various aspects of governance activity. Another non-economic benefit relates to the addressing of wider policy priorities, which, in the context of developing countries could constitute the priorities as articulated in the UNDP Millennium Development Goals. III. MEASUREMENT FRAMEWORK AND METHODOLOGY USED IN THE STUDY Two approaches - MAREVA developed by the ADAE (Agence pour le Development de l’Administration Electronique - Electronic Administration Development Agency) in France with the help of Bearing Point and the WiBe Economic Efficiency Assessment methodology being used by the German federal administration [11] were useful in developing the framework proposed in this chapter. These methodologies developed by two Governments in EU countries focus broadly on the same dimensions. They offer two levels of impact assessment: first, in terms of how the project provides a business case justification of expenditure

185

and second, in terms of how the project meets the goals of the agency concerned and, in turn, how this helps in meeting wider government strategies. Guiding these assessment frameworks are strategic outcomes pursued by respective governments representing broad policy priorities that drive the direction of government. These considerations have been incorporated in the assessment framework proposed for egovernment projects in this section. 8 Borrowing from the above approaches and the work done in EU countries, the framework used for the study reported in the paper focused on the idea of measuring the total value delivered by a project to different stakeholders and takes a balanced approach between case study and quantitative analysis. It recognizes that some part of the value for each stakeholder can be monetized and other part needs to be assessed qualitatively. Most of the assessment models discussed in the literature have not been used in the context of a developing country and many have not yet been applied in practice even in developed countries. Nor do they account for a variety of delivery models used in developing countries such as common service centers and franchised outlets that can retail e-services offered by the government. Practical issues of paucity of data have not been taken into account, particularly in case of a developing country where baseline surveys are not done and monitoring and evaluation systems are weak. Recognizing that adequate data for quantitative assessment is not available for most of the projects, collection of such data using a standard measurement framework for sufficiently large number of projects is the first task that needs to be undertaken. Since impact assessment is the key objective, establishing counterfactuals is an important element of the proposed measurement framework. A number of empirical studies suggest that ICT has had an impact in improving the performance of private sector organizations particularly in developed countries. However, regarding ICT investments by the public sector in developing countries, many researchers have noted that past evaluation studies have not used a common framework or methodology and that rates of success/failure have been declared based on purposive samples [12]. Since ICT is introduced mostly in the context of governance reform to improve efficiency, effectiveness, and transparency of governments, a crucial first stage is to ascertain to what extent these intended outcomes from e-government applications have been achieved. A common measurement framework evolved on the basis 8

The indicative items are based on a review of the following documents: i) Performance Reference Model of the US Federal enterprise Architecture framework used by the office of Management and Budgets in US Federal Government; ii) European Commission, eGovernment Economics Project (eGEP), Measurement Framework Interim Version Deliverable (D.2.2), 2005; iii) TP Rama Rao, V Venkata Rao; SC Bhatnagar; J Satyanarayana, ‘eGovernance Assessment Frameworks’, 2004, http://egov.mit.gov.in; iv) Edwin Lau, ‘Electronic Government and the drive for growth and equity’, OECD, 2005; and v) Lanvin Bruno, ‘METER E-strategies Monitoring and Evaluation Toolkit’, in Robert Schware (ed), E-Development: From Excitement to Effectiveness, Prepared for the World Summit on the Information Society Tunis, 2005, World Bank.

of existing frameworks has been used in this study. An eservice delivery project impacts three groups of stakeholders: (i) clients receiving the service; (ii) agency (including implementation partners) that delivers the service; and (iii) the larger society consisting of citizens, businesses, government as a whole and civil society. The impact can be assessed in terms of a variety of outcomes experienced by each type of stakeholder. Table I lists key dimensions of outcomes for each type of stakeholder. TABLE I KEY OUTCOME DIMENSIONS Stakeholders

Key Dimension of Impact

Client

▪ Economic (Direct and indirect) ▪ Governance (Corruption, accountability, transparency, participation) ▪ Quality of service (Decency, fairness, convenience, etc.)

Agency (Including partners in implementation)

▪ Economic (Direct and indirect) ▪ Governance (Corruption, accountability, transparency, participation) ▪ Performance on key non-economic objectives ▪ Process improvements

Society Government as a whole

▪ Long term impact on Millennium Development Goals ▪ Image of the government

The primary objective of the study was to measure the impact of computerization on clients (users) of selected service delivery projects and to test the applicability of the framework across a variety of projects. For the purpose of this study a sample of eight mature projects shown in Table II was selected. These projects covered services to rural as well as urban citizens (G2C), services to businesses (G2B) and services for internal government users (G2G). Most of the services are offered by state-level agencies except in eSeva where services from the federal government are also offered. In eProcurement services are accessed through a portal. In the remaining projects service delivery is through assisted computerized counters set up by the agencies at various locations. For each project, the measurement framework was converted into a set of data collection instruments including: i) a profile of the project identifying services, clients and other stakeholders; ii) agency level data on activity levels, investments and operating costs; iii) a client survey questionnaire covering direct cost of access, quality of service and governance, and a few measures of overall satisfaction; and iv) an employee survey for perceived impact on work, efficiency and effectiveness. The survey assessed both the manual system and the computerized system that replaced it on all the above dimensions. An analysis of the differences between the old and the new system provided a measure of impact. Random samples of about 30 users were chosen from 8 different locations in a state. Locations were chosen to represent different levels of activity at the service centers, and different levels of development of the region. Thus nearly 250 respondents were surveyed for each project. The questionnaires (translated into local languages) were

186

administered at the home of the sampled users by investigators who were trained to understand the nature of projects and to interpret individual items in the instrument. The survey was combined with secondary documentation and primary data gathered through field visits, interviews, surveys and opinions solicited through e-mail. TABLE II LIST OF EIGHT PROJECTS FOR DETAILED STUDY Project Bhoomi

Description Online issue of an RTC (among other uses, this document is required for availing crop loans from banks and as surety for bail) and filing of requests for mutation for affecting changes in land record in Karnataka at 203 Kiosks. Launched in February 2001.

Karnataka Valuation and E-Registration (KAVERI)

Key services delivered by 201 Sub Registrar’s Offices in Karnataka are: online registration of property sale/purchase deeds; issue of nonencumbrance certificate; and issue of copies of registered deeds. Launched in December 2003.

Khajane

Networking and computerization of all treasuries across Karnataka. Launched in November 2002.

Computer Aided Administration of Registration Department (CARD)

Online registration of property sale/purchase deeds; issue of non-encumbrance certificate and issue of copies of previously registered deeds at 387 Sub Registrar’s Offices in Andhra Pradesh. Launched in November 1998.

eProcurement

Online tendering for goods and services by government departments and agencies in Andhra Pradesh. Launched in January 2003.

eSeva

One stop service centers delivering 135 services from central, state, local governments and public utilities. Used monthly by 3.1 million citizens at 275 locations across 190 towns. Launched in August 2003.

Ahmedabad Municipal Corporation Civic Centers (AMC)

16 Civic Centers of AMC primarily deliver three services: annual collection of property tax; issue of birth and death certificates; and issue of shop licences. Launched in September 2002.

Computerized Interstate Check Posts

10 computerized check posts use electronic weigh bridges to levy fines for overloading and overdimensioning of commercial vehicles passing through them, inspect vehicles to check for damaged headlights and non-standard license plates, and verify essential documents. Launched in March 2000.

IV. RESULTS OF CLIENT IMPACT ASSESSMENT This section presents an analysis of the eight projects from the perspective of the clients. Impacts for the key dimensions of cost, quality, governance and overall satisfaction are analyzed. A. Cost to Users In all the eight projects, e-government has reduced the number of trips users have to make to complete all transactions for a service (Fig. 1). Greater formalization of processes after process reform, quicker retrieval of data from computerized databases, and automation of tasks such as writing and copying documents, as compared with manual methods, has reduced the total processing time of a service

request in all applications. 9 This has resulted in increased predictability in service delivery and reduced the number of trips. 4.00 Manual Computerised 3.37

3.50

3.00 2.50 2.00

2.81

2.71

2.74 2.42

2.32

2.20

2.12

2.09

1.85 1.62 1.43

1.50

1.54

1.22

1.41 1.13

1.00 0.50 0.00 Bhoomi

KAVERI

Khajane DDO

Khajane Payee

CARD

eProcurement

eSeva

AMC

Saving

0.47

1.20

1.08

0.90

1.38

0.86

0.28

0.65

S.E.

0.15

0.12

0.09

0.10

0.11

0.26

0.09

0.14

Fig. 1. Number of trips. A paired t-test was applied to assess whether the differences between the number of trips for the computerized and manual systems were significantly different.

A reduction in the number of trips implies a reduction in travel costs. The only exception is the Bhoomi project where travel costs have increased primarily because the location of office that issues the RTC has been shifted from the village level to a taluka 10 level government office. The opportunity cost for users, which is represented by wage loss due to travel, is therefore, larger. However, 800 rural kiosks are being put up by a private operator in Karnataka to issues RTCs. These kiosks will access Bhoomi’s central server (where all records on the day’s transactions in various taluka level kiosks are updated) to issue RTCs, which would reduce the customers’ cost of access. Waiting time at the service centers has reduced in all the projects (Fig. 2). Among those projects in which the reduction in waiting time were statistically significant, the savings ranged from 16.16 minutes (AMC) to 96.24 minutes (CARD), amounting to nearly 50 percent reduction from the waiting time in the manual system. Reduction in waiting time has direct impact on user costs through reduction in foregone wages as indicated in Table III.

9 In the case of AMC, the time elapsed has reduced from 9.8 days to 5.3 days for certain types of transactions while in CARD, the total time required for registrations has come down from 4.69 to 1.82 days. 10 Taluka (or sub-district) is an administrative division consisting of a city or town that serves as its headquarters, possibly additional towns, and a number of villages.

187 200.00

reducing additional work for the staff.

184.79

180.00

Manual Computerised

5.00 Manual Computerised

162.49

160.00

4.00

3.81 3.65

3.55

3.50

87.78

3.08

3.06

3.00

68.06

3.92

3.63

100.68

100.00 80.00

4.22

4.19

3.79

114.95 102.03

4.51

4.50 4.27

4.18

4.11

4.03

140.00 120.00

4.51

4.50

63.16

2.75

60.06

60.00

2.50

40.00

32.96

24.67

21.76

29.32

14.47

20.00

29.66 20.73

14.63

2.00

0.00

1.50

0.00 Bhoomi

KAVERI

Khajane DDO

Khajane Payee

CARD

eProcurement

eSeva

AMC

Checkpost

Bhoomi

Saving 41.21 62.91 41.40 35.40 96.24 114.95 18.50 16.16

8.87

S.E.

1.82

9.15

7.00

6.52

4.44

1.00

7.95

7.58

1.64

1.58

Fig. 2. Waiting time (in minutes). A paired t-test was applied to assess whether the differences between the waiting time in case of computerized and manual systems were significantly different.

Kaveri

Khajane DDO

Khajane Payee

CARD

eProcurement

eSeva

AMC

Checkpost

Diff.

0.95

0.32

0.40

0.55

0.48

0.27

0.95

0.70

0.57

S.E.

0.08

0.04

0.07

0.07

0.05

0.05

0.04

0.05

0.04

Fig. 3. Overall service quality score (5-point scale). A paired t-test was applied to assess whether the differences between the overall service quality scores for the computerized and manual systems were significantly different.

TABLE III WAGE LOSS (RS) FOR MANUAL AND COMPUTERIZED SYSTEMS Bhoomi RTC Mutation Wage loss (Rs) Manual

KAVERI

Khajane CARD eSeva AMC - Payee

61.70

55.44

382.93

N

110

91

123

9

S.E.

8.45

4.85

67.68

8.82

Wage loss 98.25 (Rs) Computerized

94.66

262.38

N S.E.

66.67 190.21 80.63 125.18

16 14.94 14

27

34

10

32.90 15.35

23.94

8

36.25 161.75 65.00

88.33

68

59

122

8

10.69

13.50

24.38

6.73

63

22

24

32.79 22.55

20.81

N: Sample size

11.81

8.26

2

B. User Perception of Service Quality For all eight projects, users rated the computerized services higher in overall quality (Fig. 3). The differences were statistically significant. eSeva has shown a significant improvement in service quality whereas eProcurement has shown a marginal improvement. Bhoomi also shows very significant improvement in service quality, amongst the highest in the projects assessed in the study. The rate of error in documents (as reported by the users), which is an important measure of quality, was also lower for all the projects except the issue of RTCs (see Fig. 4). The number of RTCs issued has gone up many-fold after computerization and even minor mistakes in the name are now being noticed and reported. Reduction in errors saves additional trips for users and improves productivity by

BhoomiMutation

3.56 1.98

1.52 0.74

0

4.74 3.88

3.49

3.31

Bhoomi-RTC

While the difference between manual and computerized systems on all the cost indicators are statistically significant (in most cases at a confidence level of 99 percent), the accuracy of the estimate of the difference in direct costs can be improved by using larger samples.

8.02

6 4

81

12.79

12

1.14 KAVERI

KhajaneDDO Manual

KhajanePayee

CARD

eSeva

1.67 1.26 AMC

Computerised

Fig. 4. Error Rate (in percentage)

C. User Perception of Governance and Corruption The proportion of users paying bribes has generally declined. The decline is higher in some projects than in others (Fig. 5). For example, in the Bhoomi project, while about 30 percent of users were paying bribes in the manual system, less than one percent is continuing with the practice in the computerized system. The proportion of transactions in which a bribe was paid in KAVERI continues to be high at 21.61 percent in the computerized system. Out of the seven agencies (and eight projects) where significant corruption was reported in the manual system, four were able to eliminate or significantly reduce corruption through computerization. The impact was marginal in the remaining three agencies as corruption continued at a significant level even in the computerized systems. Although, in reducing corruption the

188

outcome is mixed, e-government seems to have the potential for significant reduction in corruption in service delivery, as shown by one project. 40.00 Manual 35.00 30.00

Computeris ed

34.32

33.82 29.71

28.02 25.00

23.71 21.61

20.00

20.42

15.00

14.48

14.17

10.00 5.68

5.00

3.35 0.84

2.71 0.84

0.00

Bhoomi-RTC

0.74 BhoomiMutation

0.00 0.00 Khajane DDO Khajane Payee

Kaveri

CARD

0.40 0.00 eSeva

eProcurement

AMC

Checkpost

Fig. 5. Proportion paying bribes (percentage).

An interesting phenomenon is that the amount of bribe users pay is generally higher in computerized systems in all the projects. This can perhaps partly be attributed to general inflation in the price levels. It may also suggest that the complexity of the types of favor requested through bribes has increased. It must be noted that the estimates of bribe amounts are less accurate than other results as the sample of respondents paying bribes was small resulting in high standard error of the estimates. The overall governance rating is higher for computerized systems than manual systems (Fig. 6). In the case of Bhoomi, there was a marked improvement in transparency and fairness whereas KAVERI showed hardly any improvement in the quality of governance. 5.00 Manual Computerised

4.48

4.50

4.20 4.03

3.34 3.15

3.25

3.74

3.61

3.58

3.50

4.21

3.99

3.94

4.00

3.50

4.31

4.19

3.51

3.33

3.33

3.00 2.50

2.00

1.50

1.00 Bhoomi

Kaveri

Khajane DDO

Khajane Payee

CARD

eProcurement

eSeva

AMC

Checkpost

Diff.

0.76

0.19

0.70

0.61

0.61

0.38

0.79

0.75

0.88

S.E.

0.07

0.04

0.06

0.06

0.04

0.04

0.04

0.04

0.06

Fig. 6. Overall governance score (5-point scale). A paired t-test was applied to assess whether the differences between the overall governance scores for the computerized and manual systems were significantly different.

D. Comparison of Projects on Overall Client Impact Respondents were asked to rate each project on 18 common attributes encompassing the three dimensions discussed earlier. They were also asked to select the three attributes that they considered most important. A weighted composite score that factors in the importance and rating of the attributes was calculated for each project. Table IV compares the composite scores across projects on a 5-point scale. It indicates that Bhoomi shows the maximum improvement as its manual version had the lowest score. eSeva also shows a very significant improvement as the computerized counters are rated close to very good in the composite score. KAVERI and CARD indicate only marginal improvement over the manual system. Ranking of projects using the composite rating, and particularly its improvement, can represent the degree of success of the project from the point of view of clients. Bhoomi and eSeva can be rated as very successful projects, AMC as moderately successful and KAVERI and CARD as projects where there is considerable scope for improvement. TABLE IV DESCENDING ORDER OF IMPROVEMENT IN COMPOSITE SCORES Manual

S.E.

Computerized

S.E.

Difference

S.E.

Bhoomi

2.86

0.07

4.46

0.04

1.60

0.09

eSeva

3.39

0.04

4.66

0.03

1.27

0.05

Khajane DDO

3.24

0.08

4.43

0.05

1.19

0.10

Khajane Payee

3.08

0.07

4.19

0.05

1.10

0.10

eProcurement

3.22

0.04

4.26

0.04

1.04

0.05

Checkpost

3.48

0.05

4.32

0.04

0.84

0.05

AMC

3.37

0.04

4.12

0.06

0.75

0.06

KAVERI

3.35

0.06

3.89

0.05

0.55

0.05

CARD

3.78

0.03

3.93

0.03

0.15

0.03

As shown in Table V, three kinds of attributes have been selected more often than others. They include attributes pertaining to: i) transactional efficiency; ii) improved governance; and iii) quality as measured by error rate and convenience. An important conclusion from this analysis is that important attributes are different for different projects. If such an exercise was to be done before designing an application it would provide useful insights into benefits that need to be targeted and the kind of process reform required. It is rarely done in practice either before or after the project implementation.

189 TABLE V TOP FOUR DESIRED ATTRIBUTIONS OF SERVICES Project Bhoomi

Desired Attribution of Services Error free transaction

No delay in transaction Greater transparency

Khajane Payee

No delay in transaction

Convenient Good time schedule location

Error free transaction

CARD

Less time and effort required No corruption

Less waiting Less corruption time

Fair treatment

Easy access

No need to visit Government office

KAVERI Khajane DDO

eProcurement

Less waiting Fewer visits time Less Error free Less waiting transaction corruption time Simplicity of Convenient Friendly Error free time schedule attitude of transaction procedures officers

Equal opportunity to all

eSeva

Less time and effort required

Fair Less waiting Convenient time schedule treatment time

AMC

Less time and effort required

Less corruption

Greater transparency

Good complaint handling system

Checkpost

No delay in transaction

Error free receipt

Error free transaction

Proper queue system

Legend: Underline - Improved Governance; Bold - Transactional Efficiency; Italics - Quality

with manpower, as was seen in the case of KAVERI. The human resource and other costs incurred by the government have remained within a narrow range for the manual and computerized systems. In terms of the impact on agencies, the ability to cope with growth in transactions was enhanced in all cases. In some cases, computerization helped in the growth of transactions. Data from employees on perceptions about impact on work load and managerial processes was analyzed. The staff in these agencies did not perceive that cost had been reduced. However, they felt that there was a significant positive impact on the quality of governance. Computerization was also seen to be partially responsible for improved tax collections by some agencies. VI. KEY FINDINGS AND IMPLICATIONS Respondents (citizens and businesses) who have used both the manual and computerized systems indicated an overwhelming preference for the computerized service delivery in most projects (see Table VI). However, in the case of Bhoomi, 79 percent of respondents preferred the computerized system whereas in most other projects the figure is 95 to 99 percent. TABLE VI PREFERENCE FOR COMPUTERIZATION (PERCENTAGE) Project Bhoomi

V. RESULTS OF IMPACT ON AGENCIES Data on transaction volumes, operating costs, investments, tax collection (if applicable), and revenues from transaction fees was to be collected for three years prior to the introduction of computerized system and for the entire period since computerization. However, it was very difficult to collect accurate data for pre-computerization as often agencies did not have an MIS that was reporting such data. Postcomputerization data on transactions and revenues was easier to collect as it had at least been recorded in individual agencies. Often, the prevalent costing systems do not permit identification of operating costs for the computerization project alone. Many costs are joint costs being incurred to support multiple activities. For different agencies, computerized delivery of services touches different portions of the overall activity portfolio. For example, AMC is an organization which delivers large scale physical services such as cleaning of roads and public areas, public lighting and maintenance of urban infrastructure. Computerized delivery of services impacts less than one percent of all employees. For agencies such as offices of the sub-registrar the impact is more widespread as the core service is computerized. By and large, the major cost component was manpower. The study revealed that the staff was not reduced in any project after computerization. In some projects work load was reduced and therefore, additional time could be devoted to other tasks. There could be cost implications in the future as the transactions are growing at a much faster rate as compared

Preference for Computerization (%) 79.34

eProcurement

83.71

Checkpost

91.25

eSeva

96.84

CARD

96.98

AMC

97.49

KAVERI

98.31

A profiling 11 of those who preferred the manual system revealed that higher-income farmers with large holdings in urban areas preferred it to the computerized system. Perhaps such farmers were able to manipulate the flexibility in the system to their advantage - something that needs to be further investigated. Preference for the computerized system is backed with specific areas where concrete benefits have accrued to them. In most cases the cost of accessing service was reduced because the number of trips that needed to be made to the concerned offices saw a significant reduction and the waiting time came down by nearly fifty percent. Corruption was significantly reduced or eliminated in five projects. Quality of service delivery and quality of governance were also perceived to have improved significantly with computerization in most cases. There is considerable variability in the composite scores across the eight projects. The scores range from a 0.15 point 11 An interaction tree was constructed by considering preference for the computerized/manual system as a dependent variable. The independent variables comprised those that are indicative of the user’s profile, viz. occupation, income and urban/rural status.

190

difference, which represents virtually no improvement, to 1.6 (on a 5-point scale), which is quite significant. Other studies have shown a similar variation for the same project implemented across different ststes [13]. This suggests that projects are not being designed using learning from other states. The distinct preference for computerized systems by the clients of poorly rated projects seems to suggest that even small gains for the users can trigger major positive change in perception about service delivery systems. Incremental operating costs for agencies (including amortized investment) per transaction could be compared with the reduction in direct cost of accessing the service reported by the clients. This would indicate the level of user fee that can be charged (should be less than the monetized gain to the user) and the degree to which the project could be made economically viable. The fact that many projects can become self-sustaining through revenues from user fee indicates that private sector investment can be tapped. Five of the eight projects have private partners and the expansion of the sixth project in rural areas is being planned through private partnership. Further analysis of data indicated significant variability of impact across different delivery centers in a given project. This variability should be a cause for concern for delivery models in which physical service centers are created. Part of this variability is explained by poor infrastructure (power and connectivity), particularly in rural areas. Also, given the large variation in activity levels at different centers, it is often difficult to match the capacity to the demand at each of these centers. Portal-based delivery accessed via the Internet can be a solution. However, unequal access to Internet can put some users at a disadvantage in such systems. A detailed study of one of the centers in KAVERI indicated that any type of system breakdown leads to corruption. The breakdown can be on account of an overload of demand in comparison to the capacity of the system to process registrations. Systematizing queues by appointments helps prevent break-down. Agents play a key role in promoting corruption. Private operators also exhibit rentseeking behavior if given an opportunity.

VII. LIMITATIONS OF THE STUDY AND AREAS FOR FURTHER RESEARCH The study was exploratory in nature undertaken with limited resources. More countries could have been covered and a greater variety of applications could have been assessed for generalizing the conclusions. The study used a sample size that was considered relatively small (and barely adequate) at the stage of planning the field work. While the difference between manual and computerized systems on all the performance indicators are found to be statistically significant (in most cases at a confidence level of 99 percent), the accuracy of the actual estimate of the difference in direct costs could be improved by the use of larger samples. A larger sample size in subsequent studies would also permit analysis

of disaggregated data at the level of each service location or for different types of clients. The conclusions on impact on agency are not as robust as those for impact on citizens because time series data on different types of costs and revenue streams related to the service being investigated could not be collected for sufficient time periods for most of the projects. The overall positive assessment of projects should be seen in the context of the eight projects that were selected for assessment. These were mature projects, with wide reach and scope of activity. These projects have been operational for at least three years, implying successful implementation. Nearly all the projects serve urban clients. Projects that serve rural clients could have a very different cost structure and demand pattern. Rural projects also face a greater challenge in the maintenance of infrastructure. For all the projects improvements were measured in comparison with the manual systems that were replaced. When countries move from computerized systems in departments to a 24x7 service delivery through portals, the degree of perceived improvement could be very different. For understanding the costs and benefit of ICT investments at a macro level, more projects from different contexts need to be evaluated and failed projects (that could not be successfully implemented) would have to be included in the analysis. It would therefore be hasty to generalize the overall conclusions of positive impact and economic viability of electronic service delivery projects from this study. Further studies need to be undertaken to explain the variation in impact on various dimensions; differences across locations for a project; and differences across projects. Studies could also be undertaken to understand the effectiveness of different delivery models and implementation modalities such as the use of public private partnerships. More data (quantitative as well as qualitative) will need to be collected from the agencies to undertake such studies. The establishment of acceptable counter-factual was a challenge for the study team. All the projects had discontinued manual delivery of service and had mandated the use of the computerized systems for the citizens. For the assessment of manual system, respondents needed to rely on memory. In case of systems that have been operational for a large number of years, such recall can be prone to error. There has been no benchmarking of the service delivery in a manual system prior to implementing a new computerized system something that needs to be done for new projects that are taken up in the future. Other forms of counterfactuals such as a user group from an adjoining state could be used to avoid the problem of recall. However, use of a different context could result in other types of biases. The study assessed direct economic impact in terms of cost of accessing the service. It does not measure the impact on the inherent value of efficient delivery of the service for the client. For example, copy of a land title is required for obtaining a farm loan, getting a bail in a court case, insuring crops and for purposes of checking the veracity of the record. A farmer would attach different values to efficient delivery of

191

land title depending on the purpose for it was required. Subsequent studies should also attempt to estimate the impact on indirect economic value. REFERENCES [1] [2]

[3] [4] [5]

[6] [7] [8] [9] [10] [11]

[12] [13]

R. Heeks. (2008, May 8). Success and failure in eGovernment projects. Available: http://www.egov4dev.org/success/index.shtml R. Basant, S. Commander, R. Harrison, and N. Menezes-Filho, “ICT adoption and productivity in developing countries: New firm level evidence from Brazil and India," IZA Discussion Paper No 2294, Germany, Sep. 2006. “World public sector report 2003: E-government at the crossroads,” United Nations, New York, 2003. “Information and communication technologies: A World Bank Group strategy,” World Bank Group, 2002. bridges.org (2005, May). E-ready for what? E-readiness in developing countries: Current status and prospects toward the millennium development goals. InfoDev [Online]. Available: http://www.infodev.org/files/2049_file_InfoDev_E_Rdnss_Rpt_rev11 May05.pdf “Value creation in eGovernment projects: An exploratory analysis conducted for the Danish presidency of the eGovernment work group of the Directors General,” European Union, 2003. “eGovernment: Realizing the vision,” Third Annual eGovernment Benchmarking Report, Accenture, 2002. “Is IT worth it - Presenting the public sector business case,” Gartner, 2002. E. Lau, “Electronic government and the drive for growth and equity,” OECD E-Government Project, 2005. “eGovernment economics project”, DG Information Society and Media, European Commission., 2005. “WiBe 4.0 - Recommendations on economic efficiency assessments in the German federal administration, in particular with regard to the use of information technology,” Federal Ministry of the Interior, Germany, 2004. R. M. Peters, M. Janssen, and T. van Engers, “Measuring e-Government Impact: Existing Practices and Shortcomings,” in Proc. 6th International Conference on Electronic Commerce, Delft, The Netherlands, 2004. “State level e-governance projects in India: Overall assessment of impact on citizens,” Centre for E-Governance, Indian Institute of Management, Ahmedabad, unpublished.

192

The contribution of user-based subsidies to the impact and sustainability of telecenters – the eCenter project in Kyrgyzstan 1

Michael L. Best, Dhanaraj Thakur and Beth E. Kolko

Abstract—We examine the extent to which user-based subsidies can promote the sustainability and development impact of telecenters, where sustainability is defined in financial and social terms. We do this by looking at a coupon scheme used by the USAID funded eCenter network in Kyrgyzstan. The network consisted of partnerships with existing commercial computer centers which provided fee-based ICT services to their communities. The eCenter program temporarily provided subsidized coupons for Internet access and computer training to users of these centers. Using a mixed-method approach, we found that user-based subsidies have to a certain degree aided financial sustainability by bringing new users to the centers, some of who will, conceivably, become long-term customers bringing ongoing revenues. The subsidies also helped to “kick-start” operations during the initial stage of the project. However we found that the distribution of the coupons did not lead to any significant enhancements to social sustainability by, for instance, encouraging users from underrepresented social groups (though we did find that women were marginally more likely to take advantage of the program). Moreover, we found that the distribution of both Internet and training coupons favored more regular users of the eCenters. Finally, looking beyond sustainability to impact within the community, we found that the coupon program had a limited development impact on participating communities. For example, the population of users who reported economic benefits from using the eCenter (eg. subsequent employment or starting a new business), did not make more use of the coupon program than the user population at large, suggesting that the coupon program itself did not account for this benefit. We argue that, even in the presence of such modest positive effects, user-based subsidies still offer an intriguing model. We believe that if the eCenters had narrowly targeted particular participants for the coupon programs it is likely that the benefits

of the program can be enhanced. For example, better social targeting of the coupons could have brought in more women or under represented groups. From a financial sustainability point of view, explicitly targeting new users (as opposed to people already using the center) could have also been more effective. In terms of impact, coupons could have been distributed, for example, specifically to entrepreneurs looking to start a new business. A method of incentivizing eCenter management to perform such targeting is probably required. Index Terms — Telecenters, Kyrgyzstan, Central Asia

Sustainability,

Subsidy,

I. INTRODUCTION

Michael L. Best is an Assistant Professor with the Sam Nunn School of International Affairs and the School of Interactive Computing, Georgia Institute of Technology, Atlanta, GA, 30332 USA. (e-mail: [email protected]). Dhanaraj Thakur is a PhD student with the School of Public Policy, Georgia Institute of Technology, Atlanta, GA, 30332 USA. (e-mail: [email protected]). Beth Kolko is an Associate Professor with the Department of Human Centered Design & Engineering, University of Washington, Seattle, WA, 98195 USA. (e-mail: [email protected]).

As a method of improving access to information and communication technologies (ICTs), the telecenter is popular among donors and governments in many different countries [1, 2]. The concept of the telecenter emerged from a community driven movement in Scandinavia in the 1980s [3]. For poor rural communities, telecenters can provide access to communication and content. This can include access to market and crop prices, to financial information/services, and communication with friends, family, and business colleagues [4]. Telecenters can also support the delivery of government services [5]. Today, telecenters can be found in many countries and are referred to by a plethora of terms: tele-cottages, public information access points, public internet access points or multi-purpose communications centers. While each of these terms represent slight differences in the object being discussed, the common characteristic of telecenters, as used in this paper, are publicly accessible ICT resources in one or more physical locations that are available with or without fees. As with any other development intervention, a critical objective is to make the telecenter sustainable. Another objective is to address the problem of poverty and other socioeconomic concerns within the community at large; thus a successful solution should also have real impact [6, 7]2. However, many telecenter projects have not realized either of these objectives for a variety of reasons such as a lack of income, inappropriate services, little or no local content [8], inadequate infrastructure (power, roads, telecoms, etc.)

1 This paper is based on the results of an evaluation provided to the Academy for Educational Development (AED) of the USAID funded eCenter project in Kyrgyzstan in June 2007.

2 In fact, a successful telecenter program might be one that has had significant impact (e.g. spun-off a few small businesses) but has no long-term sustainability (e.g. the program transitions and closes after a few years).

193

particularly in remote areas [9], lack of political, social, or managerial sustainability [10] or the challenge of identifying local and global partners that can move the telecenter to a larger scale [11]. These challenges are enormous and have often relegated telecenter initiatives to the “forever pilot syndrome” [11 pg.11]. Thus, sustainability and impact remain the two principal questions related to rural telecenters. In response, various strategies have been developed to advance the sustainability and impact of telecenter projects. One approach is the provision of subsidies for telecenter users. These are often applied for a limited period and enable the use of ICT services by community members who might not have otherwise been able to do so. While not necessarily a new approach, there is little research that examines the connection between userbased subsidies and sustainability or impact. The question that we wish to explore in this paper is to what extent can userbased subsidy programs promote the sustainability and impact of telecenters. We do this by examining a user-subsidy coupon scheme used by the USAID funded eCenter network in Kyrgyzstan. The rest of the paper is divided into several sections. First, we discuss issues surrounding the impact and sustainability of telecenters. We then articulate the methodology used in addressing our research question. Finally, we describe the eCenter project, present our results and analyses based on our definition of sustainability and impact, and posit some conclusions.

II. THE IMPACT AND SUSTAINABILITY OF TELECENTERS As with any other development intervention, a critical objective is to make the telecenter sustainable. Typically, the intention is to make a telecenter financially independent and solvent. This can be in terms of meeting maintenance costs, the recovery of initial investments, acquiring sufficient human resources/staff, or adequate service delivery [12, 13]. Sustainability can also be viewed from a social point of view. Thus, sustainability hinges on having local champions, sufficient community acceptance/awareness and involvement in the running of centers, a range of users that is nondiscriminatory and balanced, and beneficiary participation in project design [4, 9]. Additionally, there are political factors to be considered such as accessing local and national political support for the project [5]. This includes having a policy and regulatory environment which is conducive to the development of telecommunications infrastructure and a commercial internet market [14]. Investigators have proposed a number of theoretical frameworks to improve our understanding of sustainability including the critical success factor (CSF) and critical failure factor (CFF) models [15, 16]; the ‘design-actuality’ [17] or ‘design-reality’ gaps [18]; scenario analysis for long-term sustainability problems [19]; economic and financial sustainability models [20]; and political and institutional models that underline the lack of commitment on the part of political leaders and public managers. Previous works by one of us (Best) [5, 10] have presented a sustainability failure

model built upon the work of Heeks and Bhatnagar to help explain why projects that succeed initially, can still fail to enjoy long-term sustainability. While the problems of sustainability are fairly well documented (and theorized) the ultimate goals of community impact (and the measurement and evaluation of such impact) may be less conclusively observed. To date most literature has focused on formative or process evaluation as opposed to summative or impact evaluations [21, 22]. And the literature that does focus on social or economic impact assessments of telecenters to their broader community have yet to demonstrate an absolutely conclusive link (see [23] for a review). From the above discussion and in tandem with previous research, [see for example 5, 10, 24, 25], we have selected two main dimensions from which we examine sustainability: 1. Financial – This includes the financial independence, business performance, and solvency of the project over time. 2. Social – This refers to the equitable distribution of benefits among target groups, equal access and use, and locally relevant content/services. In addition, we study the center’s development impact as it extends into the communities themselves. We can observe this in both economic and social forms such as new educational opportunities, community empowerment, job creation, local economic development, etc. Various initiatives have strived to manage the inter-related issues of impact and sustainability. The literature consists of many general prescriptive reports and descriptive case studies of such initiatives [see for example 8, 26-28]. Others have reported on specific or novel approaches such as incorporating business incubators into the telecenter and sharing resulting profits with the telecenter organization [29]. Another approach is to provide subsidies to a telecenter initiative that is typically applied to initial infrastructure or staffing costs [13]. Alternatively, user-based subsidies can lower the costs that the users themselves pay for services they seek. The hope is that if well conceived, user-based subsidy programs can enhance both financial sustainability (by creating an early flow of income while developing a customer base over the long term) and social sustainability (by providing subsidies that target under-served or over-looked populations, for example women). Additionally, if the subsidies are for activities that are likely to lead to economic, social or political growth within the community at large then the impact of the telecenter should, ultimately, be enhanced. This approach has been used in a variety of settings including the Cotahuasi Internet Cabina project in Peru in 1997. In that case, the use of the telecenter’s services by local community leaders was paid for by donor funds [4]. Another example of user-based subsidies was the PC3 project in Bulgaria. Pre-paid coupons were distributed in communities where the PC3 centers were located. The main goal was to promote the centers while quickly developing a client base. This helped to reduce the financial risk faced by the new PC3 centers [30]. The eCenter project in Kyrgyzstan employed a similar logic in the use of coupons as an incentive both for the local eCenters and clients.

194

III.

METHOD

We employed a mixed-method approach to study the effect of the coupon program on the sustainability and impact of the eCenters. Our research consisted of site visits to all telecenters that were part of the eCenter program; 7 in total. These visits were done in collaboration with local researchers and took place between March and June 2007. Our research instruments included a user survey and interviews with all the center managers, available members of their staff, and local businesses. These data were supplemented by a review of user logs, project reports and updates, and business proposal guidelines. The user survey consisted of sixty-two questions developed around the issues of sustainability and impact. Specifically it explored how the computer training and Internet coupons were used, the ways users engaged with ICTs at the eCenters, and the perceived economic impact of using the centers. The majority of the questions were close-ended with a few openended questions to capture opinions on issues related to the centers. To determine our sample size we estimated the overall population size as the number of users at each eCenter over the period for which data was available: January 2006 (start of the project) to January 2007. We defined users as those who participated in the subsidy (coupon) program of the centers for either Internet access or computer training. Given this approach, we approximated the total user population at 9,497 people and thus ensuring a confidence level of 95% required a sample size of 369 users. To identify subjects from each of the centers, user contact lists were obtained from center staff. Users were then randomly selected from each list. In many cases, however, these lists were incomplete with either missing contact information or incorrect contact details. To account for this difficulty, research teams substituted or augmented random sampling with subjects obtained by opportunistic sampling at the eCenters during site visits. In addition, the researchers worked through the social network of users at each site in order to find both previous and current users of the eCenters. Thus, user surveys were first collected using standard random sampling techniques when possible and then with a combination of convenience and snowball sampling. In addition to sampling the appropriate number of users we also needed to ensure that our sampled subjects were representative of the population across each center. To account for this we stratified the sample size according to the proportion of users from each location. Table 1 gives the targeted sample size, and actual number of subjects surveyed, for all seven eCenters; in almost all cases we were able to over-sample the population. The one center that was under-sampled (Bosteri) was closed for renovations during the data collection period, complicating researchers’ attempts to contact users.

Table 1 – Targeted and Actual sample size of user surveys by eCenter Targeted Sample Actual number of eCenter Size collected user surveys Naryn 70 79 Bosteri 50 40 Karakol 75 87 Karasuu 67 70 Nookat 66 72 Talas 22 27 Ivanovka 18 19 Total 369 394

IV.

BACKGROUND – THE ECENTER PROJECT IN KYRGYZSTAN

Kyrgyzstan is a small Central Asian country that was part of the former Soviet Union. Although it is landlocked with limited resources, it has achieved economic growth of around 4% between 2000-2005 [31]. Nevertheless, the poverty rate continues to be of major concern. The national estimate of people living below the poverty line was 43% in 2005 with larger percentages in rural areas [32]. In addition, the unemployment rate was estimated at 8.1% in 2005, with 13% in urban areas [32]. This implies that a significant part of the population, though employed, is still poor. In 2002, the government approved a “National Strategy for ICT Development in the Kyrgyz Republic” as part of its plan to use ICTs to address development issues. In general, this emphasis follows what Ure [33] notes is a more open approach to the diffusion and use of ICTs by the Kyrgyz government when compared to its neighbors. However, the growth of the Internet has been hampered by the monopoly held by the state telecommunications company Kyrgyztelecom, which is deeply in debt to the World Bank and seeking to return to profitability. One consequence of this is that Internet subscription costs are high relative to average incomes. Recent estimates put Internet user rates at a relatively low level of 13.3% [34]. In addition, PC ownership was estimated at 10.4% [34]. Given the larger social and economic context, Internet access via private, home-based means is not likely to grow substantially anytime soon and outside the capital, public access is limited, expensive, and usually not fast. Given the problem of poverty and the acknowledged importance of information resources for economic development, the need to provide alternatives to private use and to enhance public ICT venues was viewed as important. The eCenter Project The eCenter project was launched in Kyrgyzstan in July 2005. It was funded through the Last Mile Initiative of the United States Agency for International Development (USAID) that seeks to promote greater access to information and communication technologies, particularly in rural and underserved areas. The goal of the project was to augment and network a group of telecenters across the country with the aim of promoting local economic development. Each telecenter was established within a pre-existing business. In this way, the

195

Nookat

centers sought to improve local access to ICTs, stimulate local business creation, improve computer skills, and increase opportunities in non-traditional employment training and job creation among the local population. The local project manager in Kyrgyzstan was the Civil Initiative on Internet Policy (CIIP)3, a Kyrgyz non-governmental organization that focuses on the promotion of civil society interests in the development of national ICT policy. The project ran from July 2005 to September 2006 with a total budget of US$390,000. During this time, a group of seven eCenters were established across the country (see Fig.1): 1. Karakol 2. Bosteri 3. Naryn 4. Nookat 5. Karasuu/Osh4 6. Ivanovka 7. Talas Each of the eCenters provides a variety of fee-based services including Internet access and email, printing, scanning, copying, faxing, multi-media services and IPtelephony. The exact suite of services offered varies from center to center. Each center also delivers a curriculum of computer literacy courses which consist of several modules: Windows, Microsoft Word, Microsoft Excel and using the Internet. Additionally, some centers offer accounting courses and one offers leadership training. Table 2 lists the estimated number of competing Internet cafés within each of the seven communities which had eCenters. It also includes the total estimated population size 3 4

See http://www.gipi.kg/ for an overview of CIIP. The Karasuu eCenter moved to Osh city in April/May 2007.

for those communities. These figures help to sketch the overall competitive landscape of the centers. By looking at the number of employees we also get a sense of the relative size of each establishment. Note that while this table focuses on Internet provision there were also other businesses providing related business services such as printing or photocopying. Table 2 – Summary characteristics of the eCenters eCenter Number of Competing Population Employees Internet Estimate Centers 1999 Karakol 7 10 64,322 Bosteri 4 0 12,000 Ivanovka 6 0 20,000 Talas 3 2 32,638 Nookat 4 0 30,000 Karasuu 3 0 19,143 (Osh) (80) (220,000) Naryn 4 10 40,050 User-based subsidies There were two main components to the eCenter program. First, there were subsidies granted to the eCenters for the provision of computer literacy/accounting courses and second, free or discounted Internet access was offered to local community members. The subsidies for computer literacy or accounting courses came in the form of reimbursements to the eCenter operator for the training costs of those who passed each course. People were able to participate in the courses by receiving coupons. Similarly, Internet access was subsidized through the use of five-hour coupons that were distributed to users. The subsidy level on both the computer training courses

196

and Internet coupons changed over time. Subsidy amounts were reduced 20% each quarter; thus while the project paid for 100% of the cost of a training course in the first quarter, by the fourth quarter it paid only 20% of the cost and required the user to pay 80%. Partnering with local businesses Each eCenter was established as part of an existing local business rather than through the creation of new centers. Suitable local business partners (including pre-existing telecenters) were selected through a competitive bidding process based on criteria such as related prior business experience, existing level of investment, relevant telecom experience, and the potential for further expansion of programs. Subsequent to a center’s selection, CIIP provided technical support on the use of the subsidy program and delivered training to the business owners and their staff. While some of the selected eCenter sites were already providing Internet services, others were engaged in computer graphics and photocopying services, gaming, computer repair or providing computer training courses. There was one final component of the eCenter project – a land grant program where suitable land for investment in office space and technological parks could be linked to each eCenter. It was envisioned that such investment would be supported by the success of the eCenters; this component of the project, however, did not achieve expected outcomes and was eventually dropped. While there are several interesting aspects to the eCenter project as a whole, we have focused only on the user-based subsidies in this study. The following sections present our findings and analysis of the user-based subsidies component of the eCenter project. V. THE SUSTAINABILITY OF THE ECENTER PROJECT IN KYRGYZSTAN5 Financial sustainability All the eCenters enthusiastically implemented the coupon program and several have since sought to expand their offering of ICT services by increasing the scope and size of their businesses. In all cases, the businesses stated that their financial situation was better off following the change to an eCenter. On average, according to the managers this transition increased their revenues by an estimated 57% (with a low of 15% reported at Karakol and a high of 100% reported at Talas). In addition, almost all of the center managers agreed that their clientele increased (some very significantly) after becoming an eCenter. The main sources of revenue varied from center to center and were based on local demand. For example, Internet access was reported to be the main source of revenue in only two of the seven centers. Other revenue sources included computer training, typing services, IP telephony and the sale of mobile phone credit. 5 Statistical results mentioned in this paper are all significant to at least the 0.05 level.

The managers themselves were quick to praise the eCenter program and saw a direct connection between their success and the use of the coupons. Some of the managers estimated that 70-90% of their current customers would not have come to the center without having been part of the coupon program. Interestingly enough, a significant percentage of users did not actually rely on coupons. Of those surveyed, only 54% and 43% reported actually receiving coupons for Internet access and computer training respectively. This implies that perhaps the managers had an exaggerated belief of the effect of the coupon program on their businesses. When asked if they would have used the eCenter services if there were no coupons, approximately 47% of respondents said yes and 19.5% said they would not have used the center in that case. It should also be noted that the subsidies were only provided for a limited period. Thus, by the end of the project users were paying for the full cost of Internet access and computer training. The existing competitive environment (Table 2) appeared to influence the scale of impact of the coupons. For example, we found that if the manager identified more competitors in the immediate community, the eCenter users were more aware of and more likely to have used other cyber cafés. If there were fewer competing centers in the catchment then users were more likely to have used the eCenter prior to the coupon program. In other words, when the level of competition was heavy the coupon program was more successful in drawing in new users. In general, 66% of those who received Internet coupons had used the eCenter before while 33% of those who had received a coupon had not used the eCenter before. The results are very similar for training coupons. Thus, during the subsidy period, coupon use was associated with the introduction of new users, which can contribute to financial sustainability. It should be noted that the decision to partner with existing and successful local businesses was also important for the success of the project. Generally, there is little evidence to suggest that the one-off donor grant approach can work [24] and so some have argued for a business centered approach to telecenters [13]. The rationale behind this approach is that the profit motive can promote financial sustainability. In the case of the eCenter project, it was in the interest of the local business partners to make sure that their centers remained in operation and were successful. The coupon program supported the partnership with local businesses by ensuring a sufficient number of customers during the initial subsidy period. It was critical therefore in “kick starting” the operations of the centers after which several of the local business partners were able to use this momentum to expand their businesses and tailor their suite of ICT services to their community needs. Social Sustainability – Types of Users We examine social sustainability primarily in terms of the equitable distribution of benefits, access and use within the community. Ideally, a more balanced and representative user base should represent broader community acceptance of the center and limited social exclusion. Table 3 below summarizes the general user distribution across the various centers. We

197

found that the majority of users were women except in Nookat where the more traditional culture in the south seems to limit female participation at the eCenters. The majority of users in the sample were typically young, suggesting the propensity to use ICTs is linked to age particularly where the general diffusion of such ICTs is low (i.e. rural areas). This young user group is consistent with a nationally young population (31% below 15 years in 2005 - [35]) and also partly explains the low marriage rates among our sampled population. In general, older customers used more of the basic ICT services such as photocopying or printing and requested the assistance of younger persons or staff to help them when using the Internet. Older users (above thirty years) were, however, just as likely to attend training courses. The education rates are similar to those nationally. Also, the user distribution was approximately proportional to the religious and ethnic composition of each community Table 3 – Summary of user characteristics eCenter Male Female Avg. Bachelors % % age or higher % Karakol 44 56 21 48 Bosteri 25 75 18 10 Ivanovka 53 47 18 5 Talas 33 67 23 15 Nookat 67 33 24 24 Karasuu 41 59 21 30 Naryn 31 69 24 47 Total for all users 43 57 21 32

Married % 10 3 0 19 39 14 25 19

Social Sustainability – Targeting and Distribution of Coupons Based on the above characteristics, we found few statistically significant differences among users who received coupons and those who did not. Among those who received Internet coupons approximately 51% were female. This could have had a marginal effect on the larger proportion of female users at the centers overall. Other variation in user characteristics (ethnicity, educational level, etc.) among those users who benefited from the coupon program and those who did not were not statistically significant. Also, there was no difference between those who got Internet coupons and those who did not in terms of their perceptions of the importance of the Internet to their jobs or schools. In terms of the training courses there were also no discernable demographic differences between those who received these coupons and those who did not. What was different was whether or not the respondent had used the eCenter prior to receiving the coupon. As mentioned above, 66% of those who received Internet coupons had used the eCenter before. Similarly, 60% of those who received training coupons had already visited the eCenter. This suggests that prior users were in a better position to learn about the program and participate. The concern is that while the coupon component of the project is now complete, the opportunity to attract larger numbers of new users to ICTs might not have

been exploited completely. Thus while the eCenters have financially benefited from the coupon program; the long term social sustainability of the project may have not. Two factors could help explain this type of coupon distribution among users. First, the implied goal of the project was to focus on persons who had the least access to the Internet and computer courses. However the way the coupon program was operationalized did not appear to be systematic or consistent across centers. Moreover, the eCenters received no particular incentives to narrowly target users and so some may have simply distributed their coupons to people already at their center. Thus, there was a significant difference among receivers of both Internet and training coupons from center to center though there was no discernable pattern to these differences. Some centers seemed to offer coupons to whoever was easily available. Indeed, some were more concerned with issuing the coupons rather than targeting and as such relied on informal means of distribution. One example of this is a neighboring Internet café owner in Karakol who said he received a training coupon for the computer literacy course. He was already exposed to more advanced computer courses but still chose to attend the eCenter course since it was free at that time. Alternatively, more formal means of distribution were used at other centers. For example, in Naryn the intention was to target as wide an audience as possible. This was done through advertising in local media and actually did result in lines literally going out of the door of the eCenter. However, it is unclear what type of users this form of marketing attracted. Second, the eCenters were encouraged but not compelled to give one Internet coupon per person, so as to increase the reach of the program. This suggestion, however, was not generally followed. For example, in Nookat users were given one coupon per quarter since, according to the manager there, it was difficult to continually find new Internet users. Alternatively, in Karasuu the strategy was one coupon per individual. Thus both targeting and distribution seemed to be dependent on the individual eCenters and was the result of a compromise required to enable a private business to undertake socially oriented goals.

VI. DEVELOPMENT IMPACTS OF THE COUPON PROGRAM Internet Access Although this represents a preliminary and indirect impact of the program, the Internet coupons did, of course, increase access to the net among users. This access can be viewed as a form of impact. Indeed, in most cases overall access increased in participating communities. For example, several of the centers are located in rural areas with limited access to ICTs. Thus, the introduction of the eCenters helped to meet the latent local demand for ICT services and training. Given the relatively high subscription costs for the Internet, this also meant that the eCenter could have been the only source of Internet access for some users. This was the case in at least three of the communities. For many, the communication function of the Internet was most important, particularly where regular post mail services

198

were infrequent. The most common online activities were sending/receiving emails and instant messaging. Other important uses included reading news, doing school related work/research, downloading music and participating in chat rooms. Several users proffered experiences of how accessing the Internet had helped them keep in touch with friends and family abroad, support their work and even religious education. Finally, in addition to providing increased quality or availability of Internet access, the eCenters have also been an important place for users to gain Internet-related skills. Approximately 70% of those who received Internet coupons said that they had learned to use the Internet at their eCenter. Development Impact – Computer training The acquisition of relevant skills is a clearer form of community development impact and the eCenters provided computer training courses which were well received by participants. Perhaps more than anything else, participants in training courses were quick to inform us of the benefits of receiving the computer literacy training. The courses were basic and allowed the centers to bring those with no understanding or experience with computers up to at least a foundational level of use. Completion was contingent on passing an evaluation which tested their ability to use the computer and some of the Microsoft applications they had studied. The majority of trainees were able to pass the test; the failure rate was approximately 10%. For many users there was a sense of pride in completing the course. For many, these courses provided the basis for the acquisition of job-related ICT skills. Eighty-five percent of all respondents stated that they acquired important job skills from their eCenter. Furthermore, some users reported actually getting jobs (usually clerical) based on these skills. Approximately 15% of all users reported finding a job as a result of using the eCenter. As noted earlier, the country has a high national poverty rate of 43% and an unemployment rate of 8.1% (2005). Therefore such impacts from the program, though small, will be valuable to both the unemployed and the working poor. We did not find, however, a link between users who participated in the coupon program and those who reported finding jobs due to training at the center. Approximately 61% of trainees who received user-subsidy coupons for their classes said they would pay for the service if there was no subsidy, demonstrating the value placed on the services and experience provided by the eCenters. Many of the eCenters are in fact planning to continue and even expand their course offerings past the lifespan of the subsidy program. For at least one eCenter this is also the main source of revenue. Content was an important factor in promoting the demand for the computer literacy courses. Much of the design and content for the courses came from the initial demand studies done prior to the start of the project. However, a few users suggested that having completed these courses they now want the center to offer more advanced computer courses. Continually modifying content to meet the needs of users will be a factor in supporting long term development impact.

Development Impact – Local economic development In terms of local economic development, we observed two types of impacts related to the introduction of user-subsidies at the eCenters. First, the eCenters have stimulated the local market for ICT services. For instance, there are approximately ten other commercial computer centers in Naryn that were established after the eCenter opened its doors. Second, the opposite is also possible. In Nookat, prior to the introduction of the eCenter there, there was one other business offering Internet services. This closed soon after the opening of the eCenter as it could not compete with the initial subsidies being provided for Internet access. That other business now operates as a café. One of the objectives of the eCenter project was to support local business development. Of all users, only 5% said that they were able to utilize either the facilities or courses at the eCenter to establish a new business. Most of the businesses were located in the services sector and a few specifically in the ICT sector. Of these users, there were no significant differences in terms of gender or levels of education. They were not necessarily in a higher income group either; for example they were no more likely to own a car than those who did not start a business. Finally, in terms of coupon use, there was again relatively little difference between those who reported starting a new business and those who did not. Whether we refer to the efficacy of the training courses or improved Internet access, one qualifying factor to note is the initial limited diffusion of ICTs in the targeted communities. This initial lack of ICTs in these mostly rural/semi-urban communities meant that the potential for the subsidies to have some modicum of impact was there from the start. We cannot be sure whether the development impact would be similar in other contexts. However, we can indicatively point to the example of the eCenter in Osh city where there are some 80 other cybercafés, various documentation centers, and in general better infrastructure for the delivery of ICTs. In an environment such as this, the eCenter modified its strategy to focus on the provision of training courses as there were few organizations providing this service. Thus, the business focus of the eCenters coupled with the goal of providing relevant ICT services to targeted communities can potentially create an impact in a variety of contexts.

VII. SUMMARY AND CONCLUSIONS The eCenter program provided subsidized coupons for computer courses and Internet access to users of already established commercial computer centers. Thus, the center owners provided the physical infrastructure for each eCenter and the subsidies helped them to access, and ostensibly expand the local market for training and Internet services. The project, therefore, represented a merger between public goals of increasing ICT access and literacy and private goals of increasing profit and market share. We focused on the subsidy component of this project, as an innovative approach to telecenter development. As such we attempt to address a gap in current research by exploring the relationship between user-based subsidies and the

199

sustainability and impact of telecenter initiatives. In this case we examined sustainability in terms of its financial and social dimensions. By financial sustainability we look at an eCenter’s financial independence and solvency; by social sustainability we are primarily concerned with the distribution of access and benefits to the community. There are of course other dimensions that could have been included, but we wanted to limit our analysis in such a way as to make it focused and useful. In terms of financial sustainability, we found that the coupon program brought some new users to the centers during the subsidy period. Self-reporting indicates that a third of Internet coupon users and 40% of the training coupon users were new. More importantly, 19.5% of users stated that they would have not have used the services of the eCenter without the coupons. The coupons also helped the local businesses thrive with customers during the initial subsidy period; it was critical, therefore, in “kick starting” the operations of the centers. Thus, we argue that the coupons have to some degree aided financial sustainability by bringing new users (and therefore additional revenue and an expanded long-term customer base) to the centers and creating a stable source of income during the initial stage of operations. Obviously, temporary subsidy programs may actually weaken long-term financial self-sustainability as the program winds down. But in this case, while we observed the end of the program we did not find evidence for an immediate financial downturn. In terms of social sustainability, the distribution of the coupons did not lead to any significant differences in the type of users except a marginal increase in female participation. We find that the distribution of both Internet and training coupons favored more regular users of the eCenters. This implies that the opportunity to expose as many new users as possible to ICTs may have been missed and this could have limited the wider social impact of the project. One of the main reasons for this was the lack of effective targeting strategies among the centers. Finally, the coupon program had a limited development impact on participating communities. One preliminary impact was the enabling of Internet access for users and the learning of Internet related skills. The coupons also facilitated computer skills training. However, among those who reported economic benefits (subsequent employment or starting a new business) after acquiring these skills, there was little or no difference between those who had been coupon recipients and just regular users who had not availed of the program. We suspect that as the targeting and distribution was left up to the local eCenters, there might have been less emphasis on sustainability and impact and more emphasis on short-term profit. We argue that both the sustainability and impact of the coupon program could have been improved with a more focused targeting strategy. For example, better social targeting of the coupons could have included more women, particularly in the more traditional areas of the country, or other vulnerable groups. From a financial sustainability point of view, targeting new users could have also been more effective. In terms of impact, coupons could have been distributed, for example, specifically to young entrepreneurs looking for skills or access

to grow their businesses. Future user-subsidy programs should examine ways to incentivize the local center managers to design and implement targeting programs that enhance social and financial sustainability as well as overall community impact. Additionally, some broad parameters and techniques to this targeting could be developed centrally and offered to participating center managers. Even with decades of worldwide experience constructing, operating and evaluating telecenters, the research reported here makes clear that the often sited preeminent goals of the telecenter movement – socio-economic sustainability and substantial community impact – remain difficult to obtain. Further experimentation with models and approaches, along with close observation and continued independent assessment work, is required if we are to realize these ultimate ambitions. ACKNOWLEDGMENTS We thank Medina Aitieva and her students from the American University of Central Asia, members of our local research team. We also thank the three anonymous reviewers for their helpful inputs.

REFERENCES 1.

2.

3. 4.

5.

6. 7. 8. 9.

10.

11. 12.

Siochrú, S.Ó., Global Governance Of Information And Communication Technologies: Implications For Transnational Civil Society Networking. 2003, Social Science Research Council: New York. Gómez, R. and P. Hunt, eds. Telecentre Evaluation - An International Perspective : Report of an International Meeting on Telecentre Evaluation. 1999, International Development Research Centre (IDRC): Montreal. Fuchs, R., Little Engines that Did: Case Histories from the Global Telecentre Movement. 1998, IDRC: Ottawa. Proenza, F.J., R. Bastidas-Buch, and G. Montero, Telecenters for Socioeconomic and Rural Development in Latin America and the Caribbean. 2001, Inter-American Development Bank: Washington DC. Kumar, R. and M. Best, Impact and Sustainability of EGovernment Services in Developing Countries: Lessons Learned from Tamil Nadu, India. Information Society, 2006. 22(1): p. 112. Mercer, C., Telecentres and transformations: Modernizing Tanzania through the internet. African Affairs, 2006. 105(419): p. 243-264. Dagron, A.G., Prometheus Riding a Cadillac? Telecentres as the promised flame of knowledge. Journal of Development Communication: Special Issue on Telecentres 2001. 12(2). Colle, R., Memo to Telecenter Planners. Electronic Journal of Information Systems in Developing Countries, 2005. 21(1): p. 113. Roman, R. and R.D. Colle, Themes and Issues in Telecentre Sustainability, in Development Informatics Working Paper Series. 2002, Institute for Development Policy and Management, University of Manchester.: Manchester, UK. Best, M.L. and R. Kumar, Sustainability Failure of Rural Telecenters: The Sustainable Access in Rural India Project. Information Technologies and International Development, 2008. 4(4): p. 31-45. Fillip, B. and D. Foote, Making the Connection - Scaling Telecenters for Development. 2007, Washington, DC: Academy for Education Development. Harris, R.W., A. Kumar, and V. Balaji, Sustainable Telecentres? Two Cases from India. 2000, Infobridge.

200

13. 14.

15.

16.

17. 18.

19. 20.

21. 22.

23.

24.

25. 26.

27. 28.

29. 30. 31. 32.

Proenza, F.J., Telecenter Sustainability - Myths and Opportunities. Journal of Development Communications, 2001. 12. Whyte, A., Understanding the Role of Community Telecentres in Development - A Proposed Approach to Evaluation, in Telecentre Evaluation - An International Perspective : Report of an International Meeting on Telecentre Evaluation, R. Gómez and P. Hunt, Editors. 1999, International Development Research Centre (IDRC): Montreal. Heeks, R. and S. Bhatnagar, Understanding success and failure in information age reform, in Reinventing government in the information age, R. Heeks, Editor. 1999, Routledge: London. p. 49-74. Heeks, R., Understanding e-Governance for Development, in iGovernment Working Paper Series No. 11. 2001, Institute for Development Policy and Management: University of Manchester, UK. Heeks, R., Information Systems and Developing Countries: Failure, Success, and Local Improvisations. The Information Society, 2002. 18(2): p. 101 - 112. Heeks, R., Most eGovernment-for-Development Projects Fail: How Can Risks be Reduced?, in IDPM i-Government Working Paper no.14. 2003, Institute for Development Policy and Management: University of Manchester, UK. Aichholzer, G., Scenarios of e-Government in 2010 and implications for strategy design Electronic Journal of eGovernment 2004. 2(1): p. 1-10. Best, M. and C. Maclay, Community Internet Access in Rural Areas: Solving the Economic Sustainability Puzzle, in The Global Information Technology Report 2001-2002: Readiness for the Networked World. 2002, Oxford University Press: Oxford. Hudson, H.E., From Rural Village to Global Village: Telecommunications for Development in the Information Age. 2006, Mahwah, New Jersey: Lawrence Erlbaum Associates. University of Washington Center for Internet Studies. CIS Literature Review Database on the Impact of Public Access to ICT. 2008 [Retrieved: September 9, 2008]; Available from: http://references.cisresearch.org/index.php. Kuriyan, R. and K. Toyama. Review of Research on Rural PC Kiosks. 2007 [Retrieved: September 16, 2008]; Available from: http://research.microsoft.com/enus/um/india/projects/ruralkiosks/kiosks research.doc. Harris, R.W., A. Kumar, and V. Balaji, Sustainable Telecentres? Two Cases from India, in The Digital Challenge : Information Technology in the Development Context, S. Krishna and S. Madon, Editors. 2003, Ashgate: Aldershot, Hants, England ; Burlington, VT, USA. p. 124-135. Bailur, S., Using Stakeholder Theory to Analyze Telecenter Projects. Information Technologies & International Development, 2007. 3(3): p. 61-80. Jensen, M. and A. Esterhuysen, The Community Telecentre Cookbook For Africa - Recipes For Self-Sustainability. 2001, United Nations Educational Scientific and Cultural Organization: Paris. UNDP, Telecottage Handbook: How to establish and run a successful telecentre. 2006, UNDP, Europe and the CIS: Bratislava. Badshah, A., S. Khan, and M. Garrido, eds. Connected for Development: Information Kiosks and Sustainability. United Nations Information and Communication Technologies Task Force Series. Vol. 4. 2003, United Nations Information and Communication Technologies Task Force: New York. Khelladi, Y., What Works:The Infocentros Telecenter Model, in A Digital Dividend Study. 2001, The World Resources Institute: Washington DC. Tifft, J.D., Summative Evaluation - Bulgaria Public Computer and Communication Centers (PC3s). 2003, USAID: Washington DC. IBRD, Joint Country Support Strategy (JCSS) for the Kyrgyz Republic (2007-2010). 2007, The World Bank: Almaty. IBRD, Kyrgyz Republic Poverty Assessment. 2007, The World Bank: Washington DC.

33.

34.

35.

Ure, J., ICT Sector Development in Five Central Asian Economies: A Policy Framework for Effective Investment Promotion and Facilitation, in International Conference on Strengthening Regional Cooperation for Managing Globalization,. 2005: Moscow, Russian Federation. CAICT. 2007 CAICT Survey: Topline Results 2007 [Retrieved: September 5, 2008]; Available from: http://depts.washington.edu/caict/surveys/2007toplines/TopLines_ 2007.pdf. UNDP, Human Development Report 2007/2008: Fighting climate change: Human solidarity in a divided world. 2007, Palgrave Macmillian: New York.

201

A Speech Enabled Indian Language Text to Braille Transliteration System Tirthankar Dasgupta and Anupam Basu

Abstract— In this paper we present a speech enabled bidirectional automatic Indian language text to Braille transliteration system. The system allows bridging the communication gap between a visually impaired and a sighted person. The present system can be configured to take Indian language text document as input and based on some transliteration rules, can generate the corresponding Braille output. The system is augmented by an Indian language text-tospeech (TTS) system through which a user can get instantaneous audio feedback from the input text. We further extended the system to support transliteration of Dzongkha 1 text to Braille. Finally we present an Audio QWERTY editor which allows a visually impaired person to read and write Indian language texts through a computer. Index Terms— Visual Impairment, Braille Transliteration, Audio QWERTY Editor, Indian Language

I. INTRODUCTION The Braille encoding system is the primary means of representing textual documents in a readable format for the visually impaired people [1], [2]. However, due to the scarcity of Braille printed reading materials, blind people in India face a daunting task while getting their formal education and achieve a respectable employment opportunity. The situation becomes worse due to the unavailability of low-cost technological support. The National Census of India has estimated around 21.9 million disabled people in the country [3]. Out of which more than 15 million people in India are blind. This is considered to be the highest among all other disabilities. Three out of every five children in the age group of 0-9 years have been reported to be visually impaired in India. Due to their inability in accessing information from written text documents, blind people face tremendous difficulties in communicating with sighted people in common places like post office, bank and other official places where the primary mode of communication is through writing. Manuscript received September 23, 2008. This work was supported in part by Communication Empowerment Laboratory, Indian Institute of Technology, Kharagpur and Media Lab Asia, Mumbai. T. Dasgupta is with the Indian Institute of Technology, Kharagpur, West Bengal India (corresponding author to provide phone: +91 9433852325; email: iamtirthankar@ gmail.com). A. Basu is with the Computer Science and Engineering Department, Indian Institute of Technology, Kharagpur, India (e-mail: [email protected]). 1 Official language of Bhutan

In order to provide proper information access and to bridge the communication gap between the visually impaired and the sighted community, the need to build some advance technologically supported systems like automatic Braille transliteration and screen reading systems are indispensable. Several works have been done on building automatic, bidirectional text to Braille transliteration system and speech enabled interfaces for the visually impaired community [4][8]. However, most of the systems cannot be directly used for the visually impaired community of India. This is due to the following reasons 1) Most of the systems are based on foreign languages like English, French, Germany, Spanish, Portuguese, and Swedish [9]-[12]. 2) Indian language scripts are quite different from that of European or American languages. Hence, separate rules are needed to transliterate Indian language texts to Braille. 3) Foreign systems like, Duxbury [9] and JAWS [13] are costly, given the Indian economic reality of the visually impaired population, who mostly belong to the poorer section. In order to overcome the above mentioned challenges and limitations of the existing systems, we present a speech enabled Indian language text to Braille transliteration (ILBT) system. The system provides a generic framework for the transliteration of large number of popular Indian language texts to Braille. Further, the system can also be used as an Indian language document reader where a user can select a particular text document to get the corresponding speech output. This allows the present system to be used by both a sighted person as well as a blind person. We also extended our present system to provide Braille transliteration of Dzongkha texts keeping in view the proximity of Bhutan with India. Currently our system handles only four Indian languages Hindi, Bengali, Assamese and English along with Dzongkha. However, due to the unified framework of the system, any other Indian language can be easily be added to it. II. INDIAN LANGUAGE BRAILLE ENCODING SYSTEM The Bharati Braille system is the standard technique of representing Indian language texts to Braille [1]. The system uses 6 dot cells to represent each character. The combination of these 6 dots can generate 26-1=63 different Braille characters.

202

It has been observed that for all Indian languages, the corresponding Braille code is the same. In other words, a distinct Braille cell may correspond to different Indian language character. However, a single Indian language character may require one or more than one Braille cell for its representation. This process is illustrated in Table 1. Table 1: Braille representation of different Indian language characters.

Braille

Bengali

Hindi

k

ক

क

.

খ

ख

" R

ঋ

ऋ

One of the most interesting phenomena found in Indian language scripts is the usage of composite or conjugate characters. Conjugates are often constructed by the sequential concatenation of two or more characters with a special character called halant. The construction of conjugate characters follows certain rules. Some of these rules are illustrated in Table 2. Table 2: Conjugate construction rules with examples taken from Bengali language (C=Consonant, H=Halant, V=Vowel)

Rule

Example

Braille

CHC

ঙ+◌্ +ক = ক+ ◌্ + র+ ি◌ = িk স+ ◌্ + ত+ ◌্ + র=st স+ ◌্ + ত+ ◌্ + র+ ◌্ + ◌ী = stী

@+K

CHCV CHCHC CHCHCV

@KRI @S@TR @S@TR9

From Table 2 it can be observed that the conjugate characters, as constructed by clustering of consonants and vowels, may have an entirely different visual representation. However, the corresponding transliterated Braille is represented by a sequence of Braille cells for each of the characters. III. RELATED WORKS The Sparsha toolset [2] is the only system encountered by us so far that can transliterate Indian language texts to Braille. The system provides some unique features like transliteration of mathematical symbols and tactile graphics to Braille. However, this system suffers from several limitations like, 1) The system takes significant amount of time during transliteration of large texts.

2) The system does not provide any auditory feedback to the user neither it gives a very good GUI for visualizing the transliterated output. 3) Editing of the transliterated output is very difficult and the system does not provide any interface to create or edit input text document. 4) Working of the system depends heavily upon the user’s knowledge, understanding and intelligence. As a result, the system provides high cognitive load to the user which may result in increasing error rate. Apart from computer based transliteration of Braille, a Braille writing tutor system has been presented in [14]. The prototype tutor system uses an E-slate device to capture a student’s action and tries to develop the Braille writing skills. IV. THE ILBT SYSTEM ARCHITECTURE The architecture of our proposed Indian language text to Braille transliteration (ILBT) system is shown in Fig. 1. The diagram shows the essential components of the present system. There are two major goals of the system: • Providing audio feedback to the Indian language text. • Forward or reverse transliteration of text documents. The input to the system may be a text document written in any popular word processor, or Indian language texts entered through a keyboard. Based on the user’s requirement the system can generate either a speech output or can transliterate the text to Braille. The details of the major components of the system are discussed in the subsequent sections. A. The Text-To-Speech Engine A text-to-speech (TTS) engine is at the core of the speech output generation module. It is converts strings of text to their corresponding human voice equivalent. In order to develop the proposed transliteration architecture, we used a text-tospeech engine called Shruti [15] that uses diphone concatenation to synthesize speech in Indian languages using. The Shruti text-to-speech engine currently supports two Indian languages namely Hindi and Bengali. We built a wrapper around the TTS engine so that it can be used by a number of applications including our transliteration system. The wrapper also provides a number of other vital features like 1) Conversion of text in Unicode or iTrans to ISCII 2 which is the native representation format of the TTS being used, 2) Pronunciations of individual and conjugate characters. B. Input Document Type & Keyboard Input Methods In order to create an Indian language document, a text editor is required for accepting the regional language text entered through the keyboard and perform different operations on it, like formatting, printing, and saving the text. Further, it is also desirable that the editor should support different font encodings (like, Unicode, ISCII, ASCII, and iTrans) which 2

Indian Standard Code for Information Exchange

203

are User I/P Through Keyboard/Mouse

Regional Language Fonts

Keyboard Input Methods Font Definitio

Keyboard Hook

Input Key

Operating System

Rendering Engine

Mouse Hook

Glyphs Indian Language Text Documents (UTF-8, UTF-16, ISCII, iTrans, ASCII)

Forward Transliteration

Forward Transliteration Rule

TTS Engine

Reverse Transliteration

SPEECH

Reverse Transliteration Rule

Transliteration Braille Document

Embosser

Text Document

Interface

BRAILLE PRINT

TTS Engine

SPEECH

Fig. 1: System Architecture of the Speech enabled Indian Language Text to Braille Transliteration System

popularly used to create Indian language text documents. There exists no such editor that can support all the above mentioned list of font encodings together. For example MSWord 2007 supports Unicode as well as ASCII encoding and the most common editor supporting ISCII is the iLeap editor. A solution to this problem is creating a new text editor that can support all the mentioned font encodings. However, the developing a new text editor will require a substantial time and effort. Further, it expects a user to learn the new system which is very difficult especially if the user is visually impaired. Hence, in our proposed architecture we allow a user to create a text content in any of the commercially available regional language text editors available in India. However, In case of unavailability of such kind of editors, our system provides a very basic form of a regional language text editor that is integrated along with the ILBT system. Currently the editor provides some basic functionalities as compared to a full-fledged text editors like, creating a new document, opening of documents in multiple tabs, saving documents, searching regional language texts etc. C. Indian Language Font Rendering Proper Indian language fonts and the font rendering engines are necessary for producing correct visual

representation of any regional language text. This is especially important for Indian languages. Though this may not seem important for a blind user, it is actually essential for him in order to communicate with other sighted people. The OpenType font format provides a large number of features essential for supporting regional languages, especially complex scripts [16]. Due to this reason it was found to be suitable for the mentioned purpose. In OpenType fonts all information related to the proper rendering of the font may be included within the font file. GSUB (Glyph Substitution) and GPOS (Glyph Positioning) are two tables responsible for storing such information. These tables play a critical role in displaying regional language text as, a lot of character reordering, combination of characters and glyph reshaping may be necessary. Glyph substitution is the process of replacement of one or more glyphs by a new glyph like while writing conjugates. Glyph positioning refers to process of shifting a glyph to produce the proper visual representation (E.g. while writing vowels in Indian languages called matras) as shown in Fig. 4. However, though the visual representation of the text changes in the above cases the original string of characters remains unchanged. Both the above mentioned tables find extensive use especially in Indian languages

204

D. Input and Output of the system The proposed ILBT system can accept English text in the form of plain text files and Microsoft Word documents. Apart from English the Braille transliteration system, as described, can take Indian language text as input. The input text document of the system can be of any type like: • Unicode text – generated by any standard editor supporting Unicode [17]. •

ISCII (Indian Script Code for Information Interchange) [18] documents generated by applications like iLeap [19]

The output of the ILBT system can be of two types. A user can select a particular section of the text document to get the corresponding Indian language speech output, or a user can obtain the corresponding Braille output on a large variety of commercially available Braille embossers [20][21]. The current system has been tested on the following Braille embossers: • Index Basic-S •

Index Basic-D

•

Index 4X4 PRO

•

Romeo Pro 50

•

Juliet Pro

E. Forward Transliteration The input to the transliteration unit is either an English text or an Indian language text document. The text documents are either based on Unicode, ASCII or ISCII encodings. As discussed in the previous sections, different Indian language characters shares the same Braille representation. In other words mapping rules to transliterate one Indian language can be simultaneously used to transliterate any other Indian language text to Braille. Thus, in order to implement this, we build separate code tables for each of the languages and based on the input language the corresponding code table is used. An example of the code table structure is shown in Fig. 2. The code tables are stored in a hash table due to which mapping of input text to Braille takes a constant amount of time resulting in a very first transliteration output. The said method of implementation also makes the system highly scalable and allows the inclusion of more languages in future if required. Transliteration of English text to Braille is a relatively straight forward process, where there is a direct mapping between the English character and the Braille character. However, Indian language text to Braille transliteration is not as straight forward as English. Transliteration rule for a particular Indian language character depends upon its previous and the next character. This process is illustrated

in Fig. 3. From the figure we can see that if a character is preceded by a halant then during transliteration, position of the character and the halant gets swapped. This phenomenon can be represented by the rule: Precede(X, halant) Æ swap(X, halant) Another interesting feature found in Indian language text transliteration is the handling of matras. A vowel may occur either as a matra or it may appear as a distinct vowel. If the vowel occurs as a matra then a special symbol appears before the vowel else no special symbol appears. This is illustrated in Fig. 4.

0981 0982 0983 0985 0986

0027 003B 002C 0041 003E

0045 0046 0047 0048 0049

090F 0036 0917 0038 0907

a

0947 0000 0000 0000 093F

0035 0036 0037 0038 0039

b

Fig. 2: Code Table Structure of both Forward (a) and Reverse (b) Transliteration Engine

স + ◌্ + ত + ◌্ + র + ◌ী = stী @ +S +@+ T+ R+9 Fig. 3: Transliteration of Bengali conjugate characters.

ক + ি◌ = িক

ক + i = কi

.

.

I

A

I

Special Character Fig. 4: Illustration of Rule that Handles Vowels and Matras

There are number of other rules present which are required for the Indian language text to Braille transliteration. The rule files are written in such a way that a new rule can be added at any point of time without disturbing the original system.

205

Fig. 5: Screenshot of the Indian Language Braille Transliteration System

F. Reverse Transliteration The Indian language Braille transliteration system allows reverse transliteration of Braille to Indian language text document. This enables a visually impaired to communicate comfortably with other sighted people. The Braille code to be transliterated may be entered into the computer using standard editors as discussed in the latter section. After translating the Braille code into text, the text may be given as an input to the text-to-speech engine, integrated within the transliteration system as discussed before, to allow audio based feedback. The process of reverse transliteration of the Braille text is similar to that of the forward transliteration. However, there are certain ambiguities that are needed to be resolved in order to get the proper rendering of Indian language texts. For example, as discussed above, vowels and matras have got the same Braille code. Hence, in order to distinguish between a vowel and matra, Braille code of the previous character is considered. An example of a rule that resolves the ambiguity between a vowel U+0990, a matra U+09C* and an ASCII character U+002F is given below: If Prev (U+002F) is U+0041 Then Replace (U002F, U+0990) Elseif Prev (U+002F) is U+002C Then Replace (U002F,U+002F) Else Replace (U002F, U+09C8)

We construct separate rules to handle code table for each of the supported languages. Thus, our system can be easily extended to allow reverse transliteration of any other Indian language texts. V. WORKING OF THE SYSTEM The different components of the ILBT system have been discussed at length so far. In the following paragraph the working of the entire system and how each of the above mentioned components fit into the system have been summarized. As mentioned previously, the ILBT system can be used either as an Indian language text reader or it can be used to transliterate Indian language text to Braille. The system can take input either from documents created using popular text editors or the system provides its own simplified Indian language text editor which can take inputs directly from the keyboard. The keyboard input methods are integrated with operating system in order to accept text in regional languages from the keyboard. Thus characters (entered through the keyboard) in regional languages are fed into our text editor. The editor analyses the keystrokes to identify the text characters and based on the users command they are immediately sent to either the TTS engine to be spoken out or to the transliteration engine to instantaneously get the Braille output. The GUI of the system also allows a user to select a

206

Fig. 6: Screenshot of the Indian Language Audio QWERTY Editor

particular portion of text and get instantaneous speech or Braille output. A screenshot of the ILBT system is shown in Fig. 5. Table 3: Mouse click events and their different GUI operations

Mouse click

GUI operations

Left, single

NOP

Left, double

Speak word

Left, triple

Speak line

Middle single

Open dialog

Middle double

New tab document

Middle triple

Transliterate

Right single

Close document

Right double

FREE

Right triple

FREE

Apart from keystroke analysis, the GUI of the editor also captures the mouse operations as performed by the user. Instead of searching for the shortcut keys from the keyboard, a user can send the same keyboard commands by performing different mouse operations. The ILBT system generates a mouse hook process to captures different possible mouse clicks. Each mouse click operation is mapped to a certain keyboard operation. The mapping rules are customizable and

based on the users requirements, can be changed easily. Some example of the mapping rules are shown in table 3. Operating the system with mouse is particularly helpful to a visually impaired person and our experiment has shown operating the system with mouse operations are more preferred than keyboard operations for certain users. VI. BRAILLE AUDIO QWERTY EDITOR The primary goal of the audio QWERTY editor is to allow visually challenged people to create Indian language Text/Braille documents (see Fig. 6). This requires an interface for accepting the regional language text entered through the keyboard and performing different operations on it, like formatting, printing and saving the text. The creation of a new editor interface was not warranted as it would put additional burden on the user to learn the new system. Hence, we choose to use some already existing standard editor with the required capabilities. Our investigations proved that Microsoft Word XP or its higher versions can be configured to accept text in regional languages including Indian languages. Apart from Microsoft Word, the Audio QWERTY editor plug-in can be integrated to any other Unicode enabled text editors like, Notepad, and Wordpad. However, there are several reasons for choosing Microsoft Word such as: 1) Support for Unicode – This ensures that almost all natural languages of the world can be used with the editor. It also implies that a document can contain text in more than one language.

207

2) Rendering of Fonts – It uses proper rendering engines for correct rendering of regional language fonts including glyph shaping and repositioning [22], [23]. 3)

Well documented object model – the editor in question exposes a comprehensive set of objects for interacting with it, hence eliminating the need for the editor source code. This also simplified the task of programming as it allows the programmer to obtain various information including editor settings, text being edited, formatting information, etc. by just querying the editor.

4) Ease of Use – the existing popularity of this editor predicts a low learning curve for the proposed system as it allows the user to utilize past experience with this editor. Furthermore, the editor provides a large number of keyboard-shortcuts for performing many common tasks which makes the system easy to use visually impaired users. It also promotes the use of mainstream applications by blind users. One of the unique features that makes the audio QWERTY editor different from other commercially available text editors is that, the editor is integrated with the Indian language TextTo-Speech system (Shruti) as a result each of the keyboard operation performed through this editor is followed by an Indian language voice feedback. This enables it to be used by any visually impaired person to read and write Braille texts. In order to write regional language or Braille documents the system provides different keyboard layouts to the user. By selecting a particular language from the operating system, a user actually enables the keyboard layout for that particular language. This allows the user to write text documents in the selected language. A keyboard layout or an IME (Input Method Editor) allows a person to type in text in a particular language using the keyboard [24]. In other words, a keyboard layout or IME can be defined as a mapping function between different key strokes and corresponding characters produced by them. It is usually in the form of a dynamic link library and binds strongly with the operating system. The Windows operating system can maintain a number of such keyboard layouts simultaneously. It also allows the user to switch dynamically between several keyboard layouts associated with different languages. Hence the user can type in text in different languages within the same document by switching keyboard layouts. In order to type in text in a regional language like Hindi or Bengali the corresponding keyboard layout is required. Microsoft Windows provides keyboard layouts for a number of regional languages. However, IMEs are not available for a large number of other regional languages. Such IMEs can be developed with the help of Windows DDK (Driver Development Kit). In our case such an IME for Indian language Braille was developed so as to enable the user to type Braille text directly into any Windows application.

VII. TRANSLITERATION OF DZONGKHA TEXT TO BRAILLE The Dzongkha script also known as the Bhutanese script is used to write Dzongkha which is the national language of Bhutan. The Dzongkha script has 195 alphabets out of which there are thirty consonants and four basic vowels. The thirty consonants can occur twice in either nominal position or in orthographic subjoined position [25]-[27]. This establishes the fact that Dzongkha scripts can be written either from top to bottom or from left to right [25]. This is illustrated with an example in Fig. 7. Similar to the Indian language scripts discussed above, two or more consonants can combine to form a conjugate Dzongkha character. However, these consonants are not separated by any special symbols (as in the case of Indian languages where the consonants within the conjugate are separated by a special symbol called halant). The consonant clustering in Dzongkha takes place between the consonant at the nominal position and consonant at the orthographic subjoined position. This is illustrated in Fig. 8.

Fig. 7: Dzongkha Characters arranged from top to bottom taken from [25].

ང + ྐ = ངྐ ས +ྐ +ྱ +ེ = སྐྱེ Fig. 8: Construction of Dzongkha Conjugates

Fig. 9: Construction of a Dzongkha Syllable taken from [25]

Unlike any other Indian language texts, there exist no interword separation in Dzongkha. Each of the Dzongkha words are composed of single or multiple syllables. The syllables are seperated by a special symbol called tsheg. Each of the syllables contains a root letter, and may additionally contain a prefix, suffix, vowel and post-suffix. Fig. 9 illustrates this phenomena with an example.

208

Fig. 10: Working of the Dzongkha text to Braille transliteration system

based on the above mentioned information, we have constructed separate forward transliteration rules for Dzongkha text to Braille. Most of the transliteration rules are very similar to Indian language text to Braille transliteration. However, there are few exceptional cases like, handling of some special conjugates in Dzongkha script, where the transliteration rules changes. For example, Braille representation of certain Dzongkha characters like “ra”, “la” and “sa” depends upon the occurrence of the character that follows it. This process is represented by the following rule: CONJ { PREFIX { U+0f62|U+0f63|U+0f66, U+0f90|U+0F92|U+0F94 } } Æ PutBraille (53|59|57) The rule says, if the head letter of a Dzongkha conjugate begins with U+0f62, U+0f63 or U+0f66 and the root latter belongs to U+0f90, U+0F92, or U+0F94 then the Braille representation of the head characters will changed to “53”, “59” or “57”. Fig. 11 illustrates the above rule with an example. Fig. 10 shows a screenshot of the working of the Dzongkha text to Braille transliteration system. The Dzongkha Braille allows 38 characters per line and 25 lines per page. However, the system allows the user to change the configuration if needed. As mentioned above Dzongkha script does not allow any inter word space. However, the syllables and sentences are separated by the special symbols like tsheg and she. An interesting feature found in Dzongkha Braille formatting is that, each line of a Braille document must

end with either tsheg or she. This issue is handled by an inbuilt auto formatting module of the transliteration system. The auto formatting module first analyzes the transliterated Braille output and puts 38 Braille characters per line at the Braille preview window. If the last character of a line does not belong to tsheg or she then the auto formatting module starts accumulating the previous characters into an array till it gets a tsheg or she. The array elements are then printed into the next line of the preview window. This result into the fact that after transliteration is over all the lines in the Braille output does not contain 38 characters. U+0F90

U+0F62

ར

+

ར

+

U+0F62

ྐ ྑ

U+0F35

=

རྐ Æ 5 k

=

རྑ r <

U+0F91

Æ

U+0F52

Fig. 11: Two different Braille representation of the Unicode character U+0F62 in conjunction with U+0F90 and U+0F91.

VIII. FIELD TESTING AND DEPLOYMENT The ILBT system has been field-tested and deployed to

209

A. First level of Evaluation: Transliteration accuracy In the first level of evaluation our primary objective is to achieve high transliteration accuracy. The accuracy of the system has been measured manually by people working in different blind schools and organizations. We create four Braille document based on Bengali texts and three Dzongkha Braille documents. All the documents are utf-8 encoded. The specifications of the Bengali text documents are shown in table 4.

Transliteration Errors ILBT

Text type

Doc 1

400

Only Bengali text

Doc 2

1000

Only Hindi text

Doc 3

1200

Bengali & English text mixed

Doc 4

1600

Hindi & English text mixed

These Braille documents are read by a blind person and the transliteration errors are marked manually by the users. We use the existing Sparsha toolset as a lower level baseline for our system. The transliteration output as generated by the two systems is compared automatically by a program. The comparison is based on two parameters a) transliteration similarity and b) Output alignment format. Fig. 12 shows the result of the above experiment. We can observe that the transliteration accuracy of ILBT system is better than the Sparsha system. Most of the errors are occurred due to some of the special Indian language characters like U+0950 (ॐ), U+0960 (ॠ), U+095B (ज़), and U+09F3 (৳) which are not present in the Text-Braille code table.

6

6

6

4

4 3

4 2 0

1 0 400

1000

1200

1600

Fig. 12: Comparison of Transliteration Errors between ILBT system and Sparsha Toolset

B. Second level of evaluation: Transliteration Speed In the second level of evaluation our primary aim is to evaluate the transliteration time taken by the system. For this we have created 7 Indian language text documents whose sizes ranges from 25000 characters to 75000 characters without considering the space. We apply the same evaluation technique as discussed in [2]. Each of the documents was tested in an Intel Core 2 Duo 2.4 GHz CPU with 2 GB RAM and 160 GB HDD. The result is shown in Fig. 13. Estim ated Transliteration Speed (in m sec.) ILBT

Sparsha

250 Time (in msec.)

Characters

8

8

Num ber of characters

Table 4: Braille transliteration test data specification

File

Sparsha

10 Number of transliteration Error

several institutes for the visually impaired both in India and Bhutan. Some of those institutes are: 1) Jadavpur University, Kolkata 2) Voice of World, Kolkata 3) Rotary Sadan, Kolkata 4) School for the blind, Dibrughar 5) Curriculum and Professional Support Division (CAPSD) Ministry of Education, Bhutan As a result of this deployment, the system is undergoing through a constant process of enhancement. A number of feedbacks and suggestions from the visually impaired persons have lead to the development of different features like integration of Indian language text to speech, accessing the system through the mouse, and inclusion of Dzongkha language for Bhutan. Based on the user feedback we perform a three level of evaluation for the current Indian language Braille transliteration system. This includes: 1. First level of evaluation: Transliteration Accuracy 2. Second level of evaluation: Transliteration Speed 3. Third level of evaluation: Task Execution Time

200 150

186

100 50

181

135

46

62

103 71

197

117

211

123

223

135

93

0 25000 31000 40000

50000 67000 70000 75000

Num ber of Characters

Fig. 13: Comparison of the computation time taken for Indian language Braille transliteration in Sparsha and ILBT.

From the result it can be noted that the ILBT system performs considerably better than the Sparsha toolset. The reason behind such a marginal difference may be due to the usage of separate hash table for each of the Indian languages. The usage of hash table reduces the searching time complexity to a constant time as a result the overall transliteration time reduces drastically. C. Third level of evaluation: Task Execution Time In the third level of evaluation, our primary goal is to compute the task execution time using the present GUI design. We compute the task execution time using the Keystroke Level

210

Model (KLM) [28], [29]. To accomplish a given task, a user must perform certain keyboard and mouse operations. The KLM identifies these operations and assigns a timestamp value to each of them. These timestamp values are then added to get the final task execution time. The keyboard and mouse operations with their estimated time, as discussed in [28], are discussed in table5. Table 5: Keyboard/ Mouse operations with their estimated time in KLM

Operations

Time (in sec.)

Table 6: Action Sequence for “Mouse Movement and Selection” Method for ILBT System

Action Performed Move the mouse pointer to the File menu (P)

1.1

Click over the File menu (BB)

2*.1=.2

Move the mouse to the Open menu item (P)

1.1

Move the mouse to the Bengali menu item (P)

1.1

Click on the Bengali menu item (BB)

2*.1=.2 4*.1=.4 4.1

Key press and release (K)

0.28

Move the mouse to an object on screen (P)

1.1

Double click to select the file from the open dialog box (BBBB)

Button press or release (mouse), (B)

0.1

Estimated Time

Hand from keyboard to mouse or vice versa (H)

0.4

Mental Preparation (M)

1.2

Type string of n characters (T(n))

n*K

We define the task of a user as “Transliterate a text document and view the print preview of the Braille output”. To accomplish this task, the ILBT system offers the following thee different methods: • Mouse Movement and Selection • Mouse Click Operations Table 5 and Table 6 show the different action sequences and the corresponding estimated time of the Mouse Movement and selection method for the ILBT and Sparsha system. One of the key observations we made from table 5 is that the mouse pointer movement operation is taking around 0.9 sec higher time that the mouse click operations. Hence, we tried to enhance our present method of Mouse Movement and Selection so that we can reduce the number of mouse movement operations. Consequently, we came up with a solution by which a user can perform most of the operations on the ILBT system with the help of a three button mouse. The various mouse click operations and their corresponding GUI commands are explained in Table 3 of section V. We try to perform the same task of “Transliterating a text document and view the print preview of the Braille output”. This task has been performed using the Mouse Click Operations method as explained above. Table 7 presents the different actions and the time required to accomplish the given transliteration task. From Table 8 we can see that accessing the system with the help of mouse click events considerably reduces the overall execution time of a user. We summarize the third level of evaluation between the Sparsha toolset and the ILBT system in Table 9.

Time (in sec.)

Table 7: Action sequence for “Mouse Movement and Selection” method for Sparsha system

Action Performed

Time (in sec)

Move the mouse to the file menu (P)

1.1

Click the file menu (BB)

2*.1=.2

Move the mouse to the open menu (P)

1.1

Click the open menu (BB)

2*.1=.2

Double click to select the file from the open dialog box (BBBB)

4*.1=.4

Move the mouse to the language selection dialog box (P)

1.1

Click on the language bar to select language (BB)

2*.1=.2

Click OK button to select input text language (BB)

2*.1=.2

Click OK button of the message box (BB)

2*.1=.2

Move the mouse to the transliterate menu (P)

1.1

Click the transliterate menu (BB)

2*.1=.2

Move the mouse to the “text-Braille” button (P)

1.1

Click the “text-Braille” (BB)

2*.1=.2

Move the mouse to the OK button of the transliteration dialog box (P)

1.1

Move mouse to the OK button of the message box (P)

1.1

Click the OK Button (BB)

2*.1=.2

Move the mouse to the print preview menu (P)

1.1

Click the print preview menu to get the preview of Braille document (BB)

2*.1=.2

Estimated Time

11

211

Table 8: Actions sequences and estimated time for the “Mouse Click Operation” method

Action Sequence

Time (in Sec.)

Move the mouse pointer to the language bar (P)

1.1

Click on the language bar to select the language (BB)

2*.1=.2

Middle click to open file dialog box (BB)

2*.1=.2

Double click to select the file from open dialog box and see the Braille print preview (BBBB)

4*.1=.4

person. REFERENCE [1] [2]

[3] [4] [5]

Estimated Time

1.9

[6] [7]

Table 9: comparison between Sparsha and ILBT system for different task completion methods

Estimated Execution Time Methods Mouse Selection

(in Seconds)

Movement

Mouse Click Operation

and

[8] [9] [10] [11]

ILBT

Sparsha

[12] [13]

4.1

11

[14]

1.9

Feature Not Available

[15] [16]

IX. CONCLUSION The Indian language text to Braille transliteration system is an attempt to bridge the communication gap between the blind and the sighted people. Though a considerable effort has been done previously by the Sparsha system, however, all the previous attempts have provided some limited capability to the system. One of the objectives of our present ILBT system is to improve the performance of Indian language Braille transliteration and provide more features and flexibility to the system. The system can perform both forward as well as reverse transliteration of text document. This will help improve the low literacy rate and provide more information to the visually impaired community of India. The system is integrated with an Indian language text to speech system which gives an instantaneous audio feedback corresponding to a selected text. We also discuss about the different aspects of the Dzongkha script and its difference with other Indian language or English script. We have extended our system to include Dzongkha text to Braille transliteration which is the first attempt of its kind. We have evaluated our system based on three parameters and showed that the ILBT system performs much better that the existing Indian language text to Braille transliteration systems. In the next phase of our work we will develop the web version of the ILBT system and try to integrate the system with an automatic speech recognition system (ASR) so that the system can be used directly by any visually impaired

[17] [18]

[19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29]

A. Basu, S. Roy, P. Dutta and S. Banerjee. “A PC based multi-user Braille reading system for the blind libraries”, IEEE Transactions on Rehabilitation Engineering, Vol. 6, No. 1, March 1998, pp.60—68 A. Lahiri, J. S. Chattopadhyay, A. Basu, “Sparsha: A comprehensive indian language toolset for the blind”. Proceedings of the 7th international ACM SIGACCESS conference on Computers and accessibility. 2005. http://v1.dpi.org/lang-en/resources/details?page=246 P. Blenkhorn, “A system for converting braille to print”, IEEE Transactions on Rehabilitation Engineering, Vol. 3, No. 2, June 1995, pp. 215-221 HAL. Dolphin Computer Access, www.dolphinuk.co.uk/products/hal.htm C.A. Pennington and K.F. McCoy, “Providing intelligent language feedback or augmentative communication users”, Springer-Verlag, 1998. T.V. Raman (1996). “Emacspeak – a speech interface”. Proceedings of CHI96, April 1996 http://www.braille.se/downloads/winbraille.htm Duxbury Braille Translator, 2000. www.duxburysystems.com/products.asp www.indexbraille.com NFBTRANS. National Federation of the Blind, 2004, http://www.nfb.org/nfbtrans.htm MONTY, VisuAide. http://www.visuaide.com/monty.html JAWS for Window. Freedom Scientific. http://www.freedomscientific.com/fs_products/software_jaws.asp N. Kalra, T. Lauwers, D. Dewey, T. Stepleton and M. B. Dias, “Iterative design of a Braille writing tutor to combat illiteracy”, Proceedings of the 2nd IEEE/ACM International Conference on Information and Communication Technologies and Development, December, 2007. Shruti, Media Lab Asia Research Laboratory, Indian Institute of Technology, Kharagpur. www.mla.iitkgp.ernet.in/projects/shruti.html Microsoft Typography, “Specifications : overview”, www.microsoft.com / typography/ Specifications Overview.mspx Unicode. http://www.unicode.org Technology Development for Indian Languages, Department of Information Technology, Ministry of Communication & Information Technology, Government of India. Available at http://tdil.mit.gov.in/standards.htm iLeap. Centre for Development of Advanced Computing. http://www.cdacindia.com/html/gist/products/ileap.asp A. Taylor. “Choosing your Braille embosser”, Braille Monitor, October 200. Available at www.nfb.org/bm/bm01/bm0110/bm011007.htm www.brailler.com/ J. Hudson for Microsoft Typography, “Windows glyph processing : an open type primer”, November 2000, http://www.microsoft.com/typography/glyph%20processing/intro.mspx R. Rolfe “What is an IME (Input Method Editor) and how do I use it?” http://www.microsoft.com/globaldev/handson http://www.tcllab.org/events/uploads/pema-bhutan.pdf http://www.omniglot.com/writing/tibetan.htm http://www.learntibetan.net/grammar/alphabet.htm D. Kieras, “Using the keystroke-level model to estimate execution times”. Unpublished report. University of Michigan.1993. S.Card, T. P. Morn, and A. Newell. “The keystroke-level model for user performance with interactive systems”, Communications of the ACM, 23 (1980), 396-210

212

Analyzing Statistical Relationships between Global Indicators through Visualization Prabath Gunawardane∗ , Erin Middleton† , Suresh Lodha∗ , Ben Crow† and James Davis∗ [email protected], [email protected], [email protected], [email protected], [email protected] ∗ Department of Computer Science † Department of Sociology University of California Santa Cruz

Abstract—There is a wealth of information collected about national level socio-economic indicators across all countries each year. These indicators are important in recognizing the level of development in certain aspects of a particular country, and are also essential in international policy making. However with past data spanning several decades and many hundreds of indicators evaluated, trying to get an intuitive sense of this data has in a way become more difficult. This is because simple indicatorwise visualization of data such as line/bar graphs or scatter plots does not do a very good job of analyzing the underlying associations or behavior. Therefore most of the socio-economic analysis regarding development tends to be focused on few main economic indicators. However, we believe that there are valuable insights to be gained from understanding how the multitude of social, economic, educational and health indicators relate to each other. The focus of our work is to provide an integration of statistical analysis with visualization to gain new socio-economic insights and knowledge. We compute correlation and linear regression between indicators using time-series data. We cluster countries based on indicator trends and analyze the results of the clustering to identify similarities and anomalies. The results are shown on a correlation or regression grid and can be visualized on a world map using a flexible interactive visualization system. This work provides a pathway to exploring deeper relationships between socio-economic indicators and countries in the hands of the user, and carries the potential for identifying important underpinnings of policy changes.

I. I NTRODUCTION AND M OTIVATION Visualization for the purpose of providing intuitive and deeper understanding of global inequality is an important problem. Several websites supporting these visualizations using raw data are becoming increasingly popular. However, seemingly easy to understand relationships between variables visualized using line graphs, bar graphs, and scatter plots, can sometimes provide incomplete information and may even lead to misleading or erroneous conclusions. In this work, we propose using statistical tools combined with visualization to provide a deeper and more complete understanding of relationships between the global socio-economic indicators. There is a large amount of data collected across all countries annually over a range of socio-economic indicators by various agencies including World Bank [2], United Nations [26], UNESCO [7], and [18] . For example the World Development Indicators Database [2] has data that covers 225 countries and regions, spanning 40 years for more than 500 indicators. While

having more information is definitely better, understanding and visualizing this data becomes a harder problem. Many websites are utilizing this large collection of socioeconomic indicator data to visualize global inequality. Popular websites include CISEIN [25], Gapminder [9], NationMaster [17], UC Atlas [24], and WorldMapper [5]. These websites utilize a number of classic visualization techniques including line graphs, bar graphs, scatter plots, and geographic maps to allow users to view this raw data in different ways. The temporal data is almost always visualized using animation. These visualizations take the first step to allow users to investigate a variety of questions: How does one country compare with other countries in the same geographic region or with similar GDP? How are different socio-economic indicators related to each other? What policies can be implemented to improve health nationally and globally? However, the simple indicatorwise visualization of data falls short of providing a deeper understanding of associations between various indicators and countries. In this work, a team of computer scientists and sociologists have worked together to create a novel integration of statistical tools and visualization with a view to gain new socio-economic knowledge. Our goal is to leverage mostly the familiar and well-known statistical (correlation, linear regression, and clustering) and visualization techniques (scatter plots and geographic maps) to investigate deeper relationships between socio-economic indicators and countries. Is the intuitive understanding provided by raw indicator visualization supported by the results of correlation and linear regression analysis? Are the causality claims obtained through complex multi-regression models, often used in socio-economic literature, validated or contrasted by correlation or regression analysis? We view our system as a first step towards building a bridge between the simple approach of using a raw indicator visualization and the high-powered causality or other policybased models. Our system features an easy-to-use interface where the user can interactively select and visualize multiple countries and / or indicators. We have coupled it with the Globalization-Health Nexus Database [21] to analyze the relationship between various health indicators. Furthermore, we have contrasted the observations of both raw and statistical visualizations with the causal relationships between these health indicators obtained

Submission Id : 94 213

using a sophisticated econometric model by Cornia et al. [4] This integration allows us to get a much better and deeper understanding of the similarities, anomalies, and evolution of indicators and countries.

the sorted correlation matrix (right diagram of Figure 2) evokes interest by social scientists, and the resulting mapping of one of the columns of the correlation matrix on to the geographic map (Figure 3) is of great interest to all. We have also developed a user interface that allows easy selection of indicators and countries from a variety of databases and visualizations to create customized visualizations (including zooming and data mining features that allow users to gain access to detailed underlying raw or computed data) that may be helpful in analyzing the data at hand. Our main focus is to investigate whether the integrated statistical-visualization system can provide any new socioeconomic knowledge or insights. We applied our system to investigate deeper questions regarding health variables. In Section V, we present three examples of the results of our investigation. Due to simple and familiar visualizations, social and computer scientists could share and understand the results equally well to create a meaningful dialogue. Many of these investigations validated the understanding obtained through simple means, but the system produced some new and surprising results and is also helpful in quantifying the intuitive understanding.

II. R ELATED W ORK There has been considerable advances in visualizing geographic information data using a variety of novel techniques [13]. A majority of these techniques include using a combination of texture and color to create a palette that can be used to display multivariate data [14], [15], [16]. Due to challenges associated with understanding animated data, spatiotemporal geographic data has been visualized using wedges, circles, and rings [23] and mashups [27]. Distortions of geographic areas using rectangles, cartograms and a combination of cartograms with pixelmaps [19] have also been used to convey the values of socio-economic indicators. Additional efforts to visualize geographic data include geographically weighted scale varying visualization [6], diffusion-based density equalizing maps [10], and two-tone pseudo-coloring to visualize one-dimensional data [22]. [12] presents interactive feature section for identifying subspaces together with interactive hierachical clustering to assist visualization. While many of these techniques appear promising and are very impressive from a visualization standpoint, most social scientists and users are unfamiliar with these techniques and remain wary of depending on these techniques to gain a better understanding of data. Integrating a statistical model with visualization has been also explored in the literature. Carr et. al. presented a way to integrate statistical summaries with visualization by the use of linked micromap and conditioned choropleth maps for spatially indexed data[3]. The concept of using glyphs to visualize a correlation matrix has been explored in [8] . Andrienko et. al. use an iterative interactive approach to classify and identify patterns in spatial data, by using visualization and data mining [1]. Guo et. al. have presented an approach to cluster and sort large multivariate datasets based on self-organizing maps [11]. While these are general visualization toolkits, our application is more tailored towards the needs of our target audience, social scientists, and specifically intended to study country/indicator based patterns relative to each other. In the integrated geographic statistical-visualization system that we have built, we are investigating relationships between causality, simple statistical relationships between indicators and countries, and intuitive understanding as obtained through simple visualization. We have chosen to use the causality model for global health indicators recently proposed by Cornia et al. [4], that we describe in further detail in Section IV-C. In order to contrast our results with those obtained by Cornia, we have integrated our system to draw data from the GHND database [21] that has been used by Cornia et al. in their study. We have also integrated other databases including the World Bank indicators. We have introduced a visualization of correlation and regression matrix (left diagram of Figure 2), that has been used mostly by computer scientists; However,

III. V ISUALIZATION Global socio-economic indicators can be captured in a 3D volume as illustrated in Figure 1. Although one can attempt to view all the data in 3D, social scientists are much more accustomed to familiar 2D visualizations. In this work, we first describe the typical visualizations associated with the 1D and 2D of this 3D volume.

Fig. 1. The 3-dimensional volume of indicator (I), country (C)and time (T) data, with (i) a vertical 2D slice highlighted which shows times series data for all indicators for a specific country, (ii) a horizontal 2D slice showing time series data for a single indicator over all countries, and (iii) a vertical 2D slice for all indicators for all countries at a specific time.

The three 1D slices of the 3D volume of global socioeconomic data are C-slice (one indicator, one time, all countries), T-slice (one indicator, once country, all time), and I-slice (one country, one time, all indicators). The C-slice is typically visualized by mapping the indicator values on a geographic 2

Submission Id : 94 214

map using pseudo-color and is perhaps one of the most popular geographic visualizations. The T-slice is commonly visualized as a time-series line graph. The I-slice is usually presented as a table. In addition to these 1D slice visualizations, visualization of 1 12 D data, that is 2 or more layers of these 1D slices, is very common. We will refer to two layers of C-slice as a 2C-slice. A 2C-slice may represent two indicators, one time, and all countries or one indicator, two times, and all countries. In the first case, the data is ideal for visualization on a geographic map using bivariate display techniques, although there is no one commonly accepted technique except perhaps for side-by-side display of two geographic maps. In the second case, although animation is commonly accepted, technique of small multiples is often employed in practice where two static images are displayed side-by-side. Gapminder [9] has developed a technique of visualizing a 4C-slice of 4 indicators using a scatter plot where 2 variables are mapped on the x-axis and y-axis and two additional variables are depicted through glyph size and glyph colors. A 2T-slice may represent two indicators, all times, for one country or one indicator, all times, for two countries. This 2T slice is typically visualized using line graphs or bar graphs. An NT-slice can also be visualized in similar ways within the space constraints. A 2I-slice may represent one time, two countries and all indicators or two times, one country and all indicators. The purpose of these 2 slices are very different. In the first case, the goal is to compare the two countries, while in the second case, the goal is to examine all the indicator trends for the same country. This data is typically presented in a tabular format or if a subset of indicators is chosen, then this subset can be visualized using classical visualization techniques including line or bar graphs.

countries to support or contrast the findings based on statistical or regression analysis. In addition, we have also supported multiple overlaid scatter plots (see Figure 9) where users are allowed to pick individual or some group of countries. 3) IC Slice for a Time Period: The Indicator-Time slice represents all the indicators across all the countries for a given year (Figure 1 (iii) ). We are not aware of any effective way of visualizing the whole 2D slice of this data. Again, it will be useful to reduce the dimensionality of indicator space. As we will see soon, statistical tools allow us to quantify the relationship between two rows or columns of the given volume of data and visualize them providing us with better understanding of relationships between the indicators or the countries. IV. S TATISTICAL T OOLS A. Correlation In this work, we utilize correlation in at least two ways – to compute correlation coefficient between two indicator trends for a given country (shown as A in Figure 1) and to compute correlation coefficient between two countries for a given indicator trend (shown as B in Figure 1). The first approach is useful if we want to analyze how two given socioeconomic indicators have varied over time with relation to each other. We compute their correlation for each country, which gives us a single correlation value per country which is visualized on a geographic map. This analysis can be used to answer questions such as ’Is an increase in immunization always correlated with a decrease in infant mortality?’ In the second analysis, we compute the correlation trends (over a period of time) between countries for a single indicator. Since each country pair would have a correlation value, we visualize these results using a correlation matrix (see Figure 2). A cellrow,column in the matrix represents the correlation between the indicator trend of a countrycolumn and the indicator trend of a countryrow . A single column of this matrix corresponds to correlation of the indicator trends of a specific country with the indicator trend of all the other countries. This analysis can be used to cluster countries based on indicator trends. We can get a better understanding of how close those countries match up by sorting and visualizing the correlation matrix by that particular country (right diagram of Figure 2). This analysis can also be used to determine whether a particular indicator trend, for example, the increase in life expectancy, has been uniform in all parts of the world, and allows easy identification of anomalies (see Figure 3). We use Pearsons product-movement correlation coefficient [20] as our correlation estimator. The correlation coefficient gives a measure of positive and negative linear correlation, ranging from +1 to -1.

A. Slice Visualization 1) TC Slice for an Indicator: The Time-Country slice (Figure 1 (ii)) represents the data for a single indicator spanning all countries over a period of time. Most websites visualize this slice using an animation of a world map where countries are pseudo-colored based on indicator values for that point in time. 2) TI Slice for a Country: This slice (Figure 1 (i)) is useful in understanding the evolution of socio-economic trends within a country. The full 2D slice includes all indicators and is difficult to visualize. A useful task in this case would be to reduce the dimensionality of socio-economic indicator space by identifying a subset of key indicators for a chosen country. This would mean restrict oneself to a few rows of the TI slice, typically by choosing specific indicators. These indicator trends are then visualized with line graphs and bar graphs. One can also investigate relationships between a pair of indicators for a specific country by employing a scatter plot by graphing one along the x-axis and the other one along the y-axis at different time periods. In this work, we have used this type of scatter plots (Figures 6, 7, and 8) for individual

B. Regression In addition to computing the correlation between two indicator trends for a country, we have also computed the linear regression fit for these indicator trends by taking one of the 3

Submission Id : 94 215

Fig. 2. (left)Correlation matrix for LEB (Life expectancy at birth) for years 1980-2005 between 207 countries; (right) The same correlation matrix sorted by the column for Sweden, indicating correlation of other countries with Sweden for LEB from 1980-2005.

Fig. 3. Correlation of other countries with Sweden for LEB from 1980-2005 shown on a world map.

indicators to be the independent and the other the response variable. yi = β0 + β1 xi + εi ,

consumption, smoking rates, unbalanced diet, migration rate, DPT immunization rate, wars, disasters, etc. Impact of these independent variables are studied on a cluster of health variables including u5MR (infant mortality under 5), IMR (infant mortality rate), and LEB (life expectancy at birth). To improve the goodness of fit, improve the robustness of the estimates, and avoid multi-collinearity problems, some variables were dropped, normalized or modified. One such variable is log (physicians/1000 people) which was divided by log (GDP per capita) to obtain an index of availability of distribution of health personnel relative to the GDP/c norm. The estimation was carried out for all the countries together, and also for four different groupings of countries – high income, middle income, low income, and transitional countries and for two different time periods, 1960-2005 and 19802005. Obtained results were examined for their statistical significance better than 1%, between 1 to 5%, between 10 to 15%, and not significant. Results most relevant to our work include statistically significant dependence of u5MR on DPT immunization rate for all the countries as well as for all the four subgroupings of the countries mentioned above and the dependence of LEB on Log (Physicians/1000 people)/ Log(GDP/c). In this work, we chose

where yi is the dependent or response variable, xi is the independent variable and εi is the residual. One would expect highly correlated indicators to lead to a good linear regression fit and the regression coefficients β1 and β0 (which is also referred to as the intercept and the slope in the case of linear regression) can be used to understand the relationship between the two indicators. Together with the correlation visualization for the two selected indicators, we also provide a visualization of the regression coefficients on a geographic map. C. Causality Recently, Cornia et al. [4] proposed a causality model for global health indicators investigating five different impact pathways for health. These pathways are material deprivation, progress in health technology, acute psychological stress, unhealthy lifestyle pathways, and socio-economic hierarchydisintegration. Each of these pathways are measured by a cluster of socio-economic indicators that include income, income inequality, unemployment rate, inflation rate, illiteracy rate, health expenditure, number of physicians, alcohol 4

Submission Id : 94 216

Fig. 4. Correlation coefficients between U5MR (under 5 mortality rate) and DPT immunization rate for years 1960-2005. This world map depicts that U5MR is negatively correlated with DPT imm. for most countries as expected. Anamolous countries, such as Germany, Kazakhastan, and Congo are easily detected in this visualization

Fig. 6. Scatter plot between U5MR vs DPT for Congo, from 1960 to 2000. Deviation from the norm is due to war.

Fig. 5. Under 5 mortality rate at zero intercept, after linear regression between U5MR and DPT immunization rate shown only for high and low income countries from 1960 - 2005. This map brings out relatively high u5MR for low income countries at comparable level of DPT immunization.

Fig. 7. Similar scatter plot for Germany. Deviation from the norm is due to variation in health polivy during 1980-2005.

Fig. 8. Similar scatter plot for Kazakhstan. Deviation from the norm is due to political changes.

Fig. 9. Scatter plot of u5MR vs. DPT showing 9 countries, 3 from each income group. These scatter plots reaffirm the general clustering og high, middle, and low income countries into three separate clusters, characterized by low, middle, and high u5MR at comparable DPT levels.

5

Submission Id : 94 217

to focus on these 4 variables – u5MR, DPT immunization rate, LEB, and Log (Physicians/1000 people)/ Log (GDP/c) together with GDP data for classification of countries. D. Clustering Classification and clustering of countries and indicators based on similarity is a common and useful endeavor. Existing solutions deal mainly with classifying the countries based at a fixed point in time, an example would be the world bank classification of countries in to ’high’,’middle’ and ’low’ income groups. In our system, we can provide results of clustering, using K-means algorithm, on any socio-economic indicator such as life expectancy and immunization rates. Also we allow clustering over a period of time, based on the correlation results of indicator trends. For example, if we could cluster countries that had an life expectancy trend similar to Sweden for the period 1980-2000. This allows grouping of countries with similar characteristics over a period of time, as opposed to just a single year. In most of our examples discussed later, we have clustered countries into four categories, although our system allows choosing the number of clusters.

Fig. 10. Correlation coefficients between LEB (Life expectancy at birth) and log(Physicians per 1000 people)/log(GDP per capita) for years 1960-2005. This correlation is positive for most countries including high income countries as expected.

V. A NALYSIS AND V ISUALIZATION Fig. 11. Countries with negative correlation coefficients for LEB (Life expectancy at birth) vs log(Physicians per 1000 people)/log(GDP per capita) for years 1960-2005. Most countries with negative correlation are erstwhile Russian block countries and a few African countries.

We now present three examples to illustrate how the integration of visualization with statistical tools have provided us with valuable socio-economic insights. All our examples draw from the highly reliable GHND database of socio-economic indicators [21]. We have chosen to focus primarily on health indicators so that we can contrast or validate our results against those obtained by Cornia et al. [4], which was described previously. For this work, we have chosen a subset of these indicators and variables to illustrate the utility of our integrated visualization-statistical tool. For health indicators, we have chosen u5MR (Infant mortality rate under 5) and LEB (Life expectancy at birth). For independent variables that impact health, we have chosen DPT imm (DPT immunization rate) and Log (Physicians/1000 people)/ Log (GDP per capita). In addition, we have also used GDP per capita. Most of the data is available for 137 countries for 204 indicators over the time period 1960 to 2005.

Fig. 12. Countries with negative regression slopes for LEB (Life expectancy at birth) vs log(Physicians per 1000 people)/log(GDP per capita) for years 1960-2005. These are the same set of countries as the countries with negative correlation coefficients. Correlation and regression analysis agree with each other.

Fig. 13. Causal coefficients between LEB and log(Physicians per 1000 people)/log(GDP per capita) obtained by Cornia et al. Causal relationship yields a surprising negative relationship for high income countries between the variables, which is counterintuitive.

A. Statistical Visualization: Anomalies and Similarities In this example, we focus on validating how correlation analysis and visualization may be helpful in analyzing relationship between indicators. To this purpose, we chose to explore the relationship between u5MR and DPT imm for all the countries. Correlation between these two indicators are computed for all the countries individually for a time period of 1960-2005. This correlation coefficient is then visualized on the world map in Figure 4. This map clearly brings out that there is a strong negative correlation between the two variables, as expected, for most of the countries, with few exceptions. This figure validates the common working assumption that an increase in immunization reduces u5MR. 6

Submission Id : 94 218

Anomalies in the relationship between u5MR and DPT imm is also brought out by Figure 4. These anomalies appear as negative or close to zero correlation for some countries. These countries include Congo, Germany, and Kazakhstan. Scatter plots of relationships between u5MR and DPT for these 3 countries are shown in Figures 6, 7 and 8 respectively. Reversal or decrease in DPT immunization rate in Congo from 1990 to 2005 is a result of war. Reversal of decrease in DPT imm rate between 1990 to 2000 in Germany is due to a variation in health policy that has been checked since 2000 resulting in continuance of the desirable trend. Finally, the increase in u5MR in Kazakhastan from 1990 to 2005 is due to political changes in the country. In summary, the correlation visualization on the world map quickly leads us to anomalies; supporting scatter plots quickly helps us in validating the anomalies and leads us to causes of these anomalies and points towards possible challenges or recommendations for changes in health policy. We now examine the relationship between the same variables, u5MR and DPT, using linear regression between the two variables. After a linear fit, we compute the y-intercept, that is, level of u5MR at a hypothetical zero DPT level. These levels of u5MR are then visualized only for the high and low income countries (excluding the middle income countries) in Figure 5. This visualization brings out the sharp contrast between the two groups of countries. This observation is further validated by picking 3 sample countries from each of the three groups – low, middle, and high income – and then visualizing the relationship between u5MR and DPT on a scatter plot in Figure 9. This supporting visualization using raw numbers further validates the observation that the low income countries are typically clustered towards the high range of u5MR and also saturate at higher levels of u5MR than the middle or high income countries. This observation leads to the conclusion that DPT can help reduce u5MR only up to a certain point in low and middle income countries and additional health measures need to be undertaken to reduce u5MR further. Although this observation may seem obvious after these visualizations, the causality model described by Cornia et al. [4] focus mostly on the regression slope and not making any of the observations listed above since their multi-variable regression model does not accommodate the simple intercept view of linear regression. Nevertheless, it is to be noted that most users, when browsing raw data using popular websites such as Gapminder and UC Atlas are intuitively looking for simple relationships between variables and the closest statistical analogs are typically correlation and regression analysis. In the examples discussed so far, simple visualizations including scatter plot, correlation, and regression visualization go a long way to provide valuable information regarding the relationship between these variables.

(Physicians per 1000 people)/ Log (GDP per Capita) over the period 1960-2005. We first discuss the derivations of the causality model regarding the relationship between these variables. Cornia et al. [4] derive that the regression coefficient between these two variables for middle, low, and transitional (Eastern block countries) are 11.2796, 14.2350, and 8.6528, being significant at 1% level for middle income countries and being significant between 1% to 5% level for low income and transitional countries. The relationship between these variables is also significant at 1% level for all the countries together with even higher regression coefficient of 36.89. Surprisingly, the regression coefficient between these two variables is negative, 28.9, also significant at 1% level. These regression coefficients are visualized in Figure 13, where the negative regression coefficient is mapped to the red color, while the other three coefficients are mapped to yellow, light green, and dark green in increasing magnitude of the regression coefficient. These causality results are in contrast with the correlation coefficients visualized for all the countries in Figure 10. The dark green colors in Figure 10 illustrate that the relationship between the two variables are positive, as expected, that is increasing the number of physicians (compared to GDP per capita) ’results’ in an increase in LEB. While the causality model in Figure 13 points to a hypothesis that in a multi-regression model, the overall increase in LEB attributed to other factors such as Log GDP/volatility, female education, alcohol consumption, and smoking, etc. is in fact offset by physicians to bring the model in line with the rest of the countries. This example illustrated the utility of statistical visualization in bringing deeper understanding and checks against the more sophisticated but harder to understand multi-regression causality models. Correlation and regression computations and visualizations bring further insight into the relationship between the two variables. Figures 11 and 12 show the countries with negative correlation coefficients and those with negative regression coefficients respectively. In this case, we observe that correlation and linear regression results agree with each other strongly. Furthermore, the negative relationships between the two variables are present predominantly for transitional (Eastern block countries). This is, again, a surprising result, since the causality model by Cornia et al. [4] computed a positive regression coefficient for these countries with high statistical significance. These visualization based observations lead us to believe that the relationship between these two variables is more complicated than a simple causal one and requires further investigation. C. Clustering We now present our third and final example, using clustering, to illustrate the utility of integrating statistical tools with visualization. To this purpose, we classified countries into four categories using many different indicators. Figure 14 presents the visualization of countries classified into low (red), lower middle (yellow), higher middle (light green), and high income (dark green) countries based on GDP in the

B. Correlation, Regression, and Causality We now present a second example of relationship between LEB (life expectancy at birth per 1000 children) and Log 7

Submission Id : 94 GDP(T)

u5MR

u5MR(T)

LEB

LEB(T)

DPT

DPT(T)

Phy

Phy(T)

Brazil Russia India China

GDP

219

2 2 1 1

1 3 1 1

1 1 2 1

1 2 1 1

2 2 3 2

1 4 1 1

1 1 3 1

4 4 3 3

1 1 2 1

1 1 1 1

as compared the trend of Sweden. Results of comparison of BRIC countries are presented in a table. The table along with the visualization show that India is slightly behind other BRIC countries in u5MR while Russia is slightly behind in the u5MR trend. In LEB, again, India is slightly behind other BRIC countries while Russia is slightly behind in the LEB trend. Put together, in the two health indicator trends, u5MR and LEB, India lags behind other BRIC countries, but can catch up if it maintains its trend, while Russia is at the greatest risk of falling behind in the health indicators. In terms of action or independent variables that impact health, DPT immunization rates for all BRIC countries are in the lowest two categories, Brazil and Russia being the lowest, and India and China next to the lowest. However, with respect to the DPT trend, Brazil, Russia, and China are in the strongest category (green), while India is somewhat lagging behind. In terms of the presence of physicians trends, all four BRIC countries are in the strongest category, while India slightly behind in the year 2000. Put together, all BRIC countries are likely to improve their health indicators by increasing the DPT immunization rate and need to maintain their strong growth trend regarding physicians.

TABLE I T HE RESULTS OF USING KM EANS TO CLUSTER COUNTRIES FOR A SPECIFIC YEAR (2000) AND ALSO BASED ON TRENDS FOR A SPAN OF YEARS (1980-2005). C LUSTERS 1 THROUGH 4 REPRESENTS THE ’ BEST ’ TO ’ WORST ’ CLASSIFICATIONS RESPECTIVELY.

Fig. 14. Clustering of countries based on GDP in year 2000

VI. C ONCLUSIONS AND F UTURE W ORK In this work, we proposed an integration of statistical computing with visualization to glean deeper understanding of global socio-economic indicators. We utilized correlation and linear regression to quantify relationships between pairs of variables and between pairs of countries. We utilized these tools to investigate static data for a fixed time period as well as dynamic trends over a large time period. Current state-of-the-art global inequality websites provide visualization support using raw data without the use of any statistical tools. Using three different examples, we demonstrate that correlation, linear regression, and causality models can bring out similarities and anomalies and provide better understanding of relationships between the variables by validating our intuitions based purely on raw data visualization or sometimes yields insights that are counter-intuitive or surprising. These observations or conclusions carry important implications in policy making both at national and global level. This research has opened up several new exciting opportunities. Which countries can be grouped together? Based on which indicators? Which socio-economic indicators can be clustered together? Can we reduce the dimensionality of indicators so that a profile of a country is captured by some principal socio-economic indicators? What lessons can a nation learn from a similar group of nations? Ideally, we would like to build a system so that the empowered users can explore relationships between countries and between variables using appropriate statistical tools combined with visualization. We believe that this exploration can always be used to validate or contrast the proposed policy decisions and may also lead to important underpinnings of national or global policy decisions that are not immediately obvious.

Fig. 15. Clustering based on GDP trends between Sweden and other countries for years 1980-2005.

year 2000. We will follow the results for the emerging BRIC (Brazil, Russia, India, and China) countries. In this case, India and China are still low income countries while Brazil and Russia are lower middle income countries. However, when we view the classification of GDP trends for the four BRIC countries over the period 1980-2005 in Figure 15, we observe that India, China, and Brazil are classified into the same (and strongest) category as most of the developed nations including USA, and most European countries, while Russian GDP growth is the next lower category. For the purposes of the trend classification, we computed the correlation between GDP trends between all the countries and Sweden. We chose Sweden because it was consistently in the top for most of the indicators that are we investigating in this study, including GDP, LEB, etc. Figures 20 to 23 classify all the countries on the four variables – u5MR, DPT imm, LEB, and Log (Physicians per 1000 people)/ Log (GDP per Capita) using the static data from the year 2000. Figures 16 to 19 classify the countries based on the trends of these four variables over the period 1980-2005 8

Submission Id : 94 220

Fig. 16. Clustering based on U5MR trends between reference country (Sweden) and other countries for years 1980-2005.

Fig. 17. Clustering based on DPT immunization trends between reference country (Sweden) and other countries for years 19802005.

Fig. 18. Clustering based on LEB trends between reference country (Sweden) and other countries for years 1980-2005.

Fig. 19. Clustering based on log(physicians per 1000 people) over log(GDP per capita) trends between reference country (Sweden) and other countries for years 1980-2005.

Fig. 20. Clustering of countries based on U5MR in year 2000

Fig. 21. Clustering of countries based on DPT immunization rates in year 2000

Fig. 22.

Fig. 23. Clustering of countries based on log(physicians per 1000 people) over log(GDP per capita) for year 2000

Clustering of countries based on LEB in year 2000

9

Submission Id : 94 221

VII. ACKNOWLEDGEMENTS

[23] P. Shanbhag, P. Rheingans, and M. desJardins. Temporal visualization of planning polygons for efficient partitioning of geo-spatial data. IEEE Symposium on Information Visualization, 2005. [24] UCSC. UC atlas. Website, 2008. http://ucatlas.ucsc.edu/. [25] C. University and W. Bank. Global poverty mapping project. Website, 2008. http://www.cisein.org/povmap/atlas.html. [26] UNSCB. United nations common database. Website, 2008. http:// unstats.un.org/unsd/cdb/cdb help/cdb quick start.asp. [27] J. Wood, J. Dykes, A. Slingsby, and K. Clarke. Interactive visual exploration of a large spatio-temporal dataset: Reflections on a geovisualization mashup. IEEE Transactions on Visualization and Computer Graphics, 13(6):1176–1183, 2007.

We would like to thank Brian Fulfrost for providing us with valuable feedback at various stages of this project. R EFERENCES [1] G. L. Andrienko and N. V. Andrienko. Data mining with C4.5 and interactive cartographic visualization. user interfaces to data intensive systems. G. T. Los Alamitos, CA, IEEE Computer Society, pages 162– 165, 1999. [2] W. Bank. World development indicators. Website, 2008. http://www. worldbank.org/data/. [3] D. B. Carr, J. Chen, B. S. Bell, L. Pickle, and Y. Zhang. Interactive linked micromap plots and dynamically conditioned choropleth maps. In Proceedings of the 2002 Annual National Conference on Digital Government Research, pages 1–7. Digital Government Society of North America, 2002. [4] G. A. Cornia, S. Rosignoli, and L. Tiberti. Globalisation and health: impact pathways and recent evidence. In Proceedings of Conference on Mapping Global Inequality, 2007. [5] D. Dorling, A. Barford, and M. Newman. Worldmapper: The world as you’ve never seen it before. IEEE Transactions on Visualization and Computer Graphics, 12(5):757–764, 2006. [6] J. Dykes and C. Brunsdon. Geographically weighted visualization: Interactive graphics for scale varying exploratory analysis. IEEE Transactions on Visualization and Computer Graphics, 13(6):1161–1168, 2007. [7] U. I. for Statistics. Global statistics. Website, 2008. http://www.uis. unesco.org. [8] M. Friendly. Corrgrams: Exploratory displays for correlation matrices. The American Statistician, 56:316–324, November 2002. [9] Gapminder. Gapminder world 2006. Website, March 2008. http://tools. google.com/gapminder. [10] M. T. Gastner and M. E. J. Newman. Diffusion-based method for producing density-equalizing maps. Proceedings of the National Academy of Sciences, 101(20):7499–7504, 2004. [11] D. Guo. Coordinating computational and visual approaches for interactive feature selection and multivariate clustering. Information Visualization, 2(4):232–246, 2003. [12] D. Guo, J. Chen, A. M. MacEachren, and K. Liao. A visualization system for space-time and multivariate patterns (vis-stamp). IEEE Transactions on Visualization and Computer Graphics, 12(6):1461– 1474, 2006. [13] D. Guo, M. Gahegan, A. M. MacEachren, and B. Zhou. Multivariate Analysis and Geovisualization with an Integrated Geographic Knowledge Discovery Approach. Cartography and Geographic Information Science, 32(2):113–133, 2005. [14] H. Hagh-Shenas, S. Kim, V. Interrante, and C. Healey. Weaving versus blending: a quantitative assessment of the information carrying capacities of two alternative methods for conveying multivariate data with color. IEEE Transactions on Visualization and Computer Graphics, 13(6):1270–1277, 2007. [15] C. G. Healey and J. T. Enns. Large datasets at a glance: Combining textures and colors in scientific visualization. IEEE Transactions on Visualization and Computer Graphics, 5(2):145–167, 1999. [16] V. Interrante. Harnessing natural textures for multivariate visualization. IEEE Computer Graphics and Applications, pages 6–11, NovemberDecember 2000. [17] Nationmaster. Nations of the world. Website, 2008. http://www. nationmaster.com/. [18] OECD. Organisation for economic co-operation and development. Website, 2008. http://www.oecd.org. [19] M.-C. Panse, M.-M. Sips, M.-D. Keim, and S. M.-S. North. Visualization of geo-spatial point sets via global shape transformation and local pixel placement. IEEE Transactions on Visualization and Computer Graphics, 12(5):749–756, 2006. [20] J. L. Rodgers and A. W. Nicewander. Thirteen ways to look at the correlation coefficient. The American Statistician, 42(1):59–66, 1988. [21] S. Rosignoli, L. Tiberti, and G. A. Cornia. The globalization-health nexus database (ghnd). Website, February 2007. http://www.unifi.it/ dpssec/sviluppo/database.html. [22] T. Saito, H. N. Miyamura, M. Yamamoto, H. Saito, Y. Hoshiya, and T. Kaseda. Two-tone pseudo coloring: Compact visualization for onedimensional data. Proceedings of Information Visualization, 0, 2005.

10

222

ATMosphere: A System for ATM Microdeposit Services in Rural Contexts Michael Paik New York University [email protected]

Lakshminarayanan Subramanian New York University [email protected]

Abstract—This paper describes strategies to lower the cost of providing Automated Teller Machine microdeposit services in rural contexts. Microdeposits represent a growing market in the developing world, but the cost of running a conventional ATM network is prohibitive due to the capital investment required to deploy networks and terminals. Our novel contributions are to use the Short Message Service (SMS) over high-penetration GSM cellular networks in conjunction with a system using location awareness to intelligently distribute available balances among machines. This allows us to provide high levels of service while simultaneously reducing risk to the financial institution and lowering per-transaction cost. Using a simulation of ATM usage patterns and distributions, our primary results under our model are: (1) transaction cost per user per year can be optimized to less than USD 0.18 given an SMS loss rate of approximately 10% while (2) customer withdrawal success rate can be maintained at approximately 98% with (3) a maximum of 5% of funds on deposit available in cash in ATMs at any given time. These results make wide deployment of rural ATM services by financial institutions feasible and economically viable in the near term using existing commodity technology. Index Terms—Finance, Simulation.

R

Financial

data

processing,

1 INTRODUCTION

EADY ATM access at the village level has been an open problem in the microfinance field, as some means for making deposits and withdrawals of small amounts of capital is necessary for the inception and growth of microfinance markets and personal accumulation of wealth for the rural poor. However, for-profit institutional financial entities have found the idea of providing such access to populations living outside of urban areas with both high teledensity and population density unattractive because the amount of capital represented by rural microdepositors is only institutionally significant once aggregated, and the cost of providing access to these customers has historically been far greater than any revenue gained by doing so. This is particularly true with the high capital investment required for traditional brick-and mortar retail banks, which also incur the cost of trained staff. More recently, systems such as MPESA[27], operated by Safaricom in Kenya have made person-to-person payment using mobile phones viable.

However these, while providing a convenient method to send funds from the city to more rural areas, do not provide significant capacity for savings as they are intended for person-to-person money transfer rather than the accumulation or retention of wealth. This paper discusses findings from a simulation of ATM transaction behavior using real-world population and geographical data from Arua, Moyo, Yumbe, and Adjumani Districts in Northwestern Uganda, bordering Southern Sudan and the Democratic Republic of the Congo, both of which have been in political upheaval and/or civil war during the past decade. This area was selected because 1) the population represents the poorest of the poor, being made up primarily of refugees from conflicts in bordering nations who have fled to an otherwise sparsely populated region[11] and 2) GSM mobile coverage of the area is relatively dense[23] as seen in Figure 2. As such, this milieu is a real-world example of settlement patterns among the most disenfranchised. The results presented here focus on satisfying a high level of consumer withdrawal demand without imposing artificially low withdrawal limits. We present a novel scheme for balance partitioning and message flow control which makes use of location awareness of ATMs alongside other optimizations to provide a solution that is performant both in terms of risk to the institution and service to the end customer. 2 THE PROBLEM 2.1 The Economic Challenge While the precise cost breakdown of deploying an ATM in a rural environment will vary from context to context, a significant part of the marginal cost overhead is in the creation and use of the network to support communication between the managing financial institution and the ATM. There is little in the way of existing network infrastructure (including wired telephone service) and the cost involved in laying even a single link given high costs of materials, transportation, and scarcity of skilled labor, construction equipment, and telecoms equipment is often prohibitive. The traditional wired model (used e.g. in the United States) is infeasible due to the capital investment required to lay network cables to sparsely populated regions. This is also a

223

Fig. 1. Population centers in northwestern Uganda

primary factor in the proliferation of self-contained cellular telephone towers which uplink via satellite or fixed wireless transmission. Other than the existing wired model, there have been efforts to connect ATMs to financial institutions using long-distance wireless links[6] which also provide telephony and internet services, but the case for capital investment of this sort for areas in which cellular telephone service already exists is more difficult to make as some degree of connectivity is already present. Any solution which proposes to address this problem must take into account lack of network capacity as well as keeping cost of deployment and operation low enough to be practicable for national banks in the region to afford. We use the wired ATM metaphor rather than simpler kiosks which read smartcards with encoded balances or paper passbooks because these methods are vulnerable to counterfeiting, theft, loss, and damage and have poor provisions for centralized auditing. 2.2 The Scope of our Approach The approach presented in this paper is aimed squarely at reducing cost and making maximum use of existing capacity in the form of the installed base of GSM network cells. As the network’s reliability, latency and bandwidth are not subject to any guarantees, we create a protocol which is designed to be resilient against extended outage conditions and slow message delivery while simultaneously maintaining an extremely low message budget and high quality of service. 2.3 The Technical Challenge The use of the installed GSM base is a major benefit to the design of a solution to this problem, but there are

Fig. 2. GSM network coverage in northwestern Uganda (darker areas indicate better coverage)

underlying difficulties, not least of which is cost. A naïve approach to the problem might be to send a series of messages back and forth between the ATM and the financial institution – one to request a withdrawal, one to issue an identity authorization challenge, another to supply the response, yet another to issue an approval, one to acknowledge the approval, etc. etc. While the absolute cost of these messages is not extraordinarily high, the cost does not scale well as the user base increases. If the institution pays for the messaging, the cost eats into any financial gain the client may realize from interest on his savings if he is expected to pay, and in either case this approach does not scale well as loss rates may be expected to increase if message traffic increases substantially while network capacity remains fixed. 3 DESIGN Our initial objective is merely to provide some form of ATM service to rural customers with small amounts of personal savings. Our simulated testbed contains 46 population centers of sizes varying between 2001 and 70872, totaling 637952 in total. We simplify this by considering households as accountholders, using the average household size across Uganda of 4.7, yielding 136541 households if we consider each center separately and round down. Having ruled out wired infrastructure, our next observation is that long-distance[10] wireless is feasible as a physical network layer, but has the disadvantage of requiring potentially extensive infrastructure deployment depending on the geography of the region, distances between nodes, and the network topology necessary to scale with growth of the size of the network. Newer

224

technologies such as WiMAX[30] may allow connectivity between adjacent villages with commodity hardware, but how much effective bandwidth a rural WiMAX mesh installation will have is still an open question, and there is still the question of deployment lead time and network maintenance costs. GSM cellular technology is widespread in the developing world, and reaches into some of the most sparsely populated regions. While the General Packet Radio Service (GPRS) is part of GSM Release 97 and newer (colloquially referred to as 2.5G and 3G), it is not a viable option for data transport as much of the installed base in rural areas is comprised of cheaper 2G cellular hardware. Due to these factors, we choose to use SMS. Once connectivity is established, a natural design principle for our system is that we want to reduce network traffic as much as possible, as this represents a marginal cost to our system, and cost is precisely what we wish to reduce. In opposition to this is a desire to maintain overall responsiveness of the system as perceived by the customer. This perception is directly related to satisfaction rate of withdrawal requests, and tied only weakly to the immediacy of funds availability from deposits, as existing microdeposit schemes often take up to a month[13] to register updated balances. A final critical design principle is that, given the intermittent network communication enforced by our traffic reduction, we need to intelligently distribute balances in such a way as to prevent any user from being able to game the system by visiting multiple ATMs within a given update window and withdraw more than his account balance.

techniques currently used in other devices, such as credit card swipe readers. A possible solution to this problem is to use a similar model to the M-PESA system and locate ATM kiosks within vendors’ shops, allowing the vendors to charge a small fee for their use. We assume that any losses incurred through physical attacks on the machine are covered by the financial institution, as with machines in urban locations. In the same vein we neglect considerations about powering the machine (solar is an option for low-powered embedded computers), how deposits will be marked, validated, and conveyed to the host institution, and the technical details of the tokens used to identify the users of the machine other than that these tokens contain some cryptographically secure writable medium, one such device being a smart card (ICC). Finally, we assume that we may neglect certain transaction metadata, such as the exact timestamp on each transaction, or that several transactions occurred within a given timeframe, preferring rather to reduce the communication budget required to send information about a given transaction by reducing timestamp granularity to an hour or day-level timescale based on the aggregation window of transactions and potentially aggregating multiple transactions which occurred close together in time into one larger transaction (e.g. five transactions that happened at the same ATM on the same account within 15 minutes of each other may be represented as one transaction which is the sum of those individual transactions, at an approximate time.)

3.1 Simplifying Assumptions In this paper we make several simplifying assumptions. Firstly, we assume that the only parties in communication for these rural ATM systems are the machines themselves and some server-side system controlled by the host institution. This neglects considerations of legal regulations which may enforce some neutrality or transparency upon financial institutions. We also assume that because of this simple two-party communication when considered in the realm of a single ATM, that communication scales linearly with transactions (O(n)) for a given ATM. In addition we observe that there is a limit to the number of transactions per day based on human factors. Assuming that an ATM is available for 12 hours per day from 8AM to 8PM, and a transaction takes approximately one minute to execute, we estimate a rough upper bound of 720 transactions per day per ATM. We further consider all physical security considerations outside the scope of our problem; we assume that any machine is located in some safe environment where the likelihood of tampering is low, and also that any tampering of the machine will be made evident using standard

3.2 SMS The Short Message Service (SMS) was defined in 1985 as part of the draft GSM specification[22] and as part of the GSM standard, shares GSM’s ubiquity. As GSM is the cellular technology most used in the developing world, SMS as a data protocol has a significant installed base that can be leveraged with little cost. The SMS format can carry 140 bytes per message, at a cost of approximately 0.05 US Dollars per message in the region our testbed represents[25]. It is important to note, however, that SMS does not guarantee delivery, nor does it guarantee in-order delivery for those messages that are delivered, though it is a store-and-forward system. As such, our protocol has to be robust against sporadic dropped packets and out-of-order delivery. Additionally, the extremely small payload of the messages requires us to address the issues of packet fragmentation and careful data management. GSM does in theory support SMS concatenation of up to 255 messages, but in practice most systems do not support any more than 10, providing a theoretical aggregate payload of 1400 bytes. However, SMS concatenation has a per-message payload cost of 7 bytes per packet for every message after

225

the first, which is a significant amount of overhead given the payload size. Also, in our testing, 1337 bytes proved to be too few to take advantage of any generalized compression algorithms, with compression of most payloads resulting in an increased size of between 5 and 60 percent, depending on the algorithm and the structure of the data. In areas where GPRS is available, most providers in the developing world provide options for Multimedia Message Service (MMS) messaging, an extension of SMS which allows arbitrary payload sizes. MTN provides MMS messages of up to 100 kilobytes for approximately USD 0.125 in urban areas of Uganda, which represents close to a 300-fold increase in messaging budget per unit cost. Naturally, as this technology expands, marginal costs will decrease commensurately. Furthermore, even where GSM coverage is not available, SMS service is available via satellite phone coverage (e.g. Thuraya[29]), which allows any system built using an SMS metaphor to be generalized to practically any location on the globe where a clear line of sight to the sky is available, albeit at far greater cost. 3.2.1 SMS Delivery and Submission Rates On-field tests conducted once every 30 minutes from Adjumani (labeled in the northeast quadrant of Figure 2) for the 12 hour window from 8AM to 8PM to a handset in Kampala using a handset sending 50 pre-written 140-byte (160 characters with septet encoding) dummy messages by hand yielded a minimum send rate of approximately 8 messages per minute during peak times. Tests using a script submitting the same 50 messages over the same period on a subsequent day to MTN’s internet SMS gateway[24] bound for the same handset yielded a minimum receipt rate of approximately 12 messages per minute. We consider these to be sane limits when the network is operational as we assume that the handset equipment used has a lowerpowered radio and significantly less processing power than an ATM might, but we do not maintain that this rate is an absolute lower bound. Loss rates during these tests were 0%, but we do not treat this result as conclusive due to the small sample size and the fact that these tests were conducted only across one day in each direction. Additionally, all messages were delivered in-order, which we also do not treat as conclusive, for similar reasons. Tseng et al. found a loss rate of 0.66%[15] after sending 915 messages over a 38 day period in 2005 in Taiwan. As Taiwan has high GSM penetration and a more modern cellular infrastructure (all but two of Taiwan’s GSM providers had high-bandwidth 3G networks by October 2005) we assume that we will see greater loss rates on the whole in our testbed. The median latency for our group of messages was asymmetric, at approximately 36 seconds from Adjumani to Kampala, and approximately 21 seconds in the other

TABLE I

TP-VP GRANULARITY AND RANGE TP-VP 0 - 143 144 - 167 168 - 196 197 - 255

Implied Validity Period (TP – VP +1) * 5 minutes 12 hours + (TP-VP – 143) * 30 minutes (TP – VP – 166) days (TP – VP – 192) weeks

direction, including any latency introduced by the internet SMS gateway. These latency and throughput numbers are used only to garner an overall idea of the timescale these messages operate on and are not assumed to be statistically significant. 3.2.2 Payload Optimizations At a minimum, the message payloads must contain some information about depositor accounts and their balances. Based on our design principle of reducing network traffic, we wish for each payload to contain information about more than one transaction. Due to the small payload size allowed in SMS messages, we made several design decisions to decrease the size of data passed via SMS. We observed that the smallest denomination available in Uganda is 50 Ugandan Shillings (Ushs), or approximately USD 0.03, so we chose to encode balance information as multiples of this amount using 20 bits, which gives us a maximum amount of Ushs26214350 or approximately USD 15,000, fully an order of magnitude more than any microdepositor is ever likely to have on deposit, while maintaining signing. Ugandan checking account numbers are mandated to have exactly ten digits[26], which would require 34 bits to encode, but we can establish a mapping from account numbers to a smaller unique integer; in this case we use 20 bits which gives us a maximum of 1048575 discrete accounts. This gives us a total of 40 bits for a (account number, amount) pair which sets an upper bound of 28 of these pairs per 140-byte payload. 3.3 Protocol Our overall strategy is to encode a withdrawal transaction by indicating a negative delta to the balance indicated on a given machine, and encode balance updates from the financial institution’s central authority (CA) reflecting verified deposits and redistributions of balances among machines with positive or negative deltas to the balance. The CA maintains tables containing the GSM phone numbers of every ATM (by which incoming SMS messages are identified), precomputed pairwise distance metrics of each machine to every other machine, cash balance in each machine, home ATM(s) for each account, and account and balance records.

226

Fig. 3. Record structure

3.3.1 Message Structure As SMS does not guarantee delivery or in-order delivery, the addition of a sequence number at some level is critical to error-free communication. While SMS has no built-in sequence number, a sent SMS message does have a user-specified validity period indicator (TP-VP[12]) not contained within the user data area (and therefore essentially free) which guarantees expiration of the message if it cannot be delivered to the destination system within a given period of time. In addition, a received SMS contains a timestamp indicating when the message was received by the SMS Center (SMSC), which is in-order for a given source as a device cannot send another SMS message until one sent prior has been either acknowledged by the SMSC or timed out. We encode each (account number, amount) pair with four additional bits which indicate a version number from 0-15 which wraps around. Combined with the timestamp on delivery, this allows us to keep updates on a given account from a given source inorder. We also set the daily transaction count limit for a given account at a given ATM to 4, meaning that an account will not experience version number wraparound for 4 days, and set the TP-VP to be 2 hours, which guarantees against version number collision. Of the 1120 bits available to the payload, we use 1056 for 24 (version, account, amount) tuples of 44 bits each, and add a 20-bit checksum to protect against bit errors. We order these tuples by account number. Given our one-perminute estimate of maximum transaction execution rate due to human interface considerations, this gives us an upper bound of 30 SMS messages per day for our 12 hour, 8AM to 8PM day, with the rate of messages sent by the ATM never exceeding one per 24 minutes. 3.3.2

Message Flows

3.3.2.1 ATM to CA

Each ATM sends a message either when a set of 24 transactions has been completed or some time t has elapsed and at least one transaction is ready to send, where t is 120 minutes with a random jitter up to 5 minutes in order to prevent synchronization among ATMs. If a message is not acknowledged by the SMSC within 30 seconds, sending is reattempted at most three times, then the versions of the accounts contained in the message are flagged as nonacknowledged (NAK), and operation continues. In order to reduce the number of updates sent, multiple transactions on the same account within the window of a single message are aggregated and treated as a single update and are given

a single version number. However, each transaction still counts towards the transaction limit imposed by the ATM. During a 30-minute window after the ATM ceases public operation each day, it will attempt to resend any messages that are NAK for the duration of the window to compensate for extended periods of loss. 3.3.2.2 CA to ATM

The server at the central authority records any incoming transaction records from an ATM and applies the balance adjustments to the appropriate account, updating the version number in the process. If a version number received is not the one expected based on the CA’s record, it is applied, but the missing sequence number is noted. 3.3.2.3 Error Detection and Correction

In order to protect against silent failure of the CA server or an ATM, we add a heartbeat/echo message containing a random string which is triggered once every four hours by each ATM as necessary (i.e. after that amount of time has passed without a message having been received), jittered by up to 20 minutes. Any echo message received by the CA is replied to immediately with the same contents. If the CA server does not hear a heartbeat or transaction message from an ATM for 360 minutes, the ATM is flagged as ‘down’. An ATM being flagged as ‘down’ ensures that the operation of the local ATM continues as normal for a given account until that either account’s available balance at that ATM or the cash in the ATM is depleted, during which time connectivity may be restored. An ATM is brought back ‘up’ whenever an update is received from that ATM. Thrashing is possible, but the timescales involved reduce the problem, and no additional resources are consumed through the CA considering a machine ‘up’. Any machine which is flagged as ‘down’ during two consecutive batches is flagged as ‘disconnected’ and becomes ineligible for any balance distribution updates from the server until manual intervention can be scheduled. At the end of the day after the ATMs have ceased public operation, the CA waits for 1 hour for any incoming updates that may have been lost, and applies any received changes, then explicitly queries ATMs for any messages still missing using a series of messages containing only the missing version number and the account number. Any ATM which still has questionable data at the end of one hour after the final update request is sent is flagged as being ‘down’ and is removed from the set of ATMs eligible for balance redistribution. After the day-end synchronization batch described above, the CA sends each ATM that is ‘up’ a message containing a hash of the concatenation of all versions, accounts, and balances that ATM is known to be aware of, ordered by account number and version number. Each

227

ATM computes the same hash upon receipt, and an echo containing the ATM’s hash value is sent back. Any ATM which returns an inconsistent value is flagged as ‘disconnected’ as above. Messages which are received but fail checksum and cannot be repaired are discarded and considered lost. 3.4 Security SMS over GSM is susceptible to a variety of attacks[7]. In particular, it was indicated in a personal communication to the author by a member of MTN Uganda staff that SMS is transmitted in plaintext throughout its network, and this is the default behavior[7]. As this is the case, and as both Originating Address (OA) spoofing and realtime sniffing of SMS messages are available in the wild, it is necessary for the system to have mechanisms to protect the secrecy of balance/account information and to provide some measure of authentication of messages sent between the ATMs and the CA. In order to maintain effective payload size while providing this type of protection, a stream cipher is used, which produces ciphertext of the same length as the plaintext. This system uses RC4 due to its ease of implementation, but any such encryption system will serve. Each ATM maintains a separate 256-bit key, and the CA maintains a keyring of all of these keys, using them as appropriate in communication to a given ATM. A 32-bit timestamp is embedded into all messages in order to mitigate the problem of replay attacks; a timestamp will only be accepted once and any subsequent message from a given source bearing the same timestamp will be discarded. In order to mitigate substitution attacks, this timestamp is placed between arbitrary (version, account, amount) tuples and will be padded with 12 bits of 1’s in order both to maintain tuple offsets and to indicate which block contains the timestamp. Version number 15 then becomes reserved as not to restrict the range of account numbers which may be used. 3.5 Balance Distribution The practice of available balance distribution among ATMs is a hedge against users who may wish to game the system by trying to take advantage of network outages or periods of higher-than-normal latency in order to withdraw more than their account balance. We assume for the purposes of this practice that ATM use multifactor authentication where at least one factor is a biometric, such as a fingerprint. The hash of this biometric is stored on the card in order to defend against card duplication. We begin by associating each given user with a ‘primary’ ATM, which is the machine closest to that user’s residence, and then consider the other ATMs which are present within one 3 hour period’s travel of that location.

In our testbed, the geography is advantageous as it offers only a few inter-town river routes, and no tarmac-paved roads. We also use the intuition that (1) accounts with higher available balances are more attractive to overwithdraw, (2) people with higher balances are similarly more likely to have access to faster transport such as bodabodas, (local motorcycle taxis), matatus (intra- and intertown minibuses), and interdistrict buses or to own or be able to borrow a private vehicle, (3) that fraud associated with higher-balance accounts costs the host institution more, and (4) holders of high-balance accounts are less likely to need to withdraw all or most of their available balances at once. We consider the maximum feasible distance covered by walking over a 3 hour period (~10km) and the maximum feasible distance covered by driving along unpaved roads during that same period (~125km). Given that the average daily unskilled rural labor wage is in the neighborhood of $1.2 per day[13] (approximately Ushs2100), we arbitrarily set the ‘high balance’ mark at Ushs900000 (USD ~5000). Accounts with balances of 450000 and under have a maximum of 100% of their account balance distributed across all the ATMs in the system, decreasing linearly to 75% at 900000 and beyond. Only ATMs within a specific distance of the primary ATM are considered for this distribution, specified by the formula (dmax/2 – d0) * min(1,b/c) + d0 where dmax represents the greatest distance between any two ATMs in the network, d0 represents the minimum radius to be considered (40km in our case), b represents the balance of the account in question, and c is the cutoff ‘high balance’ mark at which maximum distribution spread is reached. This encapsulates our intuition that we want to distribute the balance for those with higher balances over a larger area, and the linear increase in radius produces an increase in area covered proportional to the square of this value. The proportion of the available balance held in the primary ATM varies according to the formula bprimary = b * pmax – ((pmax – pmin) * min(1,b/c)) where pmax and pmin are the maximum and minimum proportions, respectively (.8 to .4 in our simulation). These numbers are based on assumptions we make about the behavior of clients as listed below, and the formula again reflects our intuition that the further the client is likely to travel, the further afield the balances should be distributed. The balances to the ATMs within the effective radius defined above other than the clients’ primary ATMs are distributed proportionately to the inverse of the square of

228

the distance of the machines from the primary ATM. Any available balances less than Ushs250 and any residues of allocated balances modulo Ushs50 are removed and added to the primary ATM’s available balance. These two formulas together have the effect of concentrating more of a user’s balance nearer to his home ATM as his balance decreases. In practice the constants and proportions are adjustable to ensure that any balances under a given amount are held entirely at the user’s primary ATM. 3.5.1 Migration The initial germ idea for this project called for support for highly migratory populations to have continuous access to funds. We solve this problem by introducing a notion of carrying ‘virtual’ available balance on a credential token used to access the system, such as a smart card. This ‘virtual’ balance would involve ‘locking’ a specific amount of available balance on one ATM and writing a devicesigned token for the amount onto the card including a unique transaction ID allowing the user to take the card to some specific other ATM and withdraw the amount ‘locked’ plus any available balance on the local ATM. This could be useful for trips where larger amounts of capital are required but carrying cash is dangerous, e.g. seasonal trips to buy seed for planting or taking animals or produce to market. Using the extra balance would cause the local machine to create a new signed token with any remaining ‘locked’ balance and inserting it into the original machine would unlock any such remaining portion and treat the spent portion as a withdrawal. This has the advantage that if the card is lost or stolen, funds which are locked on the card but have not been unlocked anywhere across the ATM network can be recovered from the institution. 3.5.2 Multihoming We wish to model one commonplace situation, in which a head of household or other member may work in a remote region for most of the year and return for several months or a season. In such cases, the person who is working away from home typically wants to send money back to his household – this is the problem that approaches like M-PESA address. We model this situation by explicitly adding the single ATM closest to the worker to the set of ATMs considered for balance distribution, if it is not already in the set. While the typical usage of such an ATM will be solely to deposit funds left over after cost of living is paid for in cash, we arbitrarily allocate 20% of balance available at the primary ATM to this ATM. The rationale is the idea that the breadwinner represents, on average, 1/5 of the total size of the household based on the average family size of 4.7 and assuming a single primary source of income. This is trivially extended to accommodate arbitrary family configurations. The

proportion that is allocated is drawn from the bottom up: any available balance outside the primary ATM is consumed first, and then primary ATM balance is reallocated until the correct proportion is reached. 4 SIMULATION 4.1 Simplifying Assumptions There are several factors present in the real-world region upon which our simulated testbed is modeled which we ignore completely. One is familial seasonal migration from place to place – for the one-year duration of our simulation we assume that people’s residences remain fixed, i.e. the primary ATM of a given account never changes; individual members may go on excursions but never the entire family at once. Another is that of immigration to the region or other sources of population growth such as refugee influx or organic growth of the number of households through marriage. We assume that all members of our simulated environment start life with accounts, that is, we have a 100% adoption rate, with some proportion of accounts beginning with a zero balance. As a final detail, we have found that we do not have enough information on the overall prevalence of bitwise errors within SMS delivery, as the literature has various accounts, and have no information at all on its prevalence in the rural context, where distance may increase and technology may be older or cheaper but there are far fewer sources of electromagnetic noise, so we ignore it, assuming that those messages not correctible by the checksum are incorporated into the loss rate.

Fig. 4. Proportion of downtime versus standard deviation outage burst length for 4 values of λ.

4.2 Simulator Overview The simulator uses minute-scale timeslices over the course of a calendar year and contains metaphors for

229

customers, ATMs, the CA server, and the SMS network connecting the various components of the overall system. In our experiments we change various parameters of the system to see what effect they have on the cost model and customer request satisfaction rate. We consider a withdrawal request successful if, having specified some amount desired which is less than his account balance, the machine both has enough cash on hand to satisfy the request and has enough of the balance allocated to the given machine to allow the machine to dispense that cash. 4.3 Network Behavior Network behavior parameters are the factors most directly related to the cost model. To model bursty outage behavior, we use a Poisson distribution to model burst arrival and a truncated normal distribution (with mean at 0 in which we only consider positive values) to model burst length, and change λ and σ2, respectively for each to test the robustness of the system under various outage models, as well as how the level of customer satisfaction changes as the network and therefore the application built atop it becomes less reliable. We would also like to vary latency with a truncated normal distribution and μ set to 36 seconds as the experimental data in 3.2.1 indicated, and the σ2 likewise set to 5 to show how the effects of variable propagation delays; however the timescale we are using renders this variability negligible. As such we set μ and σ2 to 1 minute, which biases slower than experimentally detected, which cannot artificially improve results. Finally, we add an additional Poisson distribution-based loss model for ‘black hole’ loss where messages simply disappear after having been successfully sent to the SMSC.

Fig. 5. CDF of customer weekly savings rate in Ushs.

4.4 Household Behavior The 136541 households in our testbed are assigned to population centers according to real-world population levels. Each account is held by a household having certain

specific characteristics with which we attempt to reasonably emulate a real population: 1) Primary ATM, allocated randomly using a uniform distribution 2) Weekly savings rate, generated by taking an arbitrary minimum savings rate (Ushs200) and adding it to Ushs1000 multiplied by the absolute value of a random element drawn from a Gaussian distribution with mean 0 and standard deviation 3) Visit frequency, a step function generated by another random Gaussian with mean 0 and standard deviation 1, by assigning 7 days if the random value is within 1 standard deviation, 14 if within 2, and 28 if more than 2 4) Withdrawal strategy (whether the client will withdraw a proportion of his balance or of his weekly withdrawal rate, half of the clients being in each category), selected using a uniform random variable 5) Probability of withdrawing from an ATM other than his or her primary ATM, between 0% and 20%, generated using a Gaussian distribution with mean 10% and standard deviation 5%. Any random value selected outside of 0-20% is reselected 6) 0-4 arbitrarily placed seasonal brief migrations where 10% of the population has one or more migrations. Presence of a migration is decided by a uniform random variable, and the number of migrations is generated in an analogous manner to 3) and 7) multihome (second primary) ATM, where 10% of the population has a secondary home as described above. Both the presence of a second primary ATM and its location are selected using a uniform random variable. For simplicity, we register deposits proportionate to the weekly savings rate each time a client visits an ATM for withdrawal. Deposits are processed every two weeks, and only deposits received at least one week prior to processing are applied to balances. 4.5 ATM Parameters We use a parameter n to determine how many ATMs to use to saturate our network, i.e. one ATM per n users. When this figure causes the number of ATMs to exceed one per population center, we use the notion that a given account can only use one ATM within the center and any others are treated as nonexistent, to obviate complexity of balance distribution among the multiple ATMs while maintaining semantic equivalence. We also use a ratio p to specify the proportion of the sum of the total available balances from a given ATM that the ATM should have on hand in cash, which affects cash restocking. Finally we specify how many days the server waits between balance redistributions, tb, which affects message traffic volume as well as funds availability following a deposit. To limit the number of variables in the simulation, we chose to fix the frequency of cash refills at once every two weeks.

230

Fig. 6. Loss versus Cost per User in US Dollars, averaged across all runs of the simulator.

Fig. 7. Loss versus successful transaction proportion, averaged across all runs of the simulator.

Fig. 8. Cash load proportion versus successful transaction proportion, averaged across all runs of the simulator.

Fig. 9. Number of users per ATM (n) versus Cost per User in US Dollars averaged across all runs of the simulator.

5 RESULTS

Figure 7 shows that satisfaction is unaffected by the number of messages sent, which is an encouraging indicator that our protocol recovers well from loss.

In running our simulations, we are interested in seeing the effect of our various parameters on two primary results: cost per user per year, which indicates the marginal cost of adding more users to the network, and withdrawal request satisfaction rate. 5.1 Loss Rates Simulation shows that the system is affected not specifically by outage frequency or outage burst length, but by overall downtime proportion. As shown in Figure 6, as loss increases from 1% to 12%, cost increases linearly as the number of messages which need to be resent increases. It is an obvious conclusion that messaging cost would increase as loss rate increased, but it is encouraging that the cost increases linearly rather than exponentially. The jaggedness of the curve is due to our using precomputed loss models across all trials to speed the simulator up, which exacerbates certain lossy periods as they occur in the same place in every run.

5.2 Cash Exposure Figure 8 illustrates the relationship between the cash load proportion and the proportion of requests satisfied. While the difference is absolutely minute, constituting less than 0.04% difference, the trend is clear that loading the ATMs with more cash provides slightly better request satisfaction levels. However, it is encouraging to see that even with at most 5% of the cash payable across all accounts in a given ATM loaded, satisfaction is approximately 98% of all requests under this model. 5.3 ATM Density Varying the number of users served by a given ATM has a dramatic effect on the number of messages sent as each ATM sends messages after a given period of time

231

Fig. 10. Days between balance redistributions versus cost

regardless of whether the message contains the maximum number of transactions or not, given that at least one transaction has occurred during that period of time and the server must send messages to each ATM during any balance redistributions. As shown in Figure 9, cost per user drops off dramatically between 2000 users per ATM and 4000 users per ATM and levels off thereafter. This sharp elbow is explainable by the fact that many of the population centers used in the simulator have between 2000 and 4000 accounts, and therefore increasing the number of users served per ATM decreases the number of ATMs per location from 2 to 1. In our simulation, this change in density decreases the number of ATMs from 92 to 61. On the other end of the scale, few locations have more than 8,000 accounts, so there is little change by increasing users per ATM beyond that point. 5.4 Balance Distribution Frequency As shown in Figure 10, cost per user falls off dramatically with less frequent updates at the cost of user satisfaction rate. The cost, fortunately, falls off much faster than the request satisfaction rate as seen in Figure 11. As each balance redistribution accounts for one complete set of account balances sent to each ATM in the system, this sharp increase in cost as updates become more frequent is compounded if the number of users per ATM is low enough to introduce an unnecessarily large number of ATMs into the system. 5.5 Example Optimization To attempt to select a set of parameters to produce an optimal solution, we first need to observe that cost and satisfaction rate are opposed to each other, as are cash load and satisfaction rate. An implementing institution would have to determine what mixture of cost, risk, and customer satisfaction it would be comfortable with, but one example set of parameters is

Fig. 11. Days between balance redistributions versus satisfaction rate

p = 0.05 n = 16,000 tb= 7 which yields over a 98% withdrawal success rate with a messaging cost of between USD 0.10 and 0.15 per user per year even accounting for nearly a 10% message loss rate. 6 DISCUSSION

AND FUTURE WORK

The system as described has several practical weaknesses. An important limitation is the occasional need of accountholders to withdraw their entire savings. This can, however, be ameliorated by augmenting the system with an on-demand functionality whereby a user could pay for the additional SMS messages required to withdraw his or her full balance immediately. Other weaknesses of the analysis include the explicit exclusion of physical security considerations and legal restrictions, both of which are often of critical importance in the developing world. The former in particular is important with deposits, as they need to be conveyed to some physical bank branch for validation. The details of transaction auditing and reconciliation are also neglected, but it is assumed that standard practices will be followed. A baseline study on the field performance of SMS would allow for further improvements to this system. We would like to better understand outage patterns for SMSCs and cellular networks as a whole. Some national governments, such as Nigeria’s, require cellular network operators to file reports on statistics affecting quality of service such as bit error rate, end-to-end error rate, etc., knowledge of which would allow us to better simulate this aspect of the environment. Also, detailed knowledge about the location of cellular towers and their respective power levels would greatly aid in the accurate representation of the region. The next step in research for this system in particular

232

will be investigation into whether decentralization can reduce network traffic and increase robustness against network failures, particularly at the cellular network cell hosting the institution. As balance distribution is dependent on geographical location, it may prove to be wise to have nodes collude with their nearest neighbors rather than receiving updates only from a central authority. Other potential extensions of the project are increases in granularity or changes in the strategy of balance redistribution and tracking of loss and corruption rates of received messages in order to better allocate message budget. A redistribution strategy which is more adaptive to the past patterns of the user is also likely to yield positive results in terms of customer satisfaction. As crafted, the system could be trivially extended to allow for payouts of microcredit installment loans or salaries for rural workers. 7 CONCLUSION This paper presents ATMosphere, a protocol designed to provide ATM services to the world’s rural poor. The system as simulated is based on currently deployed technologies and is implementable today using low-cost hardware. The result is a solution which has low start-up and marginal costs, represents low relative risk to financial institutions, and provides sorely needed access to savings and capital markets for the marginalized. 8 ACKNOWLEDGEMENT We would like to thank various personal friends living in rural Uganda for firsthand assistance with baseline testing of MTN’s cellular network. We would also like to thank Jinyang Li for her comments. This material is supported by the National Science Foundation under Grant No. CNS-0831934. REFERENCES [1] [2] [3]

[4] [5] [6] [7] [8]

E. Brewer, “Technology Insights for Rural Connectivity,” in Wireless Communication and Development: A Global Perspective. Marina Del Rey, CA, 2005. L. Elser, A. Hannig, and S. Wisniwski, “Comparative Analysis of Savings Mobilization Strategies,” Eschborn, GTZ, 1999. W. Enck, P. Traynor, P. McDaniel, and T. La Porta, “Exploting open functionality in SMS-capable cellular networks,” on 12th ACM Conference on Computer and Communications Security. Alexandria, VA, ACM, 2005, pp. 393-404. A. Hannig, and S. Wisniwski, “Mobilizing Microsavings: The Experience of Seven Deposit-taking Institutions,” Eschborn: GTZ, 1999. G. Ivatury and M. Pickens, “Mobile phone banking and low-income customers: Evidence from South Africa,” Washington, DC, 2006 A. Jhunjhunwala, A. Ramachandran, and A. Bandyopadhyay, “nLogue: The Story of a Rural Service Provider,” The Journal of Community Informatics,1 (1), pp. 30-38. S. Lord, “Trouble at the Telco: When GSM Goes Bad,” Network Security, 2003(1), pp. 10-12. C. Olston, “Approximate Replication,” Doctoral Dissertation, Stanford University, Stanford, 2003.

[9]

[10]

[11] [12] [13]

[14]

[15] [16]

[17] [18] [19]

[20]

[21] [22] [23] [24] [25] [26] [27] [28]

[29] [30]

C. Olston and J. Widom, “Offering a Precision-Performance Tradeoff for Aggregation Queries over Replicated Data,” in 26th International Conference on Very Large Data Bases (VLDB). Cairo, Egypt, 2000, pp. 144-145. R. Patra, S, Nedevschi, S. Surana, A. Sheth, L. Subramanian, and E. Brewer, “WiLDNet: Design and Implementation of High Performance WiFi Based Long Distance Networks,” in 4th USENIX Symposium on Networked Systems Design & Implementation. Cambridge, MA, 2007. L. Payne, “Rebuilding Communities in a Refugee Settlement: A Casebook from Uganda.” Oxford: Oxfam, 1998. G. Peersman, P. Griffiths, H. Spear, S. Cvetkovic, and C. Smythe, “A tutorial overview of the short message service within GSM.” Computing & Control Engineering Journal, 11 (2), pp. 78-89. S. Staschen, “Microsavings, Microcredit and Microinsurance: Financial Products of Small Farmer Co-operatives Ltd. In Nepal.” Cologne: AEF University of Cologne, 2001. G. Taylor and M. Bekabye, “Employment Creation and LabourBased Technology in Roadworks: Ugandan Case Study, 1999.” [Online] Available: http://www4.worldbank.org/afr/ssatp/Resources/HTML/rural_transp ort/knowledge_base/English/Module%202%5C2_4b%20Case%20St udy%20-%20Labour%20Based%20Techs.pdf C. Tseng, J. Jiang, R. Lee, F. Lu, C. Ouyang, Y. Chen, and C. Chang, “Feasibility study on application of GSM-SMS technology to field data acquisition.” Computers and Electronics in Agriculture, 53 (1), pp. 45-59. S. Wisniwski, “Microsavings Compared to Other Sources of Funds.” Eschborn: GTZ, 1999. H. Yu, and A. Vahdat, “Design and Evaluation of a Continuous Consistency Model for Replicated Services,” in Fourth Symposium on Operating Systems Design and Implementation. San Diego, CA, 2000. H. Yu, and A. Vahdat, “Efficient Numerical Error Bounding for Replicated Network Services,” in 26th International Conference on Very Large Data Bases. Cairo, Egypt, 2000, pp. 144-145. M. Zeller, “Models of Rural Finance Institutions,” in Paving the Way Forward for Rural Finance: An International Conference on Best Practices. Washington, DC, 2003. ETSI GSM 3.40: ‘Digital cellular telecommunication system (Phase 2); technical realisation of the short message service point-to-point’, European Telecommunications Standards Institute TC SMG, Version 4.13.0, May 1996. ETSI GSM 3.41: ‘Digital cellular telecommunication system (Phase 2); technical realisation of the short message service cell broadcast (SMSCB)’, European Telecommunications Standards Institute TC SMG, Version 5.2.0, May 1996. GSM Doc 28/85: ‘Services and Facilities to be provided in the GSM System’ rev2, June 1985. MTN Uganda Ltd (MTN-UGANDA) – Interactive Coverage Map, 2007. [Online] Available: http://www.gsmworld.com/cgibin/ni_map.pl?cc=ug&net=mt. MTN Uganda: A Bright Y’ello World, 2007. [Online] Available: http://www.mtn.co.ug. MTN Uganda: Tariff Guide : Mobile Solutions : Business, 2008. [Online] Available: http://www.mtn.co.ug/business/talktime_tariffs.htm. The National Cheque Standard of Uganda, 1999. Available: http://www.bou.or.ug/NPSS/Cheque%20Standard.pdf. Safaricom: M-PESA, 2007. [Online] Available: http://www.safaricom.co.ke/index.php?id=745 Quality of Service (QoS) Indicators for Mobile Services, Unknown Date. Available: http://www.ncc.gov.ng/TRS/QoS_Benchmark.pdf. Thuraya Satellite Communications, 2008. [Online] Available: http://www.thuraya.com Uganda Bureau of Statistics, 2007. [Online] Available: http://www.ubos.org WiMAX Forum – Technology, 2007. [Online] Available: http://www.wimaxforum.org/technology

233

Building a Transportation Information System Using Only GPS and Basic SMS Infrastructure Ruth E. Anderson, Anthony Poon, Caitlin Lustig, Waylon Brunette, Gaetano Borriello, Beth E. Kolko

Abstract—This work consists of two main components: (a) a longitudinal ethnographic study in Kyrgyzstan that demonstrates the importance of transportation resources in the developing world and how to plan for an appropriate ICT solution, and (b) the results of a proof-of-concept system engineered to create a bottom-up, transportation information infrastructure using only GPS and SMS. Transportation is a very important shared resource; enabling efficient and effective use of such resources aids overall development goals. The system, *bus, involved the development of a hardware device (a *box) containing a GSM modem and a GPS unit, that can be installed on a vehicle and used to track its location. The *box communicates via SMS with a server connected to a basic GSM phone. The server runs route a prediction algorithm and users can send SMS messages to the server to find when a bus will arrive at their location. The paper discusses the system and early testing, as well as the development implications for a range of urban and rural environments where transportation is scarce or inefficient, and where a central authority or institution is not in a position to provide robust information resources for users. We describe how the solution is also situated within technology usage patterns common to the developing world. Index Terms—ICTD, transportation, information services, mobile phones, GPS, SMS, Kyrgyzstan, developing world

I. INTRODUCTION

T

RANSPORTATION of goods and people is key to economic and human development. This paper discusses the development of the *bus system (pronounced “Starbus”), a transportation information system that uses GPS, GSM, and SMS technologies. The *bus system was developed in response to transportation challenges experienced in

Manuscript received September 22, 2008. This work was supported in part by U.S. National Science Foundation grants #0326101 and #0219350 and by the J. D. Noe Professorship at UW. R. E. Anderson, Dep’t of Computer Science & Engineering, University of Washington, Seattle, WA 98195 USA ([email protected]). A. Poon, Dep’t of Computer Science & Engineering, University of Washington, Seattle, WA 98195 USA ([email protected]). C. Lustig, Dep’t of Computer Science & Engineering, University of Washington, Seattle, WA 98195 USA. She is now with Intermec, Inc. ([email protected]). W. Brunette is with the Computer Science & Engineering, University of Washington, Seattle, WA 98195 USA ([email protected]). G. Borriello, Dep’t of Computer Science & Engineering, University of Washington, Seattle, WA 98195 USA ([email protected]). B. E. Kolko, Dep’t of Human Centered Design & Engineering, University of Washington, Seattle, WA 98195 USA ([email protected]).

Kyrgyzstan, a developing region with poor infrastructure and limited resources; consequently, the findings in this paper are extensible to other selected developing regions with limited resources, fragile road infrastructure, resource-constrained central government authority, and ad-hoc transportation resources for both inter- and intra-city transport. II. TRANSPORTATION AND THE KYRGYZ CONTEXT A. Background on the Transportation Challenge Transportation has long been recognized as having an impact on economic development. Factors such as levels of investment in transportation have been shown to have a positive impact on economic growth [1] [2]. In the developing world, the availability of safe and reliable transportation can have even broader implications. Not only does access to transport improve access to markets and mobility of workers, but it is also critical to the timely and affordable delivery of services such as health and education, can serve to empower vulnerable groups by increasing their independence, and is key for maintaining social networks [3]-[5]. Along with providing access to transportation, it is also important to provide potential users with information about its availability. In the developed world, users can often access information about most bus and train schedules easily via printed schedules or web pages maintained by centrallyfunded transportation authorities. In some cases, users can view real time updates on the current location and expected arrival time of their bus or train via web, phone, or SMS/text message [6]-[8]. However, when a community lacks the infrastructure to provide such information resources, potential users can find themselves unable to take advantage of whatever (limited) transportation resources are available. B. Methods The background provided in the rest of this paper is based on several years of research in Central Asia by the Central Asia Information and Communication Technologies (CAICT) project at the University of Washington. Our research has included multiple periods of ethnographic research including a design ethnography, interview studies, and a longitudinal survey in four countries of Central Asia. Ethnography has been conducted since 2000 and includes a team of US and Central Asia researchers. Field notes and photography are both used to document findings. Survey results are based on a nationwide survey of 1000 respondents

234

in each of four countries (Kazakhstan, Kyrgyzstan, Tajikistan, Uzbekistan) of age 15 and older, administered in 2006, 2007, and 2008. The survey sample was based on census information for age, gender, ethnicity, and geographic location as released by each country’s government. The survey includes multiple urban and rural sampling locations, and it was administered in Russian and other regional/local languages. In addition to the general sampling scheme, a three-stage process was used to select respondents: • probability proportional to sample size of Primary sampling units (PSUs); • consecutive random sampling of households in determined PSU; and • selection of a household respondent using a Kish Grid method. The survey is a general social survey that asks a series of questions about technology usage, as well as informationgathering habits and patterns of trust and confidence. Because Internet usage rates in the region tend to be low, the survey was not designed to target Internet users specifically, but, rather, gauge general trends among the overall population. C. Inter-city and Intra-city Transportation within Kyrgyzstan Kyrgyzstan is a small, mountainous country in Central Asia. Its population is about 5.5 million and concentrated in the capital of Bishkek in the north and in the heavily agricultural Ferghana Valley in the south. The country has a youth bubble, with the median age of 24.2 years (for comparison, the US median age is 36.7). The population is defined as 98.7% literate; ethnically the population is largely Kyrgyz (64.9%), followed by Uzbek (13.8%) and Russian (12.5%). Since independence from the Soviet Union in 1991, Kyrgyzstan has struggled economically. Unemployment is high and about 40% of the population lives below the poverty line [9]. Personal vehicle ownership rates are low, and the population depends heavily on shared and public transportation for inter-city and intra-city transportation. Cities have taxis and an independently operated bus system known as the marshrutka system. Marshrutnoye taksi is a Russian phrase for ‘routed taxicab’; the marshrutka system is a shared taxi transportation system that runs on largely pre-plotted routes but without standard schedules; the standard bus is a mini-van that seats about 15 and with standing passengers can fit 25 or more. Fares are 8 som (about US$0.19) before 8pm and 10 som (about US$0.24) after 8pm for central city routes. The drivers are often independent contractors as opposed to employees of a municipal or national transportation authority. Bus routes are part of local knowledge, and learning a new route is challenging even for a city resident. Numbered routes are marked by placards placed in the front windshield of the bus, with the rough routing listed as major stops or landmarks. As buses whiz by, people try to glimpse the list of stops written on the placard to see if the bus is going in their direction. Quickly identifying an appropriate marshrutka is a challenging step when a new route is being attempted. There is a lack of official bus stops, and people stand at areas known

to be places where the marshrutkas stop. There is no official schedule, so riders are not always sure how long they will have to wait. Sometimes buses are full and will not stop, and then riders have another unknown stretch of time to wait. Routes can shift slightly, and because stops are often demarcated by landmarks rather than actual routings, users sometimes have to guess what route a bus will take to get from listed landmark A to listed landmark B. Overall, the system can work well if one is traveling a familiar route that does not get too crowded, but if one travels at peak times, or off hours, or wants to get somewhere new, the system can be frustrating to navigate because there is no clear pathway to gather the information relevant to a rider. Marshrutkas operate on inter-city as well as intra-city routes. In Kyrgyzstan the inter-city system is relatively transparent, if not especially efficient. In order to travel between towns, one goes to the bus depot where marshrutkas are lined up. In Bishkek, the bus depot is large and well signed, with destinations labeled on placards hanging over the bay where the minibuses are parked. In smaller cities the depots can be parking lots on the edge of town. At the Bishkek depot, drivers stand in front of their buses with microphones, calling out their destination to the crowd. There is no departure time listed, and drivers will not leave until the bus is full. If your destination is a popular one, you may not have long to wait. But if you are headed to a smaller town on a not particularly well-traveled route, you could face a significant wait before the bus leaves. It is these kinds of inefficiencies that make the system ripe for technologyassisted solutions that could disseminate information to users and prospective users of the transportation infrastructure.

Figure 1 A bus depot in Bishkek with marshrutkas and car taxis.

In addition to the bus depots, there are also more informal ways of traveling between cities. In neighboring Uzbekistan, for example, if you want to travel and not take the buses (which can be overcrowded and perilous on the mountain roads), you can travel to an acknowledged site – in the case of the capital Tashkent, this is a subway stop furthest in the direction in which you want to travel. Or, in Tajikistan, a mountainous country with limited public transportation resources, you similarly go to an established place and look for a driver headed in your direction; at times these trips can

235

last for days. There, drivers with cars congregate, looking for passengers to fill up their cars; some drivers are professionals, some are there to help defray the costs of getting home (often after delivering agricultural produce or visiting relatives). The characteristic these countries have in common, and that they share with many developing countries, is that shared transportation often happens outside the realm of a national or metropolitan transit authority. This means there is no central body coordinating or allocating resources to ensure efficiency or at least moderate attempts to meet the needs of citizens. Our research considered how, under such circumstances, one might aggregate and disseminate information about transportation options to make it more efficient or effective for both riders and drivers. Another way of stating the problem is how might one build an information infrastructure for a shared public resource when a public government (or other centralized authority) is not positioned to do the job – either because of resources, will, or because the resource itself is decentralized and largely part of the informal sector. D. Personal Technology Use in the Region The rate of personal ICT usage in Kyrgyzstan is similar to many developing regions; that is, computer and Internet usage remains relatively low and has seen slow growth over the past 3 years, while mobile phone use is growing at a remarkable rate. As Figure 2 shows, the mobile usage rate in Kyrgyzstan has grown from 21% to 71% from 2006 to 2008. During that same time, Internet usage has grown almost imperceptibly. Even in neighboring Kazakhstan, with a rapidly developing economy rich with oil and gas money, Internet use is holding flat, with only 19% of the population reported as Internet users in 2007 and 2008. And even though computer use climbed from 29% to 43% in Kazakhstan over the same time frame, Internet usage showed no increase. The rapid increase in mobile usage is not unique to Kyrgyzstan, and adoption of mobiles in and of itself is not enough to establish that a mobile solution is the best avenue for a transportation solution. However, in addition to mobile usage, what we also see in Kyrgyzstan is that people use their phones for personal calls more than business calls, they are using SMS in measurable amounts (about one-third of users use SMS; see Figure 3 ), and they are not using the Internet as their primary mode of ICT access. 80% 60% 40% 20% 0% Owned Computers

Used the Internet

Owned and used mobile

2006

7%

11%

21%

2007

8%

10%

45%

2008

12%

15%

71%

Figure 2 Personal Technology Use in Kyrgyzstan.

As Figure 3 illustrates, approximately one third of mobile phone users report using SMS; interestingly, this number held rather steady between 2007 and 2008, unlike overall mobile adoption numbers. This rate of SMS usage is larger than those using their phones to find information, and the percentage of people using their phones to connect to the Internet decreased, but is so low as to be irrelevant (4% in 2007, 1% in 2008). There are information numbers one can call in Kyrgyzstan to get information about local businesses, so it is possible for mobile phone users to seek information via voice. 100% 80% 60% 40% 20% 0% Work calls

Personal

Information

Text / SMS

Play games

Internet

2007 - N=451

36%

92%

16%

28%

14%

4%

13%

9%

2008 - N=712

32%

97%

21%

29%

7%

1%

13%

14%

Music

Photos

Figure 3 Activities people engage in with mobile phones

The issue of Internet use is especially key to the particular solution this team has created. Survey findings in the region illustrate that not only is mobile usage nearly five times that of Internet usage (see Figure 2), but even when people do identify themselves as Internet users, their model of usage is very different from that of the developed world. This point is crucial, and it feeds directly into the *bus system developed as part of this study. In a city like Seattle, for example, extensive information about public transportation is available on the Internet, and people access this information either through a computer or a mobile phone. Information about bus routes, timetables, fares, re-routings, etc. is available on a comprehensive and userfriendly site. However, this model of information delivery is predicated on the reality that most users of the system have regular access to Internet resources. The percentage of Internet users in the US is estimated at 85%, and over half of those users are connected on a regular basis [10]. In Kyrgyzstan, however, even among Internet users, the majority report using the Internet once a week or less (see Figure 4). Meanwhile, mobile users use their phones on a much more frequent basis, with 89% reporting using them at least once a day. What this means is that developing and delivering a robust information delivery system that relies on people having Internet access in order to take advantage of the information resource is a fundamental mismatch between a technology and user habits and abilities. Recognizing what actually constitutes an “Internet user” then, in many parts of the developing world, helps to better envision an appropriate technological solution: delivery of information via mobile phone-based SMS. In addition, it is worth noting that mobile phone usage rates are significant among rural as well as urban populations. As Figure 5 below illustrates, mobile usage is not only an urban phenomenon. In many ICT and development situations, there

236

is a gap between rural and urban users. In this project, we have taken care to include the usage behavior of both urban and rural populations in order to develop a solution that will be appropriate for the broadest swath of users. 70% 60% 50% 40% 30% 20% 10% 0% Mobile Use - N = 712

Internet Use N = 149

Computer Use - N = 280

Every few weeks or less

2%

16%

20%

At least once a week

9%

58%

39%

At least once a day

89%

26%

41%

Figure 4 Frequency of technology use in Kyrgyzstan, 2008 60% 50% 40% 30% 20% 10% 0% Mobile Use**

Internet Use**

Urban - N= 397

56%

26%

Computer Use** 43%

Rural - N= 603

41%

5%

18%

Figure 5 Urban and rural use of technology in Kyrgyzstan, 2008 ** = p <.001 based on Chi-square test

E. Design Ethnography Findings Regarding Transportation Challenges During the summer of 2006, the CAICT team conducted design ethnography work in Kyrgyzstan to better understand people’s habits of technology use within their personal communities, as well as the pain points of everyday life, which might best indicate potential solutions that leverage those preexisting technology usage patterns. Four groups participated in the design ethnography: two multi-generational family groups, and two groups of young people (18-25). Interviews were conducted in an urban area, and a smaller, regional town. At each site one family and one group of young people were interviewed. Interviews covered issues related to technology use including Internet and mobile phones, challenges of everyday life, and social networks. In particular, interviews and questionnaires focused on patterns of reciprocity of goods and information exchange within the groups, as well as whether any technologies were used to facilitate such exchanges. Findings from the design ethnography indicated mobile phones were extremely important to participants, and that both voice and SMS played a daily role in people’s lives. Of the 12 participants in the study, only one did not own a mobile phone. Transportation emerged as one of the significant problem spaces for our users. Although for most people, a general sense of lacking information resources outside of family and friend networks was prevalent, the transportation issue was of particular interest because it generated animated discussion among our users. This was the case because the way

transportation worked was in stark contrast to the public information resources available in many developed country communities, and because transportation was clearly a fundamentally shared resource in these settings. As we continued to analyze the findings from the design ethnography, we found the following: (a) pricing issues for marshrutkas were not a concern, however the failure of the marshrutka system often meant people had to take taxis which were seen at prohibitively expensive, (b) safety and security were issues, especially for women, and this related to uncertain routings, riding at night, uncertain waiting times, and buses that would be full and unable to take new passengers, (c) learning new routes was seen as daunting, (d) there was some confusion over how much information about routes was available either through maps or online, and (e) inter-city travel, especially, often created a level of schedule uncertainty. Among these issues, the ones that stood out as particularly addressable were making the shared ride system more efficient so that riders would not be forced to default to the more expensive taxi system, and increasing information awareness to give users the mechanism by which they could increase their personal safety but still make use of the shared ride system. Although the *bus system currently focuses on intracity travel, the development impact is perhaps even more significant for inter-city use where long-distance transportation resources are even more scarce. Research into transportation information systems in the developed world turned up municipal ride share systems such as 511 Rideshare in California’s Bay Area which allows riders and drivers to find one another; robust information resources available via the Internet include bus location information, and mobile-phone based systems (e.g., in Scandinavia) that provide up-to-the-minute schedule information so people can time their arrival at their bus, tram, or train stop. III. DESIGNING AN APPROPRIATE TECHNOLOGICAL SOLUTION We designed a system with the goal of improving access to transit information for potential bus riders at a minimum of cost to users and without requiring the assistance of a central authority. For our solution, we chose three existing technologies appropriate for Bishkek (and for much of the developing world as well) and augmented them with two new components we developed. We describe these engineered components later in Section V. The existing technologies we used were SMS, GPS, and GSM, all described below. A. Short Message Service (SMS) SMS, also known as text messaging, is a protocol to exchange short messages between mobile phones. Our system uses SMS as its primary communication mechanism. We chose SMS for several reasons. As discussed in Section II.D, mobile phone ownership and text messaging is common in Bishkek. Text messages are perceived as relatively cheap, costing anywhere from US$0.01 to $0.06 to send a message, and for many plans it is free to receive messages. Cell coverage in Bishkek is also excellent, and the country overall

237

has good coverage. Good urban and rural cell coverage is becoming the case in much of the developing world. Additionally, in the ICTD community, SMS-based solutions have proven robust, flexible, and valuable to multiple communities [11]-[13]. Work such as Frontline SMS and Warana Unwired demonstrate that information delivered via SMS can have a tangible, positive impact on individuals’ lives [14] [15]. Although the mobile web is a compelling solution space for some, our research in Kyrgyzstan and elsewhere points to the value of solving for cheaper devices and more lightweight technology solutions that dovetail with technology patterns already in place. Text messages, as opposed to richer mechanisms for distributing information, such as web pages, can be supported by cheap and ubiquitous phones; additionally, because of the way the mobile phone integrates into how people already get information in their daily lives, it is a better solution space than Internet-based resources [16]. B. Global Positioning System (GPS) GPS satellites transmit microwave signals to GPS receivers that use data from the signal to determine the location of the receiver. Our system relies on information provided by GPS satellites to determine the current location of buses participating in the system. GPS was chosen over other methods of location determination, such as GSM triangulation [17], for its high level of accuracy – within 8-10 meters, and its appropriateness for outdoor usage, particularly in rural areas or urban areas with few tall buildings and mostly wide streets. Although GPS units continue to decrease in price, they currently are only included as part of high end cell phones. Thus, our system attempts to minimize the number of GPS units required by allowing bus riders to leverage GPS units in buses to “tag” locations they are interested in querying about in the future (see V.B.3)). In this way, we are leveraging an expensive piece of technology that is rare in the community (GPS enabled phones are not a significant part of the market in most parts of the world) in order to make its capabilities serve the needs of a larger group. In the *bus system, multiple individuals can make use of the capabilities of GPS technology despite individuals not owning their own device. C. Global System for Mobile (GSM) GSM communications is the most popular mobile phone network standard. Most mobile phones, including those commonly used in Kyrgyzstan, are on GSM networks. In 2005, the GSM family of technologies had more than 1.5 billion subscribers, more than the total number of Internet users world-wide [18]. IV. PROPOSED SOLUTION The overall architecture of the *bus project, shown below in 6, incorporates the three existing technologies mentioned above with two new components. These engineered components are: a GPS vehicle tracking box (*box pronounced “Starbox”) and a back-end server that aggregates the location of the *boxes, runs a route prediction algorithm, Figure

and provides the querying service to riders. These are both described in more detail later in Section V. The goal of the *bus system is to track buses’ GPS coordinates and to provide users an SMS messaging system that will tell them when a bus should be arriving at their location. The *box is a low-cost vehicle tracking solution that utilizes GPS satellites to locate *box-equipped vehicles; the *box hardware includes a GPS device and a GSM modem integrated into a single package that just requires a local SIM card. The GSM modem allows the *box to send SMS messages with the GPS data to the server, taking advantage of the widespread GSM cellular phone networks. On the back end, the central server continually collects the GPS location data from all *boxes and stores it in a database. The server consists of a laptop computer connected to a mobile phone capable of sending SMS messages and that serves a gateway to the SMS services. The server does not need to be connected to the Internet, and the service does not require cooperation with mobile providers. Potential bus riders can use their cell phones to send SMS queries to the *box server and receive transit information in response. Initially, the system will just support riders' queries for transit arrival times, but it has been designed to be easily expanded to handle more complex activities such as route planning or registration for notification of bus location updates. V. ENGINEERED COMPONENTS Each bus driver participating in the *bus transportation network will need to have a *box device installed or placed on their bus with them at all times to allow riders to receive updates on their expected arrival times. At least one server, connected to a cell phone, also needs to be running at all times to receive location update messages from *boxes and to accept and respond to queries from users. That server can be located anywhere in the country with cell coverage. A. *box 1) *box Driver Interface The *box was designed to require minimal interaction with the bus driver. In order to use the *box, the driver only needs to turn it on (using a button located on the side), and then enter a route number using a keypad as shown below in Figure 7. The selected route number is displayed on a small LCD and stored locally on the *box to be appended to each location update message sent to the server. A version of the display and button labels is under development in Russian and can easily be adapted for other languages.

238

Figure 7 Prototype of the *box. Note the golden pigtail connectors on the top of the box to allow easy attachment of GSM and GPS antennas.

Figure 6 Overview of the *bus System

The notion of a route number is a good fit for the marshrutka system in Bishkek where buses have set routes (although no set timings along those routes), indicated by numbers displayed in bus windows. Changing routes is done easily by typing in the new route number. We decided to require drivers to specify their route via the keypad interface rather than forcing the server to do the more difficult job of attempting to determine it automatically based on the path of the bus, although this is also a future possibility. One could imagine a simpler device including a hard-coded route number, although we chose the keypad and LCD interface for our prototype to allow for ease of testing on different routes. The *box was designed to open easily to allow for upgrades of the internal components and changing of batteries, or inspection of device internals to allay fears of riders or drivers. Drivers can turn the *box off when not in service and safeguard the privacy of off-duty movements. Since we would expect deployment models to build in driver accountability, we do not consider theft to be a major concern. We built three prototype devices at a cost of about US$200 each, although we believe mass production would bring the cost per box closer to US$50. In the future, as GPS-enabled cell phones become less expensive, the functionality of a *box could potentially be replaced by an application running on a driver’s own cell phone. Currently, the holder of the *box, in this case the driver or private company that owns the marshrutka, would absorb the cost of the text messages sent out from the system. Our group has considered cost, and the barrier it presents to adoption. Increased efficiencies may or may not be significant enough for an individual driver to adopt the system, but there are other entities that might sponsor such a system for users, including a newly formed transit authority, mobile companies which are counting on increased SMS traffic from users, or the employing marshrutka company deciding to adopt the system.

2) *box Internals In addition to being able to accept route numbers via a keypad and display the provided route number to the driver, internally the *box must also be able to determine its location (via GPS satellites) and send a time-stamped SMS to the server. For our prototype device, we chose the Telit GM862GPS module. We chose to implement the *bus system with this very compact module because it provides a low cost solution with the additional convenience of combining GPS and GSM capabilities on one chip (see Figure 8). Because no additional work is necessary to interface the GPS and GSM components, developing on this chip is relatively simple. The Telit module communicates with the keypad via general purpose IO to obtain the driver's currently selected route number and stores the number in memory. An ATMega16 micro-controller is used to coordinate the communication between the Telit chip, keypad and the LCD. The module continuously obtains its GPS location information from GPS satellites without any action by the driver and reports this information along with its currently selected route number to the server via an SMS message. B. *bus Server The primary purpose of the server is to accept SMS messages coming from *boxes and bus riders, and to process them accordingly. System requirements are intentionally minimal: a laptop or desktop computer running a Java virtual machine and a MySQL database, connected to a phone capable of receiving SMS messages. Lower system requirements allow the server to run using older and less expensive hardware (such a system may cost around US$200 to $300). To minimize SMS latency, it is recommended that the phone connected to the server be on the same cell network as the *boxes, although this is not required. The server does not need to be connected to the Internet, facilitating its deployment in areas with low Internet connectivity. Our prototype server was implemented using MySMS [19], an application framework built on top of SMSLib [20]. MySMS provides a transport layer, allows SQL queries, notifications, remote code execution and rapid application development to be executed over SMS.

239

Figure 8 *box Internals.

The server is responsible for handling the following types of requests which arrive via SMS messages. 1) Location Update Messages from *boxes Messages sent from the *boxes to the server have the following format: !LOC,,,, The bus-id field is the unique identifier assigned to each individual bus that is equipped with a *box. We would expect that bus drivers would display this bus id (in addition to their route number) in a way that it is accessible to riders. Riders will need the bus-id in order to associate a name to a location (described later in 3)). The server records this information in a database for use in responding to user queries. 2) Arrival Time Queries from Users Information gathered about the current location of buses in the system could be used to answer a variety of queries. The server continuously receives and replies to SMS queries from users. Potential bus riders simply text to a special *bus server telephone number, and the server responds by sending an SMS back to the user’s phone number. The most basic example is to ask the server to predict the arrival time of the next bus travelling on a given route number at a given location: NEXT TO The server replies to the user with a SMS predicting: Next arrives to at about <estimated-arrival-time>. These examples demonstrate the types of queries we support in English, but the exact format and language can be changed. The system has been designed so that it can accept queries and push out information in alternate character sets – in this case, in Cyrillic to support both the Russian and Kyrgyz languages. While some issues related to a multi-lingual society pose significant challenges (for example, when a user sends a query in a mixture of two languages), the system can be modified to support linguistic variation. 3) Geo-coding Requests We expect that there will be some locations frequented by

many users that will have well-known names. These can potentially be associated with GPS coordinates in the server’s database by the server operator. However, the *bus system was designed to work in scenarios where little or no geocoded street information exists. The solution to this problem is that we allow the clients to geo-code a location, on the fly, by using the GPS equipment already installed on the *buses. When a user wants to be able to refer to their current location in future *bus queries, they can associate a name of their choosing with the GPS coordinates of their location; they do this by becoming a secondary user of the GPS on the bus. The user can instruct the server to associate an arbitrary name to their current location by waiting until a bus arrives at that location and immediately sending the following SMS message to the server: STORE AS <my-location> bus-id is the *bus id of the bus that has arrived at the user’s location. In this case, then, a user’s location that she knows only by street name is now given GPS coordinates thanks to the GPS receiver on the bus. The server takes the bus's GPS coordinates at the time that the message was sent and associates it with the location name specified by the user. Such geo-coded locations can be retrieved in the future based on the location name and the user’s phone number. Associating location names with phone numbers prevents naming conflicts between users, although other schemes are also possible whereby location names can be shared with other users, thus organically growing a set of tags for places of common interest. User tags could also be used to create maps in regions where few maps exist. This is especially important in Kyrgyzstan where the street names changed after independence from the Soviet Union and a combination of old and new names are used. The *bus system could be used to map streets in a way that is comprehensible to all residents regardless of which naming system they use. VI. PROTOTYPE DEPLOYMENT RESULTS We have implemented prototype versions of the *box hardware and the server software. In this section, we describe some preliminary data collected on the technical performance of the system. We deployed a *box on several buses in the University District area of Seattle. The *box sent SMS messages to the server containing its GPS coordinates at an interval of one message every 30 seconds. Based on this data the server composed a model of each of the routes. We were interested in examining the accuracy of our basic route prediction algorithm as well as the amount of time required for users to receive a response to their queries. Although we have not yet field tested the system in Kyrgyzstan, we have been careful to base the engineered solution on technology that will work in resource constrained environments. Specifically, fieldwork over the years has demonstrated to the research team that cell coverage and SMS latency in Bishkek is consistent with that experienced in Seattle. Additionally, the relatively open architectural space of Bishkek means that GPS readings will be easier to obtain.

240

Traffic patterns in the two cities are, of course, different, but the route prediction algorithm deployed (see next section) can be modified based on local traffic conditions. A. Route Prediction Accuracy 1) Route Model Much research has been done on route prediction algorithms for public transit and vehicles in general (e.g. [21]). For our prototype, we implemented a simple algorithm based on the concept of building a model by merging together multiple runs of the route. While one could imagine an implementation that requires the server to be pre-seeded with detailed sets of GPS coordinates for each route, our method provides the advantage of low setup costs for the *bus server, as no special action is required to enter routes into its database. Furthermore, prediction models can be made more accurate by merging additional runs of the route, and more models for varying traffic patterns can easily be created. More sophisticated algorithms can be used in future versions, such as ones which automatically account for varying traffic flow and are tolerant of changing routes. We use our route prediction model to provide responses to user queries about when the next bus travelling on a certain route is due to arrive at a particular location. To make this prediction, the server takes recent coordinates from current runs of the route and the location entered by the user and attempts to find these points in the model. The time between the corresponding locations in the model provides an estimate of the amount of time until the bus arrives at the user's location. The accuracy of these estimated arrival times depends on two things: first, the fidelity of the model to the path of the route and secondly, the timing similarity between the route model and the buses whose status is being queried. The model fidelity is largely determined by the number of points in the model. A point consists of a set of GPS coordinates and the GPS satellite-provided time of day that the data point was measured. Accuracy of data points should be quite good (GPS accuracy is within 8-10 meters). Thus, data points collected for earlier runs on a given route should provide an accurate picture of where those buses were at the time provided. Number of data points collected per route is determined by the frequency that *boxes are configured to send location update messages to the server. More frequent location update messages incur more cost and put more load on the server, so it is desirable to keep this frequency low, yet high enough to give accurate arrival estimates. Figure 9 below shows a plot of *box location data showing the level of accuracy provided by different update message frequencies. As the figure shows, in general, sending update messages more frequently is more likely to create a model that follows the actual path of the route. Particularly around curves and turns, less frequent updates can make it seem as though the bus has cut a corner.

(a) Plot of the acquired GPS data for the whole route (approximately 3 miles) (Google Earth)

(b) Detailed view of the plot of acquired GPS data (Google Earth) Figure 9 Accuracy and resolution of GPS data collected at different message send intervals: red – 10 seconds yellow – 20 seconds, blue – 30 seconds, purple -40 seconds, green – 60 seconds.

One option is to first produce one run with finer granularity, for example, a location update every 10 seconds. This can also be done manually to specify the exact path of the route. The method we employ takes points from multiple runs of the route and combines them to produce a model with greater resolution than a single run. Thus, more runs of the same route contribute to overall accuracy of the prediction. In order to provide accurate estimates of arrival times, a route model must also reflect the timing characteristics of buses whose status is currently being queried. The extent to which this is true depends on multiple factors. On a bus system such as the one in Seattle, where there are set stops, and drivers are expected to arrive at stops at given times, variance can be caused by factors such as weather or unusually high traffic. In Bishkek, although marshrutkas follow set routes, there is no schedule of any sort. Thus, we would expect that there might be a wider variance in timings for buses in Bishkek. Collecting timing data from more bus runs and using an algorithm that incorporates time of day and day of week would help to improve accuracy in this case. 2) Prediction Accuracy Results We deployed *boxes on portions of two Metro bus routes in Seattle and collected data for four runs on each of these routes for a total of eight runs. The distance traveled was 2 miles on one route and 2.8 miles on the second. Each run consisted of an average of 11 minutes of collected data. While this running time is shorter than the length of routes we would expect to use the system with, it allowed us to collect enough data to test and provide a proof of concept. To examine the accuracy of our preliminary route prediction algorithm, we did the following. For each route, we chose one to three of the four runs to create a model. The remaining runs, which did not become part of the model, were treated as a bus

241

in motion and we ran our algorithm to generate predictions about those runs. These predictions were made for five to six simulated user locations per route, and when the bus was anywhere from one to nine minutes away from that simulated user. We provided the algorithm with the model, a time to treat as the current time of the system, calculated by subtracting the expected prediction value from the actual arrival time, and a set of location updates prior to this time, accounting for location message delays (see VI. B). We compared the prediction generated to the actual timing data gathered for that run for a total of 547 measurements of error. The error in a prediction is the absolute value of the difference between the actual time it took for a bus to arrive at the destination and the time predicted by the model. The mean error from our predictions was approximately 94.9 seconds, with a standard deviation of 83.2 seconds. As seen in Figure 10, the amount of error in the prediction increases with the amount of time the bus actually is away from the user. The error can be understood as a proportion of the actual time to arrival of the bus, with the mean proportion equal to approximately 37.9% with a standard deviation of 32.2%. For example, if the bus will actually arrive at a destination in 10 minutes, we can expect the prediction of this arrival time to be, on average about 3 minutes 45 seconds off. In addition to the accuracy of our model, the amount of time required for SMS messages to travel through the network, both those giving location updates and those from user interaction, also affects the accuracy and usefulness of bus arrival time predictions. We examine these factors below. B. Message Latency 1) *box to Server Latency The accuracy of a particular prediction depends on both the accuracy of the route model and on receiving a recent, and thus accurate, location update from the bus whose status is currently being queried. Ideally, the server would have instantaneous information about the exact location of a moving bus at the point it receives a user query about that bus. In practice, the server must base its prediction on the last location update message received from that bus. Both the configured *box frequency of sending updates and any delay in that message getting from the *box to the server affects accuracy of information about the current bus location. Latency of SMS messages can be affected by a variety of factors, such as cell phone network congestion, signal coverage for the *box or the server, or the need for messages to cross networks. We recommend that the *box and the server be on the same cell phone network, as they were in our implementation, to minimize this delay.

Figure 10 Mean error in the predicted arrival time, plotted by the actual time to arrival (both axes in seconds).

We examined 381 messages sent from the *box to the server to measure this latency. The average latency, measured from the time it was sent by the *box to the time it was received by the server, is 19.96 seconds, with a range of 10 to 40 seconds (standard deviation of 10.3 seconds). We consider this an acceptable latency for our particular application. Although these measurements were taken in Seattle, fieldwork over the years has demonstrated to the research team that cell coverage and SMS latency in Bishkek is consistent with that experienced in Seattle. 2) User Query Latency Queries return information to users in the form of an estimated time (e.g. 5:30pm) that a bus is expected to arrive at their location. The usefulness of this information depends upon both its accuracy and how quickly it is returned to the user. Factors affecting query processing latency include: server load, cell phone network load, and system limitations on rate of SMS message sends. Based on our measurements of message latency in the previous section, we would predict the roundtrip time for the user to receive a response to their query to be the latency of two SMS messages (assuming time spent at the server is negligible), or approximately 40 seconds. Considering that the typical usage scenario would return predictions on buses several minutes away, we consider this latency to be acceptable. VII. CONCLUSIONS AND FUTURE WORK The *bus system is a direct response to an articulated problem in the developing world, and it takes into close consideration technology usage patterns and technology infrastructure available in such resource-constrained environments. The system recognizes that information is valuable, and that the lack of information inhibits certain kinds of growth and efficiencies. However, there are not always central authorities or institutions with the resources or motivation to supply information infrastructure for citizens, and the *bus system provides a mechanism by which decentralized operators of a

242

transportation system can create information resources for themselves. It is also a system that uses mobile technology in a way that does not require the direct participation of the mobile provider, thus cutting out one potential barrier to adoption. Finally, *bus enables community use of a single GPS system to allow for user-generated geo-coding of the environment. Next steps for *bus include testing in Kyrgyzstan in early 2009. Our goal is to test both intra-city transportation in Bishkek, and inter-city transportation elsewhere, perhaps in another country. Although the *bus system has been largely built out with the concept of urban transport in mind, the development impact for rural areas is conceivably even greater. In areas where distances are long, roads limited, and vehicles intermittent, a system such as *bus could have dramatic implications. Future work on *bus includes examining possibilities such as incorporating some way to indicate the capacity of buses or allowing communication between drivers and riders, particularly for longer inter-city routes. A richer set of queries may be developed that allows users to subscribe to a notification system, so that they only need to send one SMS message to the server in order to receive several bus arrival time messages. Another potential expansion is to allow users to share tagged location data. Further investigation into making the system more scalable by using multiple phones per server for high density areas and by running stress tests is needed. Lastly, because the current system is based on SMS usage which generally requires a certain literacy level, further work could explore the possibility of voice interfaces. Whether moving goods to market or people to medical clinics, transportation is a crucial resource for any community. Allowing individuals to find ways to maximize such shared resources is the goal of *bus. Our approach has been a method of problem-solving that is bottom-up, that conceptualizes a technology device as shared among the community, and that recognizes the importance of information in people’s everyday lives. Our goal with this paper has also been to bring to the fore the importance of transportation for both economic and human development – an area of ICT for development that has not yet received much attention. ACKNOWLEDGMENTS We would like to thank members of the original *bus and MySMS teams of the 2007-2008 CSE477 at UW: Martin Hecko, Naasir Ramji, Jordan Walke, Brad Campbell, Ruibo Li, and David St. Hilaire. We would also like to thank CAICT team members Cynthia Putnam, Emma Rose, Rebecca Walton, Erica Johnson, Mark Licata, and also Professor Medina Aitieva, our local research team, and research participants. We thank Aidai Seidakmatova, Shairbek Juraev, and Natalie Linnell for their contributions. A special thanks to Bruce Hemingway and Jesse Dosher for the design and construction of the hardware.

REFERENCES [1]

[2]

[3]

[4]

[5]

[6] [7] [8] [9] [10]

[11]

[12]

[13] [14] [15]

[16]

[17]

[18]

[19] [20] [21]

G. Weisbrod, “Models to predict the economic development impact of transportation projects: historical experience and new applications,” The Annals of Regional Science, Volume 42, Number 3 / September, 2008. S. Démurger, “Infrastructure Development and Economic Growth: An Explanation for Regional Disparities in China?” Journal of Comparative Economics, Volume 29, Issue 1, March 2001, pp. 95-117. D. Parikesit and K. Czuczman, “Transport, the missing link? A catalyst for achieving the MDGs,” id21insights, #63, July 2006, http://www.id21.org/insights/insights63/insights63.pdf “Transport's Role in Achieving the Millennium Development Goals,” Produced by the DFID Transport Resource Centre, August 2002, Available: http://siteresources.worldbank.org/INTWDR2004/Resources/22572_Tra nsportsRole.pdf A. Bradbury, “Transport, mobility and social capital in developing countries,” Proc. of the Institution of Civil Engineers, Engineering Sustainability 159, June 2006 Issue ES2, pp. 79-86. NextBus, Last accessed: 22 Sept. 2008. http://www.nextbus.com/ Metro Tracker, Last accessed: 22 Sept. 2008. http://transit.metrokc.gov/oltools/tracker.html MyBus, Last accessed: 22 Sept. 2008. http://mybus.org/ The World Factbook, CIA, Last accessed: 22 Sept. 2008. https://www.cia.gov/library/publications/the-world-factbook/ Pew Internet and American Life Project, “A typology of Information and Communication Technology Users”, May 2006. Available: http://www.pewinternet.org/PPF/r/213/report_display.asp J. Donner, “Research Approaches to Mobile Use in the Developing World: A Review of the Literature”. The Information Society 24(3), 2008. “Improving Rural Mass Transit in South Africa - Team Smile,” Last accessed: 22 Sept. 2008. http://blogs.technet.com/unlimitedpotential/archive/2008/07/03/improvi ng-rural-mass-transit-information-in-south-africa-team-smile.aspx MSR India SMS Toolkit, Last accessed: 22 Sept. 2008. http://www.codeplex.com/smstoolkit FrontlineSMS. Company Website, Last accessed: 22 Sept. 2008. http://www.frontlinesms.com/ R. Veeraraghavan, N. Yasodhar, and K. Toyama, “Warana Unwired: Replacing PCs with Mobile Phones in a Rural Sugarcane Cooperative,” in Information and Communication Technologies for Development, 2007. B. Kolko, E. Rose, and E. Johnson, “Communication as informationseeking: the case for mobile social software for developing regions,” Proceedings of the 16th international conference on World Wide Web, 2007. P. Javid and T. Parikh, “Augmenting Rural Supply Chains with a Location-Enhanced Mobile Information System,” in Information and Communication Technologies for Development 2006, May 25-26, 2006, Berkeley, CA. M. Chen et al., “Practical Metropolitan-Scale Positioning for GSM Phones,” in UbiComp 2006: Ubiquitous Computing, Springer Berlin / Heidelberg, 2006. MySMS, Last accessed: 22 Sept. 2008. http://mysms.sourceforge.net/ SMSLib, Last accessed: 22 Sept. 2008, http://smslib.org/ A. Karbassi and M. Barth, “Vehicle route prediction and time of arrival estimation techniques for improved transportation system management,” in Proc. of Intelligent Vehicles Symposium, 2003. IEEE, June 2003, pp. 511-516.

243

Challenges in Health Information Systems Integration: Zanzibar Experience Edwin NYELLA Department of Informatics, University of Oslo, Box 1080, Blinden 0316 Oslo, Norway Tel: + 47 22 85 24 10, Fax: +47 22 85 24 01, Email: [email protected] ABSTRACT -- The healthcare milieu of most developing countries is often characterized by multiplicity of health programs supported by myriad of donors geared towards reversing disease trends in these countries. However, donor policies tend to support implementation of vertical programs which maintain their own management structures and information systems. The emerging picture overtime is proliferation of multiple and uncoordinated health information systems (HIS), that are often in conflict with the primary health care goals of integrated district based health information systems. As a step towards HIS strengthening, most countries are pursuing an integration strategy of the vertical HIS. Nevertheless, the challenges presented by the vertical reporting HIS reinforced by funds from the donors renders the integration initiatives ineffective, some ending up as total failure or as mere pilot projects. The failure of the systems after implementation transcends technical fixes. This paper drew on an empirical case to analyze the challenges associated with the effort to integrate the HIS in a context characterized by multiple vertical health programs. The study revealed the tensions that exists between the ministry of health which strived to standardize and integrate the HIS and the vertical programs which pushed the agenda to maintain their systems alongside the national HIS. However, as implied from the study, attaining integration entails the ability to strike a balance between the two forces, which can be achieved by strengthening communication and collaboration linkages between the stakeholders. Index Terms -- Health information systems, Integration, Standardization, Vertical programs

I. INTRODUCTION Health information systems (HIS) in developing countries have in recent years gained more and more attention as more effort by governments, international agencies, non governmental organizations, donors and other development partners seek to improve healthcare as a way to reverse disease trends in these countries. As part of the efforts towards reversing disease burden and provision of improved healthcare services, most countries are implementing primary health care (PHC) approach. However, implementation of the PHC approach in most countries is organized in decease-focused and specific services programs (such as Malaria, HIV/AIDS, TB &Leprosy and Family Planning programs, etc), dubbed as ‘vertical programs’ [2] [5] [12] [13]. The vertical programs are mostly funded by donors who come with specific requirements related to the monitoring and evaluation of their funds and the program at large. For instance in my case study, the HIV/AIDS program alone is supported by a number of donor agencies such as the Global fund, Center for Disease Control (CDC), United Nations for Development Programmed (UNDP), United Nations Children's Fund (UNICEF), World Bank and the World Health Organization (WHO), where almost each agency has a number of indicators which require huge amounts of data to be collected.

Due to the donor pressure associated with the vertical programs, the health information systems evolved in a rather chaotic and fragmented manner with multiple and overlapping demands from both the vertical programs and the national health administrative departments and ministries. The vertical programs usually maintain their own ‘vertical’ reporting health information systems existing alongside with the national health information system where the result emerging over time is uncoordinated, disintegrated and heterogeneous collection of systems [8] [12]. As the corollary, many studies reported the dismal state of the health information systems as being predominantly ineffective, unreliable, irrelevant and therefore inadequate in providing the management with the needed information [3] [8] [12] [13] [15]. In an attempt to ensure availability and accessibility of comprehensive health information to the national health departments, districts and the vertical programs; most countries are pursuing an integration strategy of the fragmented systems. Whereas, some countries have managed to standardize and integrate some of the vertical programs into the national HIS, ensuring continue reliance and use of the integrated system by the vertical programs is still a big challenge [16]. The fragmentation of the HIS after integration, ensuing from the non reliance and compliance to the integrated system, has not received much attention

244

either empirically or analytically by the contemporary research in information systems integration. Much of the research focused on the challenges in the process of achieving integration, for instance, [1] examined the potential and challenges of integrating the HIS of Malaria, TB and HIV/AIDs programs and also integration of multiple reporting channels within each program. Another study looked at the challenges posed by the historicity (the conservative influence of historically accumulated and institutionalized practices, technologies and perceptions) and heterogeneity (lacking integration and increasing fragmentation across the collection of information systems) of information systems in the development and integration of the health information systems [8]. [9] looked on the problems of fragmentation and challenges of integration of the routine health information system and the prevention of mother to child transmission program (PMTCT). The thesis of this paper is the emphasis of the need to comprehend the nit gritty of what goes on after integration. By specifically understanding the way vertical program managers (as users of the HIS) receive and relate to the newly implemented system, can help us answer though partly as to why some health information systems integration in the context of developing country fail or run short of expectations, after implementation. The rest of the paper is organized as follows: The literature covering challenges of HIS integration in developing countries are presented on the subsequent section followed by the research context and methodology pursued in the study. The empirical basis of this paper is an integration effort in Zanzibar, which is presented next and followed by the analysis and discussion section. Finally, the conclusion section of the paper is put forth, where implications and contributions of the study are spelt out. II. CHALLENGES IN HIS INTEGRATION Health information systems integration in developing countries is considered as an approach towards the rationalization and unification of the disparate systems in these countries. The objective of integrated HIS is to provide easy and equal access to relevant information to all stakeholders [12]. However, ensuring an integrated HIS in these settings is quite a big challenge. Generally, the challenges of HIS integration emanate from both social and technical factors [13] surrounding the integration processes. It is argued that it is more so in developing countries due to

contextual particularities related to politics, institutional conditions, high resource constraints (infrastructure, human resources, financial resources), high disease burdens and the particularities of the diseases, in which all together challenge the process of integrating the HIS [1] [17]. Most of the developing countries are funded by international donor agencies such as the World Bank, Global fund, and the Clinton Foundation, in order to support provision of health services (such as Family Planning, Immunization and VCT) to the population. However, donor policies tend to support implementation of disease specific programs dubbed ‘vertical programs’ which maintain their own management structures and information systems [2], which are often in conflict with the primary healthcare goals of integrated district based health information systems. For the case of Zanzibar for instance, the HIV/AIDS program alone which maintains its own information system is supported by a number of donor agencies such as the Global fund, Center for Disease Control (CDC), United Nations for Development Programme (UNDP), United Nations Children's Fund (UNICEF), World Bank and the World Health Organization (WHO), where almost each agency has specific requirements concerning data to be collected. In a study on HIS of the disease specific programs in Mozambique, a low income country, [1] identified major challenges related to the integration of the HIS to include heterogeneity of interests among donors, managers and health reformers; multiplicity of reporting systems even within an individual program and high disease burden. [20] discussed how poor infrastructural conditions and lack of transport challenge the flow of health information from the district to the provincial levels of the health administration hierarchy. Low or lack of computer skills have also been quoted as one of the factors contributing to the challenges to attain an integrated district based HIS, especially in the rural context of most developing countries [8] [6]. Furthermore, lack of uniform infrastructure development and uneven distribution of resources (e.g. humans, computers) in most developing countries, challenge the efforts to attain comprehensive integrated health information system [18] [4]. For instance, the uneven distribution of human and technical infrastructure in Mozambique was reported as being problematic to the effort of scaling up the district-based health information systems because some of the remote districts do not

245

have electricity. Similar problems of uneven infrastructure development was experienced in Ethiopia, where according to [18], use of standardized data formats served as gateways between the paper based systems at the periphery levels and computer software at the higher levels of the health system hierarchy. Moreover, [10] alluded to some of the challenges faced by most developing countries in achieving standardized and integrated HIS to include: conflicting interests between health programs which make it difficult to reach a “final” agreement; changes being the only constant, where new needs keep popping up (e.g. HIV/AIDS); and multiple software and paper tools which are difficult to coordinate and change. Whereas, some countries have managed to standardize and integrate some of the vertical programs information systems into the national HIS, ensuring continue reliance and use of the integrated system by the vertical programs is still a big challenge [16]. As presented by [21], some of the reasons for the categorical programs refusal to support an integrated HIS include: fear that their requirements will not get the attention needed to ensure that their needs are met, and if the programs have pride of ownership in their existing vertical systems, which meet their needs. The fragmentation of the HIS after integration, ensuing from the non reliance or refusal to support the integrated system, has not received much attention empirically and analytically by the contemporary research in information systems integration. As the literature depicts, much of the research focused on the challenges in the process of achieving integration. The argument of this paper is the need to comprehend the nit gritty of what goes on after integration. By specifically understanding the way vertical program managers (as users of the HIS) receive and relate to the newly implemented system, can help us answer though partly as to why some health information systems integration in the context of developing country fail or run short of expectations, after implementation. III. RESEARCH CONTEXT AND METHODS The research study was conducted in Zanzibar. Zanzibar is a semi-autonomous region within the United Republic of Tanzania, which is made up of two main islands, Unguja and Pemba, and several others islets located in the Indian Ocean. Zanzibar has an area of 2,332 square kilometres (the total area of Tanzania is 945,000 Km), and is divided into five administrative regions, each with two districts, making a total of ten districts in the entire region. Zanzibar maintains its own health system that is administrated by a semi-autonomous Ministry of Health and Social Welfare (MoHSW). Side by side with the health system is an information system called Health Management Information System

(HMIS), meant to provide information support to all decision making processes of the entire ministry. In this paper, the term ‘Health Information System (HIS)’ is used to refer to the HMIS and the term ‘HMIS unit’ is used to refer to the national level MoHSW department, responsible for the health information system. The study was done as part of an ongoing action research on health information system restructuring process undertaken by the Health Information System Program (HISP) in collaboration with the MoHSW and other development partners in Zanzibar. HISP is a global action research and development network on health information systems by the University of Oslo in Norway and other stakeholders, which started in South Africa in 1994 and thereafter spread to other developing countries such as Ethiopia, Vietnam, India, Mozambique, Botswana, Tanzania Mainland, Malawi and Nigeria. In all these countries, the primary goal of HISP is to design, implement, and sustain Health Information System through participatory approaches to support local management of health care delivery and information flows [10]. In 2005, Zanzibar became a node in the HISP network, when the University of Oslo was contracted by the MoHSW to engage in the restructuring of the HIS. Case study methodology was drawn upon in the field during data collection and analysis. Case study is an appropriate approach for bringing an understanding of a complex issue, which could be a program, event, an activity or a process involving one or more individuals and using a variety of data collection procedures over sustained period of time [11]. The aim of the research being to develop a comprehensive understanding of the challenges related to the way users received and engaged with the integrated HIS, case study proved to be a feasible approach. The research was carried out in three health districts; two in Unguja and one in Pemba Island within a period of five months (June to November, 2006).The districts visited in Unguja were Urban and West; both of them located on the west region of the island. Most of the MoHSW main offices such as HMIS unit and vertical programs’ offices are located in these two districts. In Pemba the research was conducted at Chake Chake district, which is located on the south region of the Island. Semi structured interviews were used, where interviewees were asked open ended questions to elicit their viewpoints related to the use of the integrated HIS. At the health unit level, 38 informants were interviewed, 19 in Unguja and 19 in Pemba, where impressions and perceptions of the health workers in relation to the previous and the newly integrated HIS were gathered. The goal was to learn micro level challenges emanating from the way users responded to the integrated HIS. At the macro level the interviews involved the following informants: District medical officers, District health

246

officers, vertical programs district and general managers and central level HMIS unit officials. During the interviews different informants’ viewpoints in relation to the new datasets and tools were gathered. Table 1 depicts the number of informants interviewed. Name /Position Health unit level staff District Medical Officers Vertical programs data managers Programs general managers National level HMIS unit officials District Health Officers HISP team members TOTAL

Informants 38 3 7 2 3 3 2 58

Table 1: Interviews conducted at health facility and above

A number of documents and software tools were analysed in the field. For instance the district implementation plan was analysed in an attempt to understand the use of the new system in the preparation of the plan. The software tools were analysed to assess compliance to the newly integrated system. IV. CASE DESCRIPTION The health information system in Zanzibar was organized haphazardly and mainly shaped by the organization of fragmented vertical programs with their own information systems [14]. The vertical programs include the Family Planning (FP), Malaria, and Expanded Program on Immunization, Tuberculosis and Leprosy, Nutrition, HIV/AIDS and Safe motherhood programs. Most of the programs’ services are integrated at the health unit level, but maintain separate reporting systems. A situational analysis of the health information system disclosed plethora of problems which included scarcity of resources, gaps in data collection tools, poor analysis of data, fragmentation of the higher levels, poor feedback and lack of motivation and limited information use [14]. As an attempt to improve the situation, a strategic plan was drafted by a joint team of stakeholders; comprising both scientific and organizational researchers, major Ministry of Health and Social Welfare (MoHSW) donors (DANIDA and WHO), University of Oslo and some officials from the MoHSW (ibid). The roadmap detailed and agreed on the major activities to be undertaken. The University of Oslo under its action research program called Health Information System Program (HISP), was contracted to undertake the task of restructuring the system by way of integrating the highly fragmented HIS. HISP as explained hitherto is a research network that aims at enhancing district health information systems in developing countries through introduction and local adaptation of open source software. The software which is known as District Health Information Software (DHIS) was developed and

adapted in South Africa and subsequently adopted in other developing countries such as Malawi, Mozambique, India, Ethiopia, Namibia, Zambia, Mali, Botswana, Vietnam and Nigeria. The HMIS restructuring process followed a participatory and incremental approach in the revision of the previous datasets and creation of new once. The incremental approach paved way for the involvement of key stakeholders in each stage of the revision exercise. Among the new datasets included was the Expanded Programme on Immunization (EPI), Disease surveillance, Reproductive and child health (RCH), STI and HIV, and Maternity dataset. Almost each dataset brought together a number of stakeholders for instance the Disease surveillance dataset apart from the national HMIS, had two main stakeholders, the EPI and Malaria programs. Malaria program in Zanzibar is very strong unlike in other countries due to many interventions by donors directed towards reversing the malaria trends in the country. This made malaria related data more valuable purposely because of the need to monitor and evaluate the interventions. The disease surveillance being the main source of the malaria data, the program was involved in the design implementation stages to ensure that the dataset met the requirements needed. For instance, the program’s data manager was involved in the training of the new disease surveillance dataset. However, despite of the involvement the data manager enacted separate data collection tool behind the scene and circulated it in some districts. When asked why, the answer was very clear, “They have taken out almost all the age group categories and left what they feel will satisfy their needs, but what about us. So we tried to design it to show them how it should look like” (Manager, Vertical program, August, 2006) Likewise, the EPI Disease surveillance tool (see Figure 1) was integrated in the new dataset for disease surveillance.

Figure 1: Monthly disease surveillance report

Although the new dataset was running for more than eight months (till the time of writing), still EPI kept

247

on collecting data using their system. The reason, as explained by one official was that, “.. Until we are sure of getting our data from HMIS, we can not abandon our system” (Manager, Vertical program; July, 2006) The aftermath of that practice was duplication of work at the point of data collection. Similarly, in some districts, family planning was running in parallel with the new integrated system despite the fact that it was integrated in the RCH dataset, and monthly submitted to the district. The practice was commonplace until the time of writing. When some health workers at the health facility level were asked why, one of them answered, “.. The new forms do not have all the required data elements as the old ones. So we fill in the old one to make sure that all the required information is taken to the owners (the Family Planning program) “(RCH coordinator, July 2006). Another health officer noted, “We have not been told to abandon them; we still submit them monthly to the district.” (MCH Aides, August 2006). At the district level some data officers kept on demanding submission of the family planning report using the separate data tool. This was partly attributed to by the inadequate knowledge about what was supposed to be done as far as the old and the new reporting system was concerned. Lack of teamwork and sharing of information at the district level between those who participated in the design process and those who didn’t led to this malady. For, most of the HMIS work at the district level was in most cases done by those who were not involved in the process. HIV/AIDS was another program which despite being involved in the process of designing new dataset for HIV and STI services, kept on running their previous tools separately. This program maintained its own fragmented information system, one for VCT and another one for STI services. These subsystems were integrated into one STI and HIV/AIDs dataset. The dataset was functional for more than six months with data routinely collected and collated from almost all Health facilities providing the two services. Although most of the data was submitted to the districts and transmitted to higher levels, the data was not fetched and used by the HIV/AIDs program. Instead, the program kept on depending entirely on their previous systems. The reason given was that the new dataset does not fulfil data requirements for program management and so it was not designed for the HIV program but for the HMIS unit. “The new tools are for the higher levels only; they can not help us in any way. We need more information compared to what is on the HMIS form. It is not designed for us” (Manager, Vertical program; July, 2006)

Though the program officers participated in the design process of the dataset, the participation as explained by one officer was meant to help HMIS unit get HIV/AIDS related data. Lack of trust by the vertical program to the capacity of HIMIS unit to maintain and sustain the information system was envisioned as one of the reasons: “……. Mostly we rely on donors in almost everything which sometimes lead to mistrust by the vertical programs of our capability to maintain and sustain the information system. For instance EPI are performing well because they have enough funds. Also HIV/AIDs have many donors which imply enough funds, unlike HMIS which has very scarce resources both physical and human resources”. (HMIS Official, July, 2006) As an attempt to resolve the dilemma according to the HMIS official, the HMIS unit resorted to participatory approaches such as meetings, workshops and seminars to build consensus. An example being mobilization of concerted efforts and funds by HMIS unit to resolve availability of data collection tools problem. The problem occurred when HMIS failed to fund production of data collection tools due to financial constraints. To resolve the problem which if left unsolved would have undermined the whole system, vertical programs as one of the major stakeholders were summoned in a meeting to deliberate and agree on strategies to solve the problem. In the meeting, it was agreed that vertical programs contribute on the production of data collection tools. Most of the programs considered the idea as one of the feasible solution for the problem. Those who were at first reluctant, slowly as they saw others responding, they also followed suit. Based on the agreement, production of both primary and secondary data collection tools to be used for a period of one year was done using funds from different vertical programs. As an attempt to explain the reason and solution for the continued use of the previous systems one stakeholder from the donor community argued: “We have agreed if possible not to talk about integration. For when people working in these disparate systems hear this, they think of being robbed of their jobs and hence their salaries. This problem is more pronounced in this context where there is massive unemployment rate. We better talk about communication and collaboration between the vertical programs and the HMIS unit” (Program director, Donor Community; Nov, 2006).

248

V. ANALYSIS AND DISCUSSION The integration of the health information system (HIS) involved standardization of the datasets, data collection tools, data processing tools and associated work practices; and institutionalization of the standards to the levels of the health system. Following is the analysis and discussion of the challenges resulted from the way vertical program managers received and related to the newly integrated system. Limited use of the integrated HIS The integration initiative sought to align different disparate fragmented information systems to form an integrated data repository at the district level which gives access of data to different stakeholders. Though, this was achieved to a certain degree, some other vertical programs whose datasets were aligned with the new initiative enacted limited use of the newly integrated HIS by running some of their previous information systems and tools in parallel with the new system. For instance, the disease surveillance dataset for EPI program was aligned with the national disease surveillance standard. Though the national dataset was in operation for more than eight months until the time of writing, the vertical program managers decided to run their disease surveillance data collection tool in parallel with the national system. This vertical program however had another dataset (on immunization) which was aligned with the new initiative and which was running smoothly. The immunization dataset is an independent dataset, with the EPI program as the main stakeholder. This is different from the disease surveillance dataset which was formed by fusing together the national disease surveillance dataset and the EPI surveillance dataset. To regain control of their disease surveillance system, the program managers decided to run their dataset alongside the integrated dataset for disease surveillance. Furthermore, the limited use came to play as the result of the lack of trust by the vertical program managers to the capability of HMIS unit to ensure sustainability of the integrated HIS. This was evident from one of the program manager’s assertions that, unless we are sure of getting our data, we can not abandon our system. Therefore, the managers drew on the history of poor economic status of HMIS unit and on their future prospects about the new system’s sustainability to make the decision to maintain their system alongside the new system. Another health program whose users enacted limited use of the new system was the family planning. In some districts, family planning dataset was running in parallel with the newly integrated system despite the fact that their data elements were integrated in the RCH dataset. This resulted as users drew on their past practices of submitting huge amounts of data to the vertical programs unlike in the new standardized

RCH tool where the family planning data elements were minimized. Most of the users argued that they were using the old tools to ensure that information owners get all the information they needed. In addition, by building on their past practices, some district officers kept on enquiring about the separate family planning reports from the health facilities. Health officers in some districts drew on their limited knowledge of DHIS software to enact limited use of the integrated HIS, where preparation of monthly reports was done by aggregating manually a number of datasets, an exercise which took two to three days to finish. While at the same time, all datasets were already entered in DHIS by the data clerk, in which a monthly report could be printed out without much hurdle. The decision to go back to the manual systems can be envisioned as to have been made due to the dilemma of not knowing how to use the new systems but also by the demand of the need to compile monthly reports. Other district health officers enacted limited use by engaging themselves more with a dataset for a particular vertical program. This came to play as the result of the officers drawing on their past experiences of vertical system mindset contrary to the new integrated HIS where all the datasets needed to be afforded equal attention. This was evident from what one health officer provided as an answer when he was confronted for poor performance of some datasets, where he asserted that his responsibility was a particular dataset whose performance was good. The limited engagement with the integrated HIS was further observed at the health unit level where health workers dealt with particular datasets more than others. This limited use at the health unit or district level came to play as users acted on their past practices related to vertical programs to conceptualize the new system which entailed equal treatment of all datasets to ensure availability of comprehensive data at all levels. Completely none-use of the integrated HIS Some other actors determined to completely shun away from the newly integrated HIS in spite of the fact that, they were involved in the design process of datasets related to their program. HIV/AIDS program managers participated in the design of a new dataset called STI and HIV dataset, which aimed at collecting data related to both services. However, this dataset was operational for about five months until the time of writing, but since then the data collected was not fetched and used by the program. The idea of having both systems running in parallel may sound logical, because the setup gives every stakeholder access to the data; contrary to the previous systems where data was vertically submitted to the programs and donors. However, the setup has enormous implication on the workload to the data collectors. Rather than rationalizing the fragmented systems and minimize

249

duplication of data, the setup intensify it and ultimately jeopardize the quality of the data collected. The none-use of the integrated HIS, came to play as the program officers’ drew on a number of assumptions and on their multiple needs of data. Some of the assumptions in relation to the new system are that, the new system was for HMIS unit, and so their participation aimed at helping them get their data and that the data collected in the new system did not satisfy their needs. The assumptions can be envisioned to have resulted from users drawing on their past experiences of collecting multiple data and on their desire for data to meet future needs. The none-use enacted by the HIV program as was the case for the limited use explained above, was also mediated by the poor economic conditions of the HMIS unit, which led to mistrust of the vertical program to the capability of HMIS unit to ensure sustainability of the system over a period of time taking into account its almost total dependency on donors. Tensions in the HIS Integration The resulting picture from the limited and none-use of the newly integrated HIS, is what I dubbed as ‘pulling effect’ (see figure 2). National HIS









Struggle to standardize & integrate the HIS Advocate on minimum action-led datasets No enough funds to sustain the system Use of participatory approaches to build trust

Vertical Programs








Want to maintain their own systems (fragmentation) Advocate multiple data needs to satisfy multiple and heterogeneous donors Have funds to develop & maintain their systems

Pulling Effect Figure 2: Tensions between the national HIS and the vertical programs

On one side HMIS unit under the MoHSW is struggling to standardize and integrate the fragmented information systems and on the other side the vertical programs opted for limited and none-use of the newly integrated HIS by maintaining their own systems. The upshot of that is a pulling effect on either side where the winner is determined by the power (e.g. to argue, funds, human resources, good strategies.), which one of the two sides need to have in order to haul the opponent. In an attempt to alleviate or eliminate the tensions, the HMIS unit ventured to consensus building through participatory approaches (PA) like meetings, workshops and seminars to try to strike a balance between the two forces. This is exemplified by mobilization of concerted efforts and funds by the

HMIS unit to solve availability of data collection tools problem. Nevertheless, it must be emphasized that the tensions which resulted from limited and none-use of the newly integrated HIS were not static, but dynamic in nature. The level of dynamism was different from one user group to another and from one vertical program to another. For instance, the HIV/AIDs which initially opted for none-use of the integrated system, at the very end of the field study in an interview with the data manager; he sanctioned the software to be installed in their computers. His acceptance came due to the need to make comparison between the data they collect and the HMIS data, supplementing their data in case of some missing data in their systems. Though the manager took the decision after learning that having the new system won’t prevent him from using their systems, but I see this as a movement from none-use to limited use. Furthermore, the health units and districts officers who opted for limited engagement by drawing on their previous vertical system experiences, slowly started to change as they learned through informal trainings, feedback meetings and seminars that all the datasets needed to be afforded equal significance. The change from noneuse to limited use conforms to the argument that, other embedded practices can be changed or replaced by others over time [18]. However, contextual particularities such as the poor economic status of most healthcare settings in developing countries sometimes reinforce their existence and hence make them hard to change. The need to strengthen communication and collaboration linkages between stakeholders The vertical programs in the Zanzibar case have very strong installed base of information systems reinforced by funds from donors. Moreover, these systems have very well defined and elaborate vertical organizational structures with many people employed in there, further reinforcing their strengths and the tension towards any change attempts. For instance, the HIV/AIDS program maintains its own administrative structure and employees working on the information system. As pointed out earlier, although most of the verticalreporting systems were harmonized and aligned in the integration initiatives, some of the systems were still running alongside the integrated system. This state of inertia could be explained partly by the perceived results of integration, including fear to lose positions, those with vested interests with the old systems to lose them and the mistrust on the capability of the national HIS in managing and sustaining the integrated system. The argument as implied from the case is that communication and collaboration between all the necessary stakeholders need to be built and strengthened as a strategy to deal with the inertia of

250

the vertical reporting systems. The national HIS however, need to take a stewardship role to ensure that comprehensive data is obtained from the disparate systems in a cost effective way. This further suggests that some of the vertical systems to run alongside the national HIS, but with the mandate that the national HIS take the driver’s seat in ensuring smooth collaboration and communication between the stakeholders. This integration perspective is inline with the concept of accepting to live with a reasonable level of none integration, since no one, including the national health authorities, is in ‘control’ in any strict sense; and therefore a relevant strategy cannot be based on a planning or control approach [8]. Hand in hand with the need to strengthen communication linkages, is the need to use gateways to link between the national HIS and the vertical programs information systems which seems strong and hard enough to integrate with the national HIS. Gateways allow the continuous existence of multiple systems, each with their internal organization and logic (ibid). Consequently, gateways can ensure transfer of data from strong multiple vertical systems to the national data warehouse, and therefore make the data available to all stakeholders in a cost effective way. Hence, the use of gateways supports the concept of accepting a certain degree of none integration, while keeping every stakeholder ‘happy’. From the case, the communication and collaboration perspective is exemplified by the approach used by the HMIS Unit in resolving the availability of data collection tools problem. The problem happened when the unit failed to fund production of data collection tools due to financial constraints. To solve the problem, which if left unsolved would have rolled back the entire system; the HMIS unit summoned all key stakeholders in a workshop to deliberate on the strategies to solve the problem. In the workshop, it was unanimously agreed that each stakeholder (vertical program) contributes some funds for production of the tools. As the results, production of tools to be used for a period of one year was made possible through communication and collaboration between the stakeholders but with the HMIS unit taking the stewardship role. The national HIS using communication and collaboration processes can play the same role to ensure availability of comprehensive data. The strategy used in Zanzibar of soliciting funds from vertical programs may point to the idea of donors as the main supporters of the health programs to pool resources together as in the case of the Sector Wide Approach (SWAp) [22], to support the integrated national HIS. However, despite the SWAp policy mandate of strengthening the local capacity to manage funds from donors, evidence shows that countries such as Tanzania is facing difficulties in coordinating the funds using the so called basket fund

[23]. This is further reinforced by [21], who argues that pooling of resources by the categorical health programs to the integrated HIS is a difficult undertaking. The challenges related with the pooling of resources underline the critical need of the proposed strategy of strengthening communication and collaboration linkages and the need to make use of gateways to ensure flow of information from vertical systems to the national HIS. VI. CONCLUSION Integration of HIS in developing countries characterized by multiple vertical programs is quite a challenge. It involves economic and political processes in articulating interests, building alliances and struggling over outcomes [19]. The tensions between the national HIS which strive for integration and the vertical programs which advocate their information systems supported by donors represent both challenges and opportunities to the integration initiatives. One of the challenges as implied from the study is ensuring compliance of the integrated HIS by the vertical programs in the face of poor economic status of most developing countries. However, as implied from this study, strengthening the collaboration and communication linkages between the national HIS and the vertical programs represent an opportunity to curb the integration challenges. While the paper propose this perspective as a strategy towards HIS integration, more empirical research is needed to find out more how it can be achieved in practice. The message the paper sends to public health and other practitioners in HIS is that misconstruing or under-estimating the non-technical issues of integration can account for a very significant portion of failures of information systems in healthcare. While technically, the integrated national HIS consisted of datasets catering for the needs of the vertical programs covered by this study, the programs however kept on using their previous systems, presenting challenges to the integration initiatives. Hence, the paper underscore the need to understand the socio economic challenges related to HIS integration which goes beyond technical fixes. The study further reiterates the need to look beyond the integration process, to include the way users receive and engage with the newly integrated HIS. VII. REFERENCES [1]

[2]

B. Chilundo, Integrating Information Systems of DiseaseSpecific Health Programmes in Low Income Countries: The Case Study of Mozambique, Ph.D. thesis, Faculty of Medicine, University of Oslo, 2004. B. Chilundo and M. Aanestad, Vertical or Integrated Programmes? The consequences for the laboratory information systems in Mozambique, In Proceedings of the IFIP TC8 & TC9/ WG8.2+9.4, 2003.

251 [3]

[4]

F. Mukama, A study of health information systems at local levels in Tanzania and Mozambique, Improving the use and management of information in health districts, Master thesis, Department of Informatics, University of Oslo, 2003. J. Braa, E. Monteiro and S. Sahay, Networks of Action, MIS Quarterly 28(3): 126, 2004.

[5]

[6]

[7]

[8]

[9]

[10]

[11] [12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

[20]

J. Braa, O. Hanseth, W. Mohammed, A. Haywod, and V. Shaw, Developing Health Information Systems in Developing Countries: The Flexible Standards Strategy. In press, MISQ, Special Issue on IT and Development, 2005. J. Lungo, Data flows in Health Information Systems: An action research study of reporting routine health delivery services and implementation of computer databases in health information”. Master thesis, Department of Informatics, University of Oslo, 2003. J.J.Rubona, Routine health information systems that operate in Tanzania”. In the RHINO workshop on issues and innovation in routine health information in developing countries, The Bolger Center, Protomac, MD, USA 14-16 March 2001. Arlington, VA 22209, USA: MEASURE Evaluation, JSI Research and Training Institute, 183 –193. M. Aanestad, ., E. Monteiro, H. Kimaro, E. Macombe, G. Macueve, F. Mukama, H. Muquingue, L.J Nhampossa, and J. Lungo, Strategies for development and integration of health information systems: coping with historicity and heterogeneity. Working paper at the University of Oslo, Department of Informatics, 2005. N. H. Shidende, Challenges and approaches to the integration of HIS: Case study from Tanzania”, M.Sc. thesis. Department of Informatics, University of Oslo, 2005. Braa, J. and Humberto M (2007): “Building collaborative networks in Africa on health information systems and open source software development – Experiences from the HISP/BEANISH network”, IST Africa, 2007. R. Yin, Case study research: Design and methods (2nd ed.). Thousand Oaks, CA: Sage Publishing, 1994. RHINO, The second RHINO workshop on: Enhancing the Quality and Use of Routine Health Information at the District Level”, September 29 – October 4, 2003. World Health Organization (WHO), Health information systems in support of the Millennium Development Goals, 2006. Y.H. Sheikh, Improving health information systems at health districts in Zanzibar, M.Sc. thesis. Department of Informatics, University of Oslo, 2005. R. Sauerborn, and T. Lippeveld, Introduction. In: T. Lippeveld, R. Sauerborn, and C. Bodart, eds. Design and Implementation of Health Information System. Geneva, World Health Organisation, 1-14. M. Galimoto, Integration of Health Information Systems: Case Study from Malawi, MSc. thesis, University of Oslo 2007. Sahay, S., E. Monteiro, and M. Aanestad, “Towards a Political Perspective of Integration in IS Research: the case of Health Information Systems in India”. Working paper, Department of Informatics, University of Oslo, Norway, 2006. Lagebo, B. and S. Mikonnen (2005): “Challenges and Approaches to Scaling and Standardizing Health Information Infrastructure in Developing Countries: Case Studies from Ethiopia”. M.Sc. thesis, University of Oslo. B. Chilundo and M. Aanestad, Vertical or Integrated Programmes? The consequences for the laboratory information systems in Mozambique, In Proceedings of the IFIP TC8 & TC9/ WG8.2+9.4, 2003 Mosse, E. and S. Sahay, “Counter Networks, Communication and Health Information Systems: A Case Study from Mozambique”. In The IFIP TC8 & TC9/ WG8.2+9.4 Working Conference on Information Systems Perspectives and Challenges in the Context of Globalization. M. Korpela,

R. Montealegre and A. Poulymenakou (Eds), Athens, Greece: 35 -51, 2003. [21] G. Land, N. Hoffman and R.Peterson, ”Developing the Missouri Integrated Public Health Information System”, In O’Carroll PW, YasnoffWA,Ward ME, Ripp LH, Martin EL, eds. Public Health Informatics and Information Systems. New York: Springer; 2002, 617–64. [22] Brown A, “Integrating vertical health programmes into sector-wide approaches: Experiences and lessons”. Institute for Health Sector Development, on behalf of the Swiss Agency for Development and Co-operation (SDC) 2001, Pages: 30. [23] Smith, M., Madon, S. and Anifalaje, A. ”Integrate Health Information

Systems

in

Tanzania:

Challenges”, EJISDC (2008), 33, 1, 1-21.

Experience

and

Cross Technology Comparison for Information Services in Rural Bangladesh

252

Faheem Hussain, Member, IEEE; Rahul Tongia, Member, IEEE Abstract— This paper introduces a framework to examine the relative effectiveness and cost-effectiveness of different information and communications technologies to deliver a range of social services in rural Bangladesh. It focuses particularly on major sectors like agriculture, education, disaster response and healthcare. Expert elicitation (using both local and international group of experts) survey on ICT effectiveness by service domains shows localization as the key determining factor for any technological intervention. Community based radio broadcasting (CR) appears as the dominant option among the considered ICTs. Human intervention has been found to be crucial in both low (20%) and medium-high (60%) literacy populations. Our study also shows public funded Terrestrial TV is more effective and has higher acceptability in providing information over privately owned Satellite/Cable TV channels. Literacy doesn’t seem to significantly affect the relative effectiveness of Information Centers with Internet connection over Print and TV based options. In addition, an extensive techno-economic model for the various ICTs and stochastic evaluation of potential penetrations of the ICTs (reach) has also found CR to be the most cost-effective option. We conclude with a proposed set of policy and operating recommendations to enable effective technology based information services for rural Bangladeshi development. Index Terms— ICTD, Rural Information Service, Community Radio, Telecenter.

I. INTRODUCTION

E

nsuring the availability of community level information

services (IS) is an important priority for the general population in South Asia. With the proliferation of Information and Communication Technology (ICT) enabled communication and broadcasting options, the numerous positive impacts that community based information services can contribute become more evident [1]. Community-based information services usually go beyond the traditional “One size fits all” information access model used by the incumbent regulators and policy makers [2]. These primarily focus on the customized needs of the target communities. Of course, not all ICTs are equivalent. Different technological solutions may better meet the needs of different service sectors. For example, an option that works well for agriculture related information

This work is supported by a grant from the MacArthur Foundation, and by Carnegie Mellon University. F. Hussain is a faculty member at Carnegie Mellon University- Qatar Campus in the Department of Information Systems (phone: +974-557-2179; fax: +974-454-8410; e-mail: [email protected]). R. Tongia is a faculty member at Carnegie Mellon University, Pittsburgh, in the Program on Computation, Organizations, and Society (COS), in the School of Computer Science (ISR), and the Dept. of Engineering & Public Policy. He is also the Associate Director of the TechBridgeWorld project at Carnegie Mellon, and a Senior Fellow at the Center for Study of Science, Technology, and Policy (CSTEP), Bangalore, India (phone: +1-412-2685619; fax: +1-412-268-2338; e-mail: [email protected]).

services may not be effective for disaster response. This paper looks into the relative effectiveness and costs of different technologies to deliver different social services in rural Bangladesh, focusing on major development sectors such as agriculture, education, disaster response, healthcare etc. This analysis has several components. First, expert opinions on ICT effectiveness by service domain (health, education, etc.) for various ICTs (TV, radio, print, telecenters, etc.), with international and local (Bangladeshi) experts in communications and development were gathered via expert elicitation surveys, taking care to manage anchoring and other biases (detailed subsequently). The elicitations were for low (20%) and medium-high (60%) literacy populations. Second, we created an extensive techno-economic model for the various ICTs. Finally, we made estimates of potential penetrations of the ICTs (reach), to estimate the costeffectiveness per user of the technologies for the particular development domain. The focus here is largely on the delivery of social services (supply side); analysis of issues such as citizens' rights to information, citizen participation in governance etc. is beyond the scope of the paper. The primary objective of this study is to develop a robust framework that can facilitate choices by policymakers, practitioners, development agencies and the general public and promote better understanding regarding the applied efficiencies and relevance of different technologies in multiple information service domains for societal development. Given the difficulties comparing technologies, expert elicitation with normalization is a useful starting point for the analysis. Formal (pre/post) type analyses for effectiveness across ICTs are rarely available, given most studies consider only one technology at a time for effectiveness analyses. II. BACKGROUND OF COMMUNITY BASED INFORMATION SERVICES IN SOUTH ASIA

Historically, governments in S. Asian countries have been reluctant to open up the airwaves, lower import duties and other taxes on information technology or liberalize regulatory policies so as to create an enabling environment for micro level information service infrastructures. On the other hand, ICT for Development (ICTD) friendly stakeholders (i.e. business entities, NGOs, innovators, donors etc.) are often challenged to come up with effective methods for developing and managing systems to deliver information services (IS) [3]. The reasons for this split might have to do with how information has been (centrally) controlled, combined with a market-type approach of first focusing on those willing or able to pay. Community oriented IS have had some successes. In Nepal, a pioneer in community media activism in S. Asia, the

253

foundations for regulatory reform in broadcasting were first laid as part of the liberalization of Nepal’s constitution in 1990 [4]. During this period, Nepal witnessed the development of independent, community based radio stations (CR)1, telecenters2 and newspapers. The potential synergy between telecenters and CR stations has also been explored there to a greater extent than elsewhere in South Asia (SA) [7]. Table 1 summarizes the present CR scenario in Nepal. Table 1 Present Status of CR in Nepal [4, 8] On Air Stations New Licenses issued in last 6 month period Districts under total CR coverage area (out of 75 districts nationwide) Total % of population covered by CR broadcasting*

33 93 56 62.5%

A series of regulatory liberalization and public-private initiatives in developing localized information service mechanisms are also evident in India. In November 16, 2006, the government finally opened up the community airwaves for non-profit organizations. It also formulated a comprehensive “Community Radio Policy”, a first such step in South Asia [9]. Throughout this Indian experience, especially noteworthy are the innovative ways people have met the huge demand of localized information, even in the face of severe legal and regulatory constraints[10] [9, 11, 12]. A. Bangladesh - Promises and Problems in Community Information Service Sector The overall communication sector in Bangladesh presents an interesting situation, where the potential of low as well as high end ICT-based information access points (IAP) [13] is immense, even in the absence of a regulatory and development framework. The national literacy rate is only 43.1% [14]3, and only approximately 50% of those classified as literate can properly read and write[15]. A recent national media survey found that 28.5% [15] of the total Bangladeshi population are in the “Information Dark”, meaning, they have never listened to a radio program, watched a TV program, read a newspaper or had any similar means of getting valuable and current information needed for their livelihood and wellbeing [15]! This situation is even worse in rural areas (35.9% of rural people have no access to any information service) [16], where most of the people live. Because of the absence of localized programming and participation, the only public radio broadcaster, Bangladesh Betar, is losing its listener base 1 Community Radio: Community radio (CR) is a subset of Radio broadcasting, which is an inexpensive and popular medium for disseminating content (information, news, entertainment) that specifically focuses on the active participation of its listeners and also tries to ensure access to information by all. In addition to broadcast, there are mechanisms for manyto-many discussions and community feedback [5]. 2 Telecenter: Telecenters are shared premises where the public can access information and communication technologies. A center offering only telephone, computer or internet services is valid under this definition [6]. 3 According to a study completed by the Government of Bangladesh in 2003, the literacy rate was found to be 62.66%. But a CAMPE study in 2002 found it to be 41.4%.

rapidly (39% in 1998, 37% in 2000, 34% in 2002 and 23% in 2004) [15]. At the same time, the penetration level of cellular phones and internet users is increasing. Presently, the teledensity of the country is 27%, with 38.93 million cellular phone users and 1.25 million PSTN (landline) users [17]. The Bangladesh Telecenter Network (BTN) has also undertaken initiatives to create 40,000 telecenters nationwide by 2011 [2]. Only recently, the Government of Bangladesh (GoB) has opened up the airwaves (on an experimental basis) for community based low power FM radio broadcasting. Figure 1 shows the status of literacy and access rates of Terrestrial TV, Telephone, Public AM and National Newspaper in several South Asian countries. Nepal lead in the access of Public AM broadcasting but has the lowest teledensity among the four S. Asian countries considered. On the other hand, India has considerably higher literacy rate but lower Terrestrial TV penetration rate than its neighbors.4 III. INTERNATIONAL EXPERIENCE AND LITERATURE ON COMMUNITY-BASED INFORMATION SERVICE (IS) Over the last few decades, amid the disparity in access to information between the global north and south, the concept of alternative or community based media has emerged. The New World Information and Communication Order (NWICO), proposed by UNESCO in the late 70s and early 80s [18], had the goal of beginning to balance this information divide [19]. Studies have shown, with the introduction of new technologies, valuable information needed by any community becomes more accessible and affordable [20]. At the same time, some experts have warned against over dependence on any particular technology and have placed greater importance on the very process of providing information service [21]. There are many technology options for IS. Community based radio broadcasting has been found to be effective for strengthening democracy, development and people’s participation in the affairs of developing regions. For example, in Africa, organizations like COMNESA in East and Southern Africa, AMARC Africa, KCOMMNET in Kenya etc. have been established to share resources among a wider range of population across the continent [22]. A significant amount of research has also been done to assess the effectiveness of different technological intervention in various information service domains, i.e., education, disaster response, agriculture etc. In the many cases evaluated, experts have found radio-aided services to be quite effective for different education or training purposes.[23-25]. The effectiveness and importance of visual broadcasting media has been supported in other studies [26, 27] While choosing an appropriate technology to enhance educational delivery or other societal services, issues like access, costs, maintenance, interactivity, user-friendliness, availability and speed should be considered [26]. According to UNESCO, users and experts also focus on the specific abilities of the considered technologies in terms of delivering 4 Data is more limited on cable TV, which has substantial penetration in some regions, but content is rarely more local community oriented.

254

Figure 1 Literacy and Access Information of Different Communication Medium in South Asian Countries

Figure 1 represents the relative status of the four South Asian countries in literacy, teledensity, newspaper readership and radio penetration (data collected during 2004-2008 period).[16, 28-34] appropriate messages in a given domain. Technological efficiencies can vary in terms of outreach capability, flexibility, interactivity etc.[35]. In addition to the traditional applications of lower end ICT options like radio and TV in a disaster response scenario, recent studies also focus on the importance of computer aided, Internet based and wireless technologies [36]. The recent trend of technology specific effectiveness studies is concentrated mainly on higher end ICTs: Computers, Webbased solutions, Telecenters etc. and, to some extent, mobiles. As auxiliary tools for providing education, these ICT options often show superior performance in comparison with the other alternatives. But factors like high cost, lack of training, lack of motivation, maintenance issues, and limited infrastructure often hinder the highest level of efficiency attainable by these relatively new technological interventions [37] [38]. Methodologically, these past studies, as well as this one, struggle with the issues of penetration (which are cost-related) compared to effectiveness in and off itself. E-access enabled information access points, predominantly in rural or economically deprived areas (popularly known as Telecenters, Rural Information Centers etc.) have developed a strong following in developing regions around the world. Studies have shown the availability of local contents and affordable services has helped many telecenters to be effective and sustainable [39, 40]. Case studies in South America and Africa provide ample evidence of micro-level democratization, as one of the positive externalities of effective telecenter based applications [41].

A. Significance of this Research While much of the research described has focused (at least qualitatively) on the effectiveness and impacts of various technologies in multiple information service domains, there is a clear need for studies that examine the comparative effectiveness of a wide range of appropriate technologies across a set of information services in specific socio-economic settings. The difficulties in conceptualizing the relative effectiveness of multiple types of interventions, taking into account a very large set of application domains, is clearly daunting. The objective of the present study is to develop a robust comparative cost-effectiveness analysis framework, which will enable different concerned stakeholders (i.e., policymakers, practitioners, international and local development agencies etc.) to have a better understanding of the applied efficiencies and relevance of different technologies in multiple information service domains. While the main focus of the application is SA, and more specifically on Bangladesh, this study can also be used as a guideline in other developing regions with similar needs and aspirations. IV. RESEARCH METHOD The principal research question we are addressing through this research is: What are the most effective technological interventions for providing Information Services in the rural South Asian (Bangladeshi) setting? There is a wide diversity of Information Service domains in SA and many of these are interrelated and in some case overlap each other, in terms of target population, societal objectives and modes of communication. In order to have a better understanding of key

255

social services, we limited the number of application domains considered in this research to seven (Agriculture: General Extension Service and Market Information; Education; Disaster Response; Health Care: AIDS and STD prevention and Family Planning; and Financial Services). These domains were selected with respect to the needs and their importance in the SA region [42]. A set of technologies was then chosen, mainly on the basis of their availability, potentials in the "Communication for Development" field [43], success stories and positive impacts in the real world scenario. This work considers broadcasting (both audio and visual), print and the Internet/multimedia enable Information centers. Cellular phone based options were not considered individually, but the impact and effectiveness of cell phones have been partially accounted for in the hybrid solutions of Information centers as well as in person-to-person approaches. In order to develop a framework for a comparative costeffectiveness model, we focused on quantifying the relative "effectiveness" of each technology using the judgment of a group of international as well as local experts in the field. It took roughly six months of correspondence using regular mail, Internet and phone conversation to produce an effectiveness dataset from five different continents. We then built a cost model for calculating costs and penetration information (using field based South Asian datasets) through extensive technoeconomic analysis. Finally, all the analyzed and calculated information have been combined in a common platform to compare the relative effectiveness of the considered technologies in information service sectors of this analysis. A. Development Domains and Technologies 1) Information Service Sectors The seven application domains we selected for the analysis and expert surveys cover different types of information, generally required at the community level in Bangladesh: a. General Agricultural Extension b. Agricultural Market Information . c. Life long learning/Adult Education d. Disaster and Emergency Response. We focused on flood warning, mitigation and rehabilitation; drought mitigation and awareness and cyclone/ Tsunami related awareness and protection programs. e. AIDS and Sexually Transmitted Diseases (STD): awareness and prevention campaigns f. Family Planning: awareness related activities. g. Financial Services. This focuses on financial information services at the personal and community level. 2) Selected Technological Interventions In many cases the most effective means to deliver information is through one-on-one communication with a respected, trusted, patient and informed person in the recipient's language and in an appropriate cultural setting. However in most of the cases in the real world, this is not feasible. In addition to the face-to-face (F-to-F) option (sometimes called a peer-to-peer mechanism, though that terminology has a separate technical meaning in networking as well), technological interventions selected for the survey were (in alphabetical order):

Community Radio (low power FM); Development FM Radio; Local Newspaper; National Newspaper; Printed Brochure; Public AM Radio; Rural information center (RIC-1) equipped with computers, Internet connectivity, and one or more of the following: scanners, printers, digital cameras, cellular phones; RIC equipped similarly as RIC-1 but without any web-based facilities (RIC-2); Satellite Radio; Satellite/Cable TV; Terrestrial TV (privately operated for profit) and Terrestrial TV (government operated). B. Expert Elicitation In order to quantify the relative effectiveness of various information service technologies for delivering a variety of social services, we administered a survey to a group of international (i.e., Australian, Guatemalan, Fijian, Indonesian, American, Nepali, and Sri Lankan) and local (Bangladeshi) experts. These groups have had many years of experience in the field of communication and development. The objective of this survey was to obtain judgments of the relative effectiveness of different approaches for delivering information. We explicitly requested the experts to ignore differences in cost. At the start, twenty international and thirty local (Bangladeshi) experts were solicited to take part in this study. Ultimately, eight international and fourteen local experts provided detailed responses. In this study, we defined “information” to be most useful when it is delivered in the language of the community, addresses topics that are of interest and concern to the community and is framed in a way that is clear and understandable to members of that community. We also defined "community" as a group of a few hundred to several thousand people who live within a few tens of kilometers of each other in a developing country and share the same language and culture. Experts were asked to assess the relative effectiveness of different information delivery technologies for different types of information services in two types of communities: communities with relatively high (60% or higher) and low level (20% or lower) of adult literacy. We defined adult literacy to mean a recipient’s ability to read at an intermediate level in his or her primary language. For each application, we asked the experts to first identify the communication option that they believed would be the most effective for a community with relatively high adult literacy and assign it a score of 100. They were then asked to assign a number to each of the other technology options to indicate how relatively less effective the experts’ believe each communication option would be for the specific information delivery identified. The whole process was conducted for both high and low literacy scenarios. We also received qualitative information from the experts. As most of the experts were not as familiar with some of the technologies, development FM, satellite radio and private terrestrial TV were omitted from further analysis. C. Cost Estimations Our objective in estimating the costs of various delivery technologies is to provide comparable descriptions or mechanisms by which to deliver services and estimate their

256

costs in a way that would allow comparisons. In general, “Cost-per Person-Year” is estimated for each option, where “cost” is the monetary amount spent by the provider to deliver the information in a real world setting and “person” is the generic term that covers target listener, viewer, reader, user etc. (as appropriate for any particular technology/ medium) within the general community population. Naturally, this assumes an equivalence amongst all the persons. For community based information services, while calculating Cost-per Person-Year, both operational and amortized capital expenditure are considered. For the services with national coverage, we calculate a generalized cost-per person-year (delivery and content development cost) that cater to numerous communities and issues, in addition to providing particular information services to any target community; as detailed later, some fixed costs were considered sunk costs for such national footprints. We assume a 2, 5 and 10 year life cycle for the audio/video program, RIC equipment and CR station equipment, respectively, and used 7.5% as the interest rate for annualized capital expenditures (sensitivity analysis also considered 11% and 15% but we used 7.5% as a baseline, a common rate for development/non-profit projects in South Asia. This matches the stochastic cost modeling of CR management in [5]. We also take into account stations’ or other facilities’ monthly operational expenditure, which include staff salaries, utility bills and space rental (if applicable). For the technologies with a national presence (i.e., National Newspaper, Public AM, Terrestrial TV and Satellite TV), we considered the literacy rates as the proxy of other major factors while determining probable access rates within any community. We assumed that higher literacy will result in higher access to certain technologies, but the rate may not be linear. In order to realistically calculate the penetration of these technologies in the high and low literacy region, we used the ratio of 60% and 20% respectively compared to the national literacy rate (43.1%) [14], multiplied by the corresponding technology’s potential national penetration number and a technology-specific multiplier. The values of all the multipliers considered were estimated based on the respective technology’s dependence on literacy for mass proliferation. For example, in order to estimate the penetration of a National newspaper in a community with 20% or lower literacy, we multiply the national newspaper readership number (26%) by the ratio (20%:43.1%) and a multiplier. As the dissemination of newspapers is directly related to the literacy of a target community, we estimate the baseline value of this technology’s multiplier to be 1. Given the wide uncertainty in parameters (i.e. cost, penetration), we have used Analytica® stochastic modeling software to estimate a distribution in the total yearly cost for each station, giving us the ability to parametrically vary a wide variety of variables. Below are selected details on the cost calculations for the various technologies; full details on assumptions of capital and operating expenditures per technology are available from the authors. Given the wide variance in numbers per technology, it is beyond the scope of this paper to display these in a

succinct manner. 1) Community Radio (CR) For the CR cost estimation, we included both monthly operational and yearly capital expenditure cost (amortized). The lifetime of radio transmission equipment was assumed to be 10 years. The listener base was calculated in terms of CR penetration among typical South Asian communities in India and Nepal (there is virtually no community radio in use in Bangladesh), which usually ranges from 60%-90%[30]. After deriving the total yearly cost, we can calculate a set of “cost per listener-year” values, depending upon the probable population (density) of any community, the potential CR listener base, transmission range etc. Our observation of global CR practices shows that approximately two-thirds of total broadcast time is dedicated to societal issues [7, 44]. Of those blocks of time, we assume an equal distribution of air time among the seven different information services. Thus, the final “cost per listener-year” for any CR based service is computed as  Total Yearly Cost   Societal Program   Selected Service   × ×   Number of Listeners   Total Broadcast Time   Total No. of Services 

 Total Yearly Cost   2   1   ×   ×    Number of Listeners   3   7 

or, 

Table 4 shows the calculated cost per person-year and penetration levels for all the technology based information service in the target community. 2) Public AM In South Asian nations most Public AM stations cover the whole country. This centralized and “one way” broadcasting does not have the flexibility to air localized programs nor does it have the means to incorporate grassroots’ participation. Considering everything, our assumptions for “cost per listener-year” then focused only on specific program production and transmission costs, which in turn can be divided among the potential listener base nationwide. Empirical studies support the practice of airing “twice a week” programs on any particular societal issue. The assumptions for any radio program’s broadcasting time and frequency in a Public AM station for a certain year are stated in Table 2. Table 2 Parameters for Public AM Programming Duration of a Program

30 min

Number of Program per week

2

Number of Programs (yearly)

96

Duration of the total Program time

48 hours

In this study, we assumed that in order to distribute community based information service, we need a series of audio programs and buy the airtime from Bangladesh Betar or the regulator to broadcast it, using the cost structure available in Bangladesh. 3) Terrestrial TV (Public) and Satellite TV Here the majority of the assumptions regarding program

257

frequency, “cost per person-year”, access to programming etc. are similar to those for Public AM. we used the cost of video production and the TV transmission pricing (both terrestrial and satellite). For calculating access level, national terrestrial TV penetration rate [16] has been used alongside the literacy ratio and a variable multiplier. 4) Face-to-Face (F-to-F) Ideally, Face-to-Face service delivery should involve oneto-one contact between a facilitator and a person who is a member of the target population. But in most cases, this is not feasible or cost effective. F-to-F mechanisms actually vary, depending on the type of information being provided. The development costs of content in all the services are either insignificant in comparison to the calculated delivery cost or external to the system. We assume, with the exception of the Health Care sector, that the monthly salary of the field workers remains the same across different information domains. The total cost has been estimated for a year long program. In General Extension and Market Information sectors, the percentage of people nationally involved in agriculture has been used as the base percentage in estimating the range of agro-related population within the considered community. We assume that a group of 50 people (varied parametrically) meets with an agricultural extension worker on a regular basis. In the Education sector, the national literacy rate was considered for estimating community illiteracy. For Disaster Response and Financial Services, community based estimations were used for calculating the target population. Each face-to-face group’s base population and other estimations are similar to the ones used in agriculture sector. Due to the sensitive nature of Health Care information services, we assumed monthly one-on-one interaction between public health field workers and the target community population. 5) National Newspaper We assume that a relevant entity procures ad space to publish specific communication material, pertaining to one of the seven information service dimensions defined in this research. The assumptions for any national newspaper’s social ad campaign and publishing frequency are stated in Table 3. Table 3 Assumptions for a Newspaper based Societal Ad-Campaign Space bought in the 1/16th of the full page front page of any (Standard Newspaper Page national newspaper with 36” X 24“ dimension) Duration of the Ad (per year)

100, 300 and 350 days (vary parametrically)

Target Population

Total Community Population X National Newspaper Readership (%) [16]

6) Local Newspaper The frequency of ad-publishing is assumed to be similar to that of the national newspaper with relatively cheaper ad-rates. We also took into account high and low literacy scenarios for calculating potential readership within the target community.

We assumed the reader ceiling of 20% for the low literacy region and 60% in a high literacy community. 7) Printed Brochure (PB) Designing printed brochures and pamphlets for any target community audience has long been a basic strategy utilized by local NGOs for distributing social service information. Readership and target community estimations are similar to local newspapers. The development cost of educational materials has been estimated to be relatively higher than the rest of the information domains for PB [45]. 8) Rural Information Center-1 (RIC-1) and Rural Information Center-2 (RIC-2 = no Internet) Rural Information Centers (RICs) house computers with various information resources and, possibly, Internet access. The cost estimation of RICs are based on data collected mainly from Bangladesh, Nepal and India. The RIC model considered in our research has the following basic facilities: computers, Internet Connection (available in RIC-1 and absent in RIC-2) and facilities for basic computer training and Printer/Cellular Phone/Fax for commercial usage. The Pallitathya Kendra (PK) model, used by the NGO D.Net in Bangladesh, has been emulated in designing RIC-1 and 2’s working infrastructure. A PK type RIC usually has an outreach service, where the RIC field workers will personally go from house to house within any target community and address their information needs by connecting users through cell phone to the pre-selected expert panel or the locally developed database [46]. For the penetration estimation of RICs, we assume that RIC usage will be greater in regions with high literacy, resulting in the need to develop multiple RICs within those areas, all within walking distance of all community members (2 or 3 km radius). For a low literacy region, we estimate less usage and the presence of a single RIC for the whole community. Also, the proactive approach taken by the RIC field workers certainly has some positive impact, at least among the literate community population and beyond. We moreover assumed that the content development and distribution cost for any information service will be similar for any RIC. V. COMPARATIVE EFFECTIVENESS ANALYSIS We need to develop a comprehensive understanding about the comparative efficacies of different information service delivery options. For the relative effectiveness data, both international and local (Bangladeshi) experts’ opinions were initially considered. According to both of the expert groups, CR turns out to be the most effective technology in both high and low literacy scenarios, although there was much less familiarity with CR, and thus higher variation, in the responses of the Bangladeshi experts. All the experts showed greater confidence in Face-to-Face intervention than most of the technologies in almost all the application domains. In addition, some prominent dimensions (e.g., directionality, user cognitive requirements etc.) within the technological “effectiveness” concept have been qualitatively analyzed, based on anecdotal evidence, field data and research findings

258

Table 4 Calculated Cost Estimations* Delivery Option CR Public AM Terrestrial TV Satellite TV National Newspaper Local Newspaper RIC-1

Information Service Sector

For any of the considered Service Sectors

RIC-2

Printed Brochure

Face-to Face

Agriculture: General Extension Agriculture: Market Information Education Disaster Response Health Care: AIDS and STD Health Care: Family Planning Financial Services Agriculture: General Extension Agriculture: Market Information Education Disaster Response Health Care: AIDS and STD

Health Care: Family Planning Financial Services * Parametrically analyzed, using SA data

5

Cost per Person-Year (US$)*

Penetration Level*

≤ 20% Literacy

≥ 60% Literacy

≤ 20% Literacy

0.011 0.0025 0.0045 0.008 0.004

0.011 0.001 0.0015 0.003 0.0014

14% 23% 9% 12%

36% 67% 26% 34%

0.35

0.08

16%

70%

0.86

0.38

33%

75%

0.17 0.35

0.23 0.23

33%

75%

0.55

0.43 16%

70%

0.36

0.24

0.35

0.23

1.59

1.59

≥ 60% Literacy 68%

Ideally, 100% of the target population will covered from an availability perspective; usage and uptake may be lower.5

1.67 0.99

2.45 2.45 1.63

This challenge is similar to radio or other media, where signal availability is necessarily only one measure, lower being actual listening time.

259

Table 5 Cross Technology Comparison for Multiple Effectiveness Dimensions Technology Intervention

Penetration**

≤ 20% Literacy

CR

Directionality

≥ 60%

Cognitive Level

Access Type

Low

Single / Shared

Literacy 68%

Two-way, Participatory

Public AM

15%

36%

One-way

Low

Single / Shared

Terrestrial TV

23%

66%

One-way

Low

Single / Shared

Satellite TV

9%

27%

One-way

Low

Single / Shared

National Newspaper Local Newspaper

12%

34%

One-way

High

Mostly Individual

16%

70%

One-way

High

Mostly Individual

Printed Brochure

16%

70%

One-way

High

Mostly Individual

RIC-1

33%

75%

Two-way

Medium*

Shared

RIC-2

33%

75%

Two-way

Medium*

Shared

* With the help of informediaries (information intermediaries) **Technology Penetration estimated using stochastic modeling for high and low literacy region. There is always a challenge between availability vs. uptake

([43] and [26]). CR has been found to be a highly accessible technology that enables two-way participation based communication in a single or shared environment (two-way capability comes from word-of-mouth and combining radio with other modes of communication, such as cell phones, as seen in SA). As with CR, the international experts also considered Public AM to be an effective literacy-independent medium. The ordinal effectiveness rankings of technological interventions diverge after this point. Localized print based options (Local Newspaper and Printed Brochure) show clear dominance over the visual broadcasting options (Terrestrial TV, Satellite TV) in a higher literacy area. On the other hand, as expected, visual media are considered more effective than print media in communities with lower literacy. In both cases, Rural Information Center with internet connection (RIC-1) has been ranked in the middle and demonstrated statistically insignificant differences in its usefulness as a community intervention with respect to other technologies, a promising aspect for a relatively new information service option6. The Bangladeshi experts’ perception about the relative effectiveness of some of the technological options is significantly different from those of the international experts. CR remains the top choice but in a high literacy area the difference in effectiveness between Terrestrial TV and CR 6 Paired T-Tests were performed on the effectiveness scores of International experts in high (between Local Newspaper and RIC-1) and low (between Terrestrial TV and RIC-1) literacy region. Both times, the difference was statistically small (with t-stat of 0.1068 and p-value of 0.459 in 60% literacy and t-stat of 3.277 and p-value of 0.0084 in 20% literacy) between the experts’ perception about relative effectiveness.

was not found to be statistically significant7. Unlike the views of international experts8, local experts rank Terrestrial TV and Public AM broadcasting high in both types of literacy regions. Print based options, generally were considered highly ineffective, a contrast with the international findings. According to the local experts, RIC-2 (the information center without any internet facility) is more effective than RIC-1 in low literacy regions9, significantly dissimilar from the views of international set of experts10. For the in-depth analysis of the relative effectiveness data (both technology and application domain), we focused on the international experts’ judgment that captures the technology for development experiences in multiple developing regions. The primary reason for this decision is the Bangladeshi experts’ absence of substantial field experience with the effectiveness of a number of technology options (i.e. CR, RIC1, RIC-2, Public AM, Local Newspaper etc.) due to regulatory obstacles (CR); poor program management (Public AM); high Internet costs, lack of human resources and poor funding (RIC); and absence of press freedom and effective representation of local issues (local newspaper). The 7 A paired T-Test between the effectiveness scores of Terrestrial TV and CR by the local experts found the difference statistically small (with t-stat of 4.43 and p-value of 0.002 in a 20% literacy area) has been found. 8 Two-sample T-Tests were performed between the effectiveness scores of two groups of experts in high and low literacy region. Both times, we found statistically significant difference (with t-stat of -10.04 and p-value of 7.6E-07 in 60% literacy and t-stat of -12.54 and p-value of 9.6E-08 in 20% literacy) between the experts’ opinions. 9 With a t-stat of -22.7 and p-value=2.39E-07 in a Paired T-Test. 10 With t-stat of -8.36 and p-value=-7.92E05 in a Two Sample T-Test.

260

Figure 2 Effectiveness Ranking of Different Technologies (According to International Experts)

Figure 2: Relative effectiveness ranking of the considered technologies in both high and low literacy region, according to the judgment of the International experts. It considers average effectiveness values of different technologies across all service domains. Individually, the experts have shown higher variance in effectiveness estimations (details in Section V)

Bangladeshi sample also has much greater variance than the (Agriculture: General Extension). In these figures, the Y and international sample, especially for technologies not now in X axes represent the effectiveness scores and plausibly use in Bangladesh [47]. achievable technology penetration respectively. The spread of This scenario has been reflected in the effectiveness ranking, effectiveness scores given by individual experts for various which fails to reflect the large potential of RIC with proper delivery systems (from 0 to 100 in the Y-axis) can be seen web based opportunities. Table 6 summarizes the present together with the highlighted ones representing average access information of different technologies, in a Bangladeshi values. The size of the average value data points correspond to setting. the Cost-per Person-Year for each considered delivery In figures 4, 5, 6 and 7, we see the graphical representations technology. According to our assumptions, the effectiveness of comparative cost-effectiveness data, from both international score of both expert groups are independent of the potential and local experts, for a sample application domain technology penetration or cost scenarios in Bangladesh. Figure 3 Effectiveness Ranking of Different Technologies (According to Local Experts)

Figure 3: Relative effectiveness ranking of the considered technologies in both high and low literacy region, according to the judgment of the Bangladeshi experts. It considers average effectiveness values of different technologies across all service domains. Individually,

the experts have shown higher variance in effectiveness estimations (details in Section V).

261

Table 6 Present Access Status of Different Communication Services in Bangladesh Technology/ Penetration/Access Information Service Information [16] Option Radio 30.4% Terrestrial TV 61% Satellite TV 12.4% Newspaper 26% Telecenters 1061 (total number, nationwide)[2]

to be most effective technologies with which to develop an efficient disaster response infrastructure in both high and low literacy communities. Observation 5: In the healthcare sector, experts emphasize the higher effectiveness of proactive Face-to-Face interventions by local healthcare workers. CR remains the most effective information service option, in all literacy and service sector scenarios. Observation 6: According to the experts, CR alongside localized print based options are more suitable in providing information related to financial services among the highly literate population. For a low literacy community, broadcasting options are considered to offer superior information services.

We integrate all these different types of data in a single graphical display for each information service domain and in two types of literacy based scenarios to: 1) have a comprehensive look at the relative effectiveness of several technologies as information service providers in the target service sectors, and 2) to examine their present positioning in terms of plausible local or national penetration level. Service Sector specific Preliminary Observations In order to find out the relative effectiveness of different technologies to deliver a range of information services, we first focus on the service sector specific performance in both high and low literacy scenarios. Observation 1: For Agriculture General Extension, CR was ranked as the most effective technological intervention, independent of the literacy rate within any community. For high literacy regions, print based options dominate broadcasting alternatives. RIC-1 is ranked third overall in effectiveness, followed by Local Newspaper, Public AM radio, National Newspaper and others. For communities with low literacy, Public AM turns out to be the second most effective way to deliver general extension information among the less literate population, followed by RIC-1, Terrestrial TV, RIC-2, Local Newspaper etc. Observation 2: For Agriculture Market Information, CR remains the top choice. Here in high literacy area, Public AM radio is considered to be more effective, followed by technological options with higher localization options (RIC-1, Local Newspaper, Printed Brochure). In low literacy areas, broadcasting options (Public AM, Terrestrial TV and Satellite TV) dominate the effectiveness ranking, followed by RICs and print based options. Observation 3: In Education (Lifelong Learning), apart from CR, in higher literacy areas, with a wide range of options available using the Internet and multimedia, RIC-1 is considered to be the second most effective delivery technology followed by Public AM, Local Newspaper, Printed Brochure, RIC-2 and the other broadcasting alternatives. For low literacy regions, broadcasting options overwhelmingly lead the relative effectiveness ranking. Observation 4: When a country like Bangladesh is under threat of imminent calamities [48], radio broadcasting (CR, Public AM) and TV (Terrestrial and Satellite) are considered

A. Technology-specific Key Observations Examining which technology works best for which information service provides the basis to further examine the effectiveness trends for different types of technologies, their potential impact in future proliferation of certain information service, the regulatory and policy implications on certain technology usage etc. Here technologies here been classified within broader groups in order to come to some general observations. 1) Broadcasting based Information Service Options Observation 1: Radio broadcasting dominates the technology based service delivery mechanism with its literacy independent potentials in localized and participatory communication. Observation 2: CR leads in cost effectiveness in all service sectors through ensuring localization and community participation. Discussion: CR has been found to be an effective mode of communication, independent of service sector type and the state of any target community’s literacy. The apparent universal efficacy of CR can be attributed to its focus on localized programming, participatory communication, appropriate content development etc[5]. Observation 3: Broadcasting based information service options are the key behind any effective Disaster Response mechanism. Discussion: The effectiveness of broadcasting based disaster response mechanisms is unparalleled, primarily due to their nearly universal acceptability, point-to-multipoint (broadcast) capabilities, and speed in disseminating the required information. Observation 4: Public funded Terrestrial TV has higher effectiveness and acceptability in providing information in comparison with privately owned Satellite/Cable TV channels. Discussion: In this research, we assume that higher literacy closely correlates with higher economic affluence, an indicator for people to opt for “better” communication technologies. In Bangladesh, Satellite/Cable TV is less popular (and has lower penetration) than Terrestrial TV[16]. It shows the positive externality a nation’s broadcasting regulations can create through mandating the incumbent public broadcasting service air social service information programs. Generally, private

262

Figure 4 Cross Technology Comparison for providing Agriculture (General Extension) based Information Service in a high literacy ( ≥ 60%) area (International Experts)

E 100 f f e 80 c t i 60 v e n e 40 s s

Face-to-Face

CR Printed Brochure

Public AM

Local Newspaper

Satellite TV National Newspaper

20

RIC-1

RIC-2

Terrestrial TV

0 0%

20%

40%

60%

80%

100%

Likely Achievable Technology Penetration in Bangladesh Figure 4 represents the relative effectiveness (judgment of the international experts), penetration and cost-per person-year status of each considered technologies in providing Agriculture (General Extension) related information service in a high literacy ( ≥ 60%) community. The X and Y axes correspond to the likely achievable technology penetration and the effectiveness scores respectively. The spread of effectiveness scores for various Information Service interventions (from 0 to 100 in the Y-axis) can be seen (with multiple dots) alongside the highlighted ones representing average values. The size of the average value data points correspond to the Cost-per Person-Year for each considered technologies. CR is the most cost-effective technology based option here, with tradeoffs of penetration, cost, and effectiveness between the next choices of face-to-face, Rural Information Center with Internet connection, and local Print media.

satellite channels have fewer regulatory obligations to abide by and little or no financial incentives in airing societal programs, let alone localized versions. Satellite also has greater challenges of localization than many other technologies based on its very wide footprint. 2) Print Based Information Service Options Observation 5: Local Print based options are more effective than the national ones. Discussion: The general population can be effectively reached with locally published newspapers or printed brochures in comparison with the big newspapers with nationwide circulations. Community ownership, localization of the content development and easy accessibility are reported as some of the key factors responsible for this trend. Observation 6: Print based options can be used as effective auxiliary mechanism for societal services that require personalized attention. Discussion: In lifelong learning, financial service advising, health care and agriculture-based information services in any high literacy region, community based print options are more effective than the visual broadcasting media, according to the effectiveness ranking and comments of the international experts. Usually NGOs and development agencies in the field use this type of intervention as a supporting tool in addition to their broader door-to-door personalized campaigns.

3) Rural Information Centers Observation 7: Literacy doesn’t affect the relative effectiveness of RIC-1 over Print and TV based options. Discussion: According to the experts in this research, on average in both high and low literacy scenarios, the RIC-1 centers are (or, perhaps, can be?) more effective than some major TV broadcasting and locally published print based options. Especially with the help of its information intermediary (“ informadiary”) mechanism, local database and cellular phone integrated applications, RIC-1s have a lot of growth potential as a relatively new technological option[49]. Observation 8: RICs without Internet can still be utilized as a successful tool for providing community education. Discussion: As we have found in this research, RICs without web enabled facilities (RIC-2) are less effective than RIC-1s as well as the majority of the other interventions in both high and low literacy communities.11 But in the education domain, RIC-2 shows higher effectiveness than mainly National Newspaper and Satellite TV. Based on discussion with the development experts in S. Asia, we found that the relatively lower demand of updates in lifelong learning curriculum is considered to be one of the main reasons behind RIC-2’s better performance in this specific service sector. 11 An RIC-1 could always operate without Internet (RIC-2) in theory, but in practice, this is rarely operationalized.

263

Figure 5 Cross Technology Comparison for providing Agriculture (General Extension) based Information Service in a low literacy (≤20%) area (International Experts)

100

E f f e c t i v e n e s s

Face-to-Face

CR

80

Public AM

Terrestrial RIC-1 TV

60

40 Satellite TV

Local Newspaper RIC-2

20

Printed Brochure

National Newspaper 0 0%

20%

40%

60%

80%

100%

Likely Achievable Technology Penetration in Bangladesh Figure 5 represents the relative effectiveness (judgment of the international experts), penetration and cost-per person-year status of each considered technologies in providing Agriculture (General Extension) related information service in a low literacy community. The X and Y axes correspond to the likely achievable technology penetration and the effectiveness scores respectively. The spread of effectiveness scores for various Information Service interventions (from 0 to 100 in the Y-axis) can be seen (with multiple dots) alongside the highlighted ones representing average values. The size of the average value data points correspond to the Cost-per Person-Year for each considered technologies. CR is the most cost-effective technology based option here, with tradeoffs in cost, penetration, and effectiveness in the next choices of face-to-face, Public AM and Rural Information Center with Internet connection.

VI. RECOMMENDATIONS FOR EFFECTIVE INFORMATION SERVICES While this is a comparative analysis, in all technologies, availability of local, relevant, right-language content is a key challenge, one for which the costs can and do vary enormously. Based on the relative effectiveness analysis through expert elicitation, stochastic cost modeling, literature review, personal discussions with practitioners and anecdotal evidences, we make the following recommendations for effective technology centric information services deployment. • Direct human involvement should be an integral part of any "Communication for Development" initiative. Experts' overwhelming preference for "Face-to-Face" interventions and research findings related to the importance of human interaction justifies its importance at least as an auxiliary medium for any technology based information service options. • As the majority of the Bangladeshi population lacks access to effective information services, participatory radio based broadcasting options can play important roles as a socially sustainable development mechanism. • Over-the-air TV broadcasting has wider appeal, both as a source of entertainment and societal communication

•

•

among the general population, irrespective of the literacy and economic status. Airing a greater number of development oriented programs, opening up education channels, ensuring participation of the local population in regional TV stations are some of the ways to develop an effective TV based information service network. Especially for a disaster prone country like Bangladesh, decentralized emergency warning systems and response mechanisms need to be established. In addition to the traditional satellite based initiatives, community based micro-broadcasting systems with local participation need to be developed for a sustainable disaster response infrastructure beyond the initial warning. In recent times, the latency in getting the weather updates from a single point has resulted in the loss of thousands of lives in Bangladesh. Ideally dominant in higher literacy areas, Print media remains the backbone of education systems, one of the most important service sectors considered in our analysis. The success of Print based options as a part of any integrated delivery mechanism [43] makes it an important element for any viable communication for development option.

264

Figure 6 Cross Technology Comparison for providing Agriculture (General Extension) based Information Service in a high literacy ( ≥ 60%) area (Local Experts)

100

E f f e c t i v e n e s s

Face-to-Face

80

Public AM

Terrestrial TV

60

Satellite TV

RIC-1

Printed Brochure

National Newspaper

20

RIC-2

Local Newspaper

40

CR

0 0%

20%

40%

60%

80%

100%

Likely Achievable Technology Penetration in Bangladesh Figure 6 represents the relative effectiveness (judgment of the local experts), penetration and cost-per person-year status of each considered technologies in providing Agriculture (General Extension) related information service in a high literacy community. The X and Y axes correspond to the likely achievable technology penetration and the effectiveness scores respectively. The spread of effectiveness scores for various Information Service interventions (from 0 to 100 in the Y-axis) can be seen (with multiple dots) alongside the highlighted ones representing average values. The size of the average value data points correspond to the Cost-per Person-Year for each considered technologies. Face-to-Face is considered to be most effective. Among the technology based options, CR is the most cost-effective one, followed by Terrestrial TV and Public AM .

•

•

•

With literacy independent access efficiency, RIC-1 can provide a platform for sustainable Information (this implicitly assumes the use of informadiaries). The PK inspired proactive approach reduces the "barrier to entry" factor for local information service facility [46]. Following the successful example of Nepal [50], a balanced integration of Face-to-Face, CR, Print and Internet based multimedia activities can address the majority of the information service needs in South Asian region. Efforts to demystify the technologies used for information service centers need to be in place to ensure greater efficacy and wider participation of the common people. For example, low cost CR options such as the "suitcase radio" can be used in the field for on- the spot broadcasting and human resource development purposes. Both the literature review and collected field data [11, 51] identified "scarcity of trained human resources" as one of the major roadblocks towards establishing an effective information service network. Experiences from South America and from Nepal showed the effectiveness of decentralized support and resource centers for training information service providers on technical issues, content

development, management etc. [11, 52] Regular inservice training for personnel and resource pooling in remote places (i.e. sharing of content, technicians etc.) also can ensure the operational sustainability of any information service center. This is vital for long-term sustainability beyond the initial period. A. Additional Policy Suggestions These draw from literature, our analysis and suggestions from the surveys: • Effective government policies and regulations need to be in place to create an enabling environment for development-oriented information services. The policies should be effective in removing bottlenecks for the free flow of information by giving away incentives like lowered or zero tax rates for imported equipments, lowered license fees etc. • Information service providers engaged in societal and development work need separation from competing with commercial entities. The absence of such support mechanisms can result in unfair competition, pressure on service providers to overlook societal commitment for

265

Figure 7 Cross Technology Comparison for providing Agriculture (General Extension) based Information Service in a low literacy (≤20%) area (Local Experts) Face-to-Face 100

E f f e c t i v e n e s s

Terrestrial TV

80

Public AM

CR

60

40 Satellite TV

Printed Brochure

RIC-2

RIC-1

20

0

National Newspaper 0%

Local Newspaper 20%

40%

60%

80%

100%

Likely Achievable Technology Penetration in Bangladesh Figure 7 represents the relative effectiveness (judgment of the local experts), penetration and cost-per person-year status of each considered technologies in providing Agriculture (General Extension) related information service in a low literacy community. The X and Y axes correspond to the likely achievable technology penetration and the effectiveness scores respectively. The spread of effectiveness scores for various Information Service interventions (from 0 to 100 in the Y-axis) can be seen (with multiple dots) alongside the highlighted ones representing average values. The size of the average value data points correspond to the Cost-per Person-Year for each considered technologies. Face-to-Face is considered to be most effective. Among the technology based options, CR is the most cost-effective one, followed by Terrestrial TV and Public AM. Rural Information Center without the Internet connection has been considered to be more effective than the one with the web-enabled options.

•

•

•

greater commercial success and, at times, the abrupt closure of such initiatives[4]. A “Development Information Service Fund” can be created to provide monetary help for establishing information infrastructure in underserved communities. All the commercial broadcasting and communication enterprises can contribute to this fund on a mandatory basis. This type of mechanism is the backbone for parallel policies worldwide on Universal Service for telephony [53]. On modeling and comparing program and labor costs of community based information service providers in isolated versus more collaborative environments, we find found that national, regional and local cooperation in terms of program and resource sharing should be encouraged officially, in addition to pooling resources for learning, training, troubleshooting etc. (given labor is a significant fraction of operational costs [5]. With the trend of transferring everything to be digitized and being transferred to cellular phone media if not the Internet, the policy framework should support digital information service initiatives, content sharing and mobile wireless based information services in near future. This

•

extends to applications such as voice enabled and GSM based help-lines (for providing healthcare, market information etc.), more hybrid solutions or innovations such as Internet Radio. Hybrid solutions between technologies, including face-toface and mobiles, might offer the most promise. This implies policies and regulations should not be spoiled by technology, but, rather, facilitates such convergence. VII. CONCLUSION

This analysis provides a comparative framework and an initial estimate of the cost-effectiveness across technologies for delivery of information services for rural users in Bangladesh (and elsewhere). Effectiveness analysis using experts’ judgment and financial data modeling in this paper have demonstrated the strength of Radio based information service options as useful and applied tools for development activities in Bangladesh and S. Asia. Rural Information Centers with net connections also turn out to be very promising as hybrid information access points (under the assumption of informediaries for extending their reach). We find awareness is a challenge in choosing technologies, given the differences between international and local (Bangladeshi

266

experts) – this implies dissemination of data and best practices to be of value. Given the enormous differences in cost, effectiveness and penetration, often with a tradeoff (there is rarely a dominant solution), policy-makers will need to decide how to prioritize technologies in the form of enabling environment. Regulatory and policy decisions can strongly impact the relative costs of technologies as well as their penetration. Nonetheless, the broad portfolio of technologies available for Information Services for development implies great potential for the region, with relatively low costs per beneficiary for all the technologies (under $1 per person-year in scale).

REFERENCES [1] [2] [3] [4] [5]

[6] [7] [8]

[9] [10] [11] [12] [13]

[14] [15] [16] [17] [18] [19]

[20]

[21]

R. Rajora, Bridging the Digital Divide: Gyandoot—The Model for Community Networks. New Delhi: Tata McGraw-Hill, 2002. "Bangladesh Telecentre Network." vol. 2008. "Case Study Series on ICT-Enabled Development:I-NetworkCase Study Series on ICT-Enabled Development:I-Network." vol. 2008: bridges.org, 2004. I. Pringle and B. Subba, "Ten Years on: The State of Community Radio in Nepal," 2007. F. Hussain and R. Tongia, "Community Radio for Development in South Asia: A Sustainability Study," in IEEE/ACM International Conference on Information and Communication Technologies and Development Bangalore, India: IEEE, 2007. F. J. Proenza, "Telecenter Sustainability- Myths and Opportunities," in Bridging the rural knowledge gap: Information systems for improved livelihoods: FAO, 2001. "Radio Sagarmatha.." vol. 2008, 2007. R. Mainali, "Volunteering and Community Participation in Community Radio: Experiences from Nepal.," National Consultation for Practicing and Potential Community Radio Operators 2007. "India: Time for Community Radio has come!," UNESCO, 2007. F. Noronha, "India still waits: Rural poor not yet ready for the promise of radio." vol. 2006, 2004. C. Fraser and S. R. Estrada, Community Radio Handbook: UNESCO, 2001. J. Antin, "Cultural Assessment for Sustainable Kiosk Projects," in ICTD 2006 UC Berkeley, California, USA: IEEE, 2006. F. Hussain, "Community Broadcasting in South Asia: Hybrid Solutions for localized Information Access Points (A Bangladeshi Case Study)," in 26th Scientific Conference on the International Association for Media and Communication Research: Media and Global Divides (IAMCR 2008) University of Stockholm, Stockholm, Sweden, 2008. "CIA- The World Factbook -- Bangladesh." vol. 2008, 2008. "Media in Development: Linkages Between Poverty Reduction, Social Development and Diversified Media in Bangladesh," Bangladesh Enterprize Institute, Dhaka 2006. "National Media Survey, 2002," 2002. "Bangladesh Telecommunication Regulatory Commission." vol. 2008, 2008. "Many Voices, One World," Essex 1980. C. Rodriguez, "From Alternative Media to Citizen's Media," in Communication for Social Change, Anthology" Hostorical and Contemporary Readinngs, A. Gumucio-Dagron and T. Tufte, Eds.: CFSC, 2001, pp. 778-786. A. E. Opubar, "If Community Media is the Answer, What is the Question?," in Communication for Social Change, Anthology" Hostorical and Contemporary Readinngs, A. Gumucio-Dagron and T. Tufte, Eds.: CFSC, 1999, pp. 778-786. A. Gumicio-Dagron, "Take Five: A Handful of Essentials for ICTs in Development," in Communication for Social Change, Anthology" Hostorical and Contemporary Readinngs, A. Gumucio-Dagron and T. Tuft, Eds.: CFSC, 2003, pp. 976-988.

[22]

[23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36]

[37] [38] [39] [40] [41]

[42] [43]

[44] [45] [46] [47] [48] [49] [50] [51] [52] [53]

A. Opoku-Menash, "Radio, Conflict and Political Transition: The Future of Community Radion in Africa," in Communication for Social Change, Anthology" Hostorical and Contemporary Readinngs, A. Gumucio-Dagron and T. Tuft, Eds.: CFSC, 2000, pp. 856-963. A. Dock and J. Helwig, "Interactive Radio Instruction: Impact, Sustainability and Future Directions," World Bank, 1999. I. Pringle and M. David, "The Kothmale Model: Using Radio to Make the Internet visible," in The One to Watch: Radio, New ICT and Interactivity, B. Girard, Ed. Rome: FAO, 2003, pp. 97-114. D. R. Holmes, D. M. Kamacharya, and J. K. Mayo, "Radio Education in Nepal," in Distance Education for Teacher Training, H. Perraton, Ed. New York: Routledge, 1993, pp. 136-195. A. W. Bates, Technology, Open Learning and Distance Education. London: Routledge, 1995. T. Hulsmann, "Costing Open and Distance Learning," 2000. "India's foremost television network - Doordarshan." vol. 2008. "CIA - The World Factbook -- India." vol. 2008, 2008. "Future Tense of Dailies ". vol. 2008, 2001. "CIA - The World Factbook -- Nepal," 2008. "CIA - The World Factbook -- Pakistan," 2008. A. Riaz, "Media and Judicial Independence in Pakistan," 2004. "Teledensity growth in Pakistan almost double in FY 2006." vol. 2008, 2006. L. Tobing, How To Do Community Radio: A Primer for Community Radio Operators. New Delhi: UNESCO. R. Iannella and K. Henricksen, "Managing Information in the Disaster Coordination Centre: Lessons and Opportunities," in 4th International ISCRAM Conference (B. Van de Walle, P. Burghardt and C. Nieuwenhuis, eds.), Delft, the Netherlands, 2007. M. Carnoy, "ICT in Education: Possibilities and Challenges," 2004. L. S. Austin and K. Husted, "Cost-Effectiveness of Television, Radio, and Print Media Programs for Public Mental Health Education ". R. Kumar and M. Best, "Social Impact and Diffusion of Telecenter Use: A Study from the Sustainable Access in Rural India Project," ITID. F. J. Proenza, "Telecenter Sustainability- Myths and Opportunities," in Bridging the rural knowledge gap: Information systems for improved livelihoods: FAO, 2001. S. Finquelievich and G. Kisilevsky, "Community Democratization Of Telecommunications Community Cooperatives In Argentina: The Case Of Telpin," The Journal of Community Informatics, 2005. "South Asia Economic Report – Social Sectors in Transition," 2007. G. M. Faheem Hussain, Rahul Tongia, "Technological Interventions for BRAC's In-Service Teacher Training Program: A Comparative Analysis," International Journal of Education and Development (submitting for publication), 2008. "Radio Lumbini 96.8 MHz." vol. 2008, 2008. B. Haiplik, "The BRAC Education Program Teacher Training and Development System Quality Improvement Initiative: Report #2, Monthly Refresher Training for BRAC Teachers," 2004. "Pallitathya Kendra." vol. 2008, 2008. S. Rahman, "Internet in Bangladesh," Bangla ICT 2008. "The Most Deadly 100 Natural Disasters of the 20th Century." vol. 2008. T. Köhler, M. Reitmaier, and G. J. Schulz, "Telecenters as Instrument for Bridging the Digital Divide in Rural Areas," TU Dresden. "Community Radio Madanpokhara (CRM) Palpa." vol. 2008, 2008. "Community Broadcasting Association of Australia ". vol. 2008, 2008. "Nepal Forum of Environmental Journalists." vol. 2008, 2008. "Universal Service." vol. 2008: Federal Communications Commission, 2008.

267

Decentralization, Clientelism and Popular Participation—Is there a role for ICTs to improve local governance? Björn-Sören Gigler, London School of Economics  Abstract—This case study investigates under which conditions ICTs can play a role in fostering the empowerment of rural communities to fully participate in the decision-making processes of local governments. The analysis using empirical evidence from rural communities in Bolivia focuses hereby on the following key questions: i) to what extent can ICTs contribute to improving the efficiency and efficacy of local government? ii) does ICTs have the potential to make local governments more transparent and responsive to the needs of rural communities and iii) can ICTs support the core objectives of the Bolivian Law of Popular Participation to strengthen the role of local government in public-policy making and the implementation of development programs? The article will explore these issues using several case studies of rural municipalities in Bolivia that have participated in the USAID-funded project Enlared Municipal. It concludes that the most important factors influencing whether ICT programs can support the processes of decentralization, improved social accountability and thus improve the well-being of rural communities depends primarily on political, social and cultural factors, whereby economic and technical factors are secondary.

O

I. INTRODUCTION

ver the past few decades, most developing countries have embarked on decentralization programs with the ambitious aims of: i) increasing government‘s efficiency and efficacy; ii) enhancing the quality of service delivery in such areas as health and education; and iii) to enhance the responsiveness and poverty-focused of governments [1]-[3]. Most of these programs operate under the assumption that decentralized governments achieve more efficient results and are less corrupt since they are closer to the people, better able to understand the local dynamics of economic and social development and are thus more accountable for their programs to rural communities [4]. These programs are usually embedded into broader policy reforms of the public sector and go frequently hand in hand with major privatization programs. In Bolivia, the 1994 Law of Popular Participation (LPP) signaled a dramatic shift in national policy from a centralized

Manuscript received September, 22, 2008. Bjӧrn-Sӧren Gigler is a PhD Student at the Development Studies Institute (DESTIN), London School of Economics and Political Science, Houghton Street, London WC2A 2AE, UK (e-mail: [email protected]).

top-down policy approach to a decentralized, participatory, locally empowering approach [5]. The LPP transferred authorities and resources from the central state to local municipalities in order to make government more efficient and to improve the implementation of its policies for poverty alleviation1. At the core of the decentralization program was (i) the delegation of new responsibility to municipal governments and Territorial Grassroots Organizations (TGOs) with an allocation of 20% of national tax revenue upon approval of their Annual Operating Plans (AOPs) for the social sectors—mainly health and education and (ii) the establishment of mechanisms for social control, called Vigilance Committees (VCs), responsible for overseeing the expenditure of funds [6]-[7]. Critics of the law point out that its implementation faced severe obstacles since powerful regional elites continued to dominate local municipal planning processes and captured the resources provided by the central government [8]. Furthermore, the results from the bottom-up participatory process had to fit with national and departmental plans, and respond to the spending priorities of the national government [9]. It is however important to acknowledge both the LLP‘s shortcomings and its successes— in particular that, with the LLP, the Bolivian government finally recognized traditional indigenous forms of organization and transferred important resources to local governments. It was, in short, a formidable attempt to shift power relations within Bolivia‘s mestizo-dominated society, so that indigenous peoples and other marginalized groups would have the space to voice their concerns, interests, and needs. This case study investigates the role ICTs can play in supporting such a decentralization process and in helping to improve the well-being of indigenous peoples. In particular, the case study addresses the following questions: (i) do ICTs contribute to improving the efficiency and efficacy of local governments, thus strengthening their institutional capacity to provide basic services to local communities?; (ii) can ICTs enhance the transparency of local governments and make them more responsible and accountable to the needs of local communities?; (iii) was Enlared Municipal, the centralized government-led program, able to achieve its main objectives, and which constraints and limitations did it face in the implementation of the program?; (iv) in which dimensions 1 Bolivian Government in cooperation with the IMF and the World Bank, ―Economic Policy Framework Paper for 1997-99‖, June 1997

268

(i.e., political, organizational, and social) do ICTs have the biggest influence on local governments and to what extent do they enhance the human and social capabilities of the program‘s participants?; and (v) did the project lead to any institutional changes in local governments, thus altering the relationship between local government officials and indigenous communities? The case study investigates these central issues, based on the empirical evidence from the government-led ―Enlared Municipal‖ project—the principal ICT program in support of local governments in Bolivia, coordinated by the Federation of Municipal Associations of Bolivia (FAM). The overall objectives of the program were (i) to improve the transparency and accountability of local governments; (ii) to enhance local governance structures and to improve the relationship between local government and indigenous communities; and (iii) to strengthen the institutional capacity and thus the performance of local governments in particular in relations to the delivery of social services. Thus, the program‘s main focus was the political dimension, as its principal aim was to improve local democratic processes and to strengthen the popular participation of indigenous peoples in local governments. Nevertheless, the program also included in its overall goals institutional and social aspects of local development as evidenced through its third general objective. Based on these objectives the project focused on the following specific activities: (i) implementing an ICT capacity-building program for local government officials; (ii) developing and promoting the use of a national Portal on issues related to municipal development www.Enlared.org.bo among local, regional, and national government officials; (iii) assisting municipalities to develop and implementation their own municipal websites at the local level and (iv) helping formulate a national ICT strategy within the framework of the ongoing UNDP-sponsored ETIC-initiative2. II. THE RELATIONSHIP BETWEEN E-GOVERNMENT, GOOD GOVERNANCE AND DECENTRALIZATION

E-government is defined in the literature “as the use of ICTs to improve the activities of public sector organizations‖ [10]. More specifically, the main objectives of e-government are (i) to make government more accessible, effective and accountable; (ii) improve the relationship between government and citizens; (iii) strengthen the coordination and cooperation within the public administration; and (iv) enhance government performance in the delivery of public services [11]-12]. The proponents of e-government frequently argue that the application of ICTs in the public sector can play a catalytic role for greater democracy, improved government performance and socio-economic development more generally [13]-[16] particular, the literature frequently points to the potential of e-government programs to promote good governance by enhancing the performance and accountability 2 The Estrategia Nacional de las Tecnologías de Información y Comunicación (ETIC) is a UNDP-financed initiative which developed through a participatory process including all sectors of the Bolivian society a National Strategy for the Information Society for Bolivia (www.etic.org.bo)

of local governments [17]-[20]. Most of these authors have an optimistic view on ICTs and their role in development and thus recommend that governments proactively embrace egovernment applications in order reap the benefits from the use of ICTs within the public sector. More critical views of e-government highlight that in praxis the majority of e-government programs in developing countries have failed [21]-[23]. In fact, Heeks [24] ascertains that the empirical evidence shows that 35% of e-government programs are total failures (the program was not implemented or was immediately abandoned after its development), and 50% are partial failures (major goals were not attained and/or there were undesirable outcomes). He introduces the ‘realitydesign‘ gap framework to analyze the underlying causes for the numerous cases of failures and shows that the major reason for these failures is the common gap between the current realities within public administrations and the design of the e-government project. Heeks concludes that the larger the design-reality gap is in the following seven dimensions (information, technology, process, objectives and values, staffing and skills, management systems and structures and other resources) the greater the risk for-government programs to fail [25]. Madon demonstrates based on an interpretive set of case studies in Kerala India, that e-government programs have frequently neglected such critical factors as adequate resources, skill-levels, values, beliefs and motivations and thus poor people and communities were unable to derive real benefits out of such investments [26]. The author proposes a framework for evaluation the impact of e-governance projects based on Sen‘s capability approach in order to be able to assess the developmental impact of these programs on peoples‘ quality of live. Madon concludes that a critical barrier for the successful implementation of e-government programs is that far-reaching back-end administrative reforms need to be implemented in parallel for these programs to have any significant positive effects on local governance [27]. The literature on the e-government and decentralization has stressed that the principal value of ICTs lays in both in enhancing the participatory (good governance) and the managerial (improved performance) dimensions of local government [28]-[30]. Lawrence Pratchett emphasizes that ICT projects have the potential to contribute to support local governments in the following three dimensions of their work: (i) enhance local democracy; (ii) promote public policymaking; and (iii) improve the quality of their service delivery [31]. Most scholars however highlight that local governments use ICTs primarily as instruments to improve the delivery of services and neglect its uses to enhance participatory mechanism conducive for good governance [32-35]. In fact, an empirical study of the ICT use of 270 local governments in California highlights that most municipalities do not see ICTs as an instrument for promoting good governance at the local level, but as an additional communications tool—failing to explore their value as change agents and their potential to alter the fundamental relationship between government and citizen [36].

269

This case study draws on previous e-government studies by Ciborra (2005); Ciborra and Navarra (2005); Madon (1993, 2004), Walsham (1993) and Heeks [37]-[42] in its approach to apply Pettigrew‘s (1985; 1998) methodology of a contextual approach which emphasizes the importance of the economic, social, political, and cultural context into which the technologies are introduced [43]-[44]. These factors thus will be essential for assessing the effects Enlared program had on local governments and the well-being of indigenous peoples.

―Frequently the needs of local communities are not considered, the projects are in fact designed on the desks of the Ministries or of the international donors. A consequence of this approach is that many projects are not sustainable at the local level and finish once the financing dries up. This program must be one of those unsustainable projects—at least for my municipality there are currently no concrete benefits visible from this project‖.3

III. THE ENLARED MUNICIPAL PROGRAM The Enlared Municipal project was implemented in two distinct phases. The first phase of the project, from March 2001 to June 2004, was implemented by the International City/County Management Association (ICMA) and financed by the U.S. Agency for International Development (USAID). ICMA is a U.S.-based international professional organization for managers, administrators, and assistants in cities, towns, and counties and provides technical and managerial assistance, training, and information resources to its members and the local government community (ICMA website). During its first phase, the project developed an online central Portal for municipal government (www.Enlared.org.bo), attempted to introduce a national bidding platform for Bolivia, and provided technical support to the Mancomunidad of the Chiquitania for the development of an integrated Management Information System for the fourteen municipalities belonging to the Mancumidad of the Chiquitania [45]. However, the program‘s impact on local governments was limited by its ambitious and poorly focused objectives, its concentrated, top-down organization, and the perception among participants that the ICMA was a foreign and distant organization unconnected with the particular realities of the Bolivian municipalities. During field visits to the Chiquitania region in the Eastern lowlands in from April 20-25 and August 2-7, 2005, it became apparent that the project did not have any sustainable impact on local governments and communities. Interviews with local indigenous leaders, several government officials, mayors, and Juan Burgos, the ICT project coordinator of APCOB—a local NGO promoting indigenous people‘s development in the eastern lowlands—revealed that the municipalities and communities were not aware of the suggested Information Management System, and that in fact the fourteen communities of the Mancomunidad did not have any connectivity to the Internet. The tremendous ―designreality‖ gap between project descriptions, consultant reports, and the realities on the ground was highlighted by the evidence through the results from my fieldwork. Most people interviewed were not aware that the project existed, even though its first phase had been completed less than a year before the fieldwork was carried out. In the municipality of Concepcion, located in the Chuiqutania, approximately five hours from the city of Santa Cruz, the newly elected mayor clearly expressed his frustration with the overly centralized planning and execution of the projects:

The findings of the fieldwork showed that the first phase of the project did not have any lasting development impact on indigenous communities, so the research is focused instead on the second phase of the project. Based on its experience with the first phase of the project, USAID decided to refocus the program and strengthened its partnership with Bolivia‘s national association of local governments—the Federación de Asociaciones Municipales de Bolivia (FAM)4. During this phase—from the beginning of September 2004 to the end of November 2005—the FAM assumed responsibility for project coordination. The Federation organized a competitive bidding process in mid-2004, which was won by the Swiss-funded program Programa de Apoyo a la Democracia Municipal (PADEM)5. The program focuses on increasing the capacity, accessibility, and accountability of municipal governments in order to strengthen the participation of community-based organizations and citizens in local governance. This program was selected, in part, because of its seven-year experience in working with local governments on issues related to good governance, participation, and empowerment6. The new Enlared project team aimed to make the program more responsive to specific local needs and integrated ICTs into the ongoing local and regional processes of policymaking and citizen participation. There was a strong emphasis on finding new ways of achieving the overall development objectives of the project. A. Socio-economic, Cultural and Technological Diversity of Municipalities In order to set realistic objectives, the project team focused its technical assistance on fifteen municipalities and nine regional municipal associations, covering all of Bolivia‘s nine states.7 A major challenge for the project team, however, was 3

Interview with Justo Seoane, Mayor of Concepción, and Ex Vice-Minister of Indigenous Affairs, April 22, 2005. 4 FAM is the National Association of all the Bolivian municipalities with the exception of the major cities of La Paz, El Alto, Cochabamba and Santa Cruz. 5 PADEM stands for the Program for the support of the democratization of municipalities, which is being coordinated by the Swiss Non-governmental organization Ayuda Obrera Suiza (AOS), which works towards the strengthening of local democratic processes and the empowerment of poor communities throughout Latin America. 6 Interview with Carlos Soria, coordinator of the Enlared project. August 25th 2005. 7 The 15 municipalities include: Cobija (Pando), Trinidad (Beni), Montero and Comarapa (Santa Cruz); Sipe Sipe and Villa Tunari (Cochabamba), Monteagudo and Tarabuco (Chuquisaca); El Puente (Tarija); Batallas and

270

that the fifteen municipalities differed extremely between rural, poor indigenous communities, such as Batallas in La Paz, or El Puente in Tarija, rural more mestizo dominated municipalities such as Comarapa or Montero in Santa Cruz and large urban municipalities such as Trinidad or Cobija— the capitals of Beni and Pando, respectively. In addition to the tremendous difference in the socio-economic, political, and cultural context, these municipalities differed in their access to electricity and telecommunications services. To address and understand the heterogeneity and the local context of the fifteen participating municipalities described above, the program team initiated its activities in October 2004 with a detailed ICT needs assessment at the local level. Its main objective was to assess the differences in awareness, ICT uses, and proficiency of use among local government officials in the fifteen participating municipalities. As part of this baseline study, the team carried out community meetings, interviews, and structured surveys with mayors, council members, technical staff of the municipalities and with representatives of local Territorial Grassroots Organizations (TGOs), members of the Vigilance Committees (VCs), and other community leaders. In total, 634 people—424 men and 210 women—participated in this baseline study and provided valuable information on their information and communications needs and their current use of ICTs [46]. The assessment found that there was little awareness and uses of the Internet in any of the rural municipalities, and that, even in the urban municipalities like Tupiza or Uyuni, local governments were not using the Internet for work. In fact, out of a total of 127 local government officials, only a quarter (24%) used the Internet for work; the overwhelming twothirds majority did not use the Internet at all, and one-tenth of participants did not even use computers. The results of the needs assessment reveal profound differences in ICT readiness among the fifteen municipalities studied. B. Main Activities Based on the baseline study, the project team focused its activities on the following four strategic areas: (i) technical redesign of the Portal; (ii) content development and online services; (iii) capacity-building in the use of ICTs; and (iv) news desk and communications. Within these different areas the project was able to achieve the following specific results. First, the Portal www.enlared.org.bo was redesigned using local technology and focusing on providing technical support to the Federation of Municipal Associations, as such helping 9 municipal associations and individual municipalities in the development of their own websites. At the completion of the second phase of the project, the Portal hosted 116 sites from these organizations. The Portal placed significant effort on providing targeted and timely content for policy makers, local government officials, donor agencies and the general public interested in issue of municipal and local development. For instance, the Portal provided for the first time a comprehensive list of all newly elected mayors (from the Coroico (La Paz); Machacamarca and Pampa Aullagas (Oruro) and Uyuni and Tupiza (Postosi).

municipal elections in December 2004) and their contact information, as well as the most recent socio-economic and human development data on all 314 municipalities. In addition to the provision of content, the Portal provided more interactive features, such as a weekly forum with national or local policy makers including relevant issues of interest such as the Constitutional Assembly or political and social exclusion of the extreme poor due to their lack of appropriate documentation. Concerning the capacity-building program, the project trained 60 municipal staff in the use of ICTs (30 from the regional associations and 30 from the 15 municipalities). The staff from the municipalities participated in two one-week national training workshops, one organized in June in La Paz and the other one organized in late August in Cochabamba. The content of the training focused on (i) basic computer skills, such as word processing or use of spreadsheets; (ii) introduction to the use of the Internet; (iii) the design and development of websites and; (iv) an introduction to ‗digital‘ journalism. The main purpose of these training workshops was to create a network of technical specialists in the municipalities, who then could become the municipal focal points for Enlared. IV. EVALUATION OF THE PROJECTS IMPACT Based on the above description of the project‘s main achievement or outputs, this section will focus on evaluating the impact the project had on the well-being of rural communities. The evaluation of the project impact is hereby defined as the “systematic analysis of the lasting or significant changes— positive or negative, intended or not—in people’s lives brought about by a given action or series of action [47]. As such, the analysis aims to unpack the extent to which this project has made a difference in peoples‘ lives and has in fact met its overall development objectives by asking the following: (i) Has the project contributed to strengthen the managerial and technical capacity of local governments, thus allowing them to play a more pro-active role in public-policy making? (ii) Has the project supported the transparency of municipalities and thus promoted participatory democratic processes at the local level; and finally (iii) Did the project enhance the efficiency and efficacy of local government thus improving its delivery of services, such as health and education to its community. This categorization is based on the theoretical framework developed by Lawrence Prachett described above in the case study [48]. The article applies the alternative evaluation framework based on Sen‘s capability approach developed by the author in an earlier article to evaluate the impact of the e-government program on people‘s well-being [49]. A. The program’s Overall Impact The principal finding presented in this section is that the program had a very limited positive overall impact on its participants‘ human and social well-being. The empirical

271

results are based on two user surveys carried out by the project team and a sub-sample from the impact survey (n=22), carried out as part of my research in July 2005. The first user survey (n=190), administrated as part of the ICT needs assessment in October 2004, was designed to generate a baseline about the awareness and proficiency of ICT use in the fifteen municipalities. The second user survey (n=186), was carried out by the project team in September 2005 in the same fifteen municipalities, as part of its monitoring and evaluation activities. The main difference between these two surveys and the impact survey that I carried out as part of my research is that they did not include any questions on people‘s perceptions of the project‘s impact on their well-being and the well-being of their communities, but instead assessed the participants‘ proficiency in ICT use and on evaluating the ICT training component of the program. Before analyzing participants‘ perceptions of the project‘s impact on their human and social capabilities, we can observe several basic statistical indicators about people‘s ICT use and conclude that the project had a limited impact in the fifteen municipalities it targeted. The first critical finding from the data analysis is that the large majority of local government officials from the fifteen municipalities that participated in the project were entirely unaware of the project‘s existence. Only a small percentage of respondents to the baseline survey administrated in 2004 and the user survey from 2005 indicated that they knew of the project, and an even smaller percentage of the respondents had actually used the services provided by the project. In fact, the data from the first survey indicate that out of 127 respondents, only 16 (13%) knew about the project and only 14 (11%) were using the first municipal Portal developed by ICMA. This figure is astonishing, particularly since in October, 2004 the project had already been operating in Bolivia at a national level for three years. These data confirm that, particularly in its first phase, the project failed to make any long-term positive impact on local governments or on the participating communities. The graphic below shows that awareness of the program among local government officials significantly increased during the second phase but the project continued to suffer from relatively low awareness about its activities at the local level. The empirical data from 2005 show that out of a total of 190 respondents, 70 people (about 37%) indicated that they knew of the project—a major improvement from the extremely low figure of 13% a year earlier. The project continued to be under-utilized: only 31 respondents (17.5%) indicated that they frequently used the services provided by the project. In spite of the significant improvements during the second phase of the project, the continuously low levels of awareness and usage among local government officials raise major questions about the program‘s overall effectiveness and are a good initial indicator for the limited overall impact of the project.

Knowlede about the EnlaRed Municipal Program among local government officials (user survey: n=190)

N/A, 22, 12% Yes, 70, 40%

No , 85, 48%

Yes

No

N/A

Fig. 1: Low Awareness of the Enlared Municipal program The principal finding of this section—that the project had a very low overall impact—was also confirmed by the empirical results from the impact survey I carried out during my fieldwork. Results from this survey indicate that—based on the participants‘ self-perceptions—the program did not enhance in any significant manner the human and social capabilities of its participants. The graphic below shows that the participants perceived the Internet to have had only a very minor positive impact across all dimensions of their lives. With the exception of the organizational dimension, in all other dimensions (i.e., political, social, and economic) the majority of participants expressed serious doubts about the positive impact of the Internet on their lives, indicating that they believed that the Internet could only ―somewhat‖ or ―not at all‖ enhance their well-being. Another remarkable finding was that the Enlared Municipal project participants were very skeptical about the Internet‘s impact in all dimensions of their lives. For instance, only 56% of the Enlared participants indicated that the Internet had a positive impact on the social dimension of their well-being. The negative perceptions of the participants is particularly striking in the political and social dimension, since the core objective of the Enlared Municipal project was to improve the performance of local governments in these two dimensions. In fact, more than two-thirds (70%) of the program‘s participants believed that the Internet could only ―somewhat‖ or ―not at all‖ improve their well-being in the political sphere of their lives. In the social dimension only 31% of its participants indicate that the Internet could significantly enhance their well-being in this dimension. In sum, the empirical data reveal that the participants were particularly skeptical of the Internet‘s ability to improve their own well-being and the living conditions of their communities politically or socially, constituting the program‘s failure. On the other hand, the organizational dimension is the only dimension in which its program participants‘ have favorable views. In fact, 59% of the respondents believe that the Internet can significantly enhance their well-being in this dimension. With respect to the economic dimension, respondents‘ views are most skeptical; an overwhelming majority (70%)

272

59

Organizational

18

41

Personal

24

35

Cultural Social

31

Political

30

Economic

0%

24

23 35

35

30

25

44

35

35

6

70

20%

40% A lot

60%

Somewhat

80%

100%

None

Fig. 2: Enlared Municipal Participants‘ overall perceptions on ICT impact indicated that the Internet would not have any impact at all on either their personal or community well-being. The empirical evidence from the impact survey thus emphasizes the participants‘ overall negative perceptions of the Internet‘s effects on their well-being. In contrast with the overall results, Enlared Municipal participants were skeptical of the Internet‘s impact on their individual well-being and on the social capabilities of local governments and their communities. In fact, the data analysis demonstrates that the program failed to enhance people‘s individual capabilities and help local governments to improve their performance, accountability, and delivery of social services to the mostly rural communities. B. Impact on the Political Dimension of Well-being This section investigates the underlying reasons that the program had only a limited effect on the political dimension. This issue is central to the evaluation of e-government programs, and specifically the Enlared Municipal program, since its main objectives were political in nature. The analysis of survey results and interview responses in the previous section shows that the Enlared Municipal project failed to significantly improve participants‘ informational capabilities; this section will investigate the project‘s failure to promote ICT use as a way to improve governmental responsiveness, transparency, and accountability. A glaring result from the impact survey was that less than one-third of the respondents (30%) thought the Internet had any significant effect on the political dimension. The data indicate that 35% of the programs participants believed that the Internet had no effect at all on this dimension. The particular pessimism of Enlared project participants becomes apparent when their responses are compared with those from the overall sample, among whom 36% of respondents indicated that they were convinced that the Internet could play a significant role in this area and only 20% expressed serious doubts about its efficacy. It is critical to note again that the Enlared Municipal program was the only ICT program that specifically prioritized the use of ICTs to enhance peoples‘ political well-being. These

B.1 The Impact on Enhanced Transparency & Accountability The first critical finding is that the Enlared Municipal program had only very limited effects on the transparency and vertical accountability of local governments and thus could not significantly strengthen local democratic processes. The graphic below shows the effect the program had on the major four areas of accountability. The disaggregated results illustrate that the program helped improve people‘s access to information about their own rights and national government policies, but failed to enhance the transparency of local governments. In fact, less then one quarter of the respondents (23.5%) believed that the Internet could make an important contribution in this area, and only 25% were convinced of its value for enhancing people‘s access to information on government programs. Enlared: The internet and Vertical Accountability 60 50

51.8 45.2

41.2

40

Percentage

Enlared: Perceptions about the Impact of the Internet on peoples' well-being

findings clearly indicate the overall limited impact of the program in this dimension. What are the underlying causes that explain this finding? Which factors explain why the program was so unsuccessful in reaching its main objectives? What were the effects of the program on local governance structures and which barriers did it encounter in trying to enhance indigenous peoples‘ political well-being? The following analysis draws on the empirical results from the impact survey and presents more in-depth qualitative evidence from the two municipalities of Batallas and Comarapa. Based on Pratchett‘s above mentioned classification and the principal objectives of the Enlared program, I have broken down the political dimension into the following two areas: (i) enhanced transparency and accountability and (ii) improved local governance.

30 20

28.6

31.7 23.1

19.6

35.3

34.4

40.4

25.1

23.5

10 0 inproved information about nat. gov. policies

improved know ledge about citizen rights A lot

improved transparency

Somew hat

improved information about gov. programs

None/Na

Fig. 3: Continued lack of governmental transparency and accountability The results point to a principal incongruity in the program: while it had some positive impact through its Portal on improving the access to information related to municipal development at the national level, it did not improve information flows between municipalities and local communities. The program had an overly abstract national agenda and improved the access to information for a small elite of national policy makers, international donors and the

273

media, however failed to enhance in any significant manner the dynamics between local governments and its citizens at the community-level. This major finding was also confirmed by the semistructured interviews with local government officials, NGO representatives and community leaders in Batallas and Comarapa. In an interview, the ―official mayor‖8 of Comarapa clearly expressed his skepticism about the project and the role of the FAM in supporting local governments in general: ―We have a very diverse population in Comarapa and the surrounding communities; about half the population are mestizos and the other half are indigenous migrants from the highlands. Thus for us to improve our communications with all the communities is very important, but I don‘t see how this program can help us to better deal with this problems. The program has developed a website—How does this help us to improve our communications with more remote indigenous communities? How do these communities benefit from this—the majority of them do not even have access to electricity, then how can they access a website. I really can‘t see the value of this type of program for us.‖9 This comment needs to be understood in the context of Comarapa, a municipality that, in spite of relatively favorable socio-economic indicators, was confronted with significant social tensions between its primarily rural indigenous population and its urban mestizo population. In the statement above, the senior advisor stresses the challenge of improving his municipality‘s capacity to communicate with local communities in order to enhance the popular participation of a broad range of people in municipal planning processes. This is a critical shortcoming the implementation of the LPP has confronted in most rural communities throughout Bolivia. While the law has succeeded in significantly enhancing the participation of communities in the identification of municipal development projects through broad-based consultations, it has faced in most municipalities serious limitations in terms of improving the participation of local communities in municipal decision-making processes about critical issues of local development such as for instance the approval of planned public investment projects (Goudsmit and Blackburn, 2001). The above quote highlights that the program failed to enhance the transparency of the municipalities and to improve its communications with more remote indigenous communities. In Comarapa a critical issue furthermore was that the local government sided frequently with the mestizo 8

The ―official mayor‖ is the most senior government official in local government, who is in charge of the all processes of the public administration (own translation). 9 9 Interview with Jose Luis, official mayor of the municipality in Comarapa, July 15, 2005.

population and attempted to suppress the influence of indigenous peoples in local politics. The president of the local agricultural producers associations and a Quechua indigenous leader describes the exclusivity of local politics: ―For us it is very clear that the municipality is quite biased against our communities. It is always the same issue. We can wait for hours in the municipality with our demands, the mayor and the councilmember won‘t listen to our concerns. They are too busy with their political games and to represent the interest of the rich people of the city— we just don‘t have any voice in their decisions and we have no information at all about what type of programs and activities they are planning for this year.‘10 In short, local authorities lacked the will to improve access to information for the general population and instead allowed a small circle of mostly urban elites to dominate the local political decision-making process. The mostly rural indigenous population was entirely excluded from any political decision-making processes and was forced to choose the path of street blockages and civil disobedience to gain a voice in the political process. Questions remain, however: What was the role of Enlared Municipal project in such a complicated socio-political environment? Did the program have any influence at all on the local political process? Why did the program fail to promote in any way the popular participation of indigenous communities in the decision making process of local government. One reason was the programs‘ inability to react flexibly to the given socio-political context. Instead of working with local government officials to make the existing municipal planning processes more transparent and to gradually enhance the popular participation of indigenous communities in local decision-making processes, the project continued to promote its original concept to assist the municipality in developing a website which was supposed to enhance its transparency. This activity provided by the program clearly did not match the local socio-political context nor was it an appropriate mechanism to address the existing democratic shortcomings in Comarapa. Thus, the program did not alter the prebendal culture 11 of local government. The same few powerful groups continued to dominate the decision-making process within local government and the mostly indigenous population continued to be excluded and marginalized in political terms. In brief, the program failed (i) to improve the transparency and accountability of local government and (ii) to enhance peoples‘ political capabilities to better participate in the decision-making processes. The empirical evidence presented 10

Interview with Julio Sanchez, president of the agricultural producer association, Comarapa, May 24, 2005. 11 A prebendal local political culture is defined by local political practices that are characterized by patrimonial, clientelistic and caudillist features (Blackburn, 2000)

274

above, based on participatory observations and unstructured interviews with key policy-makers and local actors, clearly demonstrates that the local stake-holders perceived the Enlared project to be distant from their own realities and vividly illustrates the program‘s inability to meaningfully improve local democratic processes.

of our own program. Our idea was that we could share the connectivity costs with several institutions working in Comarapa and asked the mayor for his support. Unfortunately, the mayor did not see any value in this program and instead promised that the municipality will build its own a telecenter for the community. Now, two years later, Comarapa still does not have the promised telecenter and we have never received an official response from the municipality on our proposal we submitted in writing to the mayor‘s office.‖12

B.2 Impact on Enhanced Local Governance Another critical aspect of enhanced political capabilities at the local government level is the improved capacity of municipalities for public policy making in local communities. Within the increasingly fragmented and disaggregated structures of local communities many local governments face considerable challenges in playing a central role in promoting local governance structures that are conducive for economic and social development [50] The following analysis uses the concept of ―good local governance,‖ developed by Judith Tendler, which refers to a healthy ―three-way dynamics among local government, civil society, and an active central government‖ [51]. At the core of the concepts stands the notion that a good local governance structure is essential for effective public policy making. Tendler stresses that coordination between local governments, local civil society organizations, and communities, as well as coordination between local and central government, are critical for good governance. Local government is more of an effective facilitator of partnerships between different local actors and stakeholders than a sole implementer of development programs. A critical question this raises is what role ICTs can play in such a new system of local governance? The case study will focus on the extent to which the Enlared project was able to strengthen the capacity of local governments to assume the role of a ―facilitator‖ of local development. In brief, to what extent was the project successful in enhancing healthy local governance in local communities? The principal finding from both the municipalities of Comarapa and Batallas demonstrate that the Enlared project was unable to play a central role in enhancing the capabilities of local governments in this dimension, primarily because it did not effectively promote a cultural change within local governments to a more democratic and inclusive approach of local public policy making. Instead, local government continued to manage local development programs in their traditional authoritarian and bureaucratic styles. Asked about the role the Enlared program played in local politics, Claudia Camacho from ICO cited the program‘s inability to facilitate a partnership between the local government and her NGOs in order to work together on the construction of a telecenter for Comarapa.

―When we first heard about the Enlared

Municipal program and that Comarapa was selected to participate in its pilot phase, we were very excited and thought that we could closely collaborate with the municipality on promoting the use of ICTs, a key objective

This statement reflects the all too common lack of coordination among local municipalities, local NGOs, and the central government. Despite the stated interest from the municipal government and the Enlared project, after two years Comarapa still did not have the telecenter it had been promised. A key aspect in the failure of the Enlared Municipal program to promote good governance at the local level was the frequent discrepancy between the political mandate designating local governments the executive agencies of development programs like the ICT program and their actual institutional capacity and technical knowledge to actually implement such programs. The following comment by the mayor of Comarpa about the planned telecenter illustrates this point: ―For me it is very clear. We in the municipality are the executive branch of government at the local level. It is our mandate to plan and implement all programs in our municipality. In the case of the planned telecenter, I had several meetings with the staff from ICO and explained to them that it is our role to coordinate this project. I still think that it is a good idea to build a telecenter here in Comarapa, particularly for your youth, however it needs to be clear that we need to oversee all activities related to this project.‖13 This statement clearly demonstrates that the local mayor considered it the prerogative of local government to coordinate and oversee all development programs in his municipality. His defensive attitude illustrates the tensions between his local administration and the NGO. While local government officials frequently insist that the municipality needs to spearhead all the development programs at the local level, the severe crisis of the state has led to a situation where national and international NGOs have frequently filled the institutional vacuum and are de facto implementing many programs, such as in health, education or infrastructure which under normal circumstance would be carried out by the state. 12

Interview with Claudia Camacho, ICT program coordinator of ICO in Comarapa, May 23, 2005. 13 Interview with mayor of Comarapa, May 23, 2005.

275

The above-mentioned telecenter is a good example of a situation in which the local government lacks the institutional and technical capacity to implement the program and a nongovernmental organization brings its own expertise and resources to provide the population with a needed service. Another reason for the poor local governance structure in Comarapa is the role international donors frequently play in creating dependencies in both the public sector and civil society organizations at the local level. The Enlared Municipal program, for instance, had defined its priorities and operational methods through a highly donor-centric approach. In fact the program did not allow local actors to come together in order to adjust the program‘s planned activities to their particular social, political, and cultural context and to define a way on how to collaborate on its implementation in Comarapa. In fact, the provision of internet services and ICT training workshops would have lend itself to such a collaborative approach, since many organizations within the municipality had initially expressed their strong interest in sharing the connectivity services with each other. An important consequence of the common donor-driven approach is that it provides important disincentives for local actors to work together and to coordinate their activities with each other, since each of the organizations is primarily accountable to their international donor and not to local stakeholders. The example provided above—about the inability of the municipality to react to the proposal by ICO to share the costs of the Internet connectivity—demonstrates that programs like the Enlared Municipal project focus solely on achieving their centrally defined development objectives rather than on facilitating partnerships among the different local actors. In fact, the program worked exclusively together with the municipality and entirely neglected to reach out to the other organizations within the municipality. In this sense the Enlared program is exemplary of many other development programs in Bolivia in that it was unable to facilitate a partnership between the different local stakeholders that would have significantly enhanced the impact and sustainability of the program [52]. In the case of the Comarapa municipality, the local government officials even preferred to delay the municipality‘s Internet access for many months rather than develop a collaborative relationship with a local NGO. In sum, the Enlared program failed to enhance the local governance structure at the community level, since its program had an overly abstract national agenda and its centralized approach prevented it from affecting in any significant manner institutional changes at the local level. Furthermore, the program did not facilitate effective partnerships between different local actors and instead narrowly focused on trying to implement its pre-conceived objectives irrespective of the local socio-political context. In this sense the program represents the common shortcomings of similar donor-centric programs in Bolivia. Finally, the program failed because it lacked the necessary institutional credibility with local government officials based on the severe crisis of central government institutions. In brief, the program had no

significant impact on enhancing the inter-agency coordination between local government, civil society organizations, and local communities; neither did it enhance the relationship between local and central government. C. Impact on the Organizational Dimension The overall empirical results from the impact survey seem to indicate that the program had considerable positive effects on the organizational dimension. The data show that more than three-fourths of the program‘s participants (77%) rated the impact of the Internet on this dimension either as very significant (59%) or somewhat significant (18%). This section disaggregates these overall results in order to investigate the extent to which the Enlared project did indeed lead to significant organizational changes in local governments and thus had a positive impact on their institutional capability. This issue is critical for evaluating the overall impact of the program, since one of its core objectives was to enhance the efficiency and efficacy of local governments by introducing ICTs into local public administrations. First, the analysis uses the ―social capital index‖ in order to carry out a deeper analysis of the impact of the Internet on the organizational dimension. The graphic below highlights that while a two-thirds majority of participants (69%) perceived the Internet as critical to strengthening their horizontal networks with colleagues in other municipalities, only a minority of respondents (49%) believed that the Internet had a significant effect on strengthening the organizational capabilities of local government. The participants‘ skeptical views about the program‘s impact on organizational structures of local governments is also evidenced by the fact that more than one-third (35%) of respondents doubted that the Internet could play any role in this dimension. Furthermore, it is noteworthy that the large majority of respondents (64%), in spite of their skepticism about the Internet‘s ability to affect organizational change within government, had a very favorable view of the positive effects of the Internet on local organizations outside of government, such community-based organizations. This seems to indicate that the participants believed that there are major barriers to introducing ICTs into public administrations, while they consider this to be easier to be achieved outside of government. The more in-depth analysis of the data provides much richer insight into the effects of the program on the organizational dimension. In fact, the results demonstrate that the program‘s organizational impact was limited to establishing a strong social network among its participants, instead of instituting far-reaching organizational changes within local bureaucracies. Furthermore, the observed strengthening of social capital and horizontal communication networks was the result of ad hoc formation of informal networks between the program‘s participants, rather than a sign of improved institutional linkages between the different local municipalities.

276

strengthens horizontal

69

netw orks strengthens local

20

49

gov ernments

strengthens local

16

64

organizations 0

20 A lot

35

18

40

Percentage

60

Somew hat

11

18

80

100

None/NA

Fig. 4: ICT—a powerful tool for horizontal networking The above data also have to be seen in the context of the significant barriers the program participants faced in developing and promoting the use of information systems within local bureaucracies, because of local government officials‘ limited understanding of ICTs. In fact, a second critical indicator for the program‘s lack of organizational impact on local governments is the continuously low levels of Internet use in the large majority of municipalities. The following section draws on the empirical results from the project team‘s two user surveys14 carried out in October 2004 and September 2005 with local government officials in all of the fifteen participating municipalities. The analysis of this dataset adds significant value to the statistical analysis, since these surveys were carried out with a broad range of local government officials in each of the participating municipalities15, while the impact survey focused only the IT and communication specialists who directly participated in the program‘s training workshops. The first finding from the empirical evidence is that the program failed to institutionalize the use of the Internet in public administrative processes in local governments. As the graphic below illustrates, by September 2005—twelve months into the second phase of the project—only about one-third (36%) of the local government officials were using the Internet for work; although about two-thirds of respondents indicated that they had some experience in using the Internet, most continued to use the Internet for personal, rather than professional, communication. In fact, the survey results show that only one-third of the respondents who were Internet users (29%) used it at their workplace; most people (49%) used it in telecenters or Internet cafés. The failure of the program to promote the use of ICTs in local governments is confirmed by the fact that only 40% of government officials had a personal e-mail account (such as ―.yahoo‖) and none of the participants had an institutional email account (―.gov.bo‖). It is also noteworthy that the large

Internet Use for work by local government officials (user survey: n=190)

N/A, 41, 22% Yes, 69, 36%

No , 80, 42%

Yes

N/A

The percentage of respondents who used the Internet at work remained unchanged at about 36% in both years. The use of e-mail accounts actually decreased from 47% in 2004 to only 40% in 2005. This empirical evidence clearly demonstrates that the use of ICTs and the Internet continued to be low priorities in the fifteen municipalities, in spite of their nominal participation in the Enlared Municipal project. One reason for this low level of ICT use is the extremely high turnover rate of local government officials. The municipal elections in December 2004, four months into the project implementation, resulted in profound political changes in a large majority of local governments and negatively affected the project. Internet use by position in municipality 90 80 70 60 50 40 30 20 10 0

82.2

7.1

10.7

Mayor/Council member

66.9 41.8 32.7 25.5

Technical specialist

14

Both of these user surveys focused exclusively on issues related to the awareness, readiness and use of ICTs in local governments (Enlared project files). 15 The user survey from September 2004 was based on a sample of 190 respondents and the second survey from September 2005 on a sample of 186 participants.

No

Fig. 5: Limited Internet use within local governments

percentage

Enlared: Impact on Social Capital

majority (62%) of the respondents who had an e-mail account used it only once every three days. Moreover, the graphic below illustrates the resistance or apathy of key decision-makers to these technologies. In fact, the large majority of mayors and/or council members (82%) did not consider the Internet important and thus did not use it for their work. The data indicate that only among technical staff did the majority (58.2%) use the Internet for their work. Technical staff also demonstrated higher rates of e-mail use: 54% of them had personal e-mail accounts, compared to just 18% of mayors and council members. The very low local appropriation of ICTs in local government can be confirmed by comparing the results from the baseline study from October 2004 with those from the user survey from September 2005 described above.

14.818.3

Member of oversight committee

63.8

20.5 15.7

Total

Position A lot

Somew hat

None/NA

Fig. 6: Government officials‘ skepticism about Internet use

277

In fact, 68% of the respondents from the 2005 survey indicated that they had been working for less than one year in the municipality and just eight of the training participants, out of a total of ninety-four, had participated in a previous training workshop during the first phase of the project. The principal finding about the marginal role ICTs play within local governments is also confirmed by the fact that only technical staff (i.e., IT specialists and communication officers) participated in the ICT training workshops organized by the Enlared program, and that key decision-makers (i.e., senior administrative staff and/or council members) did not show any interest in participating. In sum, the program failed to reach one of its main objectives—to promote the use of ICTs in local public administrations in order to strengthen the organizational capabilities of local government. Why did the project fail to promote organizational changes? D. IMPACT ON THE SOCIAL DIMENSION OF WELL-BEING The empirical findings reveal that the Enlared program had no substantial impact on the social dimension of people‘s well-being. In fact, as the graphic above illustrates, only onethird (31%) of its participants indicated that the Internet has an significant impact on their human capabilities in the social dimension. This finding is striking, since the social dimension was the one dimension in which participants in the overall survey had the most positive views, with 60% believing that the Internet had a major impact on their lives. These data need to be seen in the context of the third major objective of the Enlared program, which aimed to improve the performance of local governments in the delivery of social services by strengthening their institutional capabilities. This aspect of municipal development issues is particularly important for indigenous peoples, since the large majority of indigenous communities lacks access to basic social services, such as water and sanitation, education and health. Furthermore, one of the principal objectives of the entire decentralization program in Bolivia was to substantially improve the delivery of social services to rural, marginalized communities. In fact, a critical argument in favor of the LPP has been that local municipalities are supposed to have a much closer relationship to communities due to their geographic proximity. The literature, however, indicates that this major objective of the LPP was not realized due to capture of local municipal planning processes through local elites (Blackburn, 2000). The empirical findings from the impact survey seem to confirm this skeptical view in the sense that the participants expressed very critical opinions about the role ICTs can play in improving the delivery of social services to indigenous communities. While at first sight this general finding seems unforeseen, it is fully consistent with the failure of the program to significantly enhance the organizational capabilities of local government. In other words, since the program did not succeed in strengthening the institutional capacities of municipalities it is to be expected that it also failed to have any positive impact on the delivery of social services.

Enlared: Impact on Social Services Improv ed access to

38

education 14

Improv ed access to health Improv ed gov ernment

13

serv ices 0

37

25

50

38

70

17 10

20

30

A lot

40

50

%

Somewhat

60

70

80

90

100

None

Fig. 7: Internet‘s limited role in improving service delivery V. CONCLUSION This case study demonstrates that the Enlared Municipal program failed to significantly impact the accountability, transparency, and performance of local governments. The reasons for this failure were both internal and external. The main internal factors were (i) the overly centralized approach taken in the implementation of the project, compounded with the project‘s assumption that the regional municipal associations would be an effective intermediary in reaching local municipalities which turned out to be false; (ii) the political nature of the Federation of Municipal Associations (FAM), which forced the project to engage with local governments that were unwilling and unprepared to participate in such a program; (iii) the overemphasis on the technical development and maintenance of a Portal on municipal development; and (iv) the lack of adequate resources and staffing to carry out a more systematic and long-term ICT capacity-building program at the local level. In addition to these internal factors, there were several key external factors: (i) the overall political instability in Bolivia, beginning in October 2003, which involved the dissolution of two governments in one year alone (2005) and which provoked disruptive forms of civic unrest; (ii) the lack of a coherent and comprehensive national strategy for developing a telecommunications and ICT infrastructure in Bolivia, particularly in rural areas, meaning that ICT projects were implemented in technically unprepared regions; and (iii) the limited institutional and technical capacity of many local governments to develop public policies and to implement development projects. In sum, the program was too centralized, which led to a critical design-reality gap that undermined its ability to enhance local governments‘ transparency, accountability, and performance. The program was never integrated into broader issues related to local governance and did not have any significant impact on its participants‘ human and social capabilities. In order to more significantly affect indigenous peoples‘ well-being, it would have to be much more precisely tailored to the specific socio-economic, political, and cultural realities of local governments and indigenous communities.

278

Bibliography [1] J. Samoff, The bureaucracy and the bourgeoisie: Decentralization and class structure in Tanzania. Comparative Study of Society and History 21 (1), pp.3062, 1979. [2] W. Oyugi, ―Decentralization for good governance and development: The unending debate‖. Regional Development Dialogue, 21 (1): 3-22. [3] J. P. Faguet, ―Decentralization and Local Government in Bolivia: An overview from the Bottom-up‖, Crisis State Programme, Working Paper No. 29, 2003. [4] World Bank, The state in a changing world: World Development report, 1997. New York: Oxford University Press. [5] Booth at al., Empowering the poor through institutional reform? An initial appraisal of the Bolivian experience. Stockholm: Department of Social Anthropology, Stockholm University. 1996. [6] B. Kohl, ―Democratizing Decentralization in Bolivia— The Law of Popular Participation Journal of Planning Education and Research 23, pp. 153-164, 2003. [7] H. Blair, ―Participation and accountability at the periphery: Democratic local governance in six countries‖, World Development 28 (1), pp.21-39, 2000. [8] J. Blackburn, Popular participation in a prebendal society : a case study of participatory municipal planning in Sucre, Bolivia, Thesis (Ph.D.)--University of Sussex, 2000 [9] Godsmith and Blackburn, 2001) Participatory Municipal Planning in Bolivia: an ambiguous experience, in: Development in Practice, Volume 11, Number 5, 1 November 2001, pp. 587-596 (10). [10] R. Heeks, Most e-Government-for-Development Projects Fail How Can Risks be Reduced? Institute for Development Policy and Management (IDPM), iGovernment Working Paper Series, Paper no. 14, 2003. [11] T. Yigitcanlar, ―Bridging the Gap between Citizens and Local Authorities via e-government‖. in Symposium on E-government, 10–12 May 2003, Muscat, Oman. [12] C. Ciborra and D. Navarra, "Good Governance, Development Theory and Aid Policy: Risks and Challenges of E-Government in Jordan." Journal of Information Technology for International Development 11(2), 2005. [13] S. Bhatnager; ―Egovernment: Lessons from Implementation in Developing Countries‖, Regional Development Dialogue, Vol. 24, UNCRD, Autumn Issue pp. 164-174, 2002. [14] S. Krishna and G. Walsham, ―Implementing Public Information Systems in Developing Countries: Learning From a Success Story‖, Information Technology for Development, 2005. Vol. 11 (2), pp. 123140. [15] N. Negoroponte, Being Digital. New York, Vintage, 1995. [16] Wilson III, Development of National Information and

Communications Services: A Comparison of Malaysia and South Africa. Journal of Developing Societies, vol. 15, pp.47-60, 1999 [17] M. Gasco, New Technologies and Institutional Change in Public Administration, Social Science Computer Review, 21, 1, 6-14, 2003. [18] R. Heeks, 1999; Information and communication technologies, poverty and development, University of Manchester. Institute for Development Policy and Management. Working Paper Paper No. 5,1999. [19] W‘O Okot-Uma, R, Electronic Governance: Reinventing Good Governance, 2001, World Bank. [20] OECD, Engaging Citizens in Policy-Making: Information, Consultation and Policy Participation, Puma Policy Brief No. 10, 2001. [21] C. Ciborra, Unveiling E-Government and Development: Governing at a distance in the new war. Department of Information Systems, Working Paper Series. London, London School of Economics and Political Science.2005; [22] R. Heeks and S. Bhatnagar, 'Understanding success and failure in information age reform', in Heeks, R. (ed.) Reinventing Government in the Information Age – International Practice in IT-enabled Public Sector Reform, London and New York: Routledge, pp. 4974, 1999. [23] R. Heeks, Most e-Government-for-Development Projects Fail How Can Risks be Reduced? Institute for Development Policy and Management (IDPM), iGovernment Working Paper Series, Paper no. 14, 2003. [24] ibid. [25] ibid. [26] S. Madon, (2004). ―Evaluating the Developmental Impact of e-Governance Initiatives: An Exploratory Framework.‖ Electronic Journal of Information Systems in Developing Countries 20(5): 1–13.adon, 2004:2 [27] ibid. p.9. [28] Moon, The evolution of e-government among municipalities: rhetoric or reality‖, Public Administration Review, Vol. 62 No. 4, pp. 424-33. 2002; [29] J.C. Musso, Weare and H. Hale, ―Designing Web Technologies for local Governance Reform: Good Management or Good Democracy?‖ Political Communication, vol. 17, no. 1, pp.1-19, 2000. [30] L. Pratchett, New Technologies and the Modernization of Local Government: An Analysis of Biases and Constraints‖ in: Public Administration, Vol. 77, no.4, p.731-750.1999 [31] ibid. [32] ibid [33] Moon, The evolution of e-government among municipalities: rhetoric or reality‖, Public Administration Review, Vol. 62 No. 4, pp. 424-33. 2002 [35] W. J. Tettley, ICT, Local Government Capacity Building, and Civic Engagement: An Evaluation of the Sample Initiative in Ghana. In: Perspectives on Global Development and Technology, Vol. 1, no. 2d, pp-165-

279

192, 2002 [36] J.C. Mussa et al, 2000. [37] C. Ciborra (2005). [38] C. Ciborra and D. Navarra (2005). [39] S. Madon ―Introducing administrative reform through the application of computer-based information systems: a case study in India.‖ Public Administration and Development 13: 37-48, 1993. [40] S. Madon, (2004) [41] G. Walsham, Interpreting Information Systems in Organizations. Chichester: John Wiley & Sons, 1993. [42] R. Heeks, (2003) [43] Pettigrew, A.M., (1985) 'Contextual Research: A Natural Way To Link Theory And Practice', in E.E. Lawler (ed.), Doing Research that is Useful in Theory and Practice, San Francisco: Jossey Bass, [44] Pettigrew, A., Success and Failure in Corporate Transformation Initiatives, In: Galliers, R.D., Baets, W.R., Information Technology and Organizational Transformation, Chichester, Wiley, 1998. [45] ICMA Final Enlared project report, ICMA 2003. [46] Enlared Municipal Consultation Report, Enlared project files, 2004. [47] C. Roche, Roche, Impact assessment for development agencies: learning to value change. Oxford: Oxfam GB, 1999. [48] L. Pratchett, New Technologies and the Modernization of Local Government: An Analysis of Biases and Constraints‖ in: Public Administration, Vol. 77, no.4, p.731-750.1999 [49] Gigler, B., Enacting and Interpreting Technology - From Usage to Well-Bring: Experience of Indigenous Peoples with ICTs. In H. Rahman (Ed.), Empowering Marginal Communities with Information Networking (pp. 124-164). London: Idea, Group Publishing, 2006. [50] L. Pratchett, 1999: p. 735. [51] J. Tendler. Good governance in the tropics. Baltimore: Johns Hopkins University Press; 1997:145 [52] Van Niekerk, Desarrollo rural en los Andes, Un studio sobre los programas de desarrollo de Organizaciones no governamentales, Leiden Development Studies, 1994.

280

Design and Deployment of a Blood Safety Monitoring Tool S. Thomas, A. Osuntogun, J. Pitman, B. Mulenga, and S. Vempala Abstract—Blood is a scarce resource critical to the management of a variety of life-threatening medical conditions. It can also be a medium for the transmission of infections, including HIV. Monitoring the quality and quantity of available blood, essential to utilizing it as effectively as possible, has been a challenge in many developing countries. This paper describes the design and implementation of a web-based tool to monitor the collection, screening and distribution of blood in developing countries. This project was conducted under the auspices of the US President's Emergency Plan for AIDS Relief (PEPFAR), which funds blood safety projects in 14 countries in Africa and the Caribbean. We report results from a usability study, formulate relevant design principles and discuss prospects for long-term sustainability. Index Terms—developing nations, health, ICT and development, Internet, medical services

I. INTRODUCTION Blood is a scarce resource worldwide. But the scarcity is particularly acute in the developing world. In 2008, the World Health Organization estimated that although 80% of the world's population lives in developing countries, these countries only collect 45% of the global blood supply [1]. Effectively managing this scarcity—deciding who gets blood and when—is a challenge for health care systems throughout the developing world. Blood is also a highly effective medium for the transmission of blood borne illnesses, including HIV, hepatitis and syphilis. The WHO estimates that between 95-10% of all HIV infections worldwide may be linked to unsafe blood transfusions. In 2004, the US President's Emergency Plan for AIDS Relief (PEPFAR), identified blood safety as a key element in its comprehensive HIV prevention strategy. Since then, PEPFAR has committed $192 million in funding to National Blood Transfusion Services (NBTS) in 14 countries (Botswana, Cote d'Ivoire, Ethiopia, Guyana, Haiti, Kenya, Mozambique, Namibia, Nigeria, Rwanda, South Africa, Tanzania, Uganda, and Zambia). To track, monitor and evaluate the impact of this funding, the Centers for Disease Control and Prevention (CDC) worked with global blood safety experts to develop a set of program indicators and a Microsoft Excel-based data collection tool. In this paper, we describe a web-based data collection and management tool S. Thomas, A. Osuntogun, and S. Vempala are with the Georgia Institute of Technology, Atlanta, GA 30332, supported in part by a Raytheon Fellowship and the NSF. J. Pitman is with the Centers for Disease Control and Prevention, Global AIDS Program, Atlanta, GA 30333 B. Mulenga is with the Zambian National Blood Transfusion Service, Lusaka, Zambia

developed as a collaboration between CDC, the Georgia Institute of Technology and the participating countries. The tool expands on the Excel-based system to facilitate the efficient and accurate collection and analysis of more than 80 indicators from 14 countries. II. RELATED WORK Several ICT projects have focused on improving the health care system in developing countries. These projects have had several different themes including development of web-based tools for information tracking, implementations of electronic medical record systems and telemedicine systems to support education of clinicians and referral of difficult cases for further review. In [2] a web based information system supported medical record keeping and tracking of HIV treatment in Haiti. While this work is similar to our work in that there is a centralized database and an online interface for certified users to enter and track data, the population and scale of users is quite different. In our work we focus on users in 14 different countries worldwide with varying levels of internet access and technical support. We also present evaluation of the system by prospective users which indicated that the system was easy to adapt to and use. Blaya et al in [3] created a web based tool, e-Chasqui, to improve the timeliness and quality of reporting tuberculosis laboratory data in Peru. This national system is able to provide access to a hierarchy of users in health centers, regional laboratories and the national laboratory to test results. These results are then used to prescribe treatment regimens for patients at the local level. In analyzing the use of web-based tools, Marquez found that such tools provide notable enhancements in quality improvements in health care [4]. The web-based tool described in this paper is an example of such tools. III. PROBLEM AND BACKGROUND Monitoring the development of a national blood service requires data from a diverse set of sources. The WHO, in its Aide Memoire for National Blood Programmes [5], advises developing countries to build national blood strategies around four "key areas", including: The establishment of a nationallycoordinated blood transfusion service; the collection of blood only from voluntary non-remunerated blood donors; universal testing of all donated blood, including screening for transfusion-transmissible infections, blood grouping and

281

compatibility testing; and the reduction of unnecessary transfusions. These broad recommendations can be further categorized by technical area: 1) Policy information is needed to track progress in standardizing procedures (e.g., laboratory testing) and establishing a blood service's legal authority over the national blood supply, 2) Social and demographic information is needed to identify, recruit and retain blood donors with low behavioral risk profiles for infection with HIV and other transfusiontransmissible infections (TTI), 3) Scientific data are necessary to track the prevalence of infectious markers in donated blood and to monitor laboratory performance, 4) Administrative data are needed to measure human resource and logistical capacity, and, in many cases, to manage large grants from bilateral or multilateral donors, and 5) Medical information is critical to ensure that blood is used appropriately and to monitor patient outcomes. A blood service could potentially collect data on hundreds of variables. With PEPFAR support, CDC and its international partners developed a limited, but comprehensive, set of programmatic indicators. These indicators are broken down into a subset of variables which cover the five technical categories described above. Collecting these data from 14 developing countries with different data systems was the principal challenge associated with this project. To begin to address this problem, CDC developed a standardized data collection tool using Microsoft Excel. This format was selected for its ease of use (data managers in all of the PEPFAR countries were familiar with Excel) and its ability to be transferred via e-mail over low bandwidth internet connections. A version of this tool, with pre-programmed formulae and skip patterns, was field tested in 2007. While countries reported the tool was easy to use and met the programs' PEPFAR-related reporting needs, the pilot phase also revealed several weaknesses in the use of a spreadsheet-based tool, including: 1) Difficulty in tracking versions of multiple files in circulation as new data were entered and re-saved by different users, 2) Countries' inability to quickly modify, clean or correct a data set after a file was submitted to CDC, 3) The potential for transcription and other errors as multiple versions of the spreadsheet were merged (e.g., from multiple regional centers) prior to submission to CDC. 4) PEPFAR requires quarterly reports from each country, but blood is collected, tested and utilized continuously. This meant the Excel-based tool could sit idle for weeks or months at a time.

These challenges increased the possibility of inaccurate, or out-of-date, reporting. To illustrate, in several countries, regional or provincial blood service offices would would fill out and email the Excel-based tool to the national manager, or save the report on a disk or flash drive and send the disk back to the national manager by road. The national office would then compile all the results from the different provinces and send a final report to the CDC. Each update from a regional center required re-entry and re-aggregation of the data, leading to multiple versions and increasing the chance of error. To address the weaknesses of the Excel-based tool, the team developed an Internet-hosted database accessible with a standard Web browser. Storing the data in a Web-based relational database provides several key benefits compared to ad hoc circulation of a spreadsheet file: 1) There is a reliable, authoritative source of reported data. (No more wondering which file among several is most accurate.) 2) Data is continuously available throughout the reporting period allowing immediate modifications and corrections. (No files "in transit" or lost via email.) 3) The system can support automated, real-time aggregation of reported data from multiple sources with less risk of transcription errors. 4) Full change tracking with audit logs is available. Additional benefits derive from the system's use of a standard web browser interface: 5) Access is ubiquitous and available on all modern computing platforms. 6) End users are familiar with web browsers (e.g., Internet Explorer, Firefox, Safari). 7) Updates and enhancements can be easily managed and deployed. The specific design principles for the system follow directly from its requirements: 1) [Network compatibility] The system must support access through low bandwidth, dial-up connections, as high bandwidth connections cannot be assumed for all users. 2) [Intuitive interface] The system must provide a user experience appropriate for blood safety staff who may be familiar with the use of standard office software (e.g., web browsers, spreadsheets, word processing), but who lack high-level IT training. 3) [Security] The system must provide appropriate security and access control for aggregated health information. 4) [Flexible and adaptable] the system must be easy to manage, adapt, and expand. For example, it must be easy for administrators (many of whom will not be trained IT managers) to add new countries, regions or users, and ensure that data entered by these new sources are properly aggregated. (Based on previous experience, the CDC did not feel the need to support languages other than English

282

in the initial deployment. The site was designed, however, to accommodate localization to alternate languages should that be desirable in the future.)

users are able to view as much or as little of the information as they wish, all on one page, as Figure 1 highlights.

The tool was designed via collaboration between CDC, Georgia Tech, and NBTS data managers in the 14 PEPFAR countries. The Zambian National Blood Transfusion Service (ZNBTS) agreed to host the pilot phase described in this paper. This partnership was made possible through a program at Georgia Tech which seeks to leverage student and faculty expertise to address social and development problems in lowincome settings. Funding for logistics during the pilot project was provided, in part, by PEPFAR and the ZNBTS. The Georgia Tech team's time and expertise were donated gratis as part of a course. After multiple rounds of testing, user feedback and usability studies (described later), it was evident that the web tool was highly welcomed by NBTS data managers in Zambia and elsewhere. Their feedback led to changes both in the user interface and in the architecture of the system. Further, data managers expressed a strong desire to use the system not just for reporting but also as a decision informing tool. We redesigned the system to allow individual countries and individual centers within countries to choose the frequency of data entry (to match the frequency at which data was actually being aggregated physically), and to choose the type of reports and views of historic data that they desired. This was all done while maintaining a uniform quarterly PEPFAR report for all participating countries. IV. INITIAL DESIGN Initial prototypes for the system used common web application techniques and technologies. MySQL database tables were defined to collect and store reported information. Scripts written in PHP provided a web-based front end to the database, allowing users to enter data and view collected information. The web-based user interface also provided a means for system administration, including defining regions and countries, managing users, and supporting data backups. A key aspect of the user interface design was its reliance on unobtrusive enhancement[6]. This strategy creates a layered user interface, where successive layers provide more functionality, but are only invoked if the user (in particular, the user's web browser) supports the required technologies. For example, many of the input pages consist of a number of individual data elements. To make it easy for the user to focus on the information in manageable pieces, the interface provides selective disclosure--information not relevant to the immediate task is hidden or de-emphasized so that it does not distract the user. This approach has significant benefits to the user compared to alternative strategies (such as making each section a separate page or screen), as users are able to decide for themselves what information is relevant to the task. For example, a user may wish to refer to other inputs when supplying a particular value. If the other inputs exist on separate pages, then the user would be forced to jump back and forth between multiple pages. With selective disclosure,

Fig. 1. Selective disclosure allows the user to show only those sections of the page that are relevant for the task at hand.

Although it is theoretically possible to implement selective disclosure strictly through PHP scripts executing on the web server, most users would find the resulting lag between their action and the interface's response to be intolerable. For a suitably responsive interface, selective disclosure is implemented solely on the web browser client, using javascript to show or hide the appropriate interface elements. However, not all web browsers can fully support javascript, and some users may choose to disable javascript even when their browser supports it. To accommodate these users, unobtrusive enhancement in the initial design adds selective disclosure as an optional, additional layer in the interface. The prototype accomplishes this by first creating a page with all interface elements available and disclosed. Figure 2 shows a screen capture for such a page.

Fig. 2. When the page loads initially, all sections are fully shown. If the browser does not support javascript, the page remains displayed in this form.

The interface then attempts to execute javascript to add selective disclosure. If the web browser does not support javascript (or if the user has disabled that option), then the page remains as in Figure 2. Although there is no selective disclosure, the interface remains fully functional. When

283

javascript is available, however, selective disclosure immediately activates and the new functionality is available to the user. Figure 3 shows a screen capture for the same page after the javascript executes. Note that in most cases the javascript execution is fast enough that the user will never even notice the initial interface of Figure 2. For most users, the only interface of which they will be aware is the interface of Figure 3.

Fig. 3. If javascript is able to execute in the user's browser, it activates selective disclosure and enables the user to hide or reveal sections independently.

This design illustrates the benefits of unobtrusive enhancement. Users without the more advanced technology (in this case, browser support for javascript) still have access to a serviceable interface. The majority of users who do have the necessary technology, however, receive a more advanced user interface. V. EVALUATION A. Methodology Prior to the implementation of the web-based system, there were several meetings between the CDC and the team from Georgia Tech. These meetings helped determine the specifications of the system including the design of the userinterface, the preferred programming language and hosting and management of the website. To evaluate the effectiveness of the tool, we conducted a series of studies involving the end-users. We selected a subset of the PEPFAR countries to use for evaluation. All the data managers in the National Blood Transfusion System in the PEPFAR countries are currently expected to use the Excel based reporting system and therefore already had some understanding of what kinds of data would be requested. Zambia was chosen as the lead country for deployment and evaluation due to the progress in using the Excel based tool and their overall blood safety effort. The team from Georgia Tech spent approximately two weeks, July 27 to August 8, 2008, in Zambia studying the local setting, usability of the system, network characteristics and gathering qualitative feedback. The usability study component was conducted in the

national blood center in Lusaka, Zambia as well as in Kitwe, the center for the Copperbelt region. While in Zambia, the team also conducted the study with data managers in Dar-esSalaam, Tanzania, over the telephone. We next describe each of the three test sites in more detail. Zambia National Center, Lusaka. The evaluation conducted at the ZNBTS consisted of 6 participants in total, some were assigned tasks as country managers while others were assigned tasks for provincial managers. The participants consisted of a physician, biomedical scientists, financial manager, information technology manager and data managers. The participants were given the web address to the web tool and were asked to complete the tasks. Two observers from Georgia Tech were available to administer the evaluation and provide help only if absolutely necessary. None of the participants had prior experience with the website so they were allowed to navigate around the website to determine how to use the features. Internet connectivity available was via satellite and evaluators used laptops with wireless connections. Zambia Copperbelt Center, Kitwe. In Kitwe, there was dial-up internet connectivity, although it had not been functioning for some months. After several attempts during the first hour of our visit, the problem was discovered and fixed, making it possible for the evaluation to happen. There were 3 participants consisting of a physician and two data managers who were all familiar with the Excel-based tool. The dial-up connection had to be reset twice during the entire evaluation. Tanzania National Center, Dar-es-Salaam. The usability study for the Tanzania data managers was conducted over the phone. The observers called in from Lusaka, Zambia and went through the tasks on the task list with each participant individually; two participants had a copy of the task list by email and were asked to indicate by speaking aloud how they were navigating the system. Fiber-based internet connectivity was available in Tanzania which made for significantly faster navigation. B. Study Design and Analysis The usability study consisted of 3 components: a pre-trial survey, an observed trial where data managers were asked to perform a set of tasks and a post-trial survey. Pre-trial survey. A survey in form of a questionnaire was conducted prior to the trial of the web-based tool to assess the expectations of the levels of ease or difficulty of using the web-based tool. The survey was conducted via the web by using a commercial web-survey tool that provided information about the date the survey was taken, the IP Address and aggregate results from all participants etc. By conducting a web-survey, we were also able to assess the response time and the availability of internet access. There were 9 participants in the pre-trial survey, all from the Zambia NBTS; 7 from the Lusaka and 2 from Kitwe. The pre-trial survey link was sent out by July 21, 2008 and all surveys except one were taken between July 22 and July 27. 2008, prior to observers’ arrival in Zambia. All the participants had some prior knowledge of the Excel-based tool. The questions in the survey are shown in

284

Table I. Each participant was asked to rate the ease of each task on a Likert scale of 1-5 when appropriate, with 1=very easy, 2=easy, 3=neither easy nor difficult, 4=difficult and 5=very difficult. Questions 4 and 5 were given out to aid in qualitative analysis. Table I. Pre-trial Questions 1 You have forgotten your password and need to log into the web-based system. How hard or easy do you expect to find this task? 2a You are only responsible for reporting some of the data, while other users report the remaining data. You need to enter the data for which you are responsible without disturbing the other data. How hard or easy do you expect to find this task? 2b At the close of a deadline, you have only partial information to report. How hard or easy do you expect to find this task? 3 Previously entered data is wrong, and you must provide the correct information. How hard or easy do you expect to find this task? 4 What are some of the advantages you expect from a web-based system compared to the old spreadsheet tool? 5 What are some of the disadvantages you expect from a web-based system compared to the old spreadsheet tool? Trial and post-trial survey. Discount usability techniques were used to evaluate the usability of the system and the interfaces. This evaluation provided a means of obtaining insights about usability problems and possible solutions. A task list was used to obtain quantitative and qualitative measurements of the ease of use of the tool. When participants completed all tasks, they were asked to rate the ease of use of the system on the same rating scale as in the pre-survey. This evaluation occurred immediately after each participant completed all tasks, during the observers visit to Zambia. The list of tasks to be completed by each evaluator is shown below in Table II. Table II: Evaluation Task List 1 Your username is country/[email protected] and the password is the word "password''. Use the system to change the password to "country/province''. 2a Enter the regional information for the current quarter. 2b You are entering data for the current quarter for your country/province. You only have data for all the nonnumeric fields. Enter and save this part of the data. 3 There is an error in the HIV prevalence field. Please correct it. 4 Export the country/province data to an XML file and view the data using Excel. C. Results Analysis of the results obtained from this evaluation, shown below in table I and figure 4, revealed that most evaluators rated most of the tasks between scale level 1 and 2. These

results indicate that user expectations of the difficulty of using the web based tool were above the actual difficulty encountered while using the tool. There was a reasonable level of ease while working with the web-based tool. Participants were also able to find some typographical errors and provide concrete suggestions for changes. The overall survey results indicate that the system was not very difficult to adapt to and use. Figure 4 shows the average difficulty levels pre- and post-trial for each site. From Figure 4, it is apparent that the difficulty experienced by users in Zambia was significantly lower than what they anticipated. Pretest

Tanzania

Zambia

3.0 2.4 1.8 1.2 0.6 0

Task 1

Task 2

Task 3

Fig 4. Pre- and post-evaluation comparisons show tasks were as or easier than expected.

The results from Tanzania, show less ease in using the system; this effect may be attributed to the additional difficulty of having to evaluate the system while on the phone with the observers. Also, the evaluators may have felt some pressure to accomplish the tasks quickly because the observers were waiting for them to complete each task while holding on the phone line. In table III, the average difficulty for each task is given. Overall, the pre-trial results are higher than the post-evaluation results, this supports the notion that the interface was welldesigned, to the extent that such a self-evaluation can assert. The "Export" task, not mentioned in the pre-trial survey was perceived to be more difficult than the other tasks; it involved transferring data from the system to the local computer in a standard format (XML) which may have been unfamiliar to many evaluators. Table III. Mean Scores for Selected Tasks Task

Pre

TZ

ZM

ZM-NTL

ZM-CB

1

2.56

2

1

1

1

2

2.22

2.5

1.17

1.33

1

3

2

2

1.29

1.67

1

Export

-

3

2.5

3

2

285

D. Qualitative findings Utility and maintenance. Respondents to the pre-trial survey felt that the web-tool would: • Result in more timely reporting. • Be quicker and less tedious. • Be more cost-effective. • Reduce the number of errors. • Be more user-friendly than a spreadsheet. • Permit easier correction of data entry errors. • Make it easy to share data between users. All respondents were strongly of the opinion that a webbased tool would be considerably superior to a spreadsheetbased report. Security. On one hand, some users felt data security is a concern since the data is no longer locally stored and vulnerable to malicious attacks; on the other hand, keeping data on a server insures against data loss in the case of problems with local computers. In any case, it was clear that users were concerned about how secure the data would be. Functionality. While the original purpose of the tool was to allow for data aggregation for the purpose of generating PEPFAR-mandated reports, data managers were quick to see that having a real-time aggregate picture would greatly enhance their ability to manage blood collection and utilization. They asked for data analysis, such as historical trends and comparisons of data from different regions to be included in the functionality of the system. Network constraints. Given the state of network infrastructure in Zambia and other participating countries, the potential pitfall of relying heavily on the internet was noted by several data managers. Would pages load too slowly? Would the system be able to handle connectivity going up and down? While Dar-es-Salaam has a fiber-based internet connection, there is no fiber connecting Zambia to the rest of the world. Thus, all internet traffic goes via satellite, imposing a strong constraint on bandwidth, and perhaps even more significantly, on latency. Indeed we did not notice much difference in response times between Kitwe, with a dial-up connection, and Lusaka, with wireless connectivity and a dedicated satellite connection. IV. REVISED DESIGN Usability testing of the initial prototype indicated that, in general, the application and its interface were suitably easy to use. None of the tasks were rated as difficult or very difficult by any user. Comments made by participants during the study, however, indicated that one of the most unsatisfying aspects of the system was the time required for the initial loading of each web page. The main cause of the lengthy delay was the quality 1

of the network infrastructure1 . Because the same infrastructure was used for all Internet access and, thus, lengthy page load times were common for nearly all web sites, participants were somewhat conditioned to expect this delay. Nonetheless, their dissatisfaction levels suggested that it would be worthwhile to investigate opportunities to reduce the page delay. The revised design significantly reduces the perceived initial loading time through the use of Asynchronous Javascript and XML (AJAX) technology[7]. With AJAX, a web page can continue to interact with the server after it is initially loaded without requiring explicit action on the part of the user. By using AJAX, the initial contents for each page can be substantially reduced so that the page loads more quickly. While the user views or interacts with the page, it continues to retrieve additional information from the server. By combining AJAX functionality with selective disclosure, the revised design provides quicker initial page loading in a manner that is seamless for users. Figures 5 and 6 illustrate the steps involved in loading a page using the approach of the revised design. The server first supplies an initial web page that is almost completely empty. However, the individual sections on the page are marked as hidden. Even though they contain no content, users do not perceive a problem because the sections are hidden.

Fig. 5. The web server initially supplies only the skeletal framework for the full page; however, since the framework contains minimal content the network can transfer it very rapidly and present the user with the appearance of a full interface.

As soon as the initial page has loaded, it begins using AJAX requests to retrieve the actual content for each section. During this time, however, the user is still viewing the initial page and deciding how to proceed. In many cases, by the time users take their first action by selecting to show one of the sections on the page, the content will have been retrieved from the server and can be displayed immediately. When that is not the case, a progress bar tells the users that the interface is acting on their requests.

While on site in Zambia, the developers were able to make measurements of network bandwidth and latency from several locations over 24hour periods. The tests measured performance from a local computer in Zambia to the prototype web site operating by a commercial web hosting company in New York. The tests showed available bandwidth ranging from 12 to 57 kb/s to New York and 88 to 116 kb/s from New York. Mean latency ranged from 5.1 s to 6.8 s.

286

use. Because the database captures all inputs to the system as new data insertions (rather than updates to existing data), there is no need to lock database access to avoid concurrent updates. Such updates are problematic for any Internet-based application, and would be especially challenging in an environment with sporadic connectivity. The system's approach resolves conflicts during report generation rather than database modification. These features ensure that the system meets the needs of both the sponsoring agency and the local health officials. V. DEPLOYMENT TIMELINE Fig. 6. While the user decides which section to view, the network is able to complete the transfer of the hidden content. When the user choses to reveal that content, it can be immediately displayed.

The timeline in figure 7 compares the perceived and actual page loading times for the initial and revised designs. Even though it actually takes longer for the entire page and all its content to load in the revised design, the perceived load time is less.

An initial prototype for the blood safety system was created during the spring of 2008. Field trials and evaluations took place during the summer, and the revised design was created in the fall. Beginning in the first quarter of 2009, the system will transition into production in 14 countries which are part of the PEPFAR program. Discussions are currently underway with the World Health Organization to deploy a version of the system worldwide.

non-AJAX design

VI. DISCUSSION AJAX design

time t1

t2

t3

initial page framework can be displayed; user perceives page loading as complete

full page is loaded; user perceives page loading as complete

hidden elements are transferred; page loading is fully complete

Fig. 7. Although the AJAX design requires longer to load because of extra content, it allows the web browser to present a functioning page sooner than the non-AJAX design; to the user, this difference gives the perception of faster loading to the AJAX design.

One consequence of the revised design is that it requires javascript. There is no way to make the page fully functional when the user's browser does not have javascript support, so unobtrusive enhancement is not possible. Based on interactions with prospective users, however, it was determined that improving page loading times was more desirable than supporting non-javascript capable browsers. In the future, should support for non-javascript browsers be necessary, a separate user interface could be developed. Another key element of the revised design directly incorporated requests from the blood safety staff in the remote locations. In many cases, local staff wished to collect and track information in addition to that required by the CDC. Also, reporting periods were sometimes more frequent than the quarterly reporting required by the CDC. By supporting collection of local information at locally-defined reporting periods, the system provides a much greater incentive for its

Overall survey results indicate that a web-based system was highly desirable and the system developed here was easy to adapt to and use. This finding is in line with other experiences with web-based tools for health projects in the developing world[8]. A very significant observation from our survey is the need to incorporate local users in the preliminary design phases of a project. The original goal of the system was to serve as a data collection tool for CDC data managers in Atlanta; while conducting surveys, we discovered a gap in the stakeholder perspectives as the local users perspective had not previously been incorporated in the design. This highlights the necessity for human-driven design and research practices as noted by Brand & Schwittay [9]. Local data managers stressed their preference for a tool that would also help in blood management by analyzing and tracking recorded data. The final design incorporates perspectives of both stakeholders; the CDC data mangers and local users. This reflects several dimensions of Information and Communications Technology for Development (ICT4D) as outlined in work by Tongia et al. [10] by addressing the original goals of the CDC’s Excelbased data entry system while also meeting the needs of the local users in the developing countries. We highlight two aspects of our design that give us confidence in its long-term sustainability and might serve as a model for similar projects in the future. Data flow constraint. The data entry requirements of the system should match the data collection, i.e., if the system expects too much data, it will be hard to maintain and impose an overhead on data collection. In technologically advanced countries, there is often an abundance of easily available data and thus it is possible to deploy systems that use vast amounts

287

of data collected rapidly. For a system to be sustainable, however, its data requirements should match the local data flow, even if this necessitates a less complex system. Flexible design. The second most important aspect of our system design, closely related to the first, is its flexibility, both in data entry and analysis. Web-based systems are, in general, designed to minimize the constraints inherent with variable rates of data entry. As we observed with the spreadsheet based data collection system, the potential for errors seemed to rise as the numbers of users, and the numbers of copies in circulation, increased. The web-based tool is less vulnerable to these errors: delays between entries will not have as great an impact on the overall quality of submissions. Because of this, users will be more confident in the database, and more likely to use it as a decision-informing tool. Our design goes further in that data managers can choose the frequency of entering data to match the frequency at which they receive data. For example, in Lusaka we observed that blood collection data arrived weekly (by fax) from each of the other centers. On the other hand, blood utilization data was not obtained so regularly and a month-long window was more realistic. The system allows users to choose views of the data for analysis and management. The views of interest depend on the local context; e.g. in Zambia blood types have roughly the same distribution across the country, but the incidence of malaria (and thus the need for blood) varies by both time of the year and region. Therefore, reports that show geographic and seasonal trends in demand for blood would be more relevant than reports that indicate blood type distributions. Blood safety monitoring is an important special case of the more general problem of monitoring and tracking large data sets which are updated frequently. The latter has many possible applications in health, emergency relief, water and other basic resources, etc. In all these cases a web-based distributed monitoring and aggregation solution seems worthy of exploration, in spite of network constraints. We believe that the insights gained from our development of a blood safety monitoring tool could be applied more generally. ACKNOWLEDGMENT We are grateful to the Zambian National Blood Transfusion Service, Tanzania National Blood Transfusion Service, the Centers for Disease Control and Prevention in Zambia and Atlanta, and the US President's Emergency Plan for AIDS Relief. We thank Sridhar Basavaraju, Michael Best and Gregory Abowd for helpful comments on earlier drafts of this paper. REFERENCES [1] World Health Organization, "Fact Sheet No. 279: Blood Safety and Donation." 2008. http://www.who.int/mediacentre/ factsheets/fs279/en/. [2] H.S.F. Fraser, D. Jazayeri, P. Nevil, Y. Karacaoglu, P.E. Farmer, E. Lyon, M.K.C.S. Fawzi, F. Leandre, S.S. Choi, and J.S. Mukherjee, “An information system and medical record to

support HIV treatment in rural Haiti,” BMJ (Clinical Research Ed.), vol. 329, Nov. 2004, pp. 1142-6. [3] J. A. Blaya, S. S. Shin, M. J. Yagui, G. Yale, C. Z. Suarez, L. L. Asencios, J. P. Cegielski, and H. S. Fraser. “A web-based laboratory information system to improve quality of care of tuberculosis patients in Peru: functional requirements, implementation and usage statistics.” BMC Med Inform Decis Mak. 2007;7:33. [4] Lani Marquez. Are Web-Based Platforms to Support Quality Improvement Feasible in Developing Countries? Experiences from Ecuador, Honduras, Nicaragua, and Rwanda, 2007. http://www.qaproject.org/news/ISQUA07/ ISQua2007%20Marquez%20handout.pdf. [5] World Health Organization, "Blood Transfusion Safety", 2004. http://www.who.int/bloodsafety. [6] Jeremy Keith, Bulletproof AJAX. New Riders Press. February 2007. [7] Jesse James Garrett. "Ajax: A New Approach to Web Applications". 2005. http://www.adaptivepath.com/ideas/ essays/archives/000385.php. [8] Jennifer Manne and Walter Curioso (2007). “Technology and health care in the developing world.” Student BMJ, December 2007. http://student.bmj.com/issues/07/12/ life/436.php [9] P Brand, A Schwittay. The Missing Piece: HumanDriven Design and Research in ICT and Development. In Proceedings of the International Conference on Information and Communications Technologies and Development, May 2006. [10] R Tongia, E Subrahmanian. Information and Communications Technology for Development (ICT4D) – A Design Challenge? In Proceedings of the International Conference on Information and Communications Technologies and Development, 2006.

288

Dimensions of IT Literacy in an Arab Region: A Study in Barkha (Oman) Sherif M. Aziz

Abstract— The transformational potential of the Information and Communication Technology lies in the ability of the people to make the best use of it. In fact, they need to be e-ready to benefit from what is known as the digital society. The rapidity of technological advances is, in fact, pushing many developing countries digitally far behind. However, as of now, it is well accepted that the digital divide is not merely about technology deprivation, but is also about many inter-related social, educational, cultural, political and economic issues. Understanding its subtle nuances in the context of different parts of the developing world is extremely important to address the problem in a meaningful way. This paper is about an action research done in the district of Barkha in the Sultanate of Oman, through a pilot IT literacy campaign. The study reveals that, for evolving a digital society in an Arab region, issues such as linguistic literacy and information literacy are far more challenging than the much publicized computer literacy. The study also concludes that Oman is at the threshold of access to technology but has challenges with regard to cognitive access, content access and an enabling environment. Index Terms— Access to ICT, digital divide, IT Literacy, Oman

I.

INTRODUCTION

I

NFORMATION and Communication Technology (ICT) has pervasively intruded into our everyday life. Perhaps, the most recent picture of this pervasiveness is demonstrated by the aggressive way in which most governments are automating their service deliveries and citizen interfaces through what is popularly known as e-government.

The Sultanate of Oman, like all its neighbours, is pursuing a national e-government initiative since the year 2002, along with a wider Digital Oman strategy [1]. The country’s Seventh Five Year Development Plan (2006-2010) lays emphasis on upgrading the information technology sector by implementing the national strategy for Oman’s digital society [2]. The Information Technology Authority of Oman, which spearheads these initiatives, provides a definition for ‘digital society’ as a modern, progressive society that is formed as a

The pilot IT literacy campaign part of this study was fully supported by the Middle East College of Information Technology, Muscat. Sherif M. Aziz is with the Public Establishment for Industrial Estates, Sultanate of Oman (e-mail: [email protected]).

result of the adoption and integration of ICT at home, work, education and recreation [3]. To move towards a digital society, there needs to be a synergy of technology infrastructure, a skilled population, and an overall enabling environment. The enabling environment should not only have technological components and costrelated incentives, but also aspects such as motivation and a sense of purpose among the citizens. In Oman, soon many government-to-citizen interfaces will be web based or through other electronic means. An Ubar Portal is designated to be the main gateway to electronic services offered by the public sector. Citizens will access the government via the Ubar Portal that links to other portals and web sites hosted by the Government entities and enables seamless execution of e-services. This portal is named after the ancient Omani city of Ubar, the “Atlantis of the desert” and a main trading gateway into Arabian Peninsula. It is envisaged that anytime, anywhere access to the portal can occur through multiple channels such as the web and mobile devices [2]. However, for the citizens to be able to derive benefits out of all these developments, they need to be ‘ready’. In other words, there is an urgent need for an overall societal preparedness and citizen e-readiness to make the whole effort successful and beneficial to the society. To achieve this, there are many essential components, one of which is the information technology (IT) literacy among the masses. Apart from providing ICT infrastructure, IT literacy is equally important to alleviate what is popularly known as the ‘digital divide’. Information technology is now taught in all schools in Oman as a separate subject from grade one to grade ten in basic education. In grade eleven, the ICDL (International Computer Driving License) is a required course and each student has the opportunity to learn IT skills [2]. Nevertheless, considering the fact that nearly one-third of the population is beyond the usual ages of formal education 1 , there is a need for alternative means of achieving basic citizen readiness to benefit from the ongoing digital society initiatives. Perhaps, some non-formal means of achieving IT 1 Estimated from the data in the Statistical Year Book- 2004 published by the Ministry of National Economy.

289

literacy would yield better reach and coverage in the shortest possible time. In the above background, a group of volunteers under the leadership of the author, and with the support of an educational institution, carried out a pilot IT literacy campaign in the district of Barkha in Oman. This campaign was planned as an action research, which provided ample insights for evolving a national strategic framework for achieving IT literacy. Objectives of the research: The overall objective of the research was to evolve a basic strategic framework for a national IT literacy campaign, with a view to enhance citizen e-readiness in the context of the Digital Oman initiative. Specific objectives: To design and carry out a pilot IT literacy campaign in a semiurban region as a test case through an action research. To evaluate the lessons learned, and other analytical findings resulting from the aforesaid objective, in the background of relevant international thinking and literature on the issues of access to ICT, IT literacy and digital divide. To draw inferences from the aforesaid stages, and to evolve a basic strategic framework for a national IT literacy campaign. Scope of the research and its limitations: With regard to the IT literacy campaign, there was a larger community of interested parties such as the Wali 2 and his office, local community leaders, schools where the initial campaigns were conducted and, of course, the national government. Therefore, from concept to implementation, subtle socio-political issues had to be handled carefully to maintain the interest and motivation of all the key actors. In such a background, apart from coping with the inherent methodological complexity, the very management of the campaign, data collection, and the research in general demanded limiting the IT literacy campaign to a pilot action research in a limited geographical region. The pilot literacy campaign also had to be short to maintain the morale and motivation of the volunteers. In the light of the above, the field study was limited to the Wilayat of Barkha. Methodology: A mixed methodology was adopted for conducting this research. This included a literature survey, selection of appropriate theoretical framework, structured and unstructured interviews, action-research through an IT literacy campaign, a survey among the participants of the campaign using pre2 Wali is the administrative head of a district, which is known as the Wilayat

tested questionnaires, a follow-up telephone interview of the participants, and focus group exercise to evolve the national strategy. The focus group consisted of a homogenous group of ten persons who had equally fair idea about the IT literacy campaign. They had sufficient understanding about the concepts of e-government and had basic awareness about the Digital Oman initiative. The IT literacy campaign as such is not a new concept. However, the locale under consideration was not very familiar with such non-formal and non-governmental interventions. Preliminary assessment of the situation demanded an action research, which follows an ‘intuitive proactive’ approach as against a ‘rational reactive’ one [4]. In the first case, practitioners who know or think they know, what needs to be done, implement an intervention program first and then see how well it is progressing. In the latter approach, the researcher examines what is occurring with a specific focus on some known problem and then draws up a program to react to what has been discovered. II. THE REAL ACCESS The information (technology) era demands a higher degree of literacy and complexity of skills to function effectively in what is known as the “digital society”. An OECD adult literacy survey [5] reported that, across 20 countries, one in four adults who participated in the study do not possess the necessary skills to manage in today’s world. Findings point to large disparities in the average level and distribution of literacy skills, both within and between countries. The report pointed out the need for policies to be directed at the work place and family settings so that an overall societal preparedness for the information age is possible. Lack of literacy skills does cause difficulties for living, working and even survival [6]. It is not the work place alone, which demands such skills. Even one’s own healthcare demands skills to access and use appropriate information. Being information-literate has been defined as "having the ability to recognize when information is needed, then to be able to locate and evaluate the appropriate information and use it effectively"[7]. These abilities have always been important to lead a successful life. However, as of now, IT skill has become an important component of information literacy, given the fact that the channels and quantum of information have multiplied phenomenally on account of the rapid technological advancements in the field of information and communication technology. The question of digital divide Digital divide is a term that emerged in the 1990s as an expression of concern about the distribution of access to and benefits from advances in communications and information infrastructure. Digital divide, in simple terms, is the differentiation between those who have access to IT and digital information (particularly, the Internet) and those who

290

do not [8]. But, with rapid technological advances, technology itself is a moving target, thereby enhancing the rate of further divide between what could be called as the ‘information haves’ and the ‘information have-nots’. The British Council’s info@UK website glossary defines digital divide as the gap between those who can effectively use new information and communication tools and those who cannot. This divide can be the result of one or many different factors. Being a member of a socially excluded or disadvantaged group, or lack of necessary skills to participate are all important factors [9]. The average OECD country, which has 11 times the per capita income of a South Asian country, has 40 times as many computers, 146 times as many mobile phones and 1036 times as many Internet hosts [10]. North America is home to only 5 percent of the world’s population, but represents 28 percent of the world’s Internet users. The Middle East with about 4 percent of the world’s population has only 1.4 percent of the world users [11]. The foregoing picture is with regard to the existing disparities. The rapidity of technological advances is, in fact, pushing the ‘have not’ countries further behind. But the disparity with respect to e-readiness is not merely due to technological components. As of now, it is well accepted that the digital divide is not merely about technology deprivation, but is also about many inter-related social, educational, cultural, political, and economic issues. An ITU opinion poll had shown that 77.3 percent of those covered by the survey citing poverty as a major barrier towards integrating to the information society, followed by lack of education (76 percent), and lack of infrastructure (72.8 percent) [12]. The UN report [13] states: As countries progress in employing ICTs for development, the challenges underpinning inequality in access have shifted from connectivity issue to encompass a wide array of economic, social, cultural, and language barriers…The issue of a digital divide is essentially one of a disparity in real access which is inequality in both physical access to ICTs and the ability, know-how and the culture to use the technology well. The Centre for International Development at Harvard University defines an e-ready society as one that has the necessary physical infrastructure (high bandwidth, reliability, and affordable prices); integrated current ICTs throughout businesses (e-commerce, local ICT sector), communities (local content, many organizations online, ICTs used in everyday life and also taught in schools); the government (egovernment); strong telecommunication competition; independent regulation with a commitment to universal access; and no limits on trade and foreign investment[14].

The foregoing, however, appears to be idealistic, and is more close to a developed country situation. In a developing country context e-readiness requires to address the issues of basic literacy, poverty, health, and other social aspects first. In any case, digital divide is increasingly a matter of universal concern, as the personal computer and the Internet have become critical to economic success and personal advancement. Therefore, understanding its subtle nuances in the context of the developing world is extremely important. The Access for Opportunity Framework ICT for development has traditionally focused more on technology infrastructure issues. It is true that, for most developing nations, creating physical networks is very much beyond their means. However, there are a few other cash-rich developing nations such as those in the Persian Gulf region, which are examples to the argument that technology infrastructure is a necessary, but not a sufficient condition for access to various aspects of a digital society. It is in the above context that the Access for Opportunity Framework, which is detailed in the United Nations Global eGovernment Readiness Report 2004 [15], becomes relevant to study the situation of developing countries such as the Sultanate of Oman. This approach emphasizes the aspect that the end goal should be access for opportunity, rather than access to ICT. The Access for Opportunity Framework lays out the parameters of real access. The salient points of this concept are: • Physical access to ICT is only the first step towards building real access. • Access must be blended with relevant and culturally appropriate content for onward transmuting into knowledge • The blended knowledge is processed and utilized to create opportunity for economic and social empowerment This framework goes further to outline the Model of AccessAcceleration, which maintains that: Technology infrastructure needs to reach some threshold level in a given nation for access to start accelerating, but only as long as other access-supporting economic, social, educational, and cultural elements are in place [15]. Thus, a nation, which is advanced in terms of educational, cultural and social support structure, is in a better position when the technology infrastructure reaches the threshold access point. The Access-Acceleration Model is based on the basic premise that there is a threshold level of `real access’, which when

291

attained, will allow the subject to enter a state of accelerated access. The threshold level of access comprises ICT infrastructure, penetration of technology, government leadership, education directly supportive of technology, culture of technology, and global language of technology [15].This model goes further to make the following Access Framework definitions: Access threshold : A static cut off point where the mix of technological, economic and social systems begins to blend together to provide a synergy which allows positive feedback cycles of technology for the utilization of knowledge, which allow for greater economic and social opportunity. Beyond the access threshold a country enters the accelerated access phase. Accelerated Access: A dynamic changing optimal mix of technology, educational skills, economic and societal conditions, which come together at any given point in time to produce threshold access, which in turn leads to real access. Real Access: The equilibrium level of access, whereby an individual has the required availability of technology, educational skills, culturally appropriate and relevant content in his/her language of choice at an affordable cost. The UN report [15] states that the vast majority of countries have not yet reached real access or even a threshold level of ICT infrastructure needed to enter the acceleration stage, and most are lagging in the other critical access elements. A detailed study, which will consider a possible maturity process to real access, was beyond the scope of this research. Therefore, in this study, the concept of real access is viewed in a very simplistic manner as an equilibrium of availability of technology (basic telecom and internet access), education and skills, culturally appropriate and relevant content in the language of choice at an affordable cost, and the motivation (sense of purpose) to use.

implies, a felt need for information, a sense of purpose, and the ability to use information, all of which put together means the realization of opportunities and empowerment due to information availability. But at least the process of seeking information is a multidimensional technology issue, one dimension being the basic or even rudimentary ability to use the related technology. However, the Internet, which is currently the ubiquitous means of global information sharing, is still substantially a text-based technology. Therefore, all the aforesaid aspects become significant only if the person is fundamentally literate in a language, which is used in the technology driven information network or environment. Therefore, being IT literate would have three almost hierarchical components viz. (a) basic literacy in a language (b) technical literacy of being able to use the related equipment or computer literacy and (c) being information literate as explained above (see Fig.1).

Information Literacy --------------------------------Computer Literacy --------------------------------Linguistic Literacy

IT Literacy redefined The information age is largely characterized by the technological innovations that have had far reaching consequences in the way people work, communicate and do business. In fact, the convergence of telecommunication, broadcasting and computers has had significant impact on the society, which is comparable with the previous major societal changes such as the industrial revolution or the advent of the printing press. As with the advent of the printing press, the technological age requires the development of higher levels of skills in the society [16]. It is not enough to be literate in the sense of being able to read and write in some language. If a person is to survive in the information age he needs to be ‘information literate’, which, simply put, means that he has the skills to seek, understand, and use information 3 . But it also 3 The American Library Association Presidential Committee on Information Literacy 1989 defines Information Literacy as: “ To be information literate, a person must be able to recognize when information is

Fig.1 Hierarchical components of being IT Literate In regions such as the Persian Gulf, and therefore in Oman, the first component itself is a matter of concern in that the abundance of information available in the so called worldwide web itself is not in the ‘language of her/his choice’, which is Arabic. On the basis of the above concept, it can also be said that being information literate also indicate a general maturity with regard to the human capital, not necessarily in terms of mere educational qualifications, but essentially the positive impact of education and technical skills to lead better quality of life. In other words, such societies will have an ‘information maturity’ to make use of the information and knowledge around them to lead a better life.

needed and have the ability to locate, evaluate and use effectively the needed information (ALA, 1989 p.1)

292

The foregoing is the other conceptual framework, which has been evolved and used in this research. III. IT LITERACY CAMPAIGN Citizen e-readiness is a basic necessity for any digital society initiative to be successful, and IT literacy is one of its primary components. Perhaps, for quite some time, it could have been taken for granted that such IT literacy will come through the formal education system. However, at least for the time being, this will be inadequate considering the fact that about 34 percent of the Omani population has already crossed the usual age of schooling [17]. Therefore, there is a necessity for some non-formal means of making the population IT literate. However, since this is an issue of the masses, a hit and try approach can result in a strategic catastrophe. Therefore, such a non-formal intervention is best implemented after a pilot project so that, based on the lessons learned; a nation-wide strategy can be evolved and implemented in the most effective manner. It is in the above background that the IT literacy campaign at Barkha was organized as an action research. Overall organisational set-up: The campaign had to be legitimized by governmental patronage. This was obtained from the district administration. This campaign had the support of the Middle East College of Information Technology, Muscat, by way of providing student volunteers and a fully equipped Internet bus. A core team headed by the researcher, five volunteer leaders, and three technical support staff was formed with full autonomy to plan and organize the campaign. About 60 volunteers actively participated in the various activities of the campaign. Design of the campaign: The IT literacy campaign was structured in the following manner: (a) Two days class-room sessions: This included introduction to the era of computers, creating a sense of purpose for gaining IT literacy, and creating awareness about the opportunities in the information era and in an e-government environment. This was followed by general familiarization with computer. Trained volunteers gave lectures quoting real life situations. Subsequently, small groups were formed to provide hands-on training. (b) Walk-in-and-use support for the next two days: Two volunteers from the neighbourhood of each venue were made available for about three hours per day to provide further helping hand to those who have already attended the campaign on the previous two days.

(c) The Internet Bus: The Internet bus was used for promoting the campaign, for providing follow-up support to the already trained participants, for providing basic training to small groups assembling in souks 4 , and to reach out to the interior villages. The volunteers were divided into three groups – (a) publicity group (b) trainers group, and (c) follow-up group. The overall approach was that of ‘training of trainers’ within the community, which was expected to result in a chain reaction and speedy coverage. Lessons learned during the campaign: Many observations were recorded, discussed on a daily basis and corrective measures taken. The major issue was that the campaign could not attract significant number of men and women above the age of 30 years, though it was primarily aimed at attracting people who have already passed the usual age of formal education. Therefore, the Internet bus was deployed to the souks to attract aspirants from higher age groups. Even with sufficient cajoling by the volunteers, older citizens were not able to relate themselves to this activity. By the second week of the campaign it was more or less clear that the campaign has to reach out to the villages in a more intrusive way to attract older women and men. Converting the campaign into a family affair – a minor breakthrough: The whole campaign actually worked on the concepts of voluntarism and local community participation. In order to break multi-dimensional (social, cultural and gender issues) hurdles in reaching out to the household level, the volunteers were encouraged to take the Internet bus to the villages, where she/he can impart basic awareness training to mostly people who are her/his relatives and neighbours, all to be done by her/him without any outsider being there. This approach generated more active participation from women, but still could attract only a very few above the age of 40 years. Survey among the participants: A survey was conducted among the participants of the IT literacy campaign. A pre-tested questionnaire was administered to all those who attended the campaign. First set of data was collected prior to imparting training and at the time of registering for the campaign. The participants were asked to fill and return the second part of the questionnaire after the training. Subsequently, 45 days later, a telephonic interview of the participants was done to cross-check certain data as well as to get some supplementary information such as actions taken after the campaign. 4 Old market place with very narrow streets, typical of the erstwhile Arab commercial centers

293

TABLE II MOTIVATION FOR ATTENDING THE CAMPAIGN

The empirical data obtained from the aforesaid survey as well as from the published documents were used to analyse the issue of citizen e-readiness in Barkha, in the background of the UN’s Access for Opportunity framework and in the context of the Digital Oman initiative.

Sl. No

Details of the analysis and findings of this primary data are given under the Study Findings in section IV.

2

1

3 IV. STUDY FINDINGS Though the campaign had targeted attracting more of those who have passed the age of formal education, most of the participants (55.1 percent) were from the age group of 16-25 years, followed by the age group of ‘up to 15 years’ (Table I). A rough comparison of this with the age distribution in the national population reveals that the campaign could not attract all the segments of the local population representatively, the maximum distortion being at the two upper age-groups. Incidentally, the first age group seems to be properly represented. TABLE I AGE DISTRIBUTION Sl. No 1 2 3 4 Total

Age Group (years) Up to 15 16-25 26-40 More than 40

Number of Participants

% to total

Corresponding representation in the national population (%) *

74

36.1

34

113 14 4

55.1 6.8 2.0

22 25.2 18.8

205

100.0

100.0

Note:* Approximate figures computed based on the national population statistics, sourced from the Oman Census Survey Report 2003. The total population is 2.34 million with more than half a million expatriates.

Apparently, computer is still seen as a modern tool meant for the younger generation. This aspect, seen along with the fact that more than 30 percent of the participants had higher secondary education (see Table III), indicate that many of them were job seekers trying to acquire more computer skills. This is also evident from the higher representation (55.1 percent) from the relatively more active job seeking age group of 16-25 years. The expressions of motivational factors for attending the campaign reinforce the aforesaid inferences (Table II).

Motivational factor Acquire computer skills To get a decent job Can’t live without computer skills

Number 64

% to total participants 31.2

74

36.1

55

26.8

4

Upgrade the skills

20

9.8

5

Others

10

4.9

Note: Figures are not mutually exclusive.

Education Majority (30.1 percent) of the participants had education up to the 12th class, followed by those with 10th class (14.9 percent) and 11th class (16.6 percent) education (Table III). Nearly 60 percent of these participants with higher secondary education did not have computer skills. In fact, no correlation was found between the level of education and computer skills or degree of awareness about Digital Oman or e-government. TABLE III EDUCATIONAL BACKGROUND Sl. No 1 2 3 4 5 Total

Level of schooling (class) 12 11 10 7 to 9 Up to 6

Number of participants

% to total

62 34 44 51 14 205

30.2 16.6 21.5 24.9 6.8 100.0

From the above observations, it is inferred that the formal school education in Oman up till now has not made significant contribution towards general IT literacy. Further, about 34 percent of the Omani population is already outside the scope of formal education. Therefore, the non-formal means of an IT Literacy campaign was an appropriate intervention. Literacy An analysis of all the hierarchical components of being IT literate (see Fig 1) has been attempted here. The Barkha survey revealed that 87.8 percent of the people could read and write Arabic (Table IV). This is fairly a high level of literacy rate in a developing country. Many Arab countries are likely to have such high Arabic literacy, as reading the Holy Quran is almost a must from the religious point of view.

294

TABLE IV LITERACY Nature of Literacy

Arabic (Read) Arabic (Read & Write) English (Read) Basic Computer Skills

No. of persons including all family members of the participants * 1471 1381 889 427

Note: *Total number of family members covered – 1573

% to total (including family members) 93.5 87.8 56.5 27.0

The published national statistics show about 80 percent literacy rate among Omanis [18]. Obviously, this is their ability to read and write in the local language, which is Arabic. In fact, the figure of 87.8 percent obtained from the Barkha sample is almost in line with the national figure. The survey also revealed that only 56.5 percent could read English. Those who had computer skills were even less – 27 percent. The English reading ability in the national population can be much below what is reflected in the Barkha sample, as this sample had a poor representation of people above the age of 40 years. Further, a substantial number of participants of the campaign had higher secondary education as well. The opportunity to use one’s own language on global information network such as the Internet determines the extent to which one can participate in the knowledge society. However, an estimated 87 percent of documents on the Internet are in English [19]. The UN Report [20] confirms that English continues to dominate the cyberspace material. Only 1.4 percent of the world online users are Arabs, whereas they represent 4.6 percent of world population. This is in spite of the fact that the Internet use in the Middle East grew 219 percent during the period 2000-2004. This is in sharp contrast to the situation of 39.4 percent of world online users using English content, where as only 5.4 percent of the world population are English speaking [20]. Thus in the case of Oman, though Arabic literacy is high, the issue of ‘linguistic literacy’ has two aspects to tackle in order to enhance IT literacy: (a) the language access barrier, with a substantial chunk of World Wide Web content being in English. (b) limited web-content in Arabic, though all governmental sites are primarily in Arabic language. The foregoing lacunae are apart from the issue of ‘relevance’ of the web-content. What is relevant is related to the local culture and other socio-economic conditions. In most developed countries, health, education, recreation and

matrimony are all relevant contents. The relevance of these contents also depend on the maturity of the services offered by the different actors in these sectors, be it in the private or public sector. For instance, an online shopping portal is of no significance if the associated online payment and the delivery mechanisms are not intact, or if the legal framework is not mature enough to support e-commerce. The above observations should be seen along with the fact that in the Barkha sample only 27 percent had basic computer skills. This is out of a sample population of predominantly youth with nearly 60 percent having higher secondary education. From the point of being IT literate, the last component is to become ‘information literate’ (see Fig.1). From the survey of those who attended the campaign to acquire IT literacy, it is not possible to assess the status of their ‘information literacy’. Such a survey needs to be carried out differently among a larger cross-section of the population. However, a sum total of all the above observations points to the possibility of weak information literacy among the population, no matter whether it is computer-based or not. Access-for-Opportunity framework – A revisit in the context of Oman: The UN Access-for-opportunity framework detailed under the section REAL ACCESS talk about the importance of a perfect blend of technological, economic and social systems to provide a synergic situation for ‘real access’. Considering the various dimensions of this approach, a slightly modified conceptual framework was evolved to include some finer elements such as ‘motivation or sense of purpose’ among the masses to make the ‘use of ICT a way of life’ (Fig.2). Such subtle nuances can look trivial but are often critical components of many development issues involving a society. In this modified conceptual framework, the elements of ‘physical access’ and ‘affordability’ are considered together along with a via-media solution of ‘public access’ points, all of which put together can provide a synergic effect in enhancing the ‘access to technology’.

CONGNITIVE ACCESS Education IT Literacy

CONTENT ACCESS

PHYSICAL ACCESS Electricity Telephone Computer Internet

REAL ACCESS

Public Access

AFFORDABILITY Per Capita income Competitive tariff

Culturally appropriate Relevant Language of choice ENABLING ENVIRONMENT Awareness Motivation/Sense of Purpose

Fig. 2. Modified Conceptual Framework for Real Access

295

U.K

Physical Access Primarily, ICT development is built upon the telecommunication infrastructure. The UN e-Government Readiness Report [21] shows strong linkage between greater telecommunication access and higher states of e-government readiness. However, in the access-for-opportunity framework, this alone is not a satisfying condition for ‘real access’. But the aspect of ‘physical access’, which is a critical component of ‘real access’, comprises the availability of telephone, computer and the internet. As of now, the need for and availability of some form of electricity is often taken for granted. The Barkha survey, for instance, revealed that 99.5 percent of the households covered by the survey had electricity connection. The other related data are shown in Table V. TABLE V PHYSICAL ACCESS COMPONENTS Sl. No

Description

1 2 3

Telephone Computer Internet

Persons per technology Published Barka Data * Survey 11.9 10 26.7 24.4 14.1 47.6

Country

Telephon e

Compute r

Interne t

Oman

11.9

26.7

14.1

Singapor e India

2.2

1.6

2.0

25.1

138.9

62.8

Syria

8.1

51.5

77.5

Bahrain

3.8

6.2

4.0

6

36.2

21.7

China

Remark s with respect to Oman Far Ahead Far Behind Ahead in Tele. lines. Far behind in other aspects Well Ahead Behind

2.5

2.1

Well Ahead

In the entire Barkha sample only 31.7 percent of the respondents had a computer at home and only 16.6 percent had internet connectivity at home. A closer look through a cross tab of the availability of computer and internet at home revealed that, out of the 65 respondents who had a computer at home, only 34 had internet connection as well. In other words, nearly half of those with computers at home did not have internet connection (Table VII). Further analysis of the reasons for not having internet connection revealed that only 2 out of the 31 persons (6 percent) had expressed money as the constraint for not taking internet connection. 3 persons (10 percent) expressed that their families feared cultural erosion. Remaining 26 persons (84 percent) did not have any specific reason, in a way indicating that they weren’t either aware or did not have a real sense of purpose in having an internet connection. Most households procured computers, because some family members’ education demanded one at home, or because they got inspired by neighbours, especially during festival promotions. TABLE VII AVAILABILITY OF COMPUTER VS INTERNET AT HOME

Note: * See UN Global e-Government Readiness Report 2004, p.96

Except in the case of internet penetration, the Barka survey figures more or less matches with the published data. A comparison of these factors in the case of Oman with those of a select few countries reveals that Oman is not at a disadvantaged position in the Asian and Middle East Region (Table VI). TABLE VI PHYSICAL ACCESS COMPONENTS – A COMPARISON (PERSONS PER TECHNOLOGY)

1.7

Availability of internet at home NO YES Availa bility NO of comp YES uter at home Total

Total

140

0

140

31 (47.7%) [ 18.1%]

34(52.3 %) [ 100]

65 [ 31.7%]

171(83.4%)

34 (16.6%)

205

Note:( ) indicate % to row total; [ ] indicate % to column total

Another interesting observation was about the gender distribution of the participants of campaign. 58 percent of the participants were women. Further, among the participants who attended the campaign with some basic idea about computers, majority (64 percent) were women. Here, it may be noted that the male-female ratio in the national population is 1.02:1. Some of the foregoing observations thwart the general perception that the conservative cultural values and gender inequalities are the deterrents to social advancements in this Arab region. On the other hand, the predominant picture that is emerging is that of general lack of a sense of purpose of having these facilities at home, as compared to material constraints to own them. A general perceptional data obtained from all the participants further strengthens the above argument (Table VIII).

296

TABLE VIII GENERAL PERCEPTIONS ABOUT COMPUTER AND THE INTERNET (AMONG ALL THE PARTICIPANTS) Sl. Description Yes No Remarks No 1 I knew what 101 104 *Nearly 50 % did computer and not have basic Internet are awareness 2 Was not 3 202 98.5% was interested in interested; very computers positive signal 3 Learning 31 174 Only 15 % computer is considered it difficult difficult; positive 4 Money is my 37 168 Only 18% constraint indicated financial constraints 5 Internet 17 188 Only 8.3 % spoils our thought so; culture against the usual perceptions about the Arab region 6 Computer is 3 202 Very positive not a necessity 7 Knows about 91 114 OK internet 8 Know Gov 40 165 Poor awareness services delivery through Internet 9 The above 34/40 6/40 Positive within will benefit those who were me aware of e-gov. 10 Going 29/40 11/40 Signs of lack of personally to confidence in gov. offices is electronic services better 11 One day I 28/40 12/40 Positive will be forced to use 12 Ever visited a 53 152 Poor website 13 Did you often 17/53 36/53 Poor; only 32% get what you appears satisfied look for from these websites 14 Knows about 20 185 Poor awareness Digital Oman 15 Any step 25 180 Not enough taken after initiative? campaign

Affordability The earlier analyses indicated that affordability or financial capability to own and access computers cannot be a matter of serious concern in the context of Oman. Oman with a per capita income of about US$ 13000 has roughly the 60th position among all the countries 5 . There are promotional schemes offering internet connection and personal computer together. However, for at least a small segment of the population, affordability, especially in relation to the money required for their mundane priorities in life, could still be a problem. But such people could be facilitated through public access points. In the years to come, as more Internet service providers come into play, the internet and telecom tariffs are likely to be more competitive. Public Access Private cyber-cafes are the only form of public internet access points available in this country. These are mostly limited to towns. There has been no significant attempt either from the government or any social institution to provide public internet access points, at least to encourage the use of the internet. The concept of self-employed IT kiosks are under government consideration, essentially to function as an interface with the IT illiterates, once e-government services are rolled out. In fact, maintaining a public internet access point will be cheaper than maintaining even a small tree in this desert land. The question of physical access as explained earlier, affordability, and the possibility of public access points throughout the country are a set of components of the ‘real access’ framework that need to be considered together. From the above perspective, Oman can be considered to have reached an appropriate threshold level with respect to access to technology (Fig. 3). This ‘access to technology’ would encompass physical access and affordability.

PHYSICAL ACCESS Has reached the threshold level PUBLIC ACCESS Will be cheaper than maintaining a small tree in the desert

REAL ACCESS

Fig.3. Access to Technology in Oman 5

Note: Incidentally, 60 percent of the participants had higher secondary education.

AFFORDABILITY Per capita income-60th rank among countries; Reasonable telecom tariff in the GCC region; Survey do not indicate finance constraints; Promotional schemes to own PCs exist.

This is as per the PPP (Purchase Power Parity) Method. As per the Atlas Method the per capita income of Oman is US$7830, with almost the same global ranking.

297

Cognitive Access Only 7.7 percent of the national population has postsecondary education [18]. It was also shown earlier that in the sample population covered by the survey, the school education has not contributed significantly towards IT Literacy among the masses. The overall picture obtained from the analyses under the subheadings Education and Literacy is that of a poor cognitive access situation, with weakness at the levels of linguistic literacy, computer literacy and information literacy, to participate and benefit from a world-wide information network. In the absence of sufficient cognitive access, the digitization of government processes, which is part of the Digital Oman movement, is unlikely to have much impact on the quality of life of the wider cross-section of the Omani society. Addressing the issue of linguistic literacy can have two possible approaches: (a) A long-term strategy of enhancing Arabic content (b) A short-term solution of popularizing the learning of English so that what is available in English in the world-wide-web is readily usable, and relevant Even computer literacy can be considerably enhanced through a multi-pronged approach of:

has strict filtering of pornographic material; and practically do not have any other control mechanism. Enabling environment - Motivation or sense of purpose 19.5 percent of the participants knew about e-government in general, mostly through media reports about global developments. Only 9.8 percent had heard about Oman egovernment initiatives. Out of the 40 participants who were aware about e-government, 29 persons still felt that going to offices personally is better to ensure speedy processing. A look at Table VIII reveals that the responses to most aspects connected with the individual and his capabilities are positive in nature (descriptions corresponding to Sl. no. 1 to 7). Others, wherein government has to play an active role in popularising or reaching out with the correct message with regard to egovernance, the people need to gain better awareness and confidence levels. But the fundamental attitude of the people is positive enough to mature into an information society, provided the related government interventions are appropriate, timely and motivating. The ‘real access’ issue in Oman All the above findings put together, it can be concluded that, among the various components of `real access’, the Omani population is almost at the threshold of access to technology, but are yet to reach an advantageous position with regard to cognitive access, content access, and the required sense of purpose to partake and benefit from the digital society era (Fig. 4).

(a) Strengthening computer education at schools and colleges (b) Non-formal means such as IT literacy campaigns, done with a missionary zeal.

PHYSICAL ACCESS Access to technology has reached the threshold level COGNITIVE ACCESS

However, attaining information literacy is a complex issue, which can be achieved only through a long drawn societal transformation at various levels. Language of one’s choice, relevance of the contents, and the possibilities to directly benefit from such electronic interfaces are all factors, that motivate the population to be information literate.

Poor; Efficacy of education not reflected in Barka sample

PUBLIC ACCESS Will be cheaper than maintaining a small tree in the desert

CONTENT ACCESS Very limited culturally appropriate content and language of choice. But no apprehensions about cultural erosion.

AFFORDABILITY

REAL ACCESS

Content Access ENABLING ENVIRONMENT

All the above analyses also reveal a situation of very limited ‘Arabic’ and ‘culturally appropriate’ content. Most government web sites, which are mostly in Arabic, do not provide significant interactive or transactional benefit to the users. The survey among the participants of the campaign revealed that among those who have visited one website or the other (i.e. 26 % of the Barka sample) only 32 percent felt that they got what they were looking for. Majority (91.7%) did not have any apprehension about cultural erosion due to the Internet. This country, of course,

Poor awareness ; some awareness of global developments. Not aware and confident about local initiatives. No evidence of motivation or sense of purpose.

Percapita income60th rank among countries; Reasonable telecom tariff in the GCC region; Survey do not indicate finance constraints; Promotional schemes to own PCs exist.

Fig.4. Assessment of ‘Real Access’ Situation in Oman. Grey boxes indicate challenges.

298

V. KEY CONCLUSIONS This section tries to highlight the key conclusions of the study.

As is evident from the figure, what has been proposed is a multi-pronged and holistic strategic framework, with the following broad considerations: a.

The predominant picture that is emerging out of this study is that the access to technology cannot be considered as a deterrent to citizen e-readiness in Oman. However, the issue of cognitive access capabilities need to be addressed from different angles and a non-formal means of training (such as the IT literacy campaign) is imperative to alleviate possible digital divide.

b.

Some of the specific conclusions are: a.

b.

c.

d. e. f. g.

h. i.

The issue of ‘linguistic literacy’ has two problems to tackle. Firstly, the inability to use the relatively huge English content available in the World Wide Web. Secondly, though Arabic literacy in Oman is high, the web-content in Arabic is limited. Therefore, for the masses to be IT literate and derive meaningful benefits of a digital society, the language access barrier may be considered as a fundamental deterrent. Addressing the issue of linguistic literacy requires a long-term strategy of enhancing Arabic content as well as a short-term solution of popularizing the learning of English. The possibility of weak information literacy among the population has to be considered while evolving digital society strategies. The findings indicate that affordability or financial capability to own and access computers cannot be a matter of serious concern in the context of Oman. There is a general lack of sense of purpose for having access to computers (say, at home), as compared to other material constraints to own them. The survey among the participants revealed that conservative cultural values and gender inequalities cannot be considered as deterrents to social advancements in the information era. It may, however, have some effect on the rate of progress. Oman can be considered to have reached an appropriate threshold level with respect to access to technology. There is a poor cognitive access situation, with weakness at the levels of linguistic literacy, computer literacy and information literacy. VI. RECOMMENDATIONS

There need to be a thrust in maximizing social inclusion, which does not necessarily mean making everybody IT literate; but it would simply mean maximizing social inclusion through a multi-pronged approach, including the use of intermediary mechanisms. Those above the ‘job seeking’ age group have to be reached out through mobile training facilities in a sustainable and friendlier manner. An appropriate communication plan is a prerequisite to create the right type of awareness about all what is happening in the name of digital society. Attempts shall be made to involve Sheiks 6 as champions who would take the ownership of IT literacy initiatives at the local level.

c.

d.

It may be emphasised that, even to achieve IT literacy and thereby e-readiness among the people, the overall environment should be conducive and encouraging to create a sense of purpose among the people to gain these new skills. This can be achieved if the related government interventions can bring demonstrable benefits that will touch upon the quality of life of the average citizen.

Barka model campaign at schools; Separate training rooms for men and women; Reach out to the village neighborhoods and family gatherings through miniinternet bus; More public access points – some free, some as IT kiosks as self-employment ventures; One or two full-fledged internet buses, pre-arranged connectivity and long duration stay at each location as per a

planned program

and

National level policy; Declared as an `IT literacy mission’; Corpus development fund; Empowered supervisory body

publicized

APPROACH TOOLS

IT LITERACY

Adult education at schools Internet Bus Public access points IT kiosks

Awareness training to government officials, community leaders (includes Sheikhs and elected members) and key private entity officials, as a prelude to mass campaigns- shall aim at transforming their attitude and their way of dealing with the public in the information era; Define their roles in the national IT literacy mission; Multi-stake holder situation- source support from the private sector and the academia; Incentives to community volunteers

ACTORS National Government Local Government Private entities People Community leaders Volunteers Academia

Focus on cognitive access and enabling environment Touch upon quality of life aspect Mostly Non-formal education Achievable targets against a timeframe

Training of trainers approach; Enhance English and computer skills at schools; Generate a sense of purpose and motivation through demonstrable projects; Immediate steps to enhance Arabic web content – in terms of quantity, relevance and reliability; Enhance trust and confidence in local web content and eservices; Quality standards for government websites; At least one IT literate person in every household in two years.

Fig.5. Strategic framework for a national IT literacy campaign

In the light of the above findings, a focus group exercise was conducted. The findings based on the analyses of the published data, field data collected during the IT literacy campaign, and the focus group exercise have been combined to evolve a strategic framework for a national IT Literacy campaign (see Fig. 5).

POLICY

Recognise the need for social change and preparedness; Maximize social inclusion; Minimise gender disparities while achieving targets; Use of ICT shall become a way of life. Consider that Oman is at the threshold of access to technology – if public access points and more promotional schemes to own computers are encouraged, access to technology is a nonissue; Government’s main focus shall be on enhancing cognitive access through a multipronged development strategy

REFERENCES [1] http://www.ita.gov.om/ITAPortal/eServices/eoman_strate gy.aspx 6

Tribal chief

299

[2] ESCWA, National Profile of the Information Society in the Sultanate of Oman. July 2007. Available from: http://www.escwa.un.org/wsis/reports/docs/Oman-07E.pdf (Accessed 07 April 2008) [3] http://www.ita.gov.om/ITAPortal/Info/FAQ.aspx [4] E.C. Wragg, An Introduction to classroom observation. London: Routledge, 1994. [5] http://www.literacytrust.org.uk/database/internat.html [6] E. Gayner, Back to basics: the role of preparing young people for the information society. Available from: http://www.emaraldinsight.com/0090-7324.htm (Accessed 20 June 2005) [7] American Library Association, Presidential Committee on Information Literacy: Final Report, 1989.Available from: http://www.ala.org/ala/mgrps/divs/acrl/publications/white papers/presidential.cfm (Accessed 18 June 2008) [8] http://www.ntia.doc.gov/ntiahome/digitaldivide/ [9] http://www.britishcouncil.org/fr/[email protected] [10] The World Bank, Information and Communication Technologies: A World Bank Group Strategy. Washington, D.C.: Global ICT Department, April 2002 [11] UN, Global e-Government Readiness Report 2004: Towards Access for Opportunity. New York: Department of Economic and Social Affairs, 2004, p. 107. [12] http://www.itu.int/osg/spu/wsis-themes/survey/index.html [13] UN, Global e-Government Readiness Report 2004: Towards Access for Opportunity. New York: Department of Economic and Social Affairs, 2004, p.8. [14] http://www.cid.harvard.edu/cr/gitrr_030202.html [15] UN, Global e-Government Readiness Report 2004: Towards Access for Opportunity. New York: Department of Economic and Social Affairs, 2004, p.76. [16] E. Gayner, Back to basics: the role of preparing young people for the information society. Available from: http://www.emaraldinsight.com/0090-7324.htm (Accessed 20 June 2005) [17] Ministry of National Economy, Statistical Year Book. Muscat: MoNE. 2004. [18] Ministry of National Economy, Oman Census Survey Report. Muscat: MoNE, 2003 [19] K. Taglang, “Content and the Digital Divide: What do People Want”, as quoted in UN Global e-Government Readiness Report 2004: Towards Access for Opportunity. New York: Department of Economic and Social Affairs, 2004. [20] UN, Global e-Government Readiness Report 2004: Towards Access for Opportunity. New York: Department of Economic and Social Affairs, 2004, p.107 [21] _____, Global e-Government Readiness Report 2004: Towards Access for Opportunity. New York: Department of Economic and Social Affairs, 2004, p.94

300

Emergency Communication and System Design: The Case of Indian Ocean Tsunami R. Chen, J. Coles, J. Lee, and H.R. Rao

Abstract—On December 26th, 2004, the largest natural disaster in recent recorded history took the lives of over 225,000 people from over 40 different nations and displaced millions more. In an increasingly global environment, these disasters are no longer isolated phenomena and must be responded to with a global perspective. The number of casualties from Indian Ocean Tsunami is a testament to the failure of humans in mitigating large scale emergency incidents and it is the consequence of ineffective information sharing and communication among key stakeholders. Using Activity Theory as the theoretical lens, we explore the communication phenomena in emergency response and we identify the major challenge facing communication practices. The paper further explores the design, implementation, and management of advanced information technologies to address the communication challenges. Index Terms—Emergency information technology, globe

E

services,

communication,

I. INTRODUCTION

mergency management is “the process of gathering resources and acting upon the problems immediately during and after a critical incident” [1]. Emergency incidents may be natural or man-made. Limited information, unpredictable disaster development, short time window, and high operation complexity renders the management of emergency incidents a challenging task [2, 3]. In order to mount an effective response to an emergency situation, it is critical that the key actors communicate effectively and timely so as to exchange task critical information and form collaborative response activities. While communication has been increasing recognized as the key to success emergency response, the practices in the field are far from satisfactory and failures are frequently documented. Fig. 1 presents a brief illustration of emergency communications involving the variety of stakeholders.

Manuscript received September 21, 2008. This work was supported in part by the National Science Foundation under Grant 0809186. Rui Chen is with Ball State University, Muncie, IN 47306 USA (phone:716880-2161; e-mail: [email protected]). John Coles is with State University of New York at Buffalo, NY 14260 USA (e-mail: [email protected]). Jinkyu Lee is with Oklahoma State University, OK 74106 USA (e-mail: [email protected]). H. Raghav Rao is with State University of New York at Buffalo, NY 14260 USA (e-mail: [email protected])

Fig. 1. Overview of Emergency Communication

The 2004 Indian Ocean earthquake took place at 00:58 UTC on December 26 2004. The earthquake was measured 9.0 on the Richter scale 1,000 km away from the Andaman Islands. It caused a powerful tsunami that killed more than 225,000 people in eleven countries, including Indonesia, Sri Lanka, India, and Thailand. The Indian Ocean Tsunami was regarded as one of the deadliest natural disasters in history. Communications in the emergency response of tsunami was a failure. During the early hours of December 26th, 2004, such effective communication was not demonstrated on many fronts, both national and international. Indian soldiers on the islands of Nicobar and Andaman sent warnings to their mainland counterparts after getting hit a full two hours before it struck the Indian coastal region. These warnings went unheeded and no evacuation orders were issued for Sri Lankan or Indian coastal regions resulting in the loss of tens of thousands of lives. The head of Thailand’s Meteorological Service refused to issue a tsunami warning despite the looming danger because he didn’t want to cause unneeded panic [4]. News of the true damage done to Aceh Indonesia was not known for hours after the initial impact possibly due to cut phone lines and poor military communication channels [5]. The goal of this research is to uncover the dynamics in emergency communication and to recognize the major challenges and barriers. More importantly we discuss the available technical instruments and propose their design and management to meet these challenges for effective inter-agency communication. The Indian Ocean Tsunami provides an opportunity to observe and analyze how the information and communication technologies impact on the socio-economic development in the world. Despite of the plethora of prior literature on the response to Indian Ocean Tsunami, little work has been done to analyze the phenomena using a theory driven approach. As a consequence, the knowledge thus gained is segregated and a complete understanding is missing. Through

301

the lens of Activity Theory, we identify the key issues pertaining to the socio-technical systems involved in emergency communication. This paper thus contributes to the improvement in information system to enable and support emergency communication in large scale disasters. This paper is organized as follows. In the subsequent section, we review Activity Theory that is used in this paper as the theoretical foundation to explore emergency communication. Next, we apply Activity Theory in emergency management using the case of tsunami disaster. We then elaborate the information technology design and management to cater to the requirements and challenges raised. We conclude this paper with limitations and future studies. II. THEORETICAL FOUNDATION With the advancement in modern information technologies, information systems designated for emergency management have been practiced and they have proved to play an important role in facilitating the incident management [6-8]. To design and implement effective systems for emergency communication, it is important that we understand the set of design requirements. The development of communication systems requires systematic approaches to elicit and analyze the internal elements, structure, and relationships of core communication elements. In this paper, we use Activity Theory to guide the requirement engineering process in communication system development [9, 10]. An approach driven by Activity Theory represents a method that has gained increasing attention in recent years [11-13]. Activity Theory provides a lens to analyze the computer-supported activity of a group or organization and to study the design of artifacts for individuals and organizations. Activity Theory suggests that human activity is directed toward a material or ideal object, mediated by artifacts or instruments, and socially constituted within the surrounding environment [14]. Activity can be understood as a systemic structure with various activities that are collated or extended away from the core activities [15]. The subject is the active element of the process and can be either an individual or a group. The object transformed by the activity can be an ideal or material object. The transformation process is enabled and supported by instruments (physical or logical). The instrument provides the subject with the experience historically collected by his/her community. During the interaction, subjects internalize and/or externalize their cognitive schemes and their understanding of the relationship between themselves and the external objects, instruments, surroundings, etc. Activity Theory also considers interaction as one critical aspect and suggests that interactions are the driving force in human interaction and system design. The interactions may also exist inside the subjects, objects, instruments, and their interactions. In Fig. 2, we present the application of Activity Theory in emergency management. That is, the emergency responders (e.g., firefighter, police, and emergency medical service personnel) use the communication systems (e.g., phone and

email) to exchange task critical information (e.g., chemical release) in an attempt to build common operating picture of the status quo of the ongoing incident and mitigation. This inter-agency communication is influenced by the rules (e.g., standard operating procedure –SOP), community (e.g., both professional communities and municipality communities), and division of labor (e.g., operations, logistics, and planning) in emergency management realm [16]. Instrument Email, Phone Real Time Information Sharing Subject Emergency Responder

Rule Standard operating procedures

Activity

Communication

Community Government, NGO, volunteer, business, news media, victims

Object Situational Awareness

Division of Labor Task & Task Assignment

Fig.2. Application of Activity Theory in Emergency Communication

III. COMMUNICATION IN EMERGENCY MANAGEMENT In this section, we employ Activity Theory to analyze emergency communication in an attempt to understand the challenges and requirements for emergency communication system design. A. Emergency Communication in General In accordance with Activity Theory, the subjects in emergency communication are the first and second emergency responders. They may include fire department, police department, emergency medical services (EMS), hazard materials workers (Haz-Mat), etc. The subjects actively communicate with each other to achieve situational awareness. Biros et al define situational awareness as “the decision-makers’ moment-by-moment ability to monitor and understand the state of a complex system and its environment” [17]. During emergency, the completeness and accuracy of decision makers’ situational awareness is crucial to their abilities in comprehending the disaster facts, hazards and risks, and mitigation capabilities, which “piece together an accurate of reality” [18]. The communication activities function on the basis of technical and non-technical instruments such as existing forms, guidelines, and alert systems. These instruments are likely to be developed under the supervision of the individual actors and with propertied designs in terms of taxonomy, notation, and ways of implementation. For example, the fire companies in neighboring counties may adopt different instruments (e.g., alert systems and reporting forms) in managing disaster

302

response. These heterogeneous and incompatible instruments are typically involved in a large scale incident response where multiple subjects are called in. Emergency communication is subject to influences from rules and division of labor. The rule systems embody the norms and conventions adopted in emergency response community and their management practices. Example may include responders’ attitudes toward risk taking and risk sharing. Division of labor, on the other hand, refers to the individual domain expertise and tasks typically performed by response subjects. Based upon the rules in place and the division of labor, subjects may behave differently in their communication behaviors. Additionally, response community impacts the emergency communication patterns. General speaking, emergency community include government agencies, non-governmental organizations (NGOs), independent volunteers, local businesses, news media, and victims [19]. Government agencies are often the actors looked to in times of disaster, thus it is critical for open and functional relationships to be established prior to a disaster in order to facilitate effective communication. With the variety of governmental levels and the red tape that must be sorted through, it is also critical for the intra-governmental communication channels to be aided by the early establishment of relationship and trust [20, 21]. Without these additional channels the government paths are dysfunctional and often avoided by other actors when possible. The issue of agency tribalism between local government sanctioned response agencies adds a new dimension of complexity to the situation. In the United States, an example of this comes up often between the individual fire companies and the police departments due to issues in funding and jurisdiction. Some governmental actors can be highly effective in emergencies by nature of their position or the information and resources to which they have access. Local elected officials are a great avenue of communication to the general populace and are most effective when they have the crucial relationships established with potential resources in the region and have the trust of the general populace. Another great resource are the government monitoring agencies such as the Pacific Tsunami Warning Center (PTWC) and the US Geological Survey (USGS), which were the agencies on the front line of tsunami detection and information dissemination in the early hours of the Indian Ocean Tsunami [4]. Other resources that can often provide crucial data or resources are the local scientific community, private for-profit organizations, and private not-for profit organizations. Local colleges and universities can be often be a great source of crucial information both prior and in response to emergencies due to the intricate and invasive approach to problem solving on a regional and international scale [22]. Private organizations are great resources for both equipment and manpower mobilization. From factory workers to church goers, the effective training and utilization of local labor is critical to a speedy and effective recovery [4]. Whenever there is a disaster, local private citizens are

involved as is the news media. The four issues that plague private citizen involvement are as follows. These four issues are exacerbated by the involvement of news media. 1. Easy numbness to warnings 2. Dramatization of information given 3. Mistrust of centralized government 4. Demand for constant information especially immediately before and after a disaster occurs News media descends upon any significant event quickly and in an unfiltered manner, however, they are also the primary source of information that private citizens turn to in emergency situations. For this reason it is crucial for the news media to be viewed as an ally and not an enemy in emergency response information gathering and dissemination. A relationship built on mutual trust in critical to the effective utilization of local and international media and should be closely guarded [23]. Table 1 summarizes a list of typical challenges to emergency communication. We elaborate an example challenge. Table 1. Emergency Communication and Challenges Emergency Response Tasks Creation of a Common Operating Perspective Formal Decision Making Procedures Collection of Relevant and Effective Data Correct Interpretation of Data Appropriate Actions Identified and Executed Coordinated Response Effort

Challenges to Effective Communication and Action Data Quality X

Scope of Time Lack of Information Decisions Pressure Trust Complexity Impact X X

X

X

X X

X

X

X X

X X

X X

X X

X

The quality of data which a disaster generates can easily be overwhelming for three major reasons: the varying lack of quality information, the time consuming task of accurate data interpretation, and the differentiation of pertinent compared to additional extraneous information [24]. The process of finding quality data sources often involves trust-bonds created prior to the disaster [25]. Thus it is crucial for emergency managers to be open to sources that could provide additional information but should also not be too accepting of all information so as not to be overwhelmed with the task of accurately interpreting the pertinent elements of data. Once the seemingly pertinent data is extracted, the issue of deciding whether the new information is immediately relevant is crucial in making informed and effective decisions. When the data quality is low, emergency management is hindered as decisions are often be short-sighted or ineffective in the larger response as a whole [20]. B. Case of Tsunami We discuss in this section the major communication failures in the response to tsunami incident. Our discussion is focused on subject and community in Activity Theory as these are

303

deemed as the major failure in emergency communication. Emergency responders, subjects, along Indian coast did not equip systems capable of modeling and predicting tsunami scales and impacts. Immediately after an earthquake, computer models can calculate how fast the waves will travel, as well as their amplitude. In an interview, Tsunami expert Tad Murty mentioned that he had persuaded the Indian government to build a fully functional tsunami warning system but was told that there were not sufficient financial resources [26]. A system like this might had saved hundreds of lives. As a matter of fact, 26 countries in the Indian Ocean were notified of the tsunami within 15 minutes after the earthquake took pace. India was unfortunately not among them and thus missed the opportunity to detect the incident. Emergency responders also experienced difficulties in sense making of the incident and how it unfolded. In spite of a lead time of almost six hours, no early warning could be issued before tidal surges lashed the Indian coasts. The main reason was the absence of systems to translate common knowledge into scientific information. The response communities were not familiar with the phenomenon of tsunami and had neither proper prior experience nor trainings. Even after the information was collected interpreted, it was difficult to pass the information from one point to another. A great majority of communication infrastructures failed in the wake of Tsunami, just at the very time when they are most needed. At Maldives Island of Sri Lanka, telecommunications infrastructure was based on a microwave terrestrial backbone network. When Tsunami hit, all public telecommunication services to the 13 atolls (163 inhabited islands) were interrupted, causing the complete breakdown of the communications [27]. Disconnections among response communities also lead to the devastation of tsunami incident. First, communication breakdown took place inside the domestic response communities. The IAF administration of India, for example, was informed that the Car Nicobar air base was hit by tsunami an hour before it reached Tamil Nadu. However the first message was not published from the Indian Meteorology Department to the government until 41 minutes later. This example reveals the lack of standard operating procedure in the local government. Standard operating procedure prescribes the control and flow of emergency information among the stakeholders. Without it, emergency information will not flow among the involved agencies smoothly and the information supply chain will break down. Breakdown also took place among international response communities. The Pacific Tsunami Warning Centre of the United States successfully recorded the earthquake in the Indian Ocean and warned that “There was no threat of tsunami to the US area of responsibility.” While the warning would be useful to alert the countries along the Indian coast, the warning did not fully reach the Asian response community which could had saved thousands of lives. Considering the broad areas affected, emergency communication activities were difficult to initiate. To initiate

communication activities, it is necessary that one determines the scope of decision impact (i.e., parties he/she needs to reach) [24]. By identifying these critical individuals, the reduction in travel time and increase in accuracy of information between actors will lead to more effective responses [21]. This broader perspective gives the basis for a much stronger international response instead of a solely national response. In the case of the tsunami, this would have significantly strengthened the early response and preparedness of other nations had India, Thailand, Indonesia, and the United States communicated their knowledge both with each other and the other nations affected in a timelier manner. However, the scope of decision impacts was hard to decide by the emergency mangers who tried to gather and disseminate information during the response to tsunami. As a consequence, responders were not aware of what information should be passed on to what parties. Without a formal central framework and method of collecting information centrally and disseminating, it is difficult for actors to receive or even where to start when attempting to acquire the information necessary for an effectively mounted response [19, 22]. Using Activity Theory (AT) framework, Table 2 summarizes the observations of Tsunami response communications along with the problematic issues raised. In the subsequent section, we propose the set of solutions. Table 2. Activity Theory Informed Analysis on Tsunami Response Communications AT Construct Problems Consequence Lack of input Failure in detecting the event Subject Lack of subject Failure in interpreting the matter knowledge raw data Community

Lack of international collaboration

Failure in information sharing among countries

Instrument

Lack of communication infrastructure

Failure in information exchange and sharing between any two points

Rule

Lack of standard Failure in domestic response operating procedure information supply chain

IV. EMERGENCY COMMUNICATION SYSTEMS The information system discipline has progressed with new systems designs and management practices developed. For example, Hale develops a layered crisis communication architecture (CCA) that enumerates the communication functionality requirement for emergency information systems [18]. The layers include connectivity, data-validation, filtering, value interpretation, organizational memory, and group process layers. Michalowski et al develop a palm-based mobile system Mobile Emergency Triage (MET) [28]. In this section, we discuss the technologies that facilitate emergency communication. As in Table 3, the discussions are driven by the analysis of principal constructs of Activity Theory and their

304

interactions. Table 3. AT-Driven Communication System Design Designs driven by AT Constructs AT Construct Problems Design Components Lack of input Early warning systems Knowledge management Subject Lack of subject systems, intelligence matter knowledge processing Community

Lack of international collaboration

International emergency communication

Instrument

Lack of communication infrastructure

Improved infrastructure management for mobility, reliability and redundancy

Rule

Lack of standard Rule-based interoperable operating procedure information supply chain

Designs driven by Interactions between AT Constructs Subject –n– Community Grid computing Instrument –n– community Open source approach

A. Early Warning Systems Early warning systems are important to disseminate incident (e.g., tsunami) occurrence to the stakeholders in the disaster affected regions. The significance of early warning for emergency management has been emphasized in major international agendas including the Barbados Plan of Action for Small Island Developing States, the Johannesburg Plan of Implementation, and the G8 summit in Gleneagles as well as major environmental agreements such as the UN Framework Convention on Climate Change and the UN Convention to Combat Desertification [29].To improve the alert system, it is important to leverage the existing systems as well develop new warning systems. A lesson learned in the response to tsunami is that the existing warning systems function below optimal performance. International Tsunami Information Centre (ITIC), for example, monitors the tsunami activities in the Pacific and disseminates tsunami information and warning messages to well over 100 points scattered across the ocean. Unfortunately ITIC was confined to its member countries before the Indian Ocean Tsunami; this led to non-member countries such as India did not receive warning information in a timely manner. A series of consultation meetings led to a general agreement by implementing agencies and donors that all countries affected by the 24 December 2004 tsunami and other countries at risk to tsunami should be involved in this project for the establishment of tsunami early warning system in the Indian Ocean Region [29]. New warning systems are also in need to augment the existing capabilities of warning systems and to utilize modern computing technologies. Oregon and RAINS (Regional Alliances for Infrastructure and Network Security) in the

United States, for example, are launching a locally-targeted emergency alerting service featuring real-time tsunami warnings with evacuation routes. RAINS' Connect & Protect service captures NOAA/National Weather Service's tsunami warnings when issued, and immediately sends localized alerts via computers, pagers and cell phones, to local citizens responsible for public safety. It will dramatically increase the speed and reach of the warnings within a community. B. Knowledge management systems During emergency response, individuals, teams, and organizations share and apply knowledge as they process information, make decisions, and act on existing knowledge [30, 31]. Knowledge may be categorized in two distinct forms: explicit knowledge which is easy to communicate and can be codified and tacit knowledge which is inextricably woven with the experiences and situational contexts [32, 33]. Knowledge management systems such as knowledge repository provide vital support to help the decision makers analyze and understand the facts. As manifested by the Tsunami case, responders may not be familiar with all types of incidents and the approaches in mitigation. The lack of training and expertise may account for failures as such. To this end, knowledge management systems may be adopted to equip the responders with the right information when incident first strikes. There exist a few public knowledge repositories for emergency response such as the Responder Knowledge Base (www.rkb.us). Private knowledge repositories are available in most commercial-off-the-shelf (COTS) emergency management systems such as DisasterLAN (disasterlan.com), E-Team (eteam.com), and WebEOC (esi911.com). Systems as such retain domain-specific knowledge that complements the knowledge repositories maintained by individual teams. C. Intelligence Processing During emergency, the completeness and accuracy of information shared and exchanged among responders is crucial to their abilities in comprehending the disaster facts, hazards and risks, and mitigation capabilities, which “piece together an accurate of reality” [18]. System designs such as information infusion are advised in this regard. As noted above, however, the information quality in emergency communication is typically low where errors are frequently observed. For example, Chen et al suggest that information on emergency incident is likely to be tainted and prone to mistakes [34]. Eye-witness accounts of the scene are often biased by their comprehension processes, background, recollection, and verbalization skills. As a consequence, triangulation-based data validation [35, 36] is an indispensable component for emergency response information systems. The validation usually relies on multiple and independent sources to validate an event, fact, or conclusion. Designs using data correlation and mining techniques have proved to be adequate solutions [37, 38]. Reference databases are also advertised for employment as they enhance the system competence in logic reasoning and inconsistency detection [39].

305

Field interviews with emergency responders find that incident information may expand rapidly during the course of mitigation; the increasing level of response enactment [40] accumulates new information on incident, responder, resource, operation, and environment from all stakeholders. This multi-dimension information cross interrelate with each other and together match the high “requisite variety” [41] of the decision problems in extreme events. Literature on Cognitive Processing Capacity (CPC) suggests that individuals are limited in cognitive resources and incapable to allocate them optimally for problem solving [42, 43]. In the face of information of large volume and high complexity, emergency managers are prone to be “overloaded.” The mental stresses resulted by time pressure, perceive risks, and concerns on “public image” further intensify the cognitive overload. To this end, tactics such as information filtering are recommended by prior emergency studies so as to reduce the amount of irrelevant data, organize related messages into cohesive and coherent sets, and prioritize message sets according to level of importance [8, 18, 34, 39]. In addition, research on Human Computer Interaction (HCI) suggests that cognitive overload may be reduced through design schemes in interface structure, information packaging, and information accessibility [44, 45]. D. Communication Infrastructure Management Timely communication among responders (i.e., “subjects”) relies on the establishment of robust and efficient communication infrastructures. Before the Indian Ocean Tsunami, many of the countries in the region did not have a well-organized disaster management system except for tropical cyclone-prone countries such as India and Bangladesh. National disaster management offices in most counties were very weak, and there were few established infrastructures (channels and platforms) for emergency communication. Modern emergency management employs a variety of communication channels including LAN, WAN, ad-hoc wireless network, mesh network, 800M Hz radio, and satellite phone [46-48]. During any typical emergency operation, hundreds of task critical updates, briefings, reports, requests, queries, and orders are circulated inside the incident management organizations [49]. These channels enable real-time information sharing and communications, establishing a “common operational picture” to keep all the decision makers on the same page [50, 51]. A common operational picture is a single identical display of relevant operational information shared by more than one actor. Through information transfer, responders working on varying aspects of the mitigation tasks are able to synchronize their visions on the focal incident, smooth out potential inconsistencies, and synthesize a complete “shared mental model” [52]. Information sharing mechanisms such as “publish / subscribe” system are recommended for this regard. Emergency communication infrastructure is subject to threats on performance, information assurance, and communication interoperability. Channel performance measures the communication capacity, connectivity reliability,

channel accessibility, and system adaptability. Take connectivity reliability for example. During incidents, “wired” communication infrastructures are likely to be destroyed by the disaster. It is important that the response community plan and implement technologies to remedy the communication breakdowns. In the case of tsunami, the lack of communication connectivity in the most tsunami-affected Indian Ocean countries negatively affected some national and local level activities. Options may include (1) mobile systems to deploy communication channels in the field, including satellite phones, portable radio systems, WAN network, and mesh networks; (2) radio-communication systems such as maritime, aeronautical, and radio determination services; and (3) redundancy plans with backup channels to support system breakdown or overload due to request and command flooding. The existing emergency system literature has identified and evaluated a set of solutions which include increased technology investment, partnership of public/private network, load balancing scheme, priority telecommunication services, and redundancy design [18, 53-56]. Information assurance, on the other hand, refers to information security and privacy. Security is a key element to emergency communication as long as sensitive information is present; it continues to be focal concern for channel design considering the increasing awareness of terrorist attacks [57]. System designs such as encryption, decryption, and intrusion detection may suffice the security requirement. Privacy is a more recent issue facing emergency management nowadays. Existing regulations such as HIPPA (Health Insurance Portability Accountability Act of 1996) require that privacy information including personal identity and medical history be carefully handled [58]. E. International Emergency Communication To mitigate large scale incident such as tsunami, emergency communication may easily span over multiple countries and districts. The presence of a coordinated communication plan during emergency is pivotal to allow task critical information and intelligence travel across the boundaries of response communities both domestic and international. Regional coordination of emergency communication should first be strengthened through activities of working groups on mitigation, preparedness and response in order to ensure integration of communication channels such as early warning systems into national and regional mitigation, preparedness and response capability building efforts within a multi-hazard framework [29]. In addition, communication coordination may be established through both country and community level approaches to ensure those good practices, including hazard and vulnerability assessment, organizational strengthening, community participation, warning system operation, capacity building, evacuation planning, and the design and construction of shelters, can be shared and exchanged in a coordinated manner to avoid duplications, confusions, and overlaps. Communication among international organizations is subject to interoperability challenge. The interoperability issue is

306

pronounced for large scale emergency management (e.g., tsunami) because the technologies adopted by participating organizations from varying countries and regions are in general incompatible for reasons ranging from the ability of local agencies to fund availability to the lack of unified guidelines for software and hardware [59-62]. To this end, it is extremely important that the emergency response community as a whole coordinate to develop consistent data standards, transmit protocols, homogenous devices, compatible application interfaces, and congruent regulations [16, 63-66]. A number of efforts have been observed national wide with as to establish the communication interoperability solutions. These examples include the National Information Exchange Model (www.niem.org), Emergency Data Exchange Language (www.comcare.org/edxl.html), Common Alert Protocol (www.incident.com/cap), SAFECOM initiative (www.safecomprogram.gov), and Federal Enterprise Architecture (www.whitehouse.gov/omb/egov). Each of them addresses a portion of the interoperability challenges and more efforts are needed before a full-fledged solution may be achieved. Without coordinated efforts, emergency communications are likely to break down when they travel across the boundaries of communities, both domestic and international. That is, relief teams from the world will not be able to communicate using their individual systems during response to large disaster such as tsunami, ultimately slowing down the response progress. F. Rule-base Information Supply Chain Standard operating procedure (SOP) prescribes the rules to control and direct emergency information among stakeholders. Unfortunately many developing countries do not have systematic SOP and they may learn from the developed countries. Emergency management in U.S. is guided by the National Incident Management System - NIMS [67]. NIMS facilitates the emergency communication in the following ways: (1) it constructs a clear organizational hierarchy that directs the information flows; (2) it standardizes the format and content of information input and output; and (3) it establishes the protocols of reporting, meeting, and cross-boundary adjustments. And thus, SOP supports the information to flow smoothly among the agencies and form an emergency response information supply chain. To this end, rule-based communication systems that share and exchange information following SOPs are desirable to the emergency responders. G. More Designs driven by Interactions between AT factors Activity theory underlines the importance of interactions between factors as an important driving force for activity system development. Emergency responders, subjects, analyze the incident facts before they can share it with the other agencies. While numerous initiatives have been started to improve information processing, local agencies are typically limited in the systems that they can offer to build and manage. To this end, the broad community of public/private firms and the general public has

offered cost-effective alternatives to help the responders. Grid computing and grid storage has recently been employed in emergency response thanks to the rapid growth in network bandwidth. The computing and storage resources of individuals and volunteer organizations can be pooled to boost the information analysis capabilities of the local responders. These systems have been used recently to study weather, pollution, and remediation, etc [68]. Telecommunication software and systems, instruments, function to support communication activates when incidents strike. The development of systems as such may be improved through the aids from the community. Open source approach offers the opportunity for technology innovations and improvements in disaster communication system designs. This approach is also known as Free and Open Source Software (FOSS) development and community mechanism, where collaborative designs are achieved through inputs from the entire community. Example FOSS systems include the Sabana disaster system [69]. V. CONCLUSION The damage caused by the tsunami was unprecedented, but the lessons learned are crucial to creating communication frameworks that can withstand such buffeting uncertainty. By identifying the key characteristics that can cause communication in emergencies to be ineffective or to be entirely void, future researchers and practitioners will have a frame of reference from which to approach the issue of emergency communications [20]. In this paper, we employ Activity Theory to study communication in emergency management and explore the key dimensions of issues. Taking Indian Ocean Tsunami as an example, we identify and discuss the pitfalls in the communication practices for incident response. The authors further explore the issues in information system design and management in an attempt to address the pending challenges in emergency communication. New knowledge gained will allow effective pro-active measures to be taken to mitigate and respond more effectively to future disasters by recognizing the key players in an emergency, effectively facilitating international cooperation, turning academic research into practical application, and avoiding pitfalls in areas needing greater research. To further improve emergency communications, it is critical that future studies extend the proposed design and management frameworks to learn from and connect with the broad spectrum of information and communication technologies and development (ICTD) applications. REFERENCE [1]

S. Y. Shen and M. J. Shaw, "Managing Coordination in Emergency Response Systems with Information Technologies," in Tenth American Conference on Information Systems, New York, 2004, pp. 2110-2120.

307

[2]

[3]

[4]

[5]

[6]

[7] [8]

[9]

[10] [11] [12] [13] [14] [15]

[16]

[17]

R. Chen, R. Sharman, R. H. Rao, and S. Upadhyaya, "Coordination in Emergency Response Management," Communications of the ACM, vol. 51, pp. 66-73, 2008. E. L. Quarantelli, "Problematical Aspects of The Information/Communication Revolution for Disaster Planning and Research: Ten Non-Technical Issues and Questions," Disaster Prevention and Management, vol. 6, pp. 94-106, 1997. J. Telford, J. Cosgrave, and R. Houghton, "Joint Evaluation of the International Response to the Indian Ocean Tsunami: Synthesis Report," Tsunami Evaluation Coalition2006. R. Samarajiva, "Policy Commentary: Mobilizing Information and Communications Technologies for Effective Disaster Warning: Lessons from the 2004 Tsunami," News Media Society, 2005. B. Van de Walle and M. Turoff, "Emergency Response Information Systems: Emerging Trends and Technologies," Communications of the ACM, vol. 50, pp. 29-31, 2007. M. Turoff, "Past and Future Emergency Response Information Systems," Communications of the ACM, vol. 45, pp. 29-32, 2002. T. X. Bui and S. R. Sankaran, "Design Considerations for A Virtual Information Center for Humanitarian Assistance/Disaster Relief Using Workflow Modeling," Decision Support Systems, vol. 31, 2001. Y. Engestrom, "Activity Theory and Individual and Social Transformation," in Perspectives on Activity Theory, R. M. a. R. P. Engeström, Ed. Cambridge, UK: Cambridge University Press, 1999, pp. 19-38. Y. Engestrom, Learning by Expanding: An Activity-Theoretical Approach to Developmental Research. Helsinki: Orienta-Konsultit, 1987. L. Uden and A. Kumaresan, "Usable Collaborative Email Requirements Using Activity Theory," Informatics, vol. 31, pp. 71-83, 2007. V. Kaptelinin and B. A. Nardi, Acting with Technology: Activity Theory and Interaction Design. Cambridge, MA: MIT Press, 2006. A. Chaudhury, D. N. Mallick, and H. R. Rao, "Web Channels in E-Commerce," Communications of the ACM, vol. 44, p. 99, 2001. L. S. Vygotsky, Mind and Society. Cambridge, MA: Harvard University Press, 1978. O. W. Bertelsen and S. Bodker, "Activity Theory," in HCI Models Theories, and Frameworks: Toward A Multidisciplinary Science, J. M. Caroll, Ed. San Francisco: Morgan Kaufmann, 2003, pp. 291-324. R. Chen, R. Sharman, N. Chakravarti, H. R. Rao, and S. Upadhyaya, "Emergency Response Information System Interoperability: Development of Chemical Incident Response Data Model," Journal of the Association for Information Systems, vol. 9, 2008. D. P. Biros, M. Daly, and G. Gunsch, "The Influence of Task Load and Automation Trust on Deception Detection," Group Decision and Negotiation, vol. 13, 2004.

[18] [19] [20] [21] [22]

[23]

[24] [25]

[26]

[27]

[28]

[29]

[30]

[31]

[32] [33]

J. Hale, "A Layered Communication Architecture for the Support of Crisis Response," Journal of Management Information Systems, vol. 14, 1997. D. Alexander, "World Disasters Report 2005: Focus on Information in Disasters," 2006. L. K. Comfort, "Crisis Management in Hindsight: Cognition, Communication, Coordination, and Control," Public Administration Review, 2007. S. Corbacioglu and N. Kapucu, "Organizational Learning and Self-Adaptation in Dynamic Disaster Environments," Disasters, vol. 30, pp. 212-233, 2006. N. Martin, "The Asian Tsunami: An Urgent Case for Improved Government Information Systems and Management," Disaster Prevention and Management, vol. 16, pp. 188-200, 2007. N. Kapucu, "Collaborative Emergency Management: Better Community Organizing, Better Public Preparedness and Response," Overseas Development Institute, 2008. M. W. Seeger, T. L. Sellnow, and R. R. Ulmer, Communication and Organizational Crisis. Westport, CT: Praeger Publishers, 2003. J. M. Kendra and T. Wachtendorf, "Elements of Resilience after the World Trade Center Disaster: Reconstituting New Year City's Emergency Operations Centre," Disaster, vol. 27, pp. 37-53, 2007. S. Jain, A. Malik, and R. D'Souza, "26 Countries Alerted in 15 Minutes, India not One of Them," in Indian Express Mumbai: Indian Express Newspapers Ltd, 2004. T. A. o. M. (TAM), "Report on the Failure of the Telecommunications Services as a result of the Asian Tsunami on 26 December 2004," Telecommunications Authority of Maldives (TAM)2005. W. Michalowski, S. Rubin, R. Slowinski, and S. Wilk, "Mobile Clinical Support System for Pediatric Emergencies," Decision Support Systems, vol. 36, 2003. D. C. Pattie, S. Dannenmann, and Y. Hagiwara, "Evaluation and Strengthening of Early Warning Systems in Countries Affected by the 26 December Tsunami," in Information Systems for Crisis Response and Management, 2008. C. Ryu, Y. J. Kim, A. Cahudhury, and H. R. Rao, "Knowledge Acquisition via Three Learning Processes in Enterprise Information Portals," MIS Quarterly, vol. 29, pp. 245-278, 2005. T. S. Raghu, A. Chaudhury, and H. R. Rao, "Business Process Change: A Coordination Mechanism Approach," Knowledge and Process Management, vol. 5, pp. 87-98, 1998. I. Nonaka, "A Dynamic Theory of Organizational Knowledge Creation," Organization Science, vol. 5, pp. 14-37, 1994. J. C. Spender, "Strategy Theorizing: Expanding the Agenda," in Advances in Strategic Management, P.

308

[34]

[35] [36] [37]

[38]

[39]

[40] [41]

[42] [43]

[44]

[45] [46] [47]

[48]

Shrivastava, A. Huff, and H. Dutton, Eds. Greenwich, CT: JAI Press, 1992, pp. 3-32. R. Chen, R. Sharman, H. R. Rao, and S. Upadhyaya, "Design Principles for Critical Incident Response Systems," Information Systems and E-Business Management, vol. 5, 2007. J. Douglas, Investigative Social Research Newbury Park, CA: SAGE Publications, 1976. E. J. Webb, D. T. Campbell, R. D. Schwartz, and L. Sechrest, Unobtrusive Measures. Chicago, IL: Rand-McNally, 1965. H. Chen, D. Zeng, H. Atabakhsh, W. Wyzga, and J. Schroeder, "COPLINK: Managing Law Enforcement Data and Knowledge," Communication of the ACM, vol. 46, pp. 28-34, 2003. H. Chen, J. Schroeder, R. Hauck, H. Ridgeway, H. Atabakhsh, H. Gupta, C. Boarman, R. K, and A. Clements, "COPLINK Connect: Information and Knowledge Management for Law Enforcement," Decision Support Systems, vol. 34, pp. 271-285, 2003. M. Turoff, M. Chumer, B. Van de Walle, and X. Yao, "The Design of A Dynamic Emergency Response Management Information System (DERMIS)," Journal of Information Technology Theory and Application, vol. 5, 2004. K. E. Weick, "Enacted Sensemaking in Crisis Situations," Journal of Management Studies, vol. 25, 1988. K. E. Weick, K. M. Sutcliffe, and D. Obstfeld, "Organizing for High Reliability: Processes of Collective Mindfulness," in Research in Organizational Behavior. vol. 21, B. M. S. a. L. L. Cummings, Ed. Greenwich, CT: JAI Press, 1999, pp. 81-123. J. Sweller, "Cognitive Load During Problem Solving: Effects on Learning," Cognitive Science, vol. 12, 1988. B. Britton and A. Tesser, "Effects of Prior Knowledge on Use of Cognitive Capacity in Three Complex Cognitive Tasks," Journal of Verbal Learning and Verbal Behaviour, vol. 21, 1982. Y. J. Kim, R. Kishore, and G. L. Sanders, "From DQ to EQ: Understanding Data Quality in the Context of E-Business Systems," Decision Support Systems, vol. forthcoming, 2003. R. Chen, "Consumer's Initial Acceptance of E-Commerce Website: A Contingency Approach," in AMCIS 2007, Keystone, CO, 2007. P. Anderson, "Information Technology: Where is It in the Coordination of Emergency Services," in Asia Pacific Police Technology Conference, 1991. G. E. G. Beroggi and W. A. Wallace, "Real-Time Decision Support for Emergency Management: An Integration of Advanced Computer and Communications Technology," Journal of Contingencies and Crisis Management, vol. 3, p. 18, 1995. S. S. Dawes, T. Birkland, G. K. Tayi, and C. A. Schneider, "Information, Technology, and

[49]

[50] [51]

[52] [53] [54]

[55] [56]

[57]

[58] [59] [60]

[61]

[62]

Coordination: Lessons from the World Trade Center Response," Center for Technology in Government, 2004. R. Chen, R. Sharman, H. R. Rao, and S. Upadhyaya, "Design Principles of Coordinated Multi-incident Emergency Response Systems," in IEEE International Conference on Intelligence and Security Informatics 2005, Atlanta, GA, 2005. J. J. Carafano, "Preparing Responders to Respond: The Challenges to Emergency Preparedness in the 21st Century," Heritage Lectures, vol. 812, 2003. S. F. Midkiff and C. W. Bostian, "Rapidly-Deployable Broadband Wireless Networksfor Disaster and Emergency Response," in The First IEEE Workshop on Disaster Recovery Networks (DIREN '02) New York, NY, 2002. T. X. Bui, "Towards a Theory of Shared Mental Model in CSCW," in 1992 Workshop on CSCW, 1992. D. Berge, The First 24 Hours: A Comprehensive Guide to Successful Crisis Communications. Cambridge, MA: Basil Blackwell, 1990. B. S. Manoj and A. H. Baker, "Communication Challenges in Emergency Response," Communications of the ACM, vol. 50, pp. 51-53, 2007. NCP, "Priority Telecommunications Services," Washington, DC: Department of Homeland Security, National Communications System, 2006. M. Turoff, S. R. Hiltz, M. Bieber, B. Whitworth, and J. Fjermestad, "Computer Mediated Communications for Group Support: Past and Future," in HCI in the New Mellennium, J. Carroll, Ed.: Addison Wesley, 2001. H. Chen, F. Y. Wang, and D. Zeng, "Intelligence and Security Informatics for Homeland Security: Information, Communication, and Transportation," IEEE Transactions on Intelligent Transportation Systems, vol. 5, pp. 329-341, 2004. t. Congress, "Health Insurance Portability and Accountability Act of 1996," in Public Law 104-191, T. Congress, Ed., 1996. BJA-DOJ, "NIEM Executive Briefing," D. o. Justice, Ed., 2007. J. Fedorowicz, U. J. Gelinas, J. L. Gogan, M. Howard, M. L. Markus, C. Usoff, and R. Vidgen, "Modeling Physical Barriers to Interorganizational System Implementation Success," International Journal of Information Technology and Management, vol. forthcoming, 2007. J. L. Gogan, C. B. Williams, and J. Fedorowicz, "Fatal Flaws in Information Sharing," International Journal of Technology, Knowledge and Society, vol. 1, pp. 93-100, 2005. C. B. Williams, J. L. Gogan, and J. Fedorowicz, "Public Safety and Cross-Boundary Data Sharing: Lessons from the CapWIN Project," IEEE Computer, vol. 38, p. 28, 2005.

309

[63] [64] [65]

[66]

[67] [68] [69]

D. Aylward and E. Jones, "Data Interoperability: Sharing Information for a Safer America." vol. 2006: COMCARE Emergency Response Alliance, 2006. DHS, "Fact Sheet: Achieving First Responder Communications Interoperability," in Department of Homeland Security Press Washington, DC, 2005. D. Frale, "Emergency Interoperability Consortium Announces Agreement with Department of Homeland Security to Promote Data Sharing During Emergencies," in PrimeZone Media Network, 2005. T. Harrison, J. R. Gil-Garcia, T. A. Pardo, and T. Fiona, "Learning about Interoperability for Emergency Response: Geographic Information Technologies and the World Trade Center Crisis," in The Thirty-Ninth Annual Hawaii International Conference on System Sciences, Hawaii, 2006. DOS, "National Incident Management System," Department of Homeland Security, Washington, DC2004. IBM, "World Community Grid," 2008. P. Currion, C. De Silva, and B. Van de Walle, "Open Source Software for Disaster Management," Communications of ACM, vol. 50, pp. 61-65, 2007.

310

Empowering Muslim Youth through Computer Education, Access, Use: A Gender Analysis Farida Khan and Rehana Ghadially Abstract – In the present information society, technical education has acquired immense importance as the Information and Communication Technologies (ICTs) hold potential for bridging socio-economic divides and empowering the marginalized such as women and minority groups. This paper explores the access and use of computer/Internet and examines how these affect empowerment levels among young men and women. This paper considers the psychological, social, educational and economic benefits following from computer education and usage of computer and Internet technology. Data was collected from 155 young girls (N=82) and boys (N=73) from three computer training centers in Mumbai. Statistical tests such as t-tests, 2-way ANOVA and chi-squares were computed to compare male and female subjects on empowerment and ownership/access and use. The figures for ownership and home Internet connection were low for the entire sample. Computer training centre and cyber café are important points of access for females and males respectively. Further, it is found that young women report higher gains from computer learning and technology use, hence, computer education can be a key gender equalizer. In light of the above, policy measures to widen access and provide subsidized training are suggested. Index Terms – access, computers/Internet, empowerment, gender

I. INTRODUCTION Today’s world is shaped by availability of information and ability to communicate, both of which are enabled through the rapid expansion of Internet and extensive use of computers. While there have been developmental efforts to increase the access and use of Information and Communication Technologies (ICTs) in general, there is still a marked difference in the technology access across different groups, resulting in the digital divide. Access considers equality of access to resources, in this case to computer and Internet technology. The uneven distribution of ICTs across the world as well as within societies gives rise to a digital divide; that results into two groups; namely, the information rich, those who have access to abundant information and the information poor, who lack such an access. Manuscript received September 22, 2008. Farida Khan is with Department of Computer Science and Engineering, Indian Institute of Technology Bombay, India (phone: 91-022-25764966, e-mail: [email protected]).

Rehana Ghadially is with Department of Humanities and Social Sciences, Indian Institute of Technology Bombay, India (e-mail: [email protected]) .

Even as India is a leading destination for outsourced IT and IT enabled work that promises employment opportunities, there are several divides within the country that leave the disadvantaged groups out of the development radar. The digital divide operates across geographic location (rural-urban), class (rich-poor), gender (male-female) and language (English-vernacular) [1]. While development effort in India has considered one or the other of these factors, they overlook the marginalized status of minorities such as Muslims. The Muslim minority is disadvantaged both educationally and economically, thus driving them off the information highway [2, 3, 4]. In addition, they face societal barriers such as limited awareness and discrimination [5]. Cost and language barriers compound the constraints further limiting the extent of computer and Internet use. Women are doubly disadvantaged as they face an additional gender barrier [6]. A coverage of ICT and India reports that ICT skills and access provide an opportunity to leapfrog and bridge the disparities in information, education and income [7]. Hence, we postulate that equipping the young members of the Muslim community, especially its women with the technical skills and access to the technology will result in a more balanced development of the country. In a developing country like India, ownership of personal computers and Internet penetration is limited at 2.5 and 4.2 per cent respectively [8]. Most users depend on shared access points, predominantly at educational institutions and workplaces. Three-fourths of Internet users in the country depend on community access points such as cyber café, telecenter, and information kiosk, with cyber cafes as the predominant public access model in urban areas [9]. It is estimated that there are around 50,000 cyber cafes used by almost 70 percent of Internet users [10]. However, the cost of surfing Internet at cyber cafes restricts the usage to the middle and high income groups [11]. In addition, research indicates that even when access is available, men are advantaged over women in the extent of computer usage [12]. In light of the above discussion on disparity in the access and use of the technology, the present study seeks to explore the different access points for the two sexes and assess gender differences in computer ownership, Internet connection at home, points of access and extent of computer/Internet use.

1

311

Education especially technical education is instrumental in expanding capacities and improving employment opportunities, thus resulting not only in economic but also personal and social empowerment such as enhancing confidence levels and social status [13]. Considering that computer skill is of immense importance in the technology driven society, this paper considers the gains resulting from having acquired basic computer education. Unlike the much emphasized economic gains, the present paper adopts a multidimensional perspective and considers four different facets of empowerment --psychological, social, educational and economic. In addition, it explores the access and use of computer and Internet among men and women and examines how these affect the different dimensions and levels of empowerment. In recent times, research in gender and development studies has focused on the concept of empowerment [14]. According to their theoretical background and parameters of assessment, empowerment is understood by different researchers in different ways. Those from the field of sociology focus on empowerment of the marginalized groups or society as a whole, and assess it in terms of political activism, advocacy and networking. Those from the field of economics focus on empowerment of the individual, group or economy as a whole and assess it in terms of improved income levels. Experts in gender studies consider women’s empowerment per se and elaborate on how altered gender relations would benefit women and assess it in terms of increased income opportunities, greater participation in decision making, social networking and advocacy. On the other hand, those from psychology and management studies concentrate on psychological gains and characterize empowerment in terms of personality (self-efficacy, internal locus of control), motivational (feeling enabled) and cognitive (meaning, competence, self-determination, impact) factors [15, 16, 17]. For the present study, a cognitive perspective of psychological empowerment is adopted. Hence, psychological empowerment is construed as --meaningfulness (commitment to the task), Competence (self-confidence), Self-determination (choice) and impact (making a difference). In the present information society, computers and Internet are the pervasive technologies that hold potential for equalizing the disparities in information, education and income [18]. With the growth of Information Technology enabled services (ITES) in India, a plethora of computer related jobs have opened up. Comparison of workparticipation rate of the ITES with other service sectors indicates that the field is more accommodating of women [19]. Research has highlighted the personal (self confidence) and social gains (status, increased connectivity) that come from ICT education [20, 21]. However, most of these studies are focused on women; hence a gender comparison does not follow. In addition, the literature is skewed towards the economic benefits thus

sidelining the other facets of empowerment. This paper attempts to provide a more complete view of empowerment by considering the psychological, social, educational and economic gains flowing from computer learning, technology access and usage.

Objectives of the study • • •

To explore gender differences, if any, in computer ownership, Internet connection, points of access and extent of computer/Internet use. To assess the empowerment flowing from computer education for males and females. To explore differences, if any, in empowerment among males and females with different levels of computer/Internet use. II. METHODOLOGY

Participants of the study comprised of 155 trainees (82 females and 73 males) from Mumbai, enrolled in a one year diploma in computer applications and multilingual desktop publishing offered by the Ministry of Human Resource Development, under the National Council for Promotion of Urdu language (NCPUL) scheme 1 . The course requires the trainees to complete a diploma in Urdu language along with the computer course. While theoretically open to anyone interested in the Urdu language, it is particularly the low income Muslim young who are drawn to it because it is highly subsidized. Hence, a large section of Muslim boys and girls who are well versed with Urdu language enroll for the course. Computation from the demographic profile revealed the following information about the sample. The average age of the subjects was 21.34 years (males = 20.46 years; females = 22.31 years) and their educational level varied from higher secondary (45.93 per cent) to undergraduates (22.67 per cent) and college graduates (31.40 per cent). Majority of the subjects (69.77 per cent) studied in Urdu medium schools, while less than one-third (26.74 per cent) had English as the medium of instruction. Majority (73.2%) of the sample belonged to low (monthly family income of less than Rupees 5000/-; US 1$ = 48.70 Rupees) and low middle income group (monthly family income of Rupees 5000/- - 10,000/-). 47.67% fathers were running their own business, 33.14% engaged in service and 19.19% had retired. 87.2% mothers were homemakers, while 12.79% engaged in home-based work

1 For a detailed description of the scheme, refer Rehana Ghadially and Farida Umrani (2004) IT Education: Initiatives among Mumbai Muslims”, i4D Information for Development, February.

312

or service. Participants report low educational levels of parents as nearly one-fourth (24%) of the fathers and half (44%) of the mothers had completed only primary education. The tools for data collection included standardized psychological scales to measure empowerment, a questionnaire to gauge access and use of computer/Internet and a demographic profile. Computer ownership was estimated by one question each for computer and Internet connection at home. Point of access was assessed by one question where subjects selected the places of access from a list provided to them. Extent of computer and Internet use was estimated in terms of the number of hours subjects spend on a computer/Internet in a week. In addition, a personal profile was estimated to get a few demographic details. Psychological empowerment was assessed by an adapted version of Spreitzer’s 12-item scale [17]. It has four subscales – meaningfulness (value of the task), competence (efficacy), self determination (choice) and impact (difference made). A sample item from the psychological empowerment scale is, ‘Learning computers has given me considerable opportunity for independence and freedom in how I do my daily tasks.’ The reliability co-efficients (internal consistency) range between 0.610.72. Only overall empowerment score is considered for analysis in the present study. For social, educational and economic empowerment, the scales were designed by the researchers. Social empowerment was measured by a 4item scale that tapped four aspects --- increased status, social comparison, being with the times and keeping in touch with friend and/or relatives. A sample item from social empowerment scale is, ‘Computer education makes me feel more up to date and current.’ Educational empowerment was measured by a 3-item scale that covered three aspects, information source on courses/colleges/universities, accessing online resources and preparation of class reports and presentations. A sample item is, ‘Learning computers has opened new ways to find information about different courses/colleges/universities for me.’ Economic empowerment was assessed by a five-item scale that tapped the following five areas --- opening of new job opportunities, application in family business, starting a computer related entrepreneurial venture, earning from home and managing family responsibilities with a job. A sample item of economic empowerment scale is ‘Learning computers has opened new job opportunities for me.’ The reliability co-efficients (Cronbach’s alpha) of the social, educational and economic empowerment scales were 0.66, 0.72, and 0.77 respectively. Each of the empowerment scale asked the subjects to indicate the extent to which they agree that learning computers has resulted in the following outcomes on a 6-point rating scale ranging from 1 = disagree to 6 = highly agree. The scales were pilot tested and in line with participants’ feedback, Hindi translation was provided along with the English version.

Permission was ought from the heads of three NCPUL computer training institutes, to conduct the study. Data was collected across twelve visits by the senior author from a group of 10-15 subjects, on completion of ten months of computer training.

Data Analysis Chi-squares and independent t-tests and were computed to compare male and female subjects on ownership/access, use and empowerment. Two 2-way ANOVAs were computed to gauge the influence of gender and differential level of computer and Internet use on the different dimensions of empowerment. III. RESULTS The study gathered information on computer ownership, Internet connection at home and points of access. It was found that 14.8 per cent subjects had a computer at home. This included 12.2 per cent females and 17.8 per cent males (χ2 =1.38, ns). 6.0 per cent of the sample (5.0 per cent females and 6.8 per cent males, χ2 =1.58, ns) had Internet connection a home. These findings indicate that although the difference between male and female subjects’ computer ownership and Internet connection is not significant, there is a trend in favour of males. The different places from where subjects accessed the technology beyond class hours at the training centre were explored. The computer institute where the subjects got their training and cyber cafe emerged as the most important points of access followed by home (own/relative/friend/neighbor) and workplace. A small group of subjects did not have access to computers beyond class hours. A gendered view of these findings indicates that females and males differ significantly on points of access (χ2 = 20.19, p<0.001). While the computer training institute emerged as a major access point for females (χ2 = 10.88, p<0.001; females 57.3 per cent and males 27.4 per cent), cyber café is more important for males (χ2 = 4.26, p<0.05; females, 12.2 per cent, males, 38.4 per cent). The other points of access --- home and workplace --- did not show any significant gender difference. 8.5 per cent females and 4.1 per cent males did not access computers beyond class hours (Table 1). The findings on the points of access indicate that computer training institutes and cyber cafes are the major access points for females and males respectively. Though not significant, the trend indicates that more males have access to home computers and twice as many females as compared to males have no access to a computer at all.

313

High use

31.0

χ2 df=2

Points of Access Computer institute Cyber café

24.5

38.4

12.2

4.26*

Home

23.2

27.4

19.5

0.44

Workplace

2.6

2.7

2.4

0

Do not access χ2 df =4

6.5

4.1

8.5

1.6

***

20.19***

p<0.005, *p<0.05

In addition to the above, the extent of computer and Internet use was also measured. On the basis of the number of hours in a week subjects worked on computers, beyond the class hours at the institute, they were classified into three groups --- no use (not at all), low use (less than 5 hours) and high use (5 hours or more). More than one-fourth of the sample did not use computers at all, two-fifth reported low use and less than one-third were in the high use category. The gender view indicates that female and male subjects differ significantly with regard to the extent of computer use. The number of females and males in no computer use category differed significantly (χ2 = 7.71, df = 1, p<0.01). Even when the women use computers beyond class hours, the extent of use is limited as compared to men. However, this difference is not significant. Thus, the findings provide evidence for the disadvantage of women participants (Table 2). TABLE 2 Extent of Computer and Internet Use for Males and Females Computer Use Internet Use χ2 χ2 df=1 df=1 T M F T M F No use

27.1 %

16.5

36.6

7.71* *

59.4

35.6

80.5

17.39**

Low use

41.9

47.9

36.6

0.38

33.5

52.1

17.1

11.08**

26.8

7.94** **

TABLE 1 Points of Access to Computers for Males and Females Total Males Females Gender (T) (M) (F) χ2 df =1 43.2 % 27.4 57.3 10.88***

35.6

0.32

7.1

12.3

2.4

4.45*

32.15***

p<0.01, * p<0.05

A similar classification was done with regard to Internet use. More than half of the sample did not use Internet at all, one-third surfed Internet for less than 5 hours and a small number surfed for more than 5 hours in a week. Female and male subjects differed significantly with regard to the extent of Internet use (χ2 = 32.15, df = 2, p<0.001). The two sexes differed significantly on each of the three categories of Internet use --- no use (χ2 = 17.39, df = 1, p < 0.01), low use (χ2 = 11.08, df=1, p < 0.01) and high use (χ2 = 4.45, df = 1, p < 0.01). 80.5 per cent females reported not using Internet at all, while the corresponding figure for males was 35.6 per cent. 17.1 per cent females and 52.1 per cent males report low Internet use; while 2.4 per cent females and 12.3 per cent males fall in the high Internet use category (Table 2). Thus, Internet use is significantly low for females as compared to males. Comparing computer and Internet usage of the two sexes indicated that the disparity in technology usage is more marked in case of Internet than the computer, thus pushing the women off the information highway. In addition to the above, the study estimated the levels of empowerment of male and female trainees. An estimate of composite empowerment was calculated by adding scores on the psychological, social, educational and economic empowerment dimensions and dividing it by four. To test whether the male and female subjects differ significantly on empowerment, independent sample t-tests were computed. The two sexes did not differ significantly on any of the dimensions of empowerment --- composite (t = 1.40, df =153), psychological (t = -0.93, df =153), social (t = 1.10, df =153), educational (t = -1.53, df =153) and economic (t = -1.27, df =153). Although no significant differences were obtained on the different dimensions of empowerment, the trend in all cases was for women to experience more empowerment than men. Basic computer education results in relatively higher composite empowerment for females (mean = 30.58, SD = 3.82) as compared to males (mean = 29.75, SD = 3.55). Similarly, females reported higher psychological (mean = 58.83, SD = 9.46) and social (mean = 21.44, SD = 2.99) empowerment than males (mean = 57.51, SD = 8.05; mean = 20.90, SD = 3.05, respectively). Following the trend, females reported relatively higher educational (mean = 16.17, SD = 2.83) and economic (mean = 25.89, SD = 4.33) empowerment as compared to males (mean = 15.53, SD = 2.27; mean = 25.04, SD = 3.93 respectively).

314

The interaction effect of gender and computer/Internet use on the empowerment was assessed by means of a two-way ANOVA (Tables 3 and 4). The combined effect of gender and computer use resulted in a significant difference in the composite [F (1, 151) = 8.02, p<0.01], psychological [F (1, 151) = 4.14, p<0.05], educational [F (1, 151) = 9.62, p<0.015] and economic [F (1, 151) = 6.12, p<0.01] empowerment. In other words, females reporting computer use obtained significantly higher scores on all aspects of empowerment as compared to males.

The trend of the scores indicated that females with computer use reported higher empowerment than males with computer use as well as females and males with no computer use (Table 5).

TABLE 3 TABLE 5 Results of Two-Way ANOVA Assessing the Effect of Gender and Computer Use on Empowerment t-test Scores of Male and Female Ss with Different Levels Dependent Source SS df MSS F of Computer Use on Empowerment Variable Composite Gender* 101.17 1 101.17 8.02** empowerment Computer use Computer Use Internet Use Emp (CE) Error 1905.34 151 12.62 M F t M F t df=111 df=61 Psychological Gender* 306.63 1 306.63 4.14* CE 29.61 31.71 29.63 31.70 -2.09* empowerment Computer use 3.22*** (PE) Error 11189.03 151 PE 57.39 61.07 -2.43* 57.23 61.19 -1.64 Gender* Social 25.04 1 25.04 2.82 empowerment Computer use SE 20.85 22.04 -2.22* 20.98 21.56 -0.70 Error 1340.46 151 8.87 (SE) Educational Gender* 58.17 1 58.17 9.62** EDE 15.43 17.02 15.19 17.00 -2.82** empowerment Computer use 4.26*** (EDE) Error 913.115 151 6.05 24.75 26.69 -2.61** 25.11 27.06 -2.03* Economic Gender* 101.84 1 101.84 6.12** ECE empowerment Computer use (ECE) Error 2512.89 151 16.64 **p<0.01;*p<0.05 (one-tailed) ** p<0.01, *p<0.05 A similar computation was done with respect to gender and Internet use. The interaction effect of gender and TABLE 4 Internet use resulted in a significant difference in educational Results of Two-Way ANOVA Assessing the Effect of empowerment [F (1, 151) = 4.40, p<0.05)]. The trend of the Gender and Internet Use on Empowerment mean scores indicates that females with Internet use reported higher empowerment than males with Internet use as well as Dependent Source SS df MSS F females and males with no Internet use (Table 5). The mean Variable scores of females (mean = 31.70, SD = 2.91) and males CE Gender* 21.69 1 21.69 1.59 (mean = 29.63, SD = 3.59) on composite empowerment were Internet use significantly different (t = -3.22, p<0.005). The mean scores Error 2065.48 151 13.68 on psychological, social, educational and economic PE Gender* 99.42 1 99.42 1.27 empowerment of the two sexes reporting computer use were Internet use found to be significantly different (t = -2.43; p<0.05, -2.22; Error 11791.48 151 78.09 p<0.05, -4.26 p<0.005, -2.61; p<0.01). Females reporting SE Gender* 0.02 1 0.02 0.002 computer use was significantly higher than males on each of Internet the dimensions (Table 5). Thus, technology use provides Error 1391.486 151 9.21 maximum benefits for women. EDE Gender* 28.90 1 28.90 4.40* ECE

*

p<0.05

Internet use Error Gender* Internet Error

IV. DISCUSSION

990.60 11.80

151 1

6.56 11.80

0.68

2607.009

151

17.26

1.27

A. Computer Ownership/Internet Connection, Access, Computer/Internet Use

315

The study assessed computer access and usage of the technology in terms of family ownership of computer, having an Internet connection at home and the various points of access to computer/ Internet. Computer ownership and home Internet connection for the sample is considerably lower as compared to ICT penetration in Mumbai where the study was conducted. The city ranks first in PC penetration with 32 per cent household owning a PC [22]. Research indicates that Mumbai and Delhi together constitute close to 70 per cent of the total installed computers in Indian homes and 46 per cent of the computer owners have an Internet connection [12]. Thus, even though the figures of computer ownership and Internet connection for the sample under study are better than the national average (2.5 and 4.2 % respectively [8]), they are much lower than the figures for Mumbai city. The sample consisting of college students and fresh graduates, have no personal disposable income to purchase a computer. Besides 80 per cent of the subjects in the study belonged to low income group and hence cannot afford a personal computer or Internet connection. In addition, due to low education levels of the parents, there is lack of awareness about the potential of the Internet. Hence, even when they purchase a computer to support education of their children, less than half of the sample has an Internet connection. Besides, technology access at home is adversely affected by space constraints. Households in Mumbai utilize an average of 2.9 square meter of floor space per person, one of the lowest in the world [23]. Given the low socio-economic levels of the sample and the pigeon-hole like houses of Muslim ethnic enclaves of Mumbai, the disadvantage for the young members of the community with regard to computer ownership and Internet connection follows. The two sexes did not differ significantly on computer ownership and home Internet connection. This is because majority of the sample belong to low income groups and are financially dependent on their families. Hence, both boys and girls were affected by economic constraints. However, the trend indicates that males are slightly advantaged as compared to females. There is lack of gender segregated data on computer ownership, hence stretching a point, a study on difference in family computer usage is considered. A study on computer usage in India indicates that 82 per cent of males compared to only 16 per cent of females are the primary users of family computer [12]. Besides, Indian families tend to make higher investments in education of the boys compared to girls due to different role expectations [24]. As computer is identified as an educational tool, it is likely that families would purchase computers for their sons rather than daughters. In addition, males identify the home computer as personally owned; while females identify it as belonging to the head of the family [25]. Cost barrier is more prominent in case of Internet connection as compared to computer ownership, in terms of recurring monthly payments. In addition, due to low education levels (one-fourth of the fathers and half of the mothers

had primary education or were illiterate), parents may not be aware of the potential benefits of Internet. Besides private access, the different points of access were explored. It was found that significantly higher number of females access computers at the training institute. Access at the institute does not require cost investment, but does not provide Internet access either. Further, the women will be left vying for access after completion of the course. Cyber café is significantly more important point of access for males than females. This difference may be explained by three reasons --- nature of use, cost and socio-cultural attitudes. Males seek to use the Internet, which the computer training centre does not provide. In urban India, cyber cafes are the predominant public access model for Internet usage. Even though the cost of surfing in the cyber cafes has reduced from Rupees 50 to Rupees 20 per hour in the last decade, this is still out of reach of the low income groups. Young girls may hesitate to use family resources for surfing the Internet, thus curtailing their visit to cyber cafés. Third, sociocultural restrictions add to the disadvantage of young women as families impose sanctions against cyber cafes as these are characterized as boys’ hangouts and lack female attendants [26]. Besides, families may be more willing to provide financial support to the boys for accessing technology, thus placing them at an advantage as far as access at cyber café is concerned. Workplace does not emerge as an important access point as the sample consists predominantly of college students and fresh graduates. Only six participants (4 men, 2 women) were working in part time jobs such as giving home tuitions, office assistant, etc. Thus, despite possessing the technical skills, the women from the current sample have lower access to the technology as compared to their male counterparts. Limited Internet connection at home and cyber café usage seriously disadvantages the women as research indicates that Internet provides women access to information, greater participation in decisionmaking and networking, thus empowering them [18]. Subjects were asked the number of hours they used computers/ Internet in a week, besides the regular class hours at the training institute. On the basis of their responses, the usage was classified into three levels --- nouse, low-use and high-use a week. A similar classification was done with regard to Internet use. Two issues are apparent with regard to computer/ Internet use; one, the disparity across technology --- computer v/s Internet; and gender --- males v/s females. While a quarter of the subjects fell in no-use category for computer use, with regard to the Internet, this rose to 60 per cent. One-third of the subjects reported high computer use but only 7.1 per cent report high Internet use. These figures bear testimony to the fact that computer use and Internet use are distinct categories and merging them into a single unit could be misleading. While using computers requires only access to the machine, for Internet use the cost gets multi-fold due

316

to investments in modems, optical fibers, charges of Internet service providers and computer maintenance. This may explain the lack of Internet usage. The other reasons for limited Internet use are the non-availability of subsidized public access points, limited English language competency and socio-cultural attitudes. Though cyber café is the main access point in urban India, the cost of usage is beyond the reach of low income groups. Research indicates that one-third of Internet users in Indian cities have an income above Rupees 25,000/- , while only 4.4 percent of the Internet users have an income below Rupees 5000/- [11]. Access at cyber cafes is beyond the reach of low economic groups. As majority of the sample are poor, they have little disposable income that could be used for Internet access. Research indicates that there is a strong relationship between the use of the Internet and ability to speak English [9, 27]. Surveying a group of students from Gujarat, it is observed that students educated in the vernacular medium at school struggled at college where the medium of instruction is English [28]. As 70 per cent of the sample in the present study had Urdu as the medium of instruction in school, their limited English language skills probably affected their Internet use. Although considerable web content is now available in a variety of Indian languages; a cursory review of the Urdu websites by the first author in 2007, revealed that they focused on literature, poetry and news rather than education and employment. Hence, tangible gains do not follow for the subjects surfing these websites. B. Computer Education, Use and Empowerment In addition to computer access and use, the study focused on analyzing gender differences in empowerment levels following from technical skills training. It was found that basic computer education yields similar empowerment gains for males as well as females. There is no research comparing men and women on empowerment flowing from computer education hence, the explanations offered are speculative in nature. Lack of gender differences on empowerment could be due to four reasons. First, the sample consisted of young, college going or fresh college graduates from a metropolitan locale on the verge of entering the job market. The homogenous nature of the sample makes for similar sources of empowerment such as marketable skills. Second, empowerment as assessed by the scales focuses on the cognitive level and reflects the subject’s estimate of the possible gains from computer education. It does not assess the tangible benefits flowing from technical skills, thus overlooking constraints in the real world that affect the actual benefits of computer education. Third, the item pool of the different empowerment scales is narrow and specific to this sample. It is likely that if a more diverse measure was adopted, the two sexes would indicate differential empowerment levels. Fourth, men and women score differentially on the items on the scales, thus canceling out the gap. This will be discussed in the following

paragraphs. Although the two sexes do not differ significantly on all measures of empowerment, the trend in all cases is for females to be more empowered than males. This is supported by research that indicates that though modernization and technological development initially disadvantaged women, ICT promises a technological Uturn [29]. Findings of this study are strengthened by evidence from a qualitative study of the impact of computer education that indicates that more women than men experience higher boost in self-confidence and secure computer related jobs [30]. It is likely that since South Asian women in general are socio-economically deprived, their expectation from something new will be higher than males who have better chances of being exposed to these technologies before hand. Analyses of the participants’ response on psychological empowerment indicate that females expressed slightly higher meaning (mean =14.91, SD =2.76), self-determination (mean = 14.83, SD = 3.18) and competence (mean = 15.11, SD = 2.57) as compared to males (mean = 14.78, 14.01, 14.48; SD = 2.32, 2.59, 2.58 respectively). In other words, computer education resulted in creating a sense of new possibilities for involvement, independence and self-confidence for women. On the other hand, males (mean = 14.23, SD = 3.00) reported slightly higher impact than females (mean =13.97, SD = 3.28), indicating that computer education provided them possibilities for enhanced control over their environment. Thus, the psychological gains for females and males focus on micro and macro levels respectively. Women reported relatively higher gains on all aspects of social empowerment --- feeling contemporary, enhanced status, social comparison, and connectivity --- as compared to males. This is in line with research that women perceive family gain, communication and social gain as the first, second and fifth most important benefit of personal computer use [20]. In addition, the social connectivity aspect is supported as women are responsible for maintaining family and kin relations [31]. Women report relatively higher educational empowerment than men with higher scores on information on courses/colleges/universities, accessing online journals/books/reports. This is corroborated with the research that when females use computers, they do so mainly for educational purposes; unlike males, who use it for general purposes or playing games [11]. More than one-third females have completed college; hence search for educational information follows. Males score slightly higher on the use of computers to prepare reports and presentation, may be because unlike the women, more than half of the men are in college. Females reported slightly higher scores on four of the five aspects of economic empowerment --- job opportunities, managing family responsibilities with work, earning from home and assisting in family business. There is a general perception

317

in urban India that a combination of college degree and computer skills enhances the possibility of employment. Since a higher number of women participants had already completed college, they felt better equipped to find a job. The economic gains for females focus on the potential of IT for combining family with work, a gain which ensures maintenance of traditional values and yet being productive. On the other hand, males reported a higher possibility of setting up personal business. This may be because belonging to business families (47%), males may be socialized to start their own business. Besides, families could be more willing to provide them the financial capital due to future role expectation. Looking at computer/Internet use and empowerment, it was found that significantly more females (more than one-third) fell in the no-use category of computer use, than males (one-fifth). Similarly, significantly more females (more than 80 per cent females) than males (one-third) report no Internet use. Although both computer and Internet use was relatively limited for females, they were more disadvantaged with regard to Internet use. This is a cause of concern, as most of the research that make a case of ICT-led empowerment of women, focus on the Internet technology and its ensuing benefits like networking, political activism and ecommerce [14, 32]. Women’s low Internet use in this sample is in line with previous findings. In a survey of Macau residents, it was found that Internet users were more likely to be males [33]. In a survey of Indian Internet users, it was found that females start using cyber cafes, a common Internet access point in urban India later; use them less often and for shorter duration per session [11]. Gender and extent of computer/Internet use significantly influence empowerment levels. Findings indicate that with the exception of social empowerment, the interaction effect of gender and computer use is significant for composite and the other three kinds of empowerment --- psychological, education and economic. The mean empowerment scores of males and females indicate that females reporting computer use are more empowered than males with computer use as well as females and males without computer use. Thus, the relevance of computer use for empowerment of females cannot be overstated. It is in line with the research that emphasizes psychosocial, educational and economic benefits of computer access and use for women [34, 35]. A study of six projects of the Information for Development Program (infoDev) found that computer skills increased self-esteem and promoted self-confidence of the young women participants [18]. Similarly, a study on novice women computer learners found that computer skills resulted in increased self-esteem and enhanced status [20]. The sample of the present research comprised of young college students; hence, educational benefits of using computers like preparing class assignments or presentations seem relevant. Further, females using computers understand that this technology offers an

avenue to become economically independent. Even though the advantage of females in social empowerment is not significant, the trend is in the same direction. Media advertisements of personal computers in India, project the women computer user as a modern, westernized individual; cues that highlight the social status and modernity associated with this technology [36]. A computer being associated with masculinity, its usage brings the females on an equal social footing, enhancing feelings of being contemporary and moving with the times. The interaction effect of gender and Internet use was found to result in significant difference on educational empowerment, with females using Internet obtaining highest mean empowerment scores as compared to their male counterparts. This is not surprising as research indicates that when women used technology, they do so for educational purposes unlike males who use it for general purposes or playing games [11]. The sample included higher number of females who had completed college; hence they are likely to explore opportunities for continuing education using Internet. Thus, young Muslim women derive educational benefits out of using Internet. With regard to other aspects of empowerment, it was found that there was no significant interaction effect of gender and Internet use. As far as psychological and social empowerment is concerned, the trend is in favor of females. This is in line with previous research that Internet use increases self-confidence, facilitates networking and expands income generation capacity [14]. There is limited comparative evidence on the differential benefit to males and females from computer and Internet usage. In light of this, the present research fills an important gap by providing empirical evidence for the benefits of computer education and technology usage for empowering the two sexes. However, more research with gender as an analytic variable is necessary before a clearer picture emerges. V. CONCLUSION AND POLICY RECOMMENDATIONS Five major conclusions can be drawn from the study. First, computer ownership and Internet connection is very low for the sample compared to figures for Mumbai city. Hence, the minority poor as other urban poor are on the wrong side of the digital divide. Second, computer training centre is the major access point for females, while cyber café emerges as the chief access point for males. The primary access model in urban India is cyber café and is underused by women. Related to this, the third conclusion is that when women do have access to computers and the Internet the extent of usage is limited as compared to males. Thus, the poor urban minority women are off the information highway. Fourth, technology use empowers females more than the males, thus making a strong case for ensuring sustained access and usage. The fifth important conclusion is that computer education is a

318

key gender equalizer as unlike other technologies women’s engagement with new technologies open’s new possibilities and empowers them as much as the men. In light of the above conclusions, policy recommendations are suggested. In order to widen access points and encourage technology use amongst those at the edge of the information society, community access points based on the tele-center model should be set up in urban locales with public-private partnerships. Government can provide tax benefits to business houses investing in such endeavors. Besides, to improve women’s access, cyber cafés can reserve special timings, provide separate space for girls, appoint female support staff and open women only cyber cafes. This will help the young women to overcome the inhibitions in visiting these places for technology access. In addition, civil society organizations can create awareness about potential benefits of technology use and facilitate higher familial support for technology use by women. In order to maximize tangible gains from technical skills the computer training center has to be diversified. The center can liaison with the local small and medium enterprises to provide placement support to its trainees. It can serve as an intermediary to secure microcredit to its students, with special concessions for girls. These finances can be used for purchase of a computer and setting up computer-related business such as a cyber café. Further, it can assist the trainees to form a self-help group to pool their resources and secure credit for computerrelated income generation. Urban Muslims being largely a petty business community, their business spirit may assure sustainability. What is required is the initial thrust from an external agency. It is likely that the impact of these policies can be delimited due to societal conditions such as discrimination and prejudice. Hence, fair disbursement of micro-credit and sensitization to overcome discrimination will help to widen the gains flowing from technology. These measures will enable the Muslim youth to compete on an equal footing and lead to their mainstreaming in India’s economic growth. As far as directions for future research are concerned, a more comprehensive study can yield representative results. The findings of the study are limited in generalizability as data was not collected from locations other than Mumbai and across socio-economic class. A comparative sample from other minority groups may provide better insight into the standing of Muslim youth on the information highway.

REFERENCES

[1] K. Kenniston, “The Four Digital Divides,” in Bridging the DigitalDivide: Lessons from India, K. Kenniston and D.K. Das, Eds. New Delhi, India: Sage, 2003. [2] Sachar Committee Report, “Social, Economic and Educational Status of the Muslim Community of India,” Prime Minister’s High Level Committee, Government of India, 2006. [3] A. Razaack and A. Gumber, “Differentials in Human Development: A Case Study of Empowerment of Muslims”, Programme of Research on Human Development, New Delhi, India: National Council of Applied Economic Research, 2003. [4] A. Shariff and A. Razzack, “Communal Relations and Social Integration,” in India Social Development Report, Council for Social Development. New Delhi, India: Oxford University Press, 2004. [5] O. Khalidi, “Ethnic Group Recruitment in the Indian Army: The Contrasting Cases of Sikhs, Muslims, Gurkhas and Others,” Pacific Affairs, vol. 74, no. 4, pp. 529-552, 2001. [6] Z. Hasan,. and R. Menon, “Unequal Citizens: A Study of Muslim Women in India”, New Delhi, India: Oxford University Press, 2004. [7] R. Heeks, “Social Outsourcing: Creating Livelihoods”, Information for Development (i4d), September, pp. 1719, R. (2006. [8] IMRB “International Market Research Bureau,” 2005. Available: http://www.imrbint.com [9] A. M. Haseloff, “Cyber cafés and their Potential as Community Development Tools in India,” The Journal of Community Informatics, vol. 1, no. 3, pp. 53-65, 2005. [10] A. Pasricha, “To Get Online, Most Indians go to Cyber cafés,” VOA News Service, 2004. Available: http://apiap.blogspot.com/2004_10_01_apiap_archive.htm l [11] A. Haseloff and R. Ghadially, “Gender, Cyber cafés and Internet Access: Bridging the Digital Divide?” in Urban Women in Contemporary India: A Reader, R. Ghadially, Ed. New Delhi, India: Sage, 2007, pp. 270283. [12] A. Venkatesh, “Computers and New Media Technology in Indian Households: Based on a Study of Eight Major Cities in India,” Centre for Research on Information Technology and Organizations, 2000. Available: http://www.crito.uci.edu/noah/paper/IndianSurvey1.pdf [13] S. Huyer, “Gender, ICT and Education,” 2003. Available: http://www.wigsat.org/engenderedICT.pdf [14] S. Mitter, “Globalization, ICTs and Economic Empowerment: A Feminist Critique,” GenderTechnology and Development, vol. 8, no. 1, pp. 5–29, 2004. [15] J. A. Conger and R. N. Kanungo, “The Empowerment Process: Integrating Theory and Practice,” Academy of Management Review, vol. 13, pp. 471–482, 1988.

319

[16] D.L. Corsun and C.A. Enz, “Predicting Psychological Empowerment among Service Workers: The Effect of Support-Based Relationships” Human Relations, vol. 52, no. 2, pp. 205-224, 1999. [17] G. M Sprietzer, “Psychological Empowerment in the Workplace: Dimensions, Measurement and Validation,” Academy of Management Journal, vol. 38, pp. 1442– 1465, 1995. [18] N. Hafkin, “Are ICTs Gender Neutral? A Gendered Analysis of Six Case-studies of Multi-donor ICT Projects,” Background paper to UN/INSTRAW Virtual Seminar Series on Gender and ICTs, 2002. Available:http://www.uninstraw.org/en/docs/gender_and_ict/Hafkin.pdf. [19] G. Kelkar, “Women in the Digital Era: Gender and Culture,” Paper presented at the International Conference on Women in the Digital Era: Opportunities and Challenges. Organized by Annamalai University and Software Technology Parks of India, at Chidambaram, India, 2004. [20] F. Umrani and R. Ghadially, “Empowering Women through ICT Education: Facilitating Computer Education,” Gender, Technology and Development, vol. 7, no. 3, pp. 359-377, 2003. [21] N. Hafkin and N. Taggart, “Gender, Information Technology, and Developing Countries: An Analytic Study,” Academy for Educational Development, for the Office of Women in Development, Bureau for Global programs, Field Support and Research, US Agency for International Development, 2001. [22] Dataquest, “Momentous years of Indian InfoTech”, Dataquest, vol. 18, no. 1, 25-28. [23] A. Bertaud, “Mumbai FSI Conundrum: The Perfect Storm: The Four Factors Restricting the Construction of New Floor Space in Mumbai,” 2004. Available: http://alainbertaud.com/AB_Files/AB_Mumbai_FSI_ conundrum.pdf [24] K. Chanana, “Interrogating Women’s Education: Bounded Visions, Expanding Horizons,” New Delhi, India: Rawat Publications, 2001. [25] R. Heimrath and A. Goulding, “Internet Perception and Use: A Gender Perspective,” International Journal of Human-Computer Studies, vol. 55, pp. 919-938, 2001. [26] F. Umrani and R. Ghadially, “Economic Impact of IT Education among Muslims,” i4D Information for Development, 2006. [27] V. Vehovar, Z. Batagelj and K. Lozar, “Language as a Barrier,” 1999. Available: http://www.isoc.org/inet99/proceedings/3i/3i_3.html

[28] V. Ramanathan, “Ambiguities about English: Ideologies and Critical Practice in Vernacularmedium College Classrooms in Gujarat, India,” Journal of Language, Identity, and Education, vol. 4, no. 1, pp. 45–65, 2005. [29] A. Goyal, “ICTs and the Technological U-turn for Women,” in Urban Women in Contemporary India: A Reader, R. Ghadially, Ed. New Delhi, India: Sage, 2007, pp. 255 – 269. [30] F. Umrani, “Computer Adoption among Mumbai Muslim Youth: Empowerment and Impact Assessment”, Unpublished Dissertation, Indian Institute of Technology Bombay: India, 2007. [31] R. Ghadially, “Women’s Friendship,” in Encyclopedia of Women and Islamic Cultures, S. Joseph, A. Najmabadi, J. Peteet, J. Siapno and J.Smith, Eds. Netherlands: Leiden, EWIC, pp. 194195, 2003. [32] S. Huyer and M. Carr, “Information and Communication Technologies: A Priority for Women,” Gender, Technology and Development, vol. 6, no. 1, pp. 85-100, 2002. [33] W. H. Cheung, “Internet Adoption in Macao,” Journal of Computer Mediated Communications, vol. 7, no. 2, 2001. Available: http://jcmc.indiana.edu/vol7/issue2/macao.html [34] S. Huyer and T. Sikoska, “Overcoming the Gender Digital Divide: Understanding ICTs and their Potential for Empowerment of Women,” Virtual Seminar Series on Gender and ICTs, United Nations International Research and Training Institute for the Advancement of Women (INSTRAW), 2003. Available: http://www.uninstraw.org/docs/gender_and_ict/Synth esis_Paper.pdf [35] N. Primo, “Gender Issues in the Information Society,” United Nations Educational, Scientific and Cultural Organization (UNESCO) Publications for the World Summit on the Information Society, 2003. Available: http://portal.unesco.org/ci/en/file_download.php/2505 61f24133814c18284feedc30bb5egender_issues.pdf [36] R. Ghadially and S. Ranganathan, “Gendering Computer Marketing: A Study of Print Media Advertisements in India,” Asian Women, vol. 22, no. 1, pp. 69-96, 2006.

320

1

eServices Provisioning in a Community Development Context Through a JADE MAS Platform Mamello Thinyane, Alfredo Terzoli, Peter Clayton Department of Computer Science, Rhodes University {m.thinyane, a.terzoli, p.clayton}@ru.ac.za

Abstract—A growing proliferation of ICT4D interventions has necessitated the exploration of innovative solutions for the provisioning of eServices in rural, marginalized communities. The challenges currently faced in these interventions include: situating the developed applications within the cultural and ethnographic context of the target communities, integrating greater levels of granularity and flexibility within the applications for increased context sensitivity, handling the intermittence and instability of supporting infrastructural services. These are the challenges that we address in the context of ICT4D intervention undertaken in a rural community in the Eastern Cape province of South Africa. We explore the design and implementation of a Multi-Agent System (MAS) for this community as a platform for provisioning of context-sensitive eServices, and highlight some observations with regards to the applicability and adequacy of the solution. Index Terms—eServices platform, ICT for development, ethnocentric applications, JADE MAS

I. I NTRODUCTION A number of factors are fueling the adoption of ICT as an enabler towards community development, poverty alleviation and the elimination of the digital divide. One of these is the realization of the benefits that are afforded to a community from participating in the global information and knowledge society, and the other is the sometimes mis-applied observation of the penetration of ICT and technology in the developed first world countries. The latter factor often leads to the technological determinism flaw, in which the ICT adopters incorrectly assume that the technology that works in the first world countries will necessarily work in third world contexts. The third world ICT4D interventions are undertaken in a context of pronounced socio-economic challenges that are often not prevalent in first world contexts. They are also undertaken in contexts of extreme rurality characterised by lack of infrastructure (roads, electricity, telecommunications), lack of service provisioning (health, education, government services), and high levels of marginalization (information, economic). It is therefore generally within this landscape of rurality that ICT interventions are explored to facilitate development and enhancement of life in rural communities. This environment of rurality presents unique challenges and opportunities. Five of these challenges that have been identified by Drabenstott and that are critical in shaping the rural economic outlook are: tapping digital technology, encouraging entrepreneurs,

leveraging new agriculture, improving human capital and sustaining the rural environment [1]. The adoption of ICT in rural communities is a phenomenon with pronounced variability and one which is still not thoroughly understood [2]. It is therefore still not very clear why some rural communities adopt ICT and why others are sometimes even resistant to the integration of ICT into their lives. However, research has found that some of the leading reasons for rural communities adopting ICT are more social than economic. These communities adopt ICT for improvement of their well being as far as access to information is concerned and as far as facilitating communication in the communities, and to that end the widely used applications of the Internet in these communities are email applications and search engines [3]. ICT is indeed providing opportunities for dealing with rural poverty and in many ways it is challenging the traditional paradigms of doing business, delivering services to citizens, and running societal institutions [4]. The dynamics associated with ICT in development are two fold: on one side its able to leap-frog community development based on the ability of the society to synergistically embrace the technology (e.g. the Asian economies: Hong Kong, Taiwan, Singapore) and on the other side, ICT can further the marginalization, participatory exclusion from global economies and retardation of communities that are not able to integrate technology into their lifesystems [5]. Technology, in and of itself is not a panacea for the underdevelopment woes of communities, it is however a pre-requisite for social development in this day and age [5]. Technology is not a target towards community development and social wellbeing [6], but rather a tool for facilitating the achievement of a desirable features in a society: well being, health, peace, and communality. To a large extent human activity depends on information and therefore a synergestic interaction of technology and information leads to a competitive advantage for societies [5]. One of key contributions to understanding the dynamics associated with the adoption of ICT in rural communities and in development contexts is made in the discipline of ethnocomputing. One of the primary tenets of ethnocomputing is the realization of the culture specific influences on computing and subsequently on the Internet [7]. The aspects of computing that are universal and that are aligned with the general human nature, are easily transplantable from one

321

community to another. Aspects that are cultural however, need to be considered, adapted and validated for adequacy within the environment where they’re being implemented [8]. ICT for development should therefore take into consideration the ethnographic considerations and expressions of a community to avoid the technology determinism flaw that has plagued many ICT4D projects [9]. The ability to adapt and localize ICT solutions into the different contexts as far as the cultural expressions, the language, the metaphors, the themes, the interaction modalities, and the associated knowledge are concerned, is becoming a key requirement for ICT4D interventions. This requirement factors directly into the sustainability and the successful integration of the solution into the life-system of the communities. Another requirement is for the solutions to address issues of infrastructure intermittence and unstable deployment environment that are typical in rural marginalized communities. This fact of intermittence necessitates that measures and mechanisms are integrated at all the levels in the service provisioning stack, to be able to recover from interruptions, and to tolerate faults in the supporting infrastructure. The other constraint as far as handling faults and interruptions in the network is concerned is the lack of technical expertise in these communities. As far as possible the services should automatically recover and self-heal without the need for a technical expert’s intervention. In this paper, we discuss a solution that we have implemented particularly to address these requirements in ICT applications. The next section introduces the project that we are undertaking, highlighting the different eServices applications being developed for the community. We then discuss the design factors that should be taken into consideration for context-sensitivity, followed by the detailed discussion of the Multi-Agent System (MAS) we have developed. The last sections provide some observations that have been made with regards to the adequacy and applicability of the platform in development contexts. II. ICT4D I NTERVENTION IN SA The ICT4D landscape in South Africa is one in which there’s a growing effort from the government, nongovernmental organizations and academic entities in undertaking community development interventions based on ICT solutions. It is within this context that a joint ICT4D project between Rhodes University and University of Fort Hare was initiated in 2006, with support from government and industry, towards exploring ICT-based interventions that can be undertaken in rural marginalized communities. A. Overview and Current Status The site of this project is a deep-rural, marginalized region of Dwesa, which is located in the former homeland of Transkei. Dwesa has an estimated population of about 15000 distributed in 2000 households. Dwesa is predominantly a subsistence farming community that characterizes the majority of rural communities in South Africa and other developing regions. Typical of many rural communities, there is a lack of infrastructural services in Dwesa: gravel roads, electricity

2

only at a few of the locations in the community, no centers of governmental services provisioning, a small under-resourced clinic and poor access to information resources for the schools and the community members at large. The initial objective of the project was to implement a prototype of an eCommerce portal in support of the entrepreneurship activities already prevalent in this community. This has been revised and enlarged to establish a distributed community communication and knowledge platform. Over the three years that the project has been running a number of activities have been undertaken and different infrastructure setup: 1) Community rapport: One of the key achievements in the project has been establishing a working relationship with the community as far as exploring ICT based solutions. The buy-in from the different stakeholders (the headman, the schools, the ICT ’champions’) in the community provides the essential support that ensures a synergy out of working together with the community. It is on the basis of this relationship that the project is transitioning into a Living Lab where the community members become the active (and key) participants in undertaking the implementation of solutions. 2) Training: Prior to the introduction of the project in this region, there was minimal computing literacy in the community. We have successfully run a number of training sessions, with the schools as the centers for training. The training that we’re undertaking is based on the openICDL curriculum and one of the training objectives is ’confidence’ not ’competence’. What this means is that we aim to get the community confident enough to learn on their own and thus encouraging a culture of life-long learning in the community. The schools have become the points of access to the ICT solutions for the community because they’re some of the few places where there’s electricity. The schools are also traditionally the centers for learning within the community, and as such we have managed to tap into that inertia to establish a training system in which we ’train the trainers’ who then continue training the other people in the community who are not as yet computer literate. We have currently setup ’labs’ in five schools in the region ranging in size from 2 computers to 15 computers. 3) WiMAX backbone: We have setup a WiMAX network between the schools where we have established computing labs. WiMAX is a suitable solution in this case due to the landscape of the region, and the distances between the different points. It also provides adequate bandwidth for the services that are provided within the network. 4) VSAT connection backhaul: The link to the Internet is via VSAT, which is setup in one of the schools and shared between the rest of the schools through the WiMAX network. This also provides a link back into the network for remote management and troubleshooting. 5) VOIP telephony: An Asterisk PBX has been setup at one of the schools to provide VOIP telephony services

322

for the whole community. In one way, this allows the community and the schools to collaborate a lot easily with each other, and in another way it also allows for the provisioning of voice based services for the community which is a relevant alternative in a community where there are people who are illiterate. The established infrastructure and activities undertaken thus far provide the basis on which to explore the provisioninig of different, relevant network services for this community. B. Associated eServices This section highlights a few of the key eServices that are being integrated into the network in Dwesa. This is therefore not an exhaustive list of all the different services deployed. Some of the ones not discussed here include: a local Wiki which the community can populate and access, access to offline versions of Wikipedia and Project Gutenberg, eJudiciary service portal for the community, a community help-desk system. These are the services that will be ported onto the JADE MAS eService platform to allow for a greater contextualization and sensitivity to the local environment. 1) eCommerce service1 : The Dwesa community has a high economic potential due to the fact that there is a nature reserve and arts and craft entrepreneurship activities in the region. There’s therefore the eco-tourism and cultural-tourism potential that can be explored and activated through the eCommerce portal. The current arts and craft entrepreneurship is also currently operating within the geographical constraints of Dwesa and the eCommerce portal opens it up to wider international markets. A basic functional eCommerce portal has been setup and is operational. This portal however has the limitations that it is not extensively contextualized to the community. The aspects under consideration for integration in the subsequent versions include: implementing an ontologies based back-end for a semantic processing of the underlying eCommerce knowledge; allowing for alternative interfaces into the portal which would be more appropriate for the community (e.g. voice-based interface for some of the illiterate arts and craft entrepreneurs, and a J2ME mobile phone interface which would make the portal more accessible for the majority of the users); and a more integrated handling of different associated multi-media content on the eCommerce portal. 2) eGovernment service2 : Government service provisioning is very minimal in Dwesa. This eService portal aims to facilitate the ease of access to government services. To a large extend this portal interfaces and aggregates the available government services to provide a one-stop shop for the community. The activities that can be undertaken through this portal include: downloading application forms (grants, birth certificates), accessing information released by the government, reporting matters to the police, and engaging in an open discussion forum around government related matters. The similar improvements to the eCommerce portal will also 1 http://www.dwesa.com 2 http://www.dwesa.com/egov

3

be afforded by porting this service onto the MAS eService platform. 3) eHealth service: One of the primary objectives of the eHealth services platform is the preservation and aggregation of the local traditional health knowledge in Dwesa. This is the knowledge that has been developed over the years and that is available in the community. The portal also provides accessibility to other sources of medical knowledge. While the portal is aimed for use by the average members of the community, it can also specifically provide the extra knowledge resources for the clinics in Dwesa. 4) Indigenous Knowledge portal: One of the reason for the slow adoption and integration of ICT into the lives of rural communities is the lack of relevant knowledge on the Internet. This portal aims to alleviate this by allowing for the community members to author, create and record the knowledge that is relevant and local to their region, to establish and enforce authorization permissions on ’their’ knowledge, and to have access to other knowledge that is available of the portal. This of all the services requires the greatest levels of flexibility and context-sensitivity due to the heterogeneity of the underlying knowledge, the diversity of user profiles and preferences, and the multiplicity of end-user devices that are utilized. III. D ESIGN C ONSIDERATIONS The development of a MAS eService platform has to address a number of challenges. Of particular significance are the challenges that pertain to rurality, processing of indigenous knowledge, keeping abreast of technological developments around web 3.0, and also ensuring ethnocentricity and context sensitivity of the developed solution. The design has to take into consideration a new way of structuring and layering the service platform to achieve the following goals: 1) Provision of an end-user device agnostic interface to the underlying knowledge - provision for handling heterogeneous device requests 2) Allowance for varied interaction modalities with the users 3) A context-senstive and ethno-centric knowledge platform - through encapsulation of local knowledge and the emulation of local knowledge system dynamics. This is towards achieving a seamless mobility between knowledge exchange in the real world and knowledge exchange in the virtual space (on the knowledge platform) 4) Handling of multimedia knowledge and media transcoding depending on the capabilities of requesting devices 5) A future-proof platform that embraces the developments in knowledge engineering technologies and service oriented architectures IV. T HE AGENT-BASED P LATFORM Application development platforms and architecture are constantly evolving to accommodate the needs within the software development domain. Evident characteristics in the nature of applications that are being developed, is that there’s

323

Figure 1.

4

A distributed JADE MAS platform Figure 2.

an increasing heterogeneity and therefore a need to operate in a context of multiplicity of protocols, languages, technologies and programming paradigms. Application development architectures are also harnessing the developments as far as distributed networking is concerned, and this has resulted in the implementation of service composition architecture implemented on the web, and also in the implementation of services that incorporate the development from the semantic web community. The provisioning of an architecture for distributed systems in heterogeneous usage contexts is one that has been pursued extensively in literature [10, 11]. One architectural paradigm that has been explored extensively is the web services architecture. The other perspective to problem solving in heterogeneous environments is through agents. Agents are different from web services in that they are [12]: problems solvers; proactive; goal-oriented; context-aware; and autonomous. One of the predominantly implemented agent standards is the IEEE Foundation for Intelligent Physical Agents (FIPA) standard, the other common standard is the ”Multi Agent Facility” (MAF) standard which is provided by the Object Management Group (OMG) [13]. A. JADE MAS JADE is the abbreviation of Java Agent DEvelopment framework and is a middle-ware and a software framework for the development of agent based applications. JADE is developed by Telecom Italia Lab (TILAB) and is written entirely in Java [14]. JADE implements the FIPA agent standard and provides a platform that can be distributed across different platforms and machines. Within the JADE MAS platform, agents exist within containers, which can be located on different hosts. Therefore to be part of community, agents have to exist in containers that join a single Main Container to form a distributed system that appears as a single platform. All the agents that are associated with a Main Container form a single agent platform. The Main Container is different from other containers in that it holds two platform specific agents: the Agent Management System (AMS) and the Directory Facilitator (DF) [14]. The eServices platform MAS is implemented as a single

Multi-host eService platform MAS container

agent platform with containers distributable across different hosts (Fig 1). Fig 2 is a depiction of the eServices platform MAS through a Remote Management Agent (RMA) showing the Main Container on one host (thepentagon) which contains most of the platform agents including the AMS and the DF, and also another container located on another host (thebox) and containing the Multi-Modal Interaction Agent (MMIA).

B. Platform Agents This eServices platform has been realized as a system of independent agents that collaborate and communicate for the achievement of the common goal. The agent architecture fits within the PIASK architecture to provide decoupled, modular, units (agents) of computation that exist within a community [15]. We developed the PIASK architectural pattern as conceptual layering of aggregated application functionality for multimodal, multimedia application deployment in heterogeneous contexts. PIASK provides a distinct categorization of functionality into presentation, interaction, access, social networking and knowledge base layers [16]. 1) Hyper-Text Transfer Protocol Agent (HTTPA): Access to the PIASK platform is facilitated by the HTTPA access layer agents whose role is to implement the specific transport level protocols for communication with end-user devices. The access layer agents provide the ability to handle the multiplicity of the end-user devices in an extensible manner and in a way that is separate from any other operational structures of the platform. The HTTPA handles request from web browsers but also handles requests from the openVXI VXML browser, which runs as an Asterisk application [?]. The HTTPA in essence provides a web server service that receives requests from any HTTP compliant browsers.

324

Figure 4.

Multi-Modal Interaction Agent

The MMIA is also responsible to managing the user sessions on the platform. This is maintained through an in-memory list of the current sessions. For every session on the platform, the MMIA queries the Social Networking Agents (SNA) to determine the relevant users details, and stores this in the session list. The details that are queried from the SNA include, the IM address of the user and the online status, the telephone extension of the user, and their email address. This information is utilized by the MMIA to determine the best channel of interaction with the users. For example, if a user is logged onto platform through a VOIP agent, and requests a download of a certain non-audio file, the MMIA agent will determine that the best channel to send the file through is via an email. As mentioned already, the MMIA is the primary agent that implements procedural logic within the eServices platform. This it does by defining various contexts of operation to which each request is forwarded. The following table (Table I) gives a summary of the different contexts that have been defined within the platform and which can be extended to accommodate for the definition of various other network services. MMIA context login.piask

Figure 3.

Hyper Text Transfer Protocol Agent

Besides handling the HTTP requests from web browsers, HTTPA also handles openVXI VXML browser requests from Asterisk. This provides an alternative access from Asterisk and also a richer set of constructs available within VXML mark-up language which are not available in AGI scripting. 2) Multi-Modal Interaction Agent (MMIA): The MMIA interaction agent plays a central role in the implementation of application service logic within the knowledge platform. It provides the logical interface between the platform users and the underlying platform knowledge services, and between the different access layer agents and the knowledge base layer agents. The key modules within the MMIA (Fig 4) collaborate to provide an end-to-end logical handling of the user requests. The request handler module is the first to receive the request from the access agents. This module first unpacks the message payload to strip off internal inter-agent platform messages. These messages encapsulate system information such as User Session IDs, request context, and variables required for the processing of the request. The AAA module is primarily responsible for the tasks of: authenticating the users on the eServices platform, by interfacing with the Social Networking Agent which is aware of the different platform users’ profiles and authentication credentials; determining and enforcing authorization for various requests on the underlying knowledge in the system; and finally for maintaining basic audit logs for the requests and processes that are handled on the platform.

5

description

this is the context for handling login requests logoff.piask this handles the request for logging off know.piask handles all the requests for knowledge on the platform update.piask handles the updating of knowledge on the platform and adding new knowledge do.piask handles requests to perform platform related operations - send an email, call an extension, notify a user of a message, etc. upload.piask handles uploading of content register.piask provides the logic for new user registrations delete.piask handles deleting of content Table I MMIA CONTEXTS

While these contexts are not an exhaustive specification of all the functionality implementable in the platform, they form the core of the procedural logic for operation on the underlying knowledge base repositories, which facilitate the provisioning of different services. The MMIA also implements the basic handling of different interaction modalities with the platform. The MMIA interacts directly with the access layer agents to receive the requests from the users. These multiple channels, existing as the various access layer agents, provide the interaction input and also are available as output channels for communication with the users 3) Social Networking Agent (SNA): The social networking agent plays the pivotal role of positioning the eServices (and knowledge) platform within the social context and environment by integrating the existing social dynamics within the community. The role of social networking agents is pivotal in the knowledge platform towards the overall objective of

325

6

requests that are sent from the MMIA to the knowledge base layer are sent directly to the KBA, which processes the request, to determine the appropriate agent to forward the request to. The KBA therefore provides an interface and an aggregation of the knowledge handling agents. The following are the modules within the KBA that enable it to achieve its role of managing the interaction with the rest of the knowledge layer agents (Fig 7):

Figure 6.

Knowledge base layer agents

achieving both a context sensitive and culture sensitive platform. The platform’s interface with the human agents (which in one role is the user of the system, but in another sense also an active participant in the life of the community of agents) is primarily through the social networking agents The SNA interacts primarily with the MMIA agent from which it typically receives requests associated with the users of the platform. One of such interaction is the request from the MMIA to process user authentication, or to determine contact details or the specific profile information of the users.

Figure 7.

Figure 5.

Social Networking Agent

The SNA encapsulates two key modules, the request processor module, and the user’s profile module (Fig 5). The user profile module directly interfaces with the underlying knowledge base layers to query the user profile ontologies. 4) Knowledge Base Agent (KBA): One of the core layers of the platform is the knowledge base layer, which is the predominant end-point of most requests that are handled on the platform. The different eService portals decompose down into the different units of knowledge that have to be accessed and that users interact with. While most of the knowledge that is handled is encapsulated in ontologies, there is also a large amount of information that is stored in other types of knowledge bases: legacy storage systems, knowledge encapsulated in folksonomies, DBMS based knowledge, and native file system based knowledge. The required flexibility and applicability on the knowlege platform, necessitates the implementation of agents that naturally handle the different knowledge bases (Fig. 6). The Knowledge Base Agent (KBA) acts as the manager agent for the rest of the knowledge base layer agents. All the

Knowledge Base Agent modules

The request analyzer receives a natural language request from the MMIA (possibly initially from the XMPPA agent which implement XMPP Instant Messaging interface into the platform. The XMPPA is not discussed in this paper), which gets stripped down into basic tokens by the NLP Analyzer. The search engine identifies the associated ontology, the subject within that ontology, and where available the required predicate. The request dispatcher then sends the formatted (by the request writer) request to the knowledge agents. Once a response is available from the knowledge agents, it get normalized to a standard content format, and then send back to the requesting MMIA. 5) Media Presentation Agent (MPA): The MPA primarily handles the heterogeneity of the content that is communicated with the platform. MPA implements a number of renderers in order the achieve the transcoding of content between different media formats and mark-up languages. The MPA is also responsible for handling interface localization in terms of allowing for customization of themes, and the translation of platform interface strings into specific languages. The role of the presentation agents is therefore to render content at two levels: the device level and the user level. At the device level, the content is formatted according to the

326

7

part of the FIPA specification. A communication protocol is implemented on top of this message system, for exchanging platform specific messages between the agents. The FIPA specification provides a number of features to facilitate communication between agents and for differentiating the messages processed: sender specification, conversation ID, replyWith string and message performative [18]. Within the platform, these features are used for fine grain identification of specific messages between the agents (Fig 10).

Figure 8.

Media Presentation Agent

requesting device capabilities (e.g. mark-up language, screensizes) and at the user level, the content is rendered according to the user preferences (e.g. UI language, aesthetics, layout, colour). The themes engine is utilized specifically for web browser based requests. The engine utilizes Cascade Style Sheets (CSS) as the theme definition language for the web pages that are rendered by PIASK. The localization engine is specifically responsible for translation of UI components into different languages. The engine utilizes the standard GNU Gettext Portable Object (.po) files which are loaded in memory (into a HashMap) when the MPA agent is initialized. The in-memory storage of the translations allows for faster processing of requests, by eliminating the file read and string comparison operations on the .po file. The utilization of the .po files also provides the added benefit of being easily editable by humans, allowing for easy localization of the platform. C. Agent Interaction The execution of end user requests is handled by different agents that communicate with each other and exchange messages. An example interaction is depicted in Fig 9 in which an access layer agent (HTTPA) receives a request from a user and then forwards the request to the MMIA. The MMIA handles the request and determine the next path of execution, in this specific case, the request is of a user authentication on the platform, and so the MMIA sends the request to the SNA agent. The response from the SNA agents is send back to the HTTPA, which then contacts the MPA agent that it is associated with in order to get the content rendered according to the requesting device capabilities and the requesting user’s preferences.

Figure 10.

Platform message exchange

The critical communication channels between agents are encrypted within platform. One of such links is the MMIA to SNA channel. The integrity, confidentiality and accuracy of this channel messages is important for the following reasons: 1) Authentication on the system is executed through messages passed from the MMIA to the SNA since the SNA is the only agent that has access to user’s credentials and preferences. 2) Authorization on the platform, which is undertaken by the MMIA, is done in dialog with the SNA which has the relationship types (see section x) that define the different authorization levels. 3) New user registrations are also handled through this channel by the MMIA. The platform utilizes an asymmetric encryption based on the RSA algorithm, with a 1024 bytes key size. The encryption public keys are pre-shared between the agents, and the associated Cipher objects are initialized during the agent setup sequence. The encryption occurs as a last stage before the content is added to the ACLMessage, and the decryption is the first process after the content has been read from the incoming message. V. O BSERVATIONS

Figure 9.

Example platform interactions

The JADE MAS platform inherently provides an Agent Communication Language (ACL) messaging mechanisms as

The following preliminary observations have been made, around the applicability and the adequacy of this eService platform in Dwesa: 1) The distributed nature of multi-agent systems provides an architecture closely aligned to the infrastructural

327

situation on the ground in Dwesa (and in most rural marginalized communities). The resources in Dwesa are distributed between different communities who share and cooperate with each other in a symbiotic manner. The eService platform can easily be distributed depending on the availability of the resources that the agents have to access. Thus the agent that provides or handles connectivity to the Internet, can be located on a gateway host in one school, and an agent that interfaces with the Asterisks PBX can be co-located with the PBX. 2) The eService platform is robust and fault-tolerant. The reality of rural communities is that the availability of different network nodes is not always guaranteed, sometimes due to the underlying infrastructural constraints (e.g. lack of electricity). The fault-tolerance within the eService platform is established through a mechanism in the JADE MAS that allows for migration of agents between different hosts on a network, and the automatic initialization and termination of agents available on the platform. The platform is therefore able to self-heal in cases where agents get terminated when a host goes down. There is also the ability for the replication of the Main Container within the JADE platform, which allows for the continuance of a MAS life in the case of one Main Container being terminated. 3) The eService platform has been validated for functional adequacy through a load testing experiment, in which the platform was observed to provide a linear scaling in performance for handling users’ requests. The performance degradation, under heavy usage loads, was also observed to be gradual. 4) The interaction modalities implemented within the eService platform provide a crucial interface that has been lacking through a number of the solutions that have been implemented in rural communities where there’s an especially higher levels of illiteracy. One such interface is the voice interface to the eService platform, provided through the available VOIP telephone infrastructure in the community. Not only is the voice interface crucial for the illiterate people, it is also import due to the nature of the knowledge that is available in these communities and also that the majority of these communities have traditionally audio based communication mediums (e.g. the African drum). These observations position the implemented eService portal to provide a solution that meets the needs and the requirements that are prevalent in rural, marginalized communities. These are preliminary observations with will be validated through the extensive usability testing and usage profiling. VI. C ONCLUSION The ICT4D research and implementation landscape is one faced with unique challenges which necessitate new and innovative solutions. We have highlighted challenges which are associated with: the fact of cultural diversity in which the ICT4D interventions have to be undertaken, and the intermittence of the infrastructure services in rural marginalized

8

communities. The developed solutions have to be implemented with a high level of flexibility to allow for situating within different cultural and environmental contexts. The services deployed on these platform also have to be relevant to the local communities and to integrate the local (indigenous) knowledge and systems dynamics with these communities. In addressing these challenges in a ICT4D context is a community in South Africa, we have developed a JADE based MAS that provides an architectural support for the provisioning of eServices for this community. The observations made thus far allude to the applicability of such a platform in the provisioning of context sensitive, ethnocentric eServices. R EFERENCES [1] Drabenstott, M., “New Policies for a New Rural America”, International Regional Science Review, vol. 24, no. 1, pp. 3, 2001 [2] Mitchell, S. and Clark, D., “Business adoption of information and communications technologies in the two-tier rural economy: some evidence from the South Midlands”, Journal of Rural Studies, vol. 15, no. 4, pp. 447–455, 1999 [3] Malecki, E.J., “Digital development in rural areas: potentials and pitfalls”, Journal of Rural Studies, vol. 19, no. 2, pp. 201–214, 2003 [4] Bhatnagar, S., “Information Technology and Development Foundation and Key Issues”, Information and Communication Technology in Rural development: Case Studies from India. World Bank Institute, pp. 1–12, 2000 [5] Castells, M., “Information technology, globalization and social development”, United Nations Research Institute for Social Development, 1999 [6] Hietanen, O., “The global challenges of e-Development - from digital divides towards empowerment and a sustainable global information society”, in Seminar of global perspectives of development communication, University of Tampere, 2004 [7] Tedre, M., Sutinen, E., Kahkonen, E. and Kommers, P., “Ethnocomputing: ICT in cultural and social context”, Communications of the ACM, vol. 49, no. 1, pp. 126–130, 2006 [8] Thinyane, M., Dalvit, L., Terzoli, A. and Clayton, P., “The Internet in rural communities: unrestricted and contextualized”, in ICT Africa Nepad Council, Addis Abba - Ethiopia, 2008 [9] Thinyane, M., Terzoli, A. and Clayton, P., “Transitions towards a knowledge society: Aspectual pre-evaluation of a culture-sensitive implementation framework”, IFIP – Learning to live in the knowledge society, July 2008, Vol. 281, pp 271-278 [10] Cybenko, G., Gray, R., Khrabrov, A. and Wu, Y., “Information theoretic principles of agents”, in Proceedings of the CIKM - Workshop on Intelligent Information Agents, Third International Conference on Information and Knowledge Management (CIKM 94), 1994 [11] Leymann, F., “Web Services: Distributed Applications without Limits”, Business, Technology and Web, Leipzig, 2003 [12] Payne, T. R., “Web Services from an Agent Perspective”, IEEE Intelligent Systems, vol. 23, no. 2, pp. 12-14, 2008 [13] Leszczyna, R., “Evaluation of agent platforms”, European Commission, Joint Research Centre, Institute for the Protection and Security of the Citizen, Ispra, Italy, Tech. Rep., June, 2004 [14] Bellifemine, F., Poggi, A. and Rimassa, G., “JADE–A FIPA-compliant agent framework”, Proceedings of PAAM, pp. 97–108, 1999 [15] Thinyane, M., Dalvit, L., Slay, H., Mapi, T., Terzoli, A. and Clayton, P., “An ontology-based, multi-modal platform for the inclusion of marginalized rural communities into the knowledge society”, Proceedings of the ACM SAICSIT conference on IT research in developing countries, pp. 143–151, 2007 [16] Thinyane, M., Dalvit, L., Terzoli, A. and Clayton, P., “Towards a Model of an Ontology Based, Multi-Modal and Multimedia Knowledge Portal for Marginalized Rural Communities.”, in Proceedings of IEEE Information Communication Technologies International Symposium, Fez Morocco, 3rd - 5th April, [17] Eberman, B., Carter, J., Meyer, D. and Goddeau, D., “Building voiceXML browsers with openVXI”, Proceedings of the 11th international conference on World Wide Web, pp. 713–717, 2002 [18] FIPA, ACL, “Message Structure Specification”, Foundation for Intelligent Physical Agents, 2000

328

Extending the Technology-CommunityManagement Model to Disaster Recovery: Assessing Vulnerability in Rural Asia Arul Chib Nanyang Technological University [email protected]

A.L.E Komathi Nanyang Technological University [email protected]

Abstract—The recent increase in natural disasters has a significant impact on the lives and livelihoods of the poor in Asia. The spread of information communication technologies (ICTs) in this region’s rural areas suggests the potential of technologies to enhance recovery efforts. While many ICT initiatives have been implemented to aid disaster management, from providing early warning to immediate relief, there exists a gap in the theoretical understanding of the role of technologies in disaster recovery and rehabilitation. We propose a conceptual framework for understanding the implementation of ICTs in recovery operations, drawing attention to vulnerability reducing potential of the initiatives. We review theories on ICT use in disaster management, and propose the Extended Technology-Community-Management model focusing on vulnerability assessment for the design and implementation of ICT programs for development in rural areas. We illustrate this model using case studies from ICT deployments in post-disaster Asia, particularly India, Indonesia, Sri Lanka and China, and suggest implications for theory and practice. Index Terms—Asia, disaster rehabilitation, vulnerability.

management,

ICTD,

I. INTRODUCTION Globally, more than 250 million people are affected by natural disasters every year, and the intensity and frequency of catastrophes have steadily been increasing in the last decade [1]. Asia is amongst the most affected and vulnerable regions, with disasters such as earthquakes, floods, tsunamis, cyclones, and droughts killing thousands of people each year, while millions have lost their homes, properties, livelihoods and families [2]-[4]. The increase in the proportion of people living below the poverty line, from 30% to 50% in post-tsunami Indonesia, suggests the impact of calamities on the poor, especially those living in less developed Asian countries [5], [6]. These marginalized groups face the challenge of coping with, and recovering from the effects of disasters as overall development and economic growth suffer major setbacks in the aftermath [7]-[10]. The recent spread of information communication technologies (ICTs) in Asia [11], [12] offers more people access to technologies that can to aid in disaster management efforts [13]-[15]. There has been a steady rise in the penetration, accessibility, and use of ICTs in Asia. India, in particular, continues to be one of the fastest growing major mobile telecom communication markets in the world, with annual growth rate of 91% [5]. With China’s 253 million internet users [16], Asia has a

15.3% internet penetration rate [17], growing from 11.8% in 2006 [5] to over 500 million internet users [17]. One should note there is a disparity in access to these technologies, as captured in the digital divide debate. Further, urban dwellers have more economic, educational and technological familiarity with these technologies compared to the rural population [18]. Despite this divide in technology access and use, the use of ICTs in disaster warning, mitigation and management shows the benefits of technology diffusion in rural areas. An examination of the literature suggests that ICTs are being used beyond merely facilitating early-warnings about impending disasters. The role that ICT systems, which range from mature technologies such as radio, television, and land-line telephony, to advanced modern technologies such as Global Positioning Systems (GPS), Geographic Information System (GIS), Very Small Aperture Terminal (VSAT), and cellular phones and satellite communications, play at various stages of disaster management has been recognized by academics, governmental agencies, civil society groups, nongovernmental organizations (NGOs), and voluntary welfare organizations in the development arena [19]-[22]. Communication technologies provide access to information that is a vital form of aid in itself. “Disasteraffected people need information as much as water, food, medicine, or shelter” [23], particularly in the post-disaster context, from immediate relief to long-term rehabilitation efforts. This suggests the potential of ICTs in various operations immediately following a disaster, from enabling prompt information dissemination to relief agencies and affected communities, to livelihood rebuilding initiatives in the recovery phase. Researchers have focused on the use of these technologies for disaster response in the mitigation and preparedness stages [24][29], with less emphasis placed on the scope of ICTs in the rehabilitation stage with the objective of exploiting the long-term and continual benefits of technologies, particularly in rural communities. There is a need to establish a canon for best practice [7]. This study aims to bridge the existing theoretical gap by proposing a conceptual framework to guide the design and implementation of ICT for development (ICTD) programs, specifically in post-disaster management. In this paper, we critically examine and categorise existing theories in disaster management and ICT. We integrate the theoretical implications of this framework in relation to the Technology-Community-Management (TCM) model [18], [30]. We propose an extended model

329

and illustrate it using primary data from India and Indonesia, and cases from China and Sri Lanka. II. REVIEW OF MODELS AND FRAMEWORKS We conducted a structural understanding of existing conceptual frameworks on disaster management (DM) and ICT usage in order to assess their theoretical impact and practical effectiveness to technology in disaster recovery (see Fig. 1). The ICT-DM framework identifies specific factors that are emphasised consistently in these studies, to draw two broad categorizations – ProcedureBased and Community-Oriented theories. ProcedureBased theories cover the temporal segmentation of disaster management and ICT initiatives, and illustrate the management structures involved. CommunityOriented theories provide insights into practically relevant issues involving the community, and suggest impediments to productive technology influence. These categories are further divided to guide the systematic review of concepts within each category, and to aid analysis in recognizing the commonalities between theories. The objective of the review is to identify gaps in theory and focus, and to propose a theoretical model that fills this research gap.

Fig. 1. Review of Disaster Management and ICT Models/ Frameworks.

A. Procedure-Based Theories Procedure-Based theories describe the processes involved in effective management of disaster and ICT projects by pointing out the key stages in planning recovery efforts, as well as identifying the stakeholders and management implications to technology implementations. The theories emphasize the importance of informational flow in influencing economic and psychological rehabilitation in recovery. While these theories are useful in understanding enabling factors, they largely fail to address the importance of community involvement, or identify barriers to success for ICT initiatives in post-disaster situations. We sub-divide Procedure-Based theories – stages and systems theories. While stages theories classify linear processes and define their scope, systems theories demonstrate what needs to be done, by whom, and the methods to use for optimal information flow. Stages theories illustrate the management of disasters in a longitudinal approach, where both disaster-based and ICT-specific frameworks craft the different stages as mutually exclusive levels of impact. By mapping a

structure of the steps that need to be taken at each level, the stages theories chart an overview of the levels through which ICT users and disaster operation managers successively progress. The longitudinal approach was initially adopted by the eight socio-temporal stages of disaster framework [31]-[33] that proposed the classification of the disaster management efforts according to specific outcome strategies, such as warning, impact, rescue, and recovery. This argument was further expanded in models, such as the six-phase disaster cycle [34], [35], [4], three stages of disaster [36], three timedimensions of rehabilitation efforts [14], and strategic disaster management cycles [37], that elaborated the relationships between specific temporal parameters and objectives guiding these efforts. Relating to practical situations, the influence-impact model [8] emphasized the importance of accurate, timely, and complete information outflow for basic level ICT impact to occur. Furthermore, the ICT step change table [25] identified three levels within projects, where the informational needs of people, such as knowledge of technology, literacy level, and skill capacity of users, determine the hierarchal levels of ICT projects. We argue that while stages theories are useful in highlighting the hierarchy of actions and measures necessary prior to, during, and after disaster contexts; they fail to specify potential barriers to the efforts implemented at each stage, particularly in ICT deployment. It is difficult to draw distinct boundaries in processes where users, especially those affected by disasters, possess varying, and often overlapping informational needs, and capacities to process information. In these cases, the boundaries between stages can be fuzzy, and stage-related behavior can be concurrent [38]. Systems theories present factors affecting information flow in recovery efforts, arguing that ICT providers in disaster contexts have to be critical of the information content disseminated to affected people. They draw attention to the management of recovery projects by pointing out ways in which governments and NGOs make critical investments in ICT projects with the support of the private sector. This illustrates the essential financial roles external bodies play in ensuring the economic livelihood of disaster affected people. Systems theories such as the CARICOM structure [39] and disaster management consortium [40] identified stakeholders involved in decision-making and coordination of disaster operations, while the optimal disaster information flow model [39] and the technical conceptual system model [41] map the roles information plays in disaster management. We suggest that systems theories can expand beyond coordination of information and interactions by drawing attention to the action-oriented information exchanges that highlight the combined efforts of stakeholders. An action-oriented strategy relates to the management style adopted by technology managers who look beyond profit making, and encourage community level participation in the management of ICT deployments. We argue that the deliberate attention to community involvement and ownership in running programmes would expand the boundaries of systems theories to look beyond traditional organizational styles in governing ICT initiatives.

330

Together, the stages and systems theories suggest that ICTs can be best capitalized in the recovery stage, where emphasis is on the psychological and physiological rehabilitation of affected communities. Most stages theories contend that ICT rehabilitation efforts, in the areas of physical and social rehabilitation, and health and educational rehabilitation, should focus on multi-year programs. This emphasizes the need for well-planned long-term recovery efforts to ensure that people regain their economic and psychological balance. Further, these models suggest that fostering of economic livelihood should be based on recognizing principles of social inclusion and gender equity, especially those involving fishing and farming communities. On the whole, procedure-based stages and systems theories describe the linear dimensional design and practical management aspects useful in planning and implementation of ICTs in disaster situations. B. Community-Oriented Theories Various studies have argued that particularly at the recovery stage, plans need to be understood, accepted, and implemented at the local level to be sustainable, and to achieve intended outcomes on communities in postdisaster situations [42], [43]. Community-Oriented theories emphasize various aspects of community involvement in disaster management; essentially, elaborating the existence of susceptibility factors that impact recovery efforts aimed at affected people. While the theories connect various practical applications recognising community needs and wants in disaster recovery, they generally ignore the technical demands of these recovery efforts, where over-emphasis had been placed upon providing a range of applied methods to enhance community well-being instead of elaborating on the know-how of running these recovery initiatives. We split Community-Oriented theories – vulnerability, and capacity theories. Vulnerability theories describe various aspects of vulnerability that both exist and arise in post-disaster contexts. Conversely, capacity theories look at ways to improve the capabilities of people in order to optimize ICT usage and benefits. Affecting people with varying magnitudes, vulnerability has been defined as the “characteristics of a person or group, and their situation that influence their capacity to anticipate, cope with, resist, and recover” from the impact of a disaster [44]. With insufficient capacity to cope, the poor are most vulnerable in the aftermath of disasters [9], [24]. Vulnerability theories analyze the role ICTs play in recovery efforts aimed at reducing vulnerabilities of the rural poor, by increasing capacities and focusing on livelihood enhancing development efforts. The theories discuss aspects of informational, socio-cultural, and psychological recovery that significantly impact economic vulnerability. In this study, we look at five inter-related vulnerability theories; (1) development-disaster framework, (2) capacities and vulnerabilities analysis, (3) pressure and release model, (4) access model, and (5) sustainable livelihoods approach. The development-disaster framework [45] highlighted the importance of information in assessing vulnerabilities, arguing that the lack of, or poor access to, information

could increase community vulnerabilities in post-disaster situations. Among the initial debate [46]-[49], it is suggested that the design of development programs should focus on decreasing vulnerability of communities to disasters and their negative consequences [50]. Supplementing this, the capacities and vulnerabilities analysis [46] provided disaster managers a framework for understanding and reducing vulnerabilities, and suggested an approach for the design and evaluation of development projects. Similar to the pressure and release model [44], and access model [44] proposed that reducing vulnerabilities involves increasing capacities of the affected people at the individual, group, and societal levels, arguing that well-organized and cohesive communities can withstand or recover from disasters better than those divided by issues of race, religion, class, caste, gender, ethnicity, or age. Reference [51] criticized the sustainable livelihoods approach [52] framework for failing to acknowledge socio-cultural determinants that influence poor people’s access to economic-enhancing resources and livelihood assets. This model emphasizes the impact of socio-cultural factors on technology adoption and use, critical in determining the equal access of an individual, group, or community to ICT resources. We suggest that vulnerability theories need to examine factors impeding community livelihood, access, and developments, and recognize ways to improve the overall community participation in vulnerability reduction. The ICT skills and discretionary slack [53] framework, a capacity-oriented theory, highlighted the importance of capacity building in vulnerability reduction by proposing a correlation between ICT skill level, and availability of time. It fails to account for the fact that the factors of skill and time are affected by both psychological and physiological determinants of users. This would affect the adoption, use, continuation, and termination of ICT deployments. For instance, discretionary time available could be affected by conditions such as the amount of training sessions offered, and cultural, gender-based and social restrictions. The psychological component affecting ICT skill level could include individual literacy levels, attitudes towards ICT adoption, and overall willingness to learn. Overall, this theory suggests that information processing capacity and access to information is dependent on the socio-cultural determinants as well as the mental state of individual to absorb and use the information. We contend that the vulnerability and capacity theories highlight the mutually dependent relationship between reducing vulnerabilities and increasing capacities in enhancing overall community involvement in postdisaster ICT use. However, we think community involvement in ICT initiatives need to go beyond mere participation, and explore opportunities for community empowerment through ownership, and active contribution in the management of ICT efforts directed towards recovering the livelihoods of the rural poor. III. EXTENDED TECHNOLOGY-COMMUNITYMANAGEMENT MODEL

The examination of disaster management and ICT literature allowed us to develop a framework for the

331

design and implementation of ICTs in disaster recovery. Procedure-Based theories raised key components in the planning and management of technology operations by illustrating various management styles and approaches involving key stakeholders in the financial planning and execution of ICT efforts. The Community-Oriented concepts provided insights on the potential of communities to participate in critical livelihood enhancing initiatives brought about by ICTs. We found that the prevailing concept of vulnerability needs to be considered in technology implementations. This analysis draws our attention to the main aspects to take into consideration in ICT planning, namely the community involvement, the management components, the overall design of technologies, and evaluation of existing vulnerabilities. Encapsulating these aspects, we propose the use of the Technology-Community-Management (TCM) model [18], [30] in this study. The TCM model proposes that the three key intersects of ICT characteristics - technology, management, and community, will lead to sustainable ICT for development (ICTD) interventions. Reference [30] argued that the technological design is the combination of software and hardware components, while management of a project requires an understanding of the financial requirements, establishment of key partnerships, and a regulatory environment. A key ingredient to success is the community involvement and participation in programs. The Community component of the TCM model was sub-divided into dimensions of basic needs of the community, modes of ownership of ICT investments and profits, and training of community users both in the use and in technology management. Here, financial and social sustainability are dependent on the replicability and scalability of projects. The TCM model presents a comprehensive illustration of the issues surrounding ICT implementations in rural areas, condensing the key conditions affecting each dimension, and its value to overall sustainability for effective long-term implementations. In disaster recovery, technology and its management prompt to community’s need to regain a sustainable livelihood and to acquire new skills to improve their capabilities. However, our analytical framework leads us to believe that the theory is inadequate in its failure to examine the issue of vulnerability, crucial in this study on disaster recovery. Consequently, we improved on this limitation by integrating the TCM model to key factors and variables relating to vulnerability in disaster recovery that were emphasised consistently in the theories reviewed. We extend the TCM model, and propose a new framework – the Extended Technology-Community-Management (Extended TCM) model (see Fig. 2), that identifies the four dimensions of vulnerabilities influencing technology implementation among the rural poor; physiological/psychological vulnerability, informational vulnerability, economic vulnerability, and socio-cultural vulnerability. We argue that effectively designed ICT for development (ICTD) programs have the potential to reduce the magnitude of these vulnerabilities in disaster recovery.

Fig. 2. Extended Technology-Community-Management (Extended TCM) Model.

The physiological and psychological vulnerabilities are micro-level dimensions involving the physical and mental well-being of affected persons, or a specific community. It is the extent to which people are susceptible to postcatastrophic emotional stress and trauma, as well as their access to depleted health infrastructure and resources. This cognitive variable influences people’s outlook on life, beliefs, and motivations to persevere in regaining their normal lives. The informational vulnerability deals with the access to and availability of information within affected communities. The lack of informational resources, ranging from personal documents, books and critical data, to opinion leaders and professional experts, affects the capabilities of people who are dependent on these sources of information. The informational vulnerability among rural people is further augmented by the low literacy levels and lack of relevant technological skills necessary to enable the learning and processing of information. The economic vulnerability is triggered by the loss of livelihood activities and equipments to financially support households and sustain economic growth in rural villages. From the destruction of houses, fishing boats and nets, and crops, to the loss of breadwinners and savings, the economic impact of disasters pose a significant problem to recovery in Asian rural communities. The destruction to the physical infrastructure, in terms of the breakdown of telephone wiring and electricity supply, amplifies the economic burden for overall development in villages. The socio-cultural vulnerability of communities is determined by the social structure and values of society that define human relationships in communities. These hierarchies affect access to resources and assets, and decision-making power of people, established by gender, age, race, religion, caste, and class egalitarianism within communities. The proposed Extended TCM model states that physiological and psychological enrichment, information optimization, economic resuscitation, and socio-cultural reconciliation can be achieved when paralleled with ICT interventions aimed at recovery. Here, it is crucial to first examine and assess the scale of each dimension of vulnerability in order to critically determine the roles technology, community, and management focused ICT initiatives can play in improving the lives of marginalized communities. We suggest that technology programs implemented encompassing objectives to reduce existing vulnerabilities, distinctive to each disaster affected

332

community, will foster sustainable development beyond disaster recovery. IV. RESEARCH FOCUS In this paper, we propose the Extended TechnologyCommunity-Management model to guide the design and implementation of ICT for development (ICTD) initiatives in disaster recovery, and illustrate the model using cases from disaster-affected communities in Asia. Specifically, we examine cases from disaster affected rural areas in India, Indonesia, Sri Lanka, and China. To do so, we look beyond the individual characteristics of technological design, management perspective and community participation, and examine their relationship with the four key dimensions of vulnerability. V. METHODOLOGY The Asian rural ICTD cases, used to highlight the Extended TCM model, were gathered from both primary and secondary sources. The primary fieldwork conducted in Banda Aceh, Indonesia, and Tamil Nadu, India from July 2005 to February 2006, included semi-structured interviews with representatives from relief agencies and NGOs, and members of affected communities. These interviews were complemented with first-hand data collected by one author, based as a research consultant at the Asia Tsunami Response Team of World Vision to record and evaluate the ICT strategies and plans of World Vision. Supplementary secondary data were gathered from various case illustrations. Cases illustrated rehabilitative measures initiated in the coastal regions of India, mainly using GPS and computers [54]. The Nanasala projects [55] and Govi Gnana system project [56] in Sri-Lanka, and the Jayagiri project [57] in Indonesia illustrated the function of ICTs, such as internet and radio, in empowering community livelihoods, and facilitating technology adoption. The midwives project [58] illustrated the use of mobile phones in responding to social issues of increasing child mortality rate in Indonesia. These pilot ICTD interventions, typically executed by NGOs, or funded by local governments, were mainly focused at rehabilitating rural populations affected by the 2004 tsunami, except for the Chinese study on drought-hit rural villages. The study in rural China [59] presented the potential of ICTs in enhancing education in rural China, while emphasizing the role of community involvement for successfully implemented projects. A combination of quantitative and qualitative research methods was used in the secondary studies. These indepth interviews, participant observations, documentary analysis, and surveys were conducted with NGO program officers, fisher-folk, local technology users, teachers, and community and government leaders. While some cases demonstrated the use of ICTs, such as GIS, Internet, radio, and multimedia technologies to purely disseminate disaster-related information to people, we note that cases selected for this study focus on initiatives involving community-level participation and use of technologies.

VI. FINDINGS We found that the implemented ICTD projects attempted to reduce existing vulnerabilities, either as over-arching objectives or as supplementary activities. The design of technology programs from the perspective of social sustainability required the addressing of inherent vulnerabilities of communities. Doing so allowed for the adoption, use, and maintenance of technologies towards overall development goals. Technological, management and community participation were critical factors in addressing key vulnerabilities. We discuss in-depth technological design elements of hardware and software, community needs and issues of ownership and training, and ICTD management variables of regulation, financials, and partnership structures. A. Physiological/Psychological Vulnerability In post-tsunami Tamil Nadu, India, computers purchased by the NGO partner, World Vision, aimed to build familiarity with technology and ultimately translate into educational benefits for children. However, the use of the same technology can alleviate the emotional distress of psychologically vulnerable communities. Children are prone to trauma from disasters, particularly when coping with the loss of their families and friends. Placed in child-friendly spaces, donated and maintained by community resources, these devices exposed children to games while introducing them to basic computing. The multimedia games eased them back into routines of play, and effectively reduced their stress level [54]. The well-being of communities is greatly enhanced by ICT projects designed with an in-depth understanding of community needs. In the coastal regions of Tamil Nadu where many fishermen lost their lives and livelihood assets to tsunami, 200 GPS navigation tools were deployed in 12 fishing villages with the primary objective of regaining economic livelihoods via improved catches. This electronic device assisted the fishermen as a compass in distant waters by showing distance and travel directions, thereby assisting the fishermen to navigate easily and more efficiently. From a vulnerabilities perspective, the project helps fishermen overcome fear of the sea. Besides increasing their confidence to fish, the psychological health of fishermen was improved when the saved time was spent interacting with family and friends [54]. B. Informational Vulnerability In post-disaster villages, technology enhanced capabilities of communities in obtaining updated information in information-constrained environments. Traditionally, local farmers faced exploitation due to the lack of access to updated market prices. Middlemen capitalized on this factor to gain high profits by usurping much of the value-added to the basic product. The Govi Gnana system project aimed to reduce rural farmers' exposure to price volatility by providing updated crop information about trading prices [56]. It enabled buyers and sellers of agricultural produce to view live transaction prices at various trade stalls in their own markets as well as prices in other areas of the country, made accessible at telecenter terminals and kiosks, and broadcasted via the telephone and internet. Thus, the reduced information

333

differential provided better economic resources to rural farmers. A mobile communications system was established in Banda Aceh, Indonesia, as a potential means of improving health services, where comparatively poor maternal and child mortality rates worsened after the tsunami [60]. The primary objective of mobile phone usage by rural midwives was to provide low cost means of improving the handling of complicated obstetrical cases, thus lowering rates of pre-natal and maternal mortality. However, a key information gap identified in the rural healthcare system was the lack of maternal health data by urban-based obstetricians. To rectify this vulnerability, a Short Message Service (SMS) software system was developed to allow midwives to upload critical health information to an online database, thus improving access, as well as enabling the rural midwife to consult senior nurses and doctors using mobile phones for particular cases [58]. Offering trainings to aid greater adoption of technology can make a significant difference to the information accessibility of rural community, beyond mere access to the technology. In drought-hit Yellow Sheep River, China, the “Town and Talent Project” equipped the agricultural village’s e-commerce centre with 25 computers. Online courses equipped farmers with relevant computer skills for better retrieval and usage of agricultural information [59]. A community centre in Jayagiri, Indonesia promoted community radio and multimedia as educational tools to encourage peoplecentered development in disaster recovery. The training series, directed at capacity-building of centre managers and volunteers running the community radio and programmes, aimed to improve technological skills for creating program content [57]. Finally, informational capacity building was greatly enhanced by community management of technology in recovery efforts. The collective-ownership of facilities can promote the dissemination and sharing of information within communities. The "Town and Talent Project" in Yellow Sheep River introduced internet-enabled computers to enrich educational resources in the local school. Collaborating with a local company, the teachers were trained to set up the scholastic network. These teachers were the central information providers on computer usage to both their peers and children [59]. The Nanasala telecenter project in Sri Lanka was managed by individual entrepreneurs within the community to provide internet access to rural villages [55]. The initiative exposed rural people to internet, computer, and telephone technologies with free training sessions to gain knowledge of technology use and online retrieval of information on new agriculture practices and techniques, and information on a variety of health issues. C. Economic Vulnerability Information communication technologies have the potential to empower communities to regain economic stability in the aftermath of disaster. However, beyond the economic outcomes, the ability of communities to engage in relatively expensive technology usage, and to manage their economic expectations, need to be considered. The use of GPS and fish-finders led to income generation in

Indian fishing communities, with reported livelihoods increasing from 20% to 50% [54]. Group ownership of the devices ensured maintenance of the equipment. The long-term environmental impact of technology usage was mitigated by inviting local government officials from the fisheries department to impart responsible fishing practices. The effects of poor informational infrastructure can be devastating on the available educational resources in affected areas. In addition to the destruction and damage of scholastic infrastructure, there is a crucial impact on human capital, in terms of the loss of teachers and parents. For children, ICTs are a source of learning opportunities that help in building critical technological skills. With the primary objective of providing educational opportunities, World Vision equipped two secondary schools in affected-villages of Tamil Nadu, India, with computers, providing lessons integrated with the school curriculum. In addition to computer-oriented skills, children learnt Tamil, English, and other academic subjects [54]. It was essential to provide an outlet for graduates to utilize their skills in an economically productive manner. Business training, including introduction to productivity software, was provided to young adults to improve their job prospects in the increasingly technologically-savvy Indian workforce. D. Socio-Cultural Vulnerability Community pressures are seen most directly in the existing networks of power within the social system. The traditional power of the higher castes (social power), the middlemen (economic power), and males (gender power) limits the opportunities available to those groups with less power. Introducing ICTs in such biased environment could work in the favour of the haves, and likely further marginalize the have-nots. Tactfully designed ICTD initiatives have the potential to champion the cause of development without widening existing social inequalities. Recognizing this, the M. S. Swaminathan Research Foundation, an Indian NGO, established knowledge centers in Dalit (lower caste) villages [61]; while World Vision trained women Self Help Groups to take ownership of the ICTD projects within their communities [62], [63]. The World Vision multimedia centres in most villages claimed no socio-cultural restrictions on access, with no separate timings for young girls, or lower castes individuals, allowing a co-mingling of different sections of society [54]. In some situations, the socio-cultural issues surrounding communities may indirectly affect people’s perceptions on their capacity to use technologies. In rural classrooms in China, elder teachers were reluctant to take on computer usage for fear of not being able to effectively learn and adopt computer skills, predominantly possessed by younger users. Training courses helped staff overcome this technophobia, and grasp basic E-literacy in daily teaching practices [59]. Indonesian midwives in the Banda Aceh mobile-phone project initially perceived the ICT-based health practice as a monitoring system; rather than a tool to assist their work. This negative perception, triggered by existing hierarchical class structure within the health infrastructure, could have encouraged them to revert to

334

the prior, more inefficient, paper-based system for reporting maternal health [64]. Another criticism was the unfamiliarity with text-based technology. To overcome these barriers, joint training seminars were held with doctors, health coordinators and rural midwives. Secondly, a JAVA-based applet was developed that functioned with scroll and select functionality, rather than using the traditional text-based, and possibly cumbersome, SMS. Constructive responses to the suggestions and feedback from midwives increased credibility of the initiative, which subsequently encouraged the use of ICT beyond existing social classbased barriers between rural midwives and hospital-based doctors. VII. DISCUSSION AND CONCLUSION An examination of Asian cases reveals that the deployment of ICTs in disaster recovery needs to take into consideration the vulnerabilities impacting social sustainability. This study extends the theoretical literature on disaster management and ICT, by addressing a gap in research on the best practices for ICT deployment in disaster recovery. The TCM model proposed that ICTD program design should include critical variables involving communities, including needs, ownership, and training, in addition to the external impacts of technology and management. We extend the TCM model to include four key vulnerability dimensions. We recognize the limitations of frameworks with clearly defined boundaries, and discuss the inter-relationships of vulnerabilities and ICTD program deployment characteristics of technology, its management and the community involvement. The case studies illustrate real world complexities, in terms of barriers to implementation and considerations for long-term program success. A critical distinction needs to be made between core development objectives and vulnerabilities, as these may not coincide, and may even be antithetical. Vulnerabilities are inherent factors inhibiting communities from accessing, adopting, and realizing the benefits from technology implementations. We believe these barriers need to be addressed in addition to core development objectives, such as those recommended by the United Nation’s MDG [65]. The choice of development objectives can influence the selection of beneficiaries, their control and use of technology, leading to an amplification of existing vulnerabilities in communities. For instance, the management of rural initiatives, largely led by village leadership councils, could lead to privileged access to technologies, in turn furthering the sociological imbalance within society. Further, the emphasis on education in post-disaster societies, often placed on basic primary education, could ignore the needs of adolescents and young adults, raising issues of age discrimination to technology access. We believe that project designers need to recognize the existence of vulnerabilities within the social system, and take steps to mitigate them, in addition to focusing on overall project objectives. Realistically, project managers have to recognize that the process of identifying existing vulnerabilities,

followed by designing initiatives, is an iterative process involving evaluation and reconfiguring of project design, and will necessarily be a long-term phenomenon. The durations of reaching this goal will vary depending on several intrinsic and extrinsic conditions affecting ICTD programs at different stages of execution. First, it is necessary to determine a methodological approach to elaborate on the criteria for assessing the vulnerabilities, determining the order of priority for tackling the obstacles, and gauging the effectiveness of measures. Secondly, beyond limiting the analysis of psychological vulnerabilities to disaster affected rural poor, the assessment should observe other extrinsic conditions such as the mental barriers of ICTD program managers and policy makers in related domains. Preconceptions of the marginalized communities’ learning capacities, adoptive abilities, enthusiasm levels, and financial capabilities to use and manage technology could impede their commitment to implement and sustain ICT initiatives. We should note that the identified variables of physiological and psychological vulnerability, informational vulnerability, economic vulnerability, and socio-cultural vulnerability, are not mutually excusive variables. The cases highlight the existence of overlapping vulnerabilities, and those that are sometimes in opposition to one another. ICT programs focusing on reducing economic susceptibility, such as the use of GPS system by fishermen, also have a positive impact on the psychological well-being of users. Educational programs for rural midwives involving cell-phone training addresses the psychological fear of technology, while simultaneously reducing the informational barriers present. However, vulnerabilities can contradict each other when the attempt to reduce one barrier magnifies the negative impact of another. For example, ICTD programs that aim to enhance information flow among communities through the usage of mobile phones could, in the long-run post-subsidy, negatively affect the financial capacities of communities which struggle to maintain and continue usage of the adopted technology. We have previously critiqued the stage-based approach of extant theories; yet we acknowledge that the establishment of clear stages can provide a guideline for ICTD program assessment, particularly for practitioners. It is possible that different vulnerabilities may take priority at certain stages of implementation, leading to specific TCM configurations. The extended TCM model does not incorporate hierarchies in management flow, community participation, or technology diffusion, suggesting their relationship to vulnerabilities occurs in a cooperative manner. The communication infrastructure within a community consists of both old and new ICTs, as well as established and evolving social networks. This study focused on implementations involving a single technology, aided largely by the fact that most ICTD interventions are designed as such. We propose that researchers examine dynamic communication and information environments, themselves embedded within fluid social networks, rather than one technology at a time. Future research can build on the extended TCM model as a theoretical framework to guide research in the design of ICTD programs, and to test the significance of

335

vulnerabilities beyond disaster recovery. We propose that the vulnerabilities assessment approach can and should be applied to varied contexts such as education, healthcare, agriculture, and conflict management. Subsequent studies in the area of disaster recovery could examine how the TCMV model can examine the impact of individual vulnerability dimensions to technology, management, and community variables. Alternatively, research could focus on specific components within the TCM model, such as community training, within the Management-Community intersect, and examine its relation to the four identified vulnerabilities. Here, the aim would be to explore thoroughly the barriers that exist and arise within specific components, and suggest solutions to overcoming them. There is no doubt that calamities have a disproportionately negative impact on the poor. However, the rapid proliferation of ICTs, particularly in Asia, has the potential to engender positive recovery to the lives and state of those affected. We believe well-designed ICTD programs have the power to increase the capacities of communities, while simultaneously reducing the vulnerabilities towards technology adoption. Finally, rural communities can witness long-term social sustainable progress.

[9] [10] [11] [12]

[13] [14] [15] [16]

[17]

REFERENCES [1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

Department for International Development, “International day for disaster reduction,” Department for International Development, 2007. [Online] Available: http://www.dfid.gov.uk/news/files/disasters-day.asp [Accessed: Aug. 15, 2008]. Centre for Research on the Epidemiology of Disasters, “Annual disaster statistical review: The numbers and trends 2007,” Centre for Research on the Epidemiology of Disasters, 2007. [Online] Available: http://www.emdat.be/Documents/Publications/Annual%20Disaster %20Statistical%20Review%202007.pdf [Accessed: Aug. 10, 2008]. Commonwealth Telecommunications Organization, “Using ICT for effective disaster management forum: A summary report on sessions and panel discussions,” Commonwealth Telecommunications Organization, 2006. [Online] Available: http://www.lirneasia.net/wpcontent/uploads/2006/07/CTO%20Report%20Uning%20ICT%20f or%20DM%20June%202006.pdf [Accessed: Aug. 14, 2008]. United Nations Economic and Social Commission for Asia and the Pacific, “ICST-enabled natural disaster management in Asia and the Pacific region,” United Nations Economic and Social Commission for Asia and the Pacific, 2005. [Online] Available: http://stdev.unctad.org/unsystem/cstd/escap8.doc [Accessed: Aug. 15, 2008]. International Telecommunication Union, “ICT statistics database,” International Telecommunication Union, 2008. [Online] Available: http://www.itu.int/ITUD/ICTEYE/Indicators/Indicators.aspx [Accessed: Aug. 15, 2008]. OneWorld South Asia, “Making MDGs disaster proof,” OneWorld South Asia, 2008. [Online] Available: http://southasia.oneworld.net/Article/making-mdgs-disaster-proof [Accessed: Aug. 15, 2008]. Centre for Research on the Epidemiology of Disasters, “An analytical review of selected data sets on natural disasters and impacts,” Centre for Research on the Epidemiology of Disasters, 2006. [Online] Available: http://www.emdat.be/Documents/Publications/TschoeglDataSetsR eview.pdf [Accessed: Aug. 15, 2008]. A. F. Gorp, “ICT as an enabler for humanitarian relief? An assessment of communication and information flow in interorganizational context,” presented at the annual meeting of the International Communication Association, New York, 2005. [Online] Available: http://test.scripts.psu.edu/users/a/f/afv103/ICA2005.pdf

[18]

[19]

[20] [21] [22]

[23] [24]

[25]

[26]

[27] [28]

[Accessed: Aug. 15, 2008]. B. K. Paul, “Disaster relief efforts: an update,” Progress in Development Studies, vol. 6, pp. 211–223, 2006. United Nations Development Programme, ICT and human development: Towards building a composite index for Asia. New Delhi: Elsevier, 2004. International Telecommunication Union, ICT and telecommunications in the least developed countries. Geneva: International Telecommunication Union, 2006. United Nations Development Programme, “Community-based networks and innovative technologies: New models to serve and empower the poor,” United Nations Development Programme, 2005. [Online] Available: http://www.undp.org/poverty/docs/ictd/ICTD-CommunityNets.pdf [Accessed: Sep. 10, 2008]. N. Gunawardene and F. Noronha, Communicating disasters: An Asia Pacific resource book. Bangkok: United Nations Development Programme Regional Centre, 2007. M. Jansz, “Disaster recovery and ICT in Sri Lanka: The day after,” Information for development, vol. 3, pp. 10–22, Jan. 2005. C. Wattegama, “ICT in disaster management,” 2007. [Online] Available: http://www.apdip.net/apdipenote/16.pdf [Accessed: Sep. 15, 2008]. China Internet Network Information Centre, “Statistical survey report on the internet development in China: Abridged edition,” China Internet Network Information Centre, 2008. [Online] Available: http://www.cnnic.cn/download/2008/CNNIC22threport-en.pdf [Accessed: Aug. 09, 2008]. Internet World Statistics, “Asia marketing research, internet usage, population statistics and information usage,” Internet World Statistics, 2008. [Online] Available: from http://www.internetworldstats.com/asia.htm [Accessed: Sep. 15, 2008]. A. Chib and J. Zhao, “Sustainability of ICT interventions: Lessons from rural projects in China and India,” in Communicating for social impact: Engaging communication theory, research, and pedagogy, L. Harter and M. J. Dutta, Eds., ICA 2008 Conference Theme Book, Hampton Press, in press. H. E. Hudson, “Universal access to the new information infrastructure,” in Handbook of New Media: Social Shaping and Consequences of ICTs, L. A. Lievrouw and S. M. Livingstone, Eds., London: SAGE, 2002, pp. 368–384. C. K. Prahalad, The fortune at the bottom of the pyramid: Eradicating poverty through profits. Philadelphia: Wharton School Publishing, 2004. United Nations Conference on Trade and Development, Information and communication technology development indices. Geneva: United Nations Publications, 2003. S. Yodmani and D. Hollister, “Disasters and communication technology: Perspectives from Asia,” in Second Tampere Conference on Disaster Communications, 2001. [Online] Available: http://www.reliefweb.int/telecoms/conference/cdc2001/dh.pdf [Accessed: Aug. 15, 2008]. J. Walter, ed., World disasters report 2005: Focus on information in disasters. Geneva: The International Federation of Red Cross and Red Crescent Societies, 2005, pp. 11. M. I. Ahmed and E. C. Lentz, “Enhancing the livelihoods of the rural poor through ICT: A Knowledge Map,” 2007. [Online] Available: http://wwwwds.worldbank.org/external/default/WDSContentServer/WDSP/IB /2008/11/26/000333037_20081126002013/Rendered/PDF/466300 NWP0Box31esh0Country0Study111.pdf [Accessed: Aug. 01, 2008]. A. Chetley, “Improving health, connecting people: The role of ICT in the health sector in developing countries. InfoDev framework paper,” 2006. [Online] Available: http://www.asksource.info/pdf/framework2.pdf [Accessed: Jul. 15, 2008]. K. Delgadillo and R. Borja, “Learning lessons from telecentres in Latin America and the Caribbean,” 1999. [Online] Available: http://www.idrc.ca/telecentre/evaluation/nn/16_Lea.html [Accessed: Aug. 30, 2008]. T. Eisensee and D. Stromberg, “Disaster relief,” The Quarterly Journal of Economics, vol. 122, pp. 693–728, 2007. A. Hovanesian and A. Cox, “ICT for tsunami recovery: Best practices and lessons learned,” presented at the E-Asia Conference, Putrajaya, Malaysia, 2007.

336

[29] O. Takizawa, “Using ICT in disaster mitigation,” National Institute of Information and Communications Technology News, vol. 344, pp. 1–2, 2004. [30] S. Lee and A. Chib, “Wireless initiatives for connecting rural areas: Developing a framework,” in Participation and media production: Critical reflections on content creation, N. Carpentier and B. De Cleen, Eds., ICA 2007 Conference Theme Book. Newcastle, United Kingdom: Cambridge Scholars Publishing, 2008. [31] R. R. Dynes, Organized behavior in disaster. Lexington, Massachusetts: Heath Lexington Books, 1970. [32] L. Palen and S. B. Liu, “Citizen communications in crisis: Anticipating a future of ICT-supported public participation,” in Proceedings of the Special Interest Group on Computer-Human Interaction (SIGCHI) Conference on Human Factors in Computing Systems, San Jose, California, 2007, pp. 727–736. [33] J. Powell, An introduction to the natural history of disaster. Baltimore: University of Maryland, Psychiatric Institution, 1954. [34] L. Buzna, L. Issacharoff, K. Peters, V. Rosato, and D. Helbing, “How can future ICT enhance disaster relief and recovery?” presented at the Information Communication Technology-Future and Emerging Technologies Workshop, Dresden University of Technology, Dresden, Germany, 2007. [35] Caribbean Information and Communications Technology Virtual Community, “Facilitating effective disaster management in the Caribbean: A response from the Caribbean Information and Communications Technology Community,” 2004. [Online] Available: http://www.devnet.org.gy/documents/CaribbeandisasterbriefFinal.pdf [Accessed: Aug. 15, 2008]. [36] D. Basu, “Disaster relief: A compendium of learnings from engagements in Afghanistan, Iraq, Liberia, Iran, Sudan, Guatemala, Indonesia, Sri Lanka, Pakistan, Lebanon,” NetHope, 2006. [Online] Available: http://nethope.org/doc/NetHope%20Disaster%20Relief%200906.p df [Accessed: Aug. 15, 2008]. [37] U. Sastrokusumo, “ICT for countermeasures and strategies against emergencies and disaster crisis: A suggestion to build Asia-Pacific pilot project,” presented at the Second Joint Project Team Meeting for Establishing a Disaster Management Support System in the Asia-Pacific Region, Bangkok, Thailand, 2006. [38] L. Palen, S. Vieweg, J. Sutton, S. B. Liu, and A. Hughes, “Crisis informatics: studying crisis in a networked world,” in Proceedings of the Third International Conference on E-Social Science, Ann Arbor, Michigan, 2007. Available: http://www.cs.colorado.edu/~palen/palen_papers/palencrisisinformatics.pdf [Accessed: Aug. 23, 2008]. [39] Caribbean Community and Common Market, “Caribbean ICT disaster management network,” 2004. [Online] Available: http://www.devnet.org.gy/documents/Final_ICT_and__DM__Prop osal_Nov_15.pdf [Accessed: Aug. 15, 2008]. [40] B. R. Balachandran and A. Haer, “Corporate involvement in disaster management: A conceptual framework,” presented at the Workshop on Disaster Management: Disaster Preparedness and Mitigation, New Delhi, India, 2004. [41] Y. S. Alghtani and S. Al-Zahrani, “A conceptual framework for information network system to exchange information to control accidents in Saudi Arabia,” presented at the Saudi Technical Conference and Exhibition, Saudi Arabia, 2006. [42] Associated Programme on Flood Management, Social aspects and stakeholder involvement in integrated flood management. Geneva: World Meteorological Organization, 2006. [43] Earthquakes and Megacities Initiative, “Disaster risk reduction in metropolitan regions through the cluster cities project,” Proceedings of the 2005 Americas Cluster Cities Meeting, Bogotá, Columbia. Available: http://www.emimegacities.org/upload/CCP_America_2005_10_Meeting_Report.p df [Accessed: Aug. 15, 2008]. [44] P. Blaikie, T. Cannon, I. Davis, and B. Wisner, At risk: Natural hazards, people's vulnerability and disasters. London: Routledge, 1994. [45] F. C. Cuny, eds., Disasters and Development. New York: Oxford University Press, 1983. [46] M. B. Anderson and P. J. Woodrow, Rising from the ashes: Development strategies at times of disaster. Paris: Westview Press, 1989. [47] A. Kreimer and M. Munasinghe, Managing natural disasters and the environment. Washington, D.C: The World Bank, 1991.

[48] A. Kreimer and M. Zador, Colloquium on disasters, sustainability and development: A look at the 90s. Washington, D.C: The World Bank, 1989. [49] A. Lavell, “The impact of disasters on development gains: Clarity or controversy Latin American social science faculty and the network for the social study of disaster prevention in Latin America,” presented at the International Decade for Natural Disaster Reduction Programme Forum, Geneva, 1999. [50] L. Pyles, “Community organizing for post-disaster social development,” International Social Work, vol. 50, pp. 321–333, 2007. [51] F. Ellis, Rural livelihoods and diversity in developing countries. London: Oxford University Press, 2000. [52] C. Ashley and D. Carney, Sustainable livelihoods: Lessons from early experience. London: Department for International Development, 1999. [53] T. Newholm, K. Keeling, P. Mcgoldrick, L. Macaulay, and J. Doherty, “The digital divide and the theory of optimal slack,” New Media and Society, vol. 110, pp. 295–319, 2008. [54] Center for Media Studies, Final Report: Evaluating ICT-4D project of World Vision India. India: World Vision India, 2007. [55] F. J. Proenza, “ICTA Nanasala Program: Field visit report,” 2005. [Online] Available: http://siteresources.worldbank.org/INTEDEVELOPMENT/Resour ces/Telecenter.doc [Accessed: Aug. 15, 2008]. [56] infoDev, “Sri Lanka-Govi Gnana system project,” 2007. [Online] Available: http://www.ict4rl.info/CaseStudies/GGS?v=fau [Accessed: Aug. 15, 2008]. [57] United Nations Educational, Scientific and Cultural Organization, “UNESCO develops community multimedia centre for education in Indonesia,” United Nations Educational, Scientific and Cultural Organization, 2008. [Online] Available: http://www.unescobkk.org/index.php?id=6800andtx_wecknowled gebase_pi1[tt_news]=1215andtx_wecknowledgebase_pi1[backPid ]=6771andcHash=96224fb29a [Accessed: Aug. 15, 2008]. [58] A. Chib, Information and communication technologies for health: Aceh Besar midwives mobile-phone project final report, tech. report, World Vision Indonesia Tsunami Response, Singapore, 2008. [59] J. Zhao, The Internet and rural development in China: The sociostructural paradigm. Bern: Peter Lang AG, 2008. [60] A. Chib, M. O. Lwin, J. Ang, H. Lin, and F. Santoso, “Mobiles and midwives: Improving healthcare communications via mobile phones in Aceh Besar, Indonesia,” Asian Journal of Communication, vol. 18, pp. 348–364, 2008. [61] United Nations Centre for Regional Development, “Information and communication technologies for human security in local development,” Regional Development Dialogue, vol. 23, pp. 65– 88, 2002. [62] A. Chib, “Information communication technologies for development: Cases from tsunami-affected communities in Indonesia and India,” 2006. [Online] Available: http://www3.ntu.edu.sg/sci/sirc/download/Arul_presentation%20re port.pdf [Accessed: Aug. 20, 2008]. [63] Environmental Planning Collaborative, Tsunami evaluation coalition: The international community’s funding of the tsunami emergency and relief,” 2006. [Online] Available: http://www.un.org.in/untrs/reports/local_response_india.pdf [Accessed: Aug. 29, 2008]. [64] A. Chib, Information and communication technologies for health: Midwives with mobile-phone project in Aceh Besar baseline report, tech. report, World Vision Asia Tsunami Response, Singapore, 2007. [65] United Nations, The Millennium Development Goals Report 2005. New York: United Nations, 2005.

337

Featherweight Multimedia for Information Dissemination Gerry Chu, Sambit Satpathy, Kentaro Toyama, Rikin Gandhi, Ravin Balakrishnan, S. Raghu Menon Abstract— Featherweight multimedia devices combine audio with non-electronic visual displays (e.g., paper). Because of their low cost, customizability, durability, storage capacity, and energy efficiency, they are well-suited for education and information dissemination among illiterate and semi-literate people. We present a taxonomy of featherweight multimedia devices and also derive design recommendations from our experiences deploying featherweight multimedia in the agriculture and health domains in India. We found that with some initial guidance, illiterate users can quickly learn to use and enjoy the device, especially if they are taught by peers. Index Terms—Audio, illiteracy, information dissemination

A

I. INTRODUCTION

major thrust towards bringing information and communication technologies to underserved populations is the development of low-cost computational devices such as the XO laptop from OLPC [1], Classmate PC [2], and even a potential $12 laptop [3]. While these initiatives and other factors such as increased penetration of mobile phones will undoubtedly have tremendous impact in providing the global poor with better communication tools and access to information, the cost of these devices remain prohibitive to the 2 billion+ people worldwide who live on less than $2 per day. Beyond monetary cost, illiteracy remains a major hurdle. Conservative estimates of illiteracy suggest that there are over one billion illiterate people worldwide [4] for whom even a free, connected computer is useless given that some level of literacy and basic education is typically required to operate them; recent attempts at text-free interfaces [5] hold the promise of partial accessibility to computing by the illiterate, but still don’t address the complexity of the device. While it is certainly worthwhile trying to grapple with the cost and literacy constraints of “full-fledged” technology for many applications, there are other applications that may only require a little bit of technology in order to add significant value. One such broad application area is the information dissemination that government and non-profit organizations do to improve the education and health of people. In these populations, traditional information dissemination methods Manuscript received September 22, 2008. G. Chu, S. Satpathy, and R. Balakrishnan were with Microsoft Research India (email: gerrychu | ravin @dgp.toronto.edu, [email protected]). K. Toyama, R. Gandhi, and S. Raghu Menon are with Microsoft Research India, “Scientia”, 196/36 2nd Main, Sadashivnagar, Bangalore, 560 080, India (phone: +91 (80) 6658-6000; email: kentoy|riking|[email protected]). R. Gandhi is additionally with Digital Green R. Balakrishnan is with the University of Toronto

include printed material, radio and television public service announcements (PSAs), and verbal instruction. These methods are certainly viable, but suffer from several shortcomings, such as requiring literacy to comprehend textual printed material, the relatively high cost of producing, distributing and consuming radio and television PSAs, and the arguably higher cost of verbal instruction by humans who might not even deliver a consistent message over time. Of these, only verbal instruction provides the possibility of interactive exchange with the consumer. In this paper, we explore the notion of featherweight multimedia” – various combinations of electronic audio devices with non-electronic visual displays – for interactive multimedia information dissemination (Fig. 1). Featherweight multimedia devices require minimal power, are more rugged, and are dramatically cheaper than low-cost computers or feature-rich mobile phones. Yet, they integrate many of the best elements of traditional techniques with the enhancement of interactivity. Our contribution is not in the technology per se, but rather in 1) mapping out the design space of featherweight multimedia, 2) exploring factors affecting their use for information dissemination to illiterate or semi-literate populations using our experiences deploying them, and 3) reporting the reactions of the target population to initial usage of several featherweight multimedia variants. Our preliminary experiments suggest that non-literate users with little formal education can quickly understand how to use featherweight multimedia, but that social support is required to maximize effectiveness.

Fig. 1. Village residents using a talking book

II. RELATED WORK Recently, there have been several initiatives adapting specific featherweight multimedia devices to the development

338

field. These evolved from devices that were originally created for the developed world. In addition, featherweight devices can trace their design histories to other multimedia initiatives that use audio and video in development. A. Existing featherweight multimedia devices in development Books of Hope is an organization that creates talking books in South Africa [6]. Each 16-page book is held within an electronic device that has 16 buttons. The user turns to a page, each of which has an icon. The user presses the button labeled with the same icon, triggering the appropriate caption to play aloud. Each book holds 5 minutes of audio, and with two AAA batteries, 200-500 plays can be achieved [7]. In an evaluation [8], a control group of grade 7 students attended a workshop about mental health, while an experimental group received talking books. Pre- and post-tests determined that both groups learned the same amount but a large percentage of students who received the talking books showed them to other students (80%), to their families (80%) and to church members (43%), pointing to possible viral dissemination benefits of cheap portable featherweight multimedia. Talking books such as LeapPad [9], WhizKid [10], and PowerTouch [11] are educational toys for children. They are operated by overlaying a piece of paper or workbook onto a touch or pen-sensitive screen. The device can identify the current page, so that when elements on the page are pressed, the device plays a recorded caption. There was an announced initiative in 2004 to deploy the LeapPad in Afghanistan [12], but nothing has been published about the deployment since. However, the PicTalk device, which uses LeapPad technology, is being used to teach Indian schoolchildren [13]. Students using PicTalk to learn English performed 40% better on post-tests than a control group learning using a traditional curriculum. The Literacy Bridge project has created an interactive digital audio player/recorder costing roughly US$5-$12 and are deploying it in Ghana [14]. Although they call it a “Talking Book” the focus is on the audio and not on the paper book component. B. Devices similar to featherweight multimedia Audio guides (e.g., [15]) are widely used in museums and galleries. They play an audio clip when a numeric code is typed in. An interesting element of these sorts of audio interfaces is the spatial decoupling of the visual printed media from the audio device. Similar to the aforementioned talking books, communication aides are used by people with developmental disabilities who cannot communicate verbally [16]. More advanced and expensive audio/paper devices such as the Audio Notebook [17] and the Pulse SmartPen [18] record audio as the user writes in a notebook. Tapping on a word plays the audio that was recorded when the word was written. The Daisy Consortium is the publisher of a file format for talking books meant for the blind or those with learning disabilities [19]. The Dolphin company creates players and authoring tools for talking books in the Daisy format [20].

Public advertising on billboards, as well as research on more interactive public ambient displays have demonstrated designs intended to engage casual bystanders in information exchange [21][22]. A featherweight multimedia instantiation of this concept might involve billboards with phone numbers that passersby can call for additional prerecorded information, or paper posters with an audio player, a design that we explore later in this paper. C. Related work in development There have been two notable audio interfaces for developing countries. Tamil Market uses a spoken dialog system that allows illiterate people to query the system by voice to hear crop prices spoken aloud [23]. Similarly, Pakistani community health workers can call HealthLine, ask it verbally for information, and hear medical information over their mobile phone [24]. Both require backend computers to provide speech recognition, hence imposing a fairly significant computational and cost requirement. A system using visual codes and mobile phone cameras, such as that explored by Parikh et al. [25] could also be used to trigger audio playback. Work on text-free user interfaces [5] has focused on replacing text in computing interfaces with audio prompts that are played when the user mouses over a graphical interface element. The focus of this work has been on making conventional computers accessible to the illiterate population, and hence the usual cost and power requirements of standard computation remains. Featherweight multimedia can be characterized as being text-free user interfaces without an electronic visual display. The concept of integrating paper and audio is supported by work by Medhi et al. [26]. They tested five different representations (video, photo, animation, static drawing, and text) of medical conditions with and without audio captions on Indian slum residents. They found that users best understood drawings that are combined with audio. Video might be considered as the “heavyweight” version of featherweight multimedia. Like featherweight multimedia, it has both audio and visual electronic displays. Video has been used to teach Indian schoolchildren [27] and has also been shown to be effective in teaching illiterate users to use a jobsearch computer application [28]. The Digital Green project uses video to teach sustainable agricultural techniques to Indian farmers [29]. It has been shown to be six times more effective than conventional agricultural extension workers. Despite these successful examples of video for information dissemination in developing countries, video requires an expensive television and DVD player that depend on electrical power that can be unreliable in many development settings. III. DESIGN SPACE At first glance, our definition of featherweight multimedia – the combination of an electronic audio device with some nonelectronic visual display – connotes a relatively simple range of possible instantiations. On closer investigation, however, we find a rich design space worthy of careful delineation.

339 TABLE 1. COMPARISON OF ELECTRONIC AUDIO PLAYERS LOW-END

SMARTPHONE

CD PLAYER

MOBILE PHONE

DEVICE COST CONTENT CUSTOMIZABILTIY CAPACITY AUDIO QUALITY

DURABILITY POWER EFFICIENCY USABILITY UBIQUITY SENSORS

PORTABLE AUDIO/MP3 PLAYER

TALKING BOOK (LEAPPAD)

CUSTOM EMBEDDED ELECTRONICS

GREETING-CARD AUDIO CHIPS

$10-50 High

$200-500 High

$10-$50 Medium

$10-$200 High

$15-$50 High

$0.05-$100+ Medium

$0.05 Medium

High Medium High Medium

High Medium Medium Low

Medium High Medium Low

High High High High

Medium Medium High Medium

Medium Medium Medium High

Low Low Medium High

Medium High Low

Medium Low High

Medium Medium Low

Medium Medium Low

High Low Low

Medium Low Low

Medium Low Low

A. Electronic audio player The audio component of featherweight multimedia can be sourced from various readily available technologies including CD players, portable MP3 players, mobile phones, museum audio guides, and embedded audio devices, including audio chips (as can be found in audio greeting cards [30]), embedded MP3 players, and other custom audio electronics. These vary along several interesting dimensions, summarized in Table 1: Device cost. This can vary considerably, ranging from just a few dollars at wholesale prices to several hundred dollars for state-of-the-art equipment. An embedded audio chip of the sort found in greeting cards cost just several cents, and one could embed many of these chips into a single featherweight multimedia device. At the other end of the spectrum, museum audio guides can cost thousands of dollars. Content customizability. CD/DVDs cost just a few cents to manufacture with content, while content on MP3 players can easily be uploaded from a computer. However, the distribution costs of getting those CD/DVDs to the user or providing a geographically distributed user base with access to a contentserving computer to update their MP3 players needs to be accounted for. Mobile phones offer the highly flexible solution of wireless access to content that may not even be stored on the phone itself, although costs for such access can vary considerably across regions. Many embedded devices only allow for a one-time write of the content; where they do allow updating of content, the costs are roughly similar to that of updating an MP3 player although updating an embedded system typically requires more specialized and complex tools than the mainstream ones available for MP3 players. Capacity. Capacity ranges from 5 seconds for some embedded electronics to days with MP3 players. Audio quality. This ranges from high-quality audio in CD and MP3 players to somewhat lower quality audio on mobile phones to tinny audio in greeting card style embedded devices. Volume also varies from personal levels in embedded devices to sharable levels in MP3 players with external speakers. Durability. With the possible exception of CD/DVD media that is prone to scratching, the remaining options tend to rely on solid-state memory and electronics that are roughly equivalently robust. However, the housing for the electronics could make a difference, with greeting-card style embedded devices being relatively fragile compared to an MP3 player. Power. Most of the devices surveyed have low power consumption requirements. However, they vary considerably

in the types of batteries used and whether or not they can be recharged and/or replaced easily. Some devices use readily available replaceable batteries, some are more specialized and can only be replaced by a technician, while others can only be recharged by plugging into mains power. Depending on the usage locale, the type of battery might be a crucial deciding factor in choice of a featherweight multimedia device. Usability. Given the focus on information dissemination in low-education target audience, usability is arguably even more crucial than in more mainstream technology. For example, many cheap audio players have surprisingly complex interfaces that require the user to decipher instructions on a tiny LCD screen, thus negating the positive value of their inexpensive cost. We believe that for most usage scenarios, ultra-simple interfaces with just a few buttons (start, stop, rewind) or with one button per audio caption are generally best. A numeric keypad can added if random access to content is provided through numeric codes. Ubiquity. Mobile phones have staggering penetration in many parts of the developing world. In contrast, more specialized embedded systems might be cheaper but might ultimately not be viable given their specialized nature. Sensors. Smart phones and custom embedded devices might have cameras, RFID readers, and other sensors that can enable automatic access to the appropriate audio clip depending on the visual content, whereas simpler sensor-free devices would rely on the user to select audio clips for a given visual display. B. Non-electronic visual display This can be considered along several key dimensions: Physical form factor and portability. The display can range from small and highly portable to larger fixed installations. Example form factors include sheets of paper, greeting cards, brochures, books, posters, and even billboards. Cost. This can range from essentially free (e.g., handscribbled content on a sheet of paper), to a few cents (printed paper), to a few dollars (printed books), to thousands of dollars (billboards). The high cost of a billboard seemingly detracts from our goal of ultra-low cost information dissemination, but the display’s cost can be amortized over the number of people that might use it. Indeed, a printed sheet of paper priced at just a few cents that is only used once by one person could cost more in aggregate than an expensive billboard that disseminates information to thousands of people over a longer period of time.

340

Electronic integration. Normal visual displays for featherweight multimedia are non-electronic. However, more sophisticated versions might be inexpensively tagged to enable tighter integration with the associated electronic audio player. For example, RFID or optical tags might enable a talking book to determine which page is currently active, or optical codes on a poster could be used to index into auditory content on a camera-equipped audio player. IV. USAGE EXPLORATIONS We explored the viability of several forms of featherweight multimedia via usage explorations in different field settings. These were not intended to be formal experiments or extensive deployments, but rather initial forays. We experimented with two different domains (agriculture extension and healthcare), five device configurations, and five qualitatively different preliminary investigations. The investigations were intended to shed light on several questions we felt were critical to determine the overall promise of featherweight multimedia: Q1. User literacy and education requirements: A major motivation for featherweight multimedia is that it relies on audio rather than text for communication. As such, we expect the technology to be usable by low-literacy populations. Is this in fact the case? Are there other cognitive barriers to use? Q2. User engagement: In contrast to more full-fledged multimedia systems that provide a rich set of highly engaging interactive content, will the simpler and minimally interactive content provided by featherweight multimedia be sufficiently engaging that users will actually feel compelled to access information through such devices? Q3. Social support requirements: Ideally, the device is usable by low-literacy populations on their own without any mediation from trained helpers or support from peers, but in reality, it is highly likely that some amount of social support will be required. To what extent is social support required for featherweight multimedia to be effective? V. EXPLORATIONS IN AGRICULTURE EXTENSION Our first exploration of featherweight multimedia was in the agriculture extension domain. We partnered with the Digital Green project [29], which seeks to teach targeted sustainable agricultural techniques to small and marginal farmers in India using video clips of their peers learning those same techniques. Their human-mediated video approach has shown higher adoption rates of the agricultural techniques as compared to traditional methods of dissemination such as paper posters or extension officers conducting 1-on-1 lessons. The requirement of a TV and DVD player, however, poses challenges in terms of cost, portability, and electrical power. Interestingly, Digital Green’s own assessments show that a version of their system where a poster is used in place of the TV/DVD can achieve much of the gains, though at a lower cost-effectiveness rate. Instead of the TV/DVD, we experimented with several kinds of featherweight multimedia, and also with the devices in mediated and non-mediated contexts. In all cases, the audio was copied from video clips used by Digital Green, while the

video was replaced with a paper poster that illustrates the agricultural technique (Digital Green staff assured us that the audio was understandable without the original video). Since video is more visually engaging, we wanted to observe how featherweight multimedia fares in comparison. Our explorations were conducted in a village in the state of Tamil Nadu, India. Attendance at the mediated meetings (Fig. 2), which take place in the evenings at a private home three times a week, stayed relatively constant at approximately 20 village residents throughout the meetings that we directly observed. The village comprises about 50 households or 160 individuals (50 men, 50 women and 60 children). Only 10% of the population could read or write.

Fig. 2. Mediator teaching agricultural information to village residents using featherweight multimedia – poster with audio player

A. 1st Prototype: Poster + Custom Audio, Mediated Our featherweight multimedia device for this experiment was a custom audio device (Fig. 3, left) that was meant to be hung alongside a poster. When pressed, each of the seven buttons on the right play an audio file stored on a removable SD card. These audio clips are captions corresponding to mediator-drawn illustrations on the poster. On the left is a button that stops playback, below which is an on/off switch. For ruggedness and portability, the device is equipped with a handle and is covered with cardboard. A µmp3 player [31], powered by three AA batteries, drives external speakers. The cost of this custom device was approximately US$150, with nearly two-thirds going to the cost of the mp3 player.

Fig. 3. Custom audio player (left). MuVo V100 (right). Scales are not equal.

341

At the time of our research, eight Digital-Green-style meetings were conducted with this device (with no video). Usage of the device varied from taking up almost all of the hour-long meeting to almost none on several occasions. We attended four meetings and spoke with the mediator twice outside of the meeting time. Observations and Feedback: Overall, both the mediator and the audience received the device well. We found that the audience did not get restless even when listening continuously to longer clips of 5 minutes or more, although this is surprising in light of previous research [29]. Half the audio clips on the SD card were played at each meeting. Although people did say they were tired of hearing the same content over and over again, this repetition has been shown to increase understanding. One village resident said that she could listen to the same audio two or three times before getting bored, and in fact needs to hear it that many times to fully understand the content. So, the featherweight device appeared to serve as a reasonable alternative to video. Interestingly, no one else besides the mediator operated the audio player, or expressed interest in doing so. It was seen as something only the mediator should operate. Meeting participants pointed out that since the device was custom-built by the authors, when it broke (for one meeting), a normal repairman could not fix it. They also commented that the device looked ugly. Perhaps the most significant design flaw was the device’s lack of a pause button. Pressing stop “rewound” the audio to the beginning of the particular audio clip. For this reason, the stop button was rarely used. On the positive side, people from neighboring villages heard about the meetings and asked the mediator about them with interest. B. 2nd Prototype: Poster + Off-the-shelf Audio, Mediated Based on the feedback from the first iteration, we abandoned the custom audio player and switched to a commercially available one (Creative Technologies MuVo V100: Fig. 3, right). The commercial device is more durable, is capable of playing more than 7 audio files (10+ hours), and has pause/rewind/fast-forward capability. The total cost of the system was US$38. Being a low-end device, the audio player’s screen cannot show the name of more than one audio file at a time. In addition, it does not have playlists and does not support titles in the local language. The audio player allows grouping of audio files by folder but the user interface for switching between folders is difficult to use. Therefore, we created a numbering scheme where the title of each audio clip consisted of two numbers separated by a hyphen. The first number is a code for the topic, for example (1=azolla, 2=vermicompost). The second number is the clip number within the topic. A printed guide was given to the mediator explaining the code. Navigation is accomplished by skipping forwards and backwards through the list of tracks by flicking a bi-directional self-centering spring-loaded switch. Since this is cumbersome with a large number of audio clips, we only loaded a few topics at a time onto the audio player. Posters were designed similar to the first experiment, and

the sessions were again mediated. Observations and Feedback: Despite some usability issues, this combination of device, poster, and mediation worked very well. In fact, Digital Green has been actively continuing the use of this device for five months, with over 50 meetings conducted to date. They report that adoption rates have been steadily increasing using our featherweight-multimedia solution, and that the rates are coming close to adoption rates with TV and video content. This, together with other benefits of the device such as portability, cost, and less reliance on power grid (Fig. 4), makes featherweight multimedia a powerful contender for this scenario. The experiment has not been without problems: even after several meetings, the mediator had not mastered the bidirectional switch. One cause of the difficulty is the fact that the switch is functionally overloaded: when flicked, it switches between tracks, but when held, it fast-forwards or rewinds. This suggests that controls should be modeless, especially to accommodate users new to electronics.

Fig. 4. Featherweight multimedia working during a power outage!

C. 3rd Prototype: Talking Book, Non-Mediated In this third exploration, we look at a more integrated featherweight multimedia solution: talking books. These devices are relatively cheap, have had some traction in developed countries for childhood education, are built robustly because they are designed for children, and provide access to audio and visuals via a single integrated package. We repurposed a VTech WhizKid [10] device for agriculture information dissemination. The WhizKid can operate in a standalone fashion, or it can be hooked up to a laptop via USB port, in which case the WhizKid tablet and pen are treated as a mouse by the operating system. In the latter mode, it is possible to customize the content and interaction completely, so that an inserted page of our own design can be mapped to audio files we choose. (Should this solution work, the intent would be to work with the manufacturer to produce agriculture content for the device in its standalone mode.) The retail cost of the device without the laptop is approximately US$25. The WhizKid has another benefit in that its robust, colorful design is less intimidating than a regular PC. During our studies, the laptop was hidden so as to not intimidate users. Screen shots from the video were used for visual content (Fig. 5), and relevant portions of audio were copied from the videos.

342

Fig. 5. An audio-augmented page for a talking book

We walked around a village on two different days, for a total of six hours, to various households and requested residents to try out the device in their homes. Besides the second author, a translator and a mediator were also present. Each user was given 20 minutes of free use to work with the device, and was allowed to continue if he/she wished. We provided no up-front instruction, but if the user failed to engage with the device at all after five minutes, the mediator or the translator provided gentle and low-grade assistance. A feedback questionnaire was administered verbally that asked for the user’s background information and opinions about the device. For the second day (N=4), we also administered a verbal test and feedback session after each use, to evaluate how much of the presented information the user had learned and to hear comments about the device. Observations and Feedback: In spite of some initial hesitation, most of the villagers who tried the device were comfortable with it and could use it with little or no assistance. Those participants who took the pre- and post-tests showed increased knowledge of the content presented. Over six hours spread over two days, we were able to find ten people (5 male, 5 female) willing to try the device to learn agricultural content (Fig. 6). Individuals were very hesitant to press anything at first. The elapsed time between when an audio clip had finished playing and the next selection was chosen was initially as long as 10 seconds. With more use, this reduced to about one second. Many users repeated aloud certain pieces of audio information after they heard it spoken from the device. Older users were slower to learn the device and slower to learn the content – they listened to all the clips at least twice. One user kept the pen pressed down while the audio was playing and never figured out that tapping would suffice. On the first day, we had black-and-white printouts of the video. Color turned out to be essential for some of the audio descriptions, and so on the second day, we returned with color-printed sheets. Users did not experiment very much. They never pressed the Pause button unless it was pointed out to them. They only played the audio in sequence, even while listening to the audio captions for the second time to review the material. When questioned later, users said that they feared that they might spoil the device.

During the post-experiment feedback session most of the users were able to answer the factual questions on the agricultural processes (mean of 75% answers correct for N=4). This confirmed that most users had been able to retain the key pieces of information that they had heard. Crowds immediately gathered when a participant used the device. Many of the bystanders often prompted the primary user. Also, there were a number of instances when people who listened to the audio information enthusiastically participated in the post-experiment discussion. One user said that she would be willing to share the device if the device was left at a common place in the village and remarked that it will be quite useful on days she missed the Digital Green video-screening sessions. This suggests that featherweight multimedia can be complementary to other forms of information dissemination.

Fig. 6. A user navigating through the talking book for information.

D. Discussion These explorations revealed some preliminary insights relative to our study questions. In terms of literacy and necessary user background, featherweight multimedia appears able to bridge challenges of illiteracy. Audio playback seems sufficient for this purpose, and the addition of even static visual images helps anchor discussion in mediated sessions. This is not a surprise, but it is assuring that there were no unexpected surprises with respect to adults with little formal education being able to make sense of these devices. Many of the participants in this phase of the study had also been previously exposed to agricultural content on video from Digital Green, yet none explicitly expressed a negative comparison with video. The simple interface of featherweight multimedia also seems to reduce barriers for the first-time user. Users related well to the use of local content and illustrations depicting local surroundings and people, enabling viable user engagement. In several cases, both in mediated and non-mediated use, it was seen that users repeated pieces of information to re-affirm their knowledge. This process of repeating the narration may enhance retention and assimilation of information. For us, it confirms that the users were paying attention in the first place. Featherweight multimedia also seems sufficiently engaging so as to enable effective information dissemination, both under mediated and non-mediated circumstances. In mediated sessions, audiences were not only willing to sit through uninterrupted audio playback of five minutes or more, in many cases, they requested repeated playback of the audio.

343

In non-mediated situations, users expressed initial trepidation handling the device. One remarked that he was afraid he might spoil the device, and others visibly hesitated in early interaction, until their confidence grew. In spite of the initial reluctance and limited direct guidance, most users were able to engage meaningfully with the device after just a few minutes of exposure to the device. During post-experiment feedback sessions, several people expressed a desire for shorter audio clips that summarized longer content. On the other hand, longer audio clips engaged a greater number of passive users, attracted more passers-by, and generated more discussion because of the clips’ longer playing time. Whereas longer clips sustain the attention of a larger group of people, shorter clips cater to an individual’s short attention span. With respect to the need for social support, a human facilitator is essential, at least in the initial introduction of the device, regardless of the device’s perceived ease-of-use. In all of the mediated sessions, the audience fully took in the content without concerns about the device. In contrast, users in the non-mediated sessions needed prompting and encouragement to use the device at all. The need for such mediation, however, appears to decrease with time and also with group interaction. When crowds gathered in non-mediated sessions, the actions of the primary user were influenced by people loitering around. They either voiced common requests or prompted him when the user was stuck, serving as encouragement. Such voluntary peer guidance further alludes to the effectiveness of social support.

findings that audio annotations enhanced comprehension of drawings provided further justification to our intent to add audio to these posters.

Fig. 7. Post-operative do’s and don’ts poster. We added numeric codes to turn it into a “talking poster” when coupled with an audio player.

A. Technology Based on our agriculture participants’ difficulties with the MP3 player interface, we decided to explore a different device for this setting. We wrote a program that displays a large numeric keypad on a HTC Touch smartphone’s touchscreen (Fig. 8). As numbers are typed in, they appear in the textbox at the top of the screen.

VI. EXPLORATIONS WITH HEALTHCARE INFORMATION In this second exploration, we sought a more institutional setting in which featherweight multimedia devices might be used. We partnered with Sankara Eye Hospital, located in Coimbatore, Tamil Nadu, India. It provides free eye-care surgeries to 700-1000 patients per week, funded by the much smaller segment of patients who can afford to pay. Sankara identifies non-paying patients by conducting 3-5 weekly outreach screening camps in villages. The patients requiring eye surgery are bused to the hospital, where they receive free treatment, room and board over several days, followed by a bus trip home. Two of the authors spent a week visiting Sankara Eye Hospital to find opportunities for deploying featherweight multimedia. We observed screening camps in rural areas and shadowed patients around the hospital. After these visits, we decided to focus on improving Sankara’s informational poster on post-operative care, which few patients were paying attention to despite the crucial nature of the information. This information is also read aloud by a nurse to the patients during discharge, indicating compatibility with featherweight multimedia. The poster is 0.9m x 0.6m in size. It explains ten things patients should do on the left-hand panel, while on the right, it lists ten things they should not do (Fig. 7). The images on the poster are static cartoon drawings along the lines recommended by Medhi et al. [26], although the poster additionally includes text captions in Tamil. That paper’s

Fig. 8. HTC Touch running featherweight multimedia application. Unnecessary buttons below the touchscreen are covered.

On the poster, numeric codes are printed in each box (Fig. 7). When one of these codes is entered via the numeric keypad, the appropriate audio caption plays in the manner of museum audio tour guides. No “enter” or “play” button is needed, since the numeric codes are all the same length (in our case, two digits). There is also no “clear” button. If an unrecognized code is entered, both the textbox and the program’s numeric buffer are cleared. These design decisions were made to simplify the user interface as much as possible.

Fig. 9. Ward at Sankara Eye Hospital where we conducted our observations.

344

B. Iterative Explorations We conducted four sessions of testing with the device, each with slightly different usage scenarios and with a different group of patients, in a ward with hundreds of beds (Fig. 9). Around 200-230 patients were registered in each ward, but occupancy ranged from 70-230 at any time. Tests were conducted in the early evening when patients were most likely to be present. A sampling of 75 patients over the four sessions showed that the majority were illiterate; the few who were literate often had difficulty reading text due to poor eyesight. All tests were conducted by one of the authors and a translator. We tried to interview anyone who interacted with the device or poster after they had finished. 1) First Session: Regular Poster, Non-Mediated To establish a baseline, we set up a non-augmented poster, and sat where we could observe people viewing the poster. Over the course of a 90-minute observation, we estimate that 150-200 people (non-unique) passed by the poster. Of those, only 7 people looked at the poster (Fig. 10); the rest passed by without a glance. We questioned 5 of the 7, and among them, only one was able to answer simple questions about the content of the poster. Among the rest, one person even asked us whether it was permitted to sleep on the operated side, despite the fact that the poster expressly prohibits this.

Fig. 11. Augmented poster set up with audio device and speakers

First, the translator used the device in an animated, demonstrative manner for three minutes, gesturing to the numbers on the poster and theatrically pressing the onscreen keyboard. Three patients gathered around, but after the translator finished demonstrating, the patients started to leave. We spoke to one of the three, and he was able to explain the purpose of the device and the correspondence between the poster, the printed numbers, and the resulting sounds. When prompted to try the device for himself, however, he demurred, saying that he was afraid to break the device. During the second two hours, we played a pre-recorded Tamil-language audio prompt, “Learn about eye care by pressing the numbers,” every time a person passed or glanced at the poster, simulating a motion-sensitive trigger. The audio prompt never once achieved its intended goal, and the patients always left without trying the device. Occasionally during the two hours, the translator again either demonstrated the device himself or actively encouraged patients to try it, and managed to get a total eleven people to try the device. Every time the poster was used, a crowd of 212 people gathered (Fig. 12), but would then disperse when the person operating the device stopped.

Fig. 10. The regular poster, with no audio annotation: This is a rare patient who is actually taking the time to look at the poster. Our translator is the gentleman in white; on the left, two women are passing by without showing any interest.

2) Second Session: Audio Poster, Occasionally Mediated We set up the poster with numeric codes in the same location as in session one, but this time we put the audio device in front of the poster on the table (Fig. 11), with speakers positioned behind. The backlight on the device was left on throughout the session. We ran this session for 3 hours, during which time a total of 230 non-unique patients passed by the poster (we estimate 6070 unique patients). Because it was immediately apparent (as with the first session) that very few patients would interact with the poster or the device unprompted, we tried a number of different things throughout the session to encourage them.

Fig. 12. A brave patient interacts with the poster (seen from the back) while a crowd looks on.

Nine of those who tried the device were interviewed. All but two had either glanced at the poster or had been in a crowd of people watching the device being used, then had been taught and encouraged to use the poster by the translator.

345

The other two were instances of patients teaching other patients how to use the device. In one instance, a young man in his mid-20s who was accompanying one of the patients used the device through encouragement by the translator. He then taught another man of the same age, who was also there to accompany a patient. He used the device for several minutes. One of them then came back an hour and fifteen minutes later to continue interacting with the device. In the other instance, a man and a woman stood in front of the poster. The woman was about to use the device, but the man tried it first. After each audio caption is played, they nodded approval together. The woman then took over, and a crowd of six gathered. She pressed and listened to a few captions. Another woman joined her, and the first woman systematically pressed all the numbers, taking 6 minutes. After each caption was played, the first woman made sure that the second woman understood the content. There were at least two people who were not able to understand the use of the device even after the translator’s explanation. For this group of patients, active human mediation seems necessary even to get to a point where patients will try the device, and even then only a small minority will do so. However, more people expressed curiosity and stayed to listen if someone else was operating the device. The audio prompt to invite usage (without mediation) was a failure. 3) Third Session: Audio Poster, Mediated In this session, we asked a nurse to explain the usage of the audio poster to all patients in the ward over the PA system. Immediately afterwards, a crowd of about 12 patients gathered around the poster, and the nurse demonstrated to that group. Individual patients in the crowd then tried using the device assisted by the nurse. Within eight minutes, the crowd dispersed. A second PA announcement an hour later did not result in significant sustained interest. 4) Fourth Session: Audio Brochure, Peer-Mediated For the fourth session, we tried a more intimate form factor and a different kind of mediation. Since patients spend a lot of time sitting on their beds waiting for treatment or recovering, we decided to take the content to them instead of having them come to the poster. To this end, we created an audio brochure, which is simply the poster shrunk onto A4-sized paper that could be passed around. All words and numbers remained legible, and the audio-guide device remained the same. On the first day of this experiment, we taught one of the male patients how to use the device. The patient was asked to use the device for himself, teach others to use the device, and then to pass the device on. The patient took on his responsibility eagerly, perhaps overly so. Over the course of 10 minutes, he demonstrated the audio brochure to 11 patients, letting them hear a few captions before moving on to another patient, without relinquishing the device. The next day, the translator taught a group of several patients to use the device and pass it along (Fig. 13). After a few minutes, the group taught a group of women sitting across from them how to use the device. Two women were then observed sitting at the feet of the men and having a discussion

about the device and the health information. One of the men in the group was a retired 70-year-old teacher. Five minutes later, the teacher was gesturing animatedly to the assembled people around him, demonstrating the device, but keeping control of it. The translator then told the teacher to encourage others to use the device rather than to use the device himself as a tool to teach the content. The teacher then moved to different parts of the ward, teaching other groups and having discussions with them (Fig. 14). Most of the time he did not relinquish control. Often, a member of the group pointed to a number on the brochure, which the teacher then typed in. A few times, other people were observed pressing the numbers themselves. Thirty-five minutes after the teacher got control, he returned the device and the brochure to us, and reported that he had completed teaching all 30 patients who were awake!

Fig. 13. Translator (standing) teaching a group of patients how to use smartphone with brochure

Fig. 14. Patient (a retired teacher) teaching other patients

We interviewed 10 people (besides the teacher) who had either tried the device or learned from the teacher. For the most part, they reported that they felt that the information was very important and were able to recall some key points. There were no negative responses, and no one said that they had difficulty in using the device. One woman even expressed a desire to buy the device, but said that unfortunately she did not have enough money on her!

346

C. Discussion The social framework that we found to work in institutional healthcare settings involves (1) approaching groups, (2) using a portable form factor (in this case the brochure/smartphone), (3) providing small-group tutorials with instructions to the group to teach others. In some cases, one member of the group will take the initiative and take charge of teaching others. We make several further observations: First, technical manipulation of the interface was not a problem for most of the patients (except for two) who experienced the device. In fact, even patients who did not directly interact with the device, but instead observed others operating it, were able to understand what it did and to associate the audio with the visual information. Second, user engagement varied greatly depending on the person and on the circumstances of device introduction. Most patients seemed reluctant, even afraid, to use the device, particularly when it was placed in a public location with a lot of patient traffic, even if they understood how to use the device. Patients said that they were fearful of breaking the device. This is consistent with previous findings on interfaces for the illiterate, where inexperience with electronic devices creates a barrier for first-time users [28]. These fears were not allayed by audio prompts from the device. On the other hand, when given private tutorials, some users would all but monopolize the device, appointing themselves both teacher and technology expert for their peers. Between these two extremes, there was a range of engagement patterns – one point of note was a tendency for crowds to gather around a person who was willing to operate the device, and then to disperse when the person stopped. Third, it seems clear that human mediation of some kind is essential for the device to be used, at least initially. The public poster itself was largely ignored, and the device was used only when our translator encouraged people to try it. On the other hand, peer tutorials were very effective in encouraging use. Fourth, the brochure was more effective than the larger poster. This may have been because patients perceived it less like hospital property, or because it was easier to pass around. Finally, we observed several usability problems with the smartphone application, but once patients were taught how to overcome the problem, they continued using the device. One patient slid his finger between numbers while another pressed too hard. There were also mis-registration problems with the touchscreen, causing numbers to be mistakenly pressed twice or the adjacent number being pressed instead. Most of these issues would be remedied by using a large, physical numeric keypad instead of a touch-screen virtual keypad.

can be easily modified to create devices of varying sizes, form factors (from A5 to poster sizes), and costs. The kit would consist of a board with an array of buttons that trigger audio playback. Sheets of paper could be overlaid onto this board with playback icons positioned over the underlying buttons. This kit would cost US$20-25 for an A4-sized "audio tablet" with a 1GB SD memory card. However, the kit would cost only US$0.50 to make an A5-sized “audio card” that contains 10-15 seconds of audio content. VIII. CONCLUSION In this paper, we presented a broad exploration of featherweight multimedia – combinations of electronic audio and static visuals – that can be put together for significantly lower cost than even very low-cost PCs, but can nevertheless provide a rich, interactive multimedia experience. Within the contexts of agriculture extension in rural villages and healthcare information dissemination in an eye-care hospital, we tried exploratory studies with five different configurations of featherweight multimedia devices, all with illiterate and semi-literate users. Over 75 people directly interacted with the devices, and more than 150 people participated in formal or informal learning sessions in which they observed the use of the devices. The results from this preliminary exploration suggest that most illiterate users can become immediately comfortable with the simple interfaces provided by featherweight multimedia, although fears of breaking the technology and such can create barriers to casual usage. At the other extreme, some users, especially when given short, private tutorials on usage, seem to delight in using these devices and showing them off to peers. We also note that, consistent with much of the literature on technology for development, that social support for the technology is an essential component of impactful usage. IX. ACKNOWLEDGEMENTS Thanks to Joice Sister, M. Saravanan, and Dr. R.V. Ramani at Sankara Eye Hospital; Mahalaxmi and Srikantamurthy at Digital Green/GREEN Foundation; S. Sivaraju, Ajay Khanna, Kalai Arasi, Meera Lakshmanan, and Avinash Setty for translating; and Sid Vishwananthan & family, Leah Findlater, and Gursharan Singh. X. REFERENCES [1] [2] [3]

[4]

VII. FUTURE WORK

[5]

We hope to better understand how to integrate featherweight multimedia into development-focused programs and organizations, as well as to measure their educational impact in comparison to other forms of media. We are also exploring a possibility of a low cost “kit” that can be used to construct featherweight multimedia devices in which the audio and visual content are customizable. This kit

[6] [7] [8] [9]

One Laptop Per Child. [Online]. laptop.org. [Accessed Aug 31, 2008]. Classmate PC. [Online]. classmatepc.com. [Accessed Aug 31, 2008]. International Team Aims to Develop a $12 Laptop During MIT Summit. [Online]. chronicle.com/wiredcampus/article/3219/international-teamaims-to-develop-a-12-laptop-during-mit-summit. [Accessed Aug 31, 2008]. Lourie, S. (1990). World literacy: where we stand today - One billion non-literates. Editorial, UNESCO Courier. July 1990. Medhi, I., Sagar, A., Toyama, K. (2006). Text-free user interfaces for illiterate and semi-literate users. Proc. ICTD. p. 72-82. Books of Hope. [Online]. booksofhope.com. [Accessed Aug 24, 2008]. Q & A about Speaking Books. [Online]. booksofhope.com/pdf/Q%20&%20A%20about%20Speaking%20Books %20.pdf. [Accessed Aug 24, 2008]. SADAG’s school based curriculum evaluation. [Online]. booksofhope.com/research/SADAG%27s%20School%20Based%20Curr iculum%20Evaluation.pdf. [Accessed Aug 24, 2008]. LeapFrog. [Online]. leapfrog.com. [Accessed Aug 24,2008].

347 [10] VTech WhizKid . [Online]. vtechwhizkid.com. [Accessed Aug 24, 2008]. [11] Fisher-Price PowerTouch [Online]. fisherprice.com/us/powertouch/default_flash.asp. [Accessed Aug 24, 2008]. [12] HHS to provide new interactive book of health information to women of Afghanistan and their families. [Online]. hhs.gov/news/press/2004pres/20040803.html. [Accessed Aug 24, 2008]. [13] PicTalk. [Online]. edindiasolutions.com/index-4.html. [Accessed Sep 20, 2008]. [14] Literacy Bridge. [Online]. literacybridge.org. [Accessed Aug 24, 2008]. [15] Tour-Mate. [Online]. tourmate.com. [Accessed Aug 24, 2009]. [16] Prentke Romich Company. [Online]. prentrom.com. [Accessed Aug 24, 2008]. [17] Stifelman, L, Arons, B., Schmandt, C. (2001). The Audio Notebook. Proc. CHI. p. 182-189. [18] Livescribe. [Online]. livescribe.com. [Accessed Aug 24, 2008]. [19] Daisy. [Online]. daisy.org. [Accessed Aug 24, 2008]. [20] Dolphin. [Online]. yourdolphin.com. [Accessed Aug 24, 2008]. [21] Tang, A., Finke, M., Blackstock, M., Leung, R., Deutscher, M., Lea, R. (2008). Designing for bystanders: reflections on building a public digital forum. Proc. CHI. p. 879-882. [22] Vogel, D., Balakrishnan, R. (2004). Interactive public ambient displays: Transitioning from implicit to explicit, public to personal, interaction with multiple users. Proc. UIST. p. 137-146.

[23] Plauche, M., Nallasamy, U., Pal, J., Wooters, C., Ramachandran, D. (2006). Speech recognition for illiterate access to information and technology. Proc. ICTD. p. 83-92. [24] Sherwani, J., Ali, N., Mirza, S., Fatma, A., Memon, Y., Karim, M., Tongia, R., Rosenfeld R. (2007). HealthLine: Speech-based access to health information by low-literate users. Proc. ICTD. p.. 131-139. [25] Parikh, T., Javid, P., Sasikumar K., Ghosh, K., Toyama, K. (2006). Mobile phones and paper documents: Evaluating a new approach for capturing microfinance data in rural India. Proc. CHI. p 551-560.. [26] Medhi, I., Prasad, A., Toyama, K. (2007). Optimal audio-visual representations for illiterate users of computers. Proc. WWW. p. 873882. [27] Digital StudyHall. [Online]. dsh.cs.washington.edu/info/papers.html. [Accessed Aug 24, 2008]. [28] Medhi, I., Toyama, K. (2007). Full-context videos for first-time, illiterate users. Proc. ICTD. p.140-148. [29] Gandhi, R., Veeraraghavan,R., Toyama, K., Ramprasad, V. (2007). Digital Green: Participatory video for agricultural extension. Proc. ICTD. p. 21-30. [30] Talking Greeting Card. [Online]. simaproducts.com/products/product_detail.php?product_id=632. [Accessed Aug 31, 2008]. [31] µMP3 Playback Module. [Online]. roguerobotics.com/products/electronics/ump3. [Accessed Aug 24, 2008].

ICT Governance in Higher Education: Case Study of the Rise and Fall of Open Source in a Gulf University

348

Sofiane M. Sahraoui

Abstract—Open source software is a natural fit in higher

education as both are based on an ‘open science’ model of knowledge development, and collaborative learning. It also helps bridge the gap between knowledge production and its use whenever academics become involved with open source development contrary to the proprietary model where software is produced within the confines of commercial vendors R&D labs. However in higher education, open source has not broken through the bottom of the software stack where it is confined to fulfilling mostly system computing requirements. At the application and visible level, open source adoption is hampered by a number of factors chief amongst these is the lack of reliable ICT governance structures. The problem is exacerbated in developing country contexts where governance in general is subjected to the powerplay between several actors of the higher education scene. The governance conundrum faced by open source adoption in developing countries is illustrated through a detailed case study of an open source project failure at GNU, an American-style university in the Gulf region. The rise and fall of GNU open source project is discussed within a general framework of ICT and open source governance in developing countries higher education sector. Lessons are drawn from the case to recommend a FOSS strategy for development. Index Terms—Case Study, Developing Countries, Gulf, Higher Education, ICT Governance, Open Source Software

T

I. INTRODUCTION

he global higher education (HE) scene has seen an increase in the adoption and use of free/open source software (FOSS). However this presence tends to be restricted to the bottom of the software stack; mainly the infrastructural components of university computing that are not visible to the users [1]. On the application side, the debate does not seem to have evolved beyond the technicalities of FOSS user-

Manuscript received September 22, 2008. S. M. Sahraoui is an Associate Professor of Management Information Systems at the American University of Sharjah, UAE. He is currently visiting with the University of Tunis LARODEC research laboratory. (phone: 216-23554-255; e-mail: [email protected]).

friendliness, availability of support, and the validity of its business model [2]. Yet it is at the application, hence visible, level that the FOSS model offers its greatest promise not only through idiosyncratic solutions that cater to the specific requirements of university and educational computing, but also by providing a model for collaborative learning and capacity building [3]. Whilst resistance to FOSS, beyond the bottom of the software stack, is thought to be a universal phenomenon, the problem is compounded in the developing world with governance issues. In the absence of proper ICT governance in developing countries’ higher education institutions (HEIs), FOSS adoption is left tributary to the whims of the power-play between users, IT departments, and university administrators with the outcome rarely in favor of FOSS and if so hardly sustainable over the long term. There has been little attention devoted to studying FOSS in HE let alone in developing countries’ HE [4]. The few studies that exist mostly focused on inhibitors and facilitators of FOSS in the university environment and few of these dealt with the context of developing countries. A landmark contribution by David and Shapiro [4] examined the extent to which FOSS is being created and used within emerging countries’ HE sector. However, hardly any study exists to depict vividly the process of FOSS adoption and its subsequent becoming in an emerging or developing country context. This paper purports to do just that by chronicling the ill-fate of an initially successful FOSS project in a Gulf university. In the specific context of this case, ICT governance, or lack thereof, is thought to be the root cause for FOSS failures. Details of the ‘rise and fall’ of a specific FOSS project, help illustrate the general governance conundrum faced by ‘innovators’ in a developing country context. The paper is structured as follows: section II places FOSS within a development discourse and section III introduces ICT governance in HE. Section IV illustrates the natural fit of FOSS within HE with a corollary treatment of developing country contexts. Section V chronicles a case study depicting the failed adoption of FOSS in a Gulf university environment. In the final two sections, lessons are drawn to help foster the adoption of FOSS in developing countries’ university environments. II. OPEN SOURCE FOR DEVELOPMENT In developing countries where traditions of free knowledge pursuits are rather restricted by the lack of resources and the inadequacy of governance systems, proprietary software

349

further limits the opportunities for skill development and knowledge transfer [4]. FOSS thus presents developing countries with an immeasurable opportunity to plunge into open and localized sources of knowledge [3] and muster the rudiments of knowledge creation and assimilation beyond the confines of the poorly endowed local environments. FOSS was thus hailed as the most promising avenue for developing countries to achieve ‘technological self-determination’ [5]. However evidence points that developing countries are failing to harness this great potential of FOSS for development [6]. The beneficial effect of FOSS is thought to materialize in developing countries only if (1) basic problems of Internet access are resolved which is far from being the case as “the average African university has bandwidth capacity equivalent to a broadband residential connection available in Europe, [and] pays 50 times more for their bandwidth than their educational counterparts in the rest of the world” [7]; (2) a critical mass of FOSS programmers is nurtured given the lack of skilled personnel available to develop hence support FOSS systems [5] 1 and (3) appropriate governance structures are enacted to follow through with FOSS-friendly policies. The first two factors are infrastructural and as such can be resolved through relevant investments. The third factor, which is governance-related is more critical, and could thwart any attempt to steer clear of the current situation of underachievement. In Brazil for instance, over a third of policies in a sample of HEIs required the procurement of FOSS software whenever available, but in the absence of appropriate governance structures to ensure due implementation of such policies, there is little accountability for their frequent violations [4]. Moreover and in countries that are considered as leaders in IT in the developing world, India being a case in point, FOSS was not found to play a major role in HE organization and delivery [4]. To the contrary, Indian HEIs dedicated a proportionally higher amount of their IT budget to software and license fees in comparison with other countries like Croatia and Bulgaria which are economically better off [4]. India’s involvement with FOSS seems rather situated within the realm of global software outsourcing with a limited knowledge creation and spillover effect. In their survey of a number of HEIs located in 8 developing and transition economies, David and Shapiro [4] found that the number of FOSS courses offered was very limited, hinting that FOSS is more of an industrial and business phenomena than any promise of a new FOSS-inspired learning model premised on a more egalitarian redistribution of knowledge production between North and South. In contrast, Glance et al. [8] survey of FOSS use in Australia, New Zeland, and the UK found it to be well integrated at all levels of university operations be it administration, teaching, laboratory, or research. This suggests the important role of overall ICT governance in a country in the uptake of FOSS initiatives. Indeed, Government policies or lack thereof could be a further hindrance to FOSS adoption in the developing world. Out of a total of 275 FOSS government initiatives in the world, only 8

1 The Global Desktop Project by UNU-IIST seeks to grow the number of FOSS programmers in East Asia (cf. http://www.iist.unu.edu/globaldesktop/)

belonged to Africa and the Middle East with many not acted upon [9]. The following section depicts the process of ICT governance in HE in the context of Open Source adoption. III.

ICT GOVERNANCE IN HIGHER EDUCATION

As ICT becomes ubiquitous permeating all facets of university computing including teaching and learning, it becomes primordial to take an institution-wide approach to identify computing requirements and plan their adoption and dissemination. In this respect, ICT governance provides the set of responsibilities and practices that are exercised to provide a strategic direction, evaluate achievements, manage risks, and generally ensure that institutional mechanisms are in place to implement the ICT strategic plan [10]. Strategic plans encompass the campus management of enterprise systems and services with an assorted strategy for software development and acquisition [2]. Planning covers administrative computing, academic computing, teaching & learning and research. Universities are characterized by a divide between the administrative and the academic with lingering conflicts between the two. This conflict contributes to the ills of ICT governance in HE. For administrative computing, faculty committees, whenever they exist, are called upon to offer their input into a university computing strategy without a real consequence when systems are implemented, as faculty are left out of the loop for such decisions. The adoption of expensive administrative systems is usually perceived by academics as a big opportunity cost because it takes away resources from the core academic mission of the university [11]. For academic computing which directly impacts teaching and learning and more generally the direct working environment of faculty, the stakes for defining an ICT strategy are more significant. ICT components that are dealt with in this regard include e-learning and related solutions, computerbased teaching, the integration of ICT in the curriculum, and the general organization of IT services throughout campus including hardware and software support and resources available for faculty-initiated IT activities in teaching and research. The faculty is usually involved in the elaboration of an academic computing strategy through different channels of university governance. It is at the level of adoption and implementation that conflicts arise as decision-making shifts to administrators especially for enterprise applications like LMS and learning portals. The disempowerment of faculty within HEI governance structures can be accentuated through a procurement system that generally does not involve them [4]. Furthermore, the adoption of innovative uses of ICT for teaching and learning, though prescribed in ICT plans, can be blocked for budgetary reasons heightening the frustration of faculty and the demise of ICT strategies. Lack of evaluation and accountability mechanisms contributes to the impunity in violating the ICT strategic plan and undermining the related governance structures. Though the above seems to depict a dysfunctional university environment, it is very close to the reality of HEIs in developing countries. Worse yet, in many countries,

350

structures of shared governance that would enable faculty participation in the elaboration of ICT strategies are sometimes inexistent. Shared governance being distributed conveys a loosening of control from the center; an idea that is resisted in environments accustomed to centralized decisionmaking [12]. Decisions about ICT for administrative as well as for academic purposes are then taken unilaterally by administrators who are sometimes only remotely connected to HEIs. In his study of HE reform in Tunisia, Sahraoui [13] illustrates how strategic planning for change in that country’s HE system was conducted centrally at the level of the ministry of HE with a token involvement of faculty within ad-hoc structures that were defined outside of the existing structures of faculty governance. Within such governance framework, initiatives to embed open source in teaching and learning become tributary to the power-play of forces between different stakeholders in the ICT governance game. Similarly and though FOSS might be extensively used within developing country contexts, it is generally in the absence of clear software policies and at best within the framework of IT ‘neutral’ policies [4] but rarely within the mandate of proFOSS ICT policies. The next section deals specifically with Open Source governance in HE highlighting the unique fit between FOSS and HE. IV. HIGHER EDUCATION AS A NATURAL HABITAT FOR OPEN SOURCE SOFTWARE At a time when organizations embark upon IT infrastructure upgrades and within increasingly austere budget environments and Internet-based open computing platforms, FOSS has emerged as a promising alternative to proprietary solutions. Whilst the cost factor and the probing quality and prominence of open computing products should logically tilt the balance to the advantage of open source over proprietary solutions [14], [15], IT infrastructures remain largely proprietary [11]. The cost, reliability and security implications for organizations with mainly proprietary platforms are significant, yet limited consideration seems to be devoted to alternative open-source platforms [2]. Moreover the response to the ‘IT paradigm shift’ engendered by open source computing remains largely limited to the bottom of the software stack [16], spearheaded mainly by the increase in server installations of the Linux operating system which stands at roughly 20% of total installations and the Apache Web server which powers more than two thirds of corporate websites [11]. Indeed and beyond this level, open source remains largely absent from the ‘application layer’ such as enterprise applications or productivity tools [17]. Whilst many factors can be ascribed to this including the gap between the technologists (developers) and the non-technologists (users) who hold different value-orientations regarding control and openness [17]; the risk attached to ‘entrepreneurial’ and individuallycentered modes of software development [18]; and the higher IT skill requirements of open source to configure and maintain software in the absence of conventional service contracts [19], this paper contends that governance factors especially within developing country contexts are by far the biggest stumbling

block against open source adoption beyond the infrastructural level [22]. These factors are examined below within the context of HE. First, we highlight the natural fit between FOSS and HE as they are both underlined by an open learning model. A. Open Learning Model The most prominent examples of FOSS initiatives in universities consists of teaching FOSS tools like Linux, PhP, Perl, and MySQL [17]. Educational portals like uPortal and learning management systems (LMS) like Moodle and Sakai have also been emerging as new foci of FOSS academic initiatives [21]. As of late, the development of the Kuali financial system as an enterprise management solution for HEIs seems to be porting the open source tide to the administrative side of academe [22]. Class wikis, wherein students and professors alike jointly construct learning artefacts within a Wikipedia-like environment are becoming a dominant collaborative learning tool and have been integrated within commercial LMS like Blackboard for instance [23]. Prominent institutions like MIT and Harvard use similar platforms to enable open access to teaching and research resources [23]. Projects of a more global nature are also being developed with open source tools including providing online textbooks to the developing world [24]. For a comprehensive list of FOSS tools used in education, see [3]2 . Open source has been making headways in HE through leading infrastructure products like Apache, Linux, MySQL, Firefox, and Tomcat. In the US, a third of HEIs use some form of open source infrastructure software although some are not aware that they are doing so [14]. However its impact beyond the bottom of the software stack has been more mitigated [1]. Moreover few academic institutions seem to have a strategy to integrate open source in their academic offerings so as to ensure students’ familiarity with an increasingly dominant reality in the business world [22]. On the administrative side of computing, few FOSS alternatives are being considered despite the greater concern and dissatisfaction with proprietary solutions [26]. The support factor especially for enterprise applications is usually invoked first against FOSS adoption although some research found support for FOSS to be equivalent or better than for proprietary solutions [24]. Doubts about the legitimacy or credibility of the development model of FOSS are also routinely expressed. This has a double cost, the first financial incurred through the payment of licensing fees and expensive upgrades, and the second is more fundamental and consists of the opportunity cost in foregoing the collaborative learning model of FOSS [24]. Even if one were to concede that the TCO of open source solutions is as high or higher than that of proprietary software, the advantage of FOSS over proprietary software is more fundamental and has to do with the open and collaborative learning environment afforded by FOSS and denied by the proprietary. Using the model of sourceforge.net or one of its offshoots in the educational field, such as schoolforge, eduforge or any other similar virtual development environment, FOSS tools are amenable to transform the 2 A repository of FOSS for http://www.iosn.net/education/software/.

HE

is

downloadable

from

351

classroom environment from a teacher-centered environment where knowledge is distilled by the instructor to a learnercentered learning environment where students develop skills and competencies beyond the mere acquisition of content knowledge [24]. The combination of access to the source code with the experiential learning within collaborative development environments affords endless possibilities to build learning artefacts that go beyond the mere use of software tools and could radically transform educational practice for teachers and learners. “this marks the emergence of global science and knowledge as a global public good that rest on an ethic of participation and collaboration based on the co-production and co-design of knowledge goods and services…the role of nonmarket and non proprietary production promotes the emergence of a new information economy that both depends upon and encourages great individual freedom, democratic participation, collaboration and interactivity” [27] With open source, the development process is as important as the outcome by enabling a higher form of learning that is embedded within a set of social interactions and culturally organized activities wherein students become knowledge builders and creators [24]. Likewise, “An academic environment where FOSS is prevalent will encourage staff and students to tinker and experiment with, and participate in the development of FOSS that may eventually lead to innovative solutions.” [3] The inherent ‘entrepreneurial’ characteristic of the FOSS model should make it more suitable to innovative cultures. Such is supposed to be the academic environment. Hence, the quest for open source in education seems ironic given that the academic community was the breeding ground for software development before programmers knew the meaning of ‘proprietary’ [28]. Most code was indeed created within academic-like environments before software engineering became an industry in the early sixties [24]. It is only the aggressive drive for profitability in a booming software market that required the closure of the source code. In this light, the discourse on intellectual property rights provided a moral mantle to what is otherwise a dominantly commercial drive. Once software development was evacuated from the university and unto the corporation, a divide soon emerged between corporate software developers and its academic users. As commercial software developers have little appreciation for HE software use [14], FOSS is amenable to bridge back this gap and reconcile academe with its original vocation of creation and innovation, hence merging technology-producing with technology-using [24]. The cultural predisposition for open review and exchange among peers in academe should make for the adoption of FOSS a natural fit [28] as it helps harness HE’s vast innovation capability [12]. Both rely on a model of ‘open science’ research collaboration for their knowledge/software creation [4, 29]. The current situation of HEIs which have evolved as ‘technology consumers’ rather than ‘technology innovators,’ contradicts the academic ethos based on academic freedom and the open dissemination of knowledge and information [3]. The efforts to adopt open source in academic institutions necessitate its integration in curricula as well as within the university environment as a whole, not only as a

toolset but also as a development ethic akin to the universalism and openness of academic environments. An added benefit for teachers is their reintegration within communities of practice as many of the open source projects for HE are being collaboratively developed across HEIs [14] 3 ; in contrast with their current isolation within the meanders of functionalized teaching [24]. The narrative below, depicting the failed adoption of open source at Gulf university is a vivid illustration of how the ills of governance can bring to a halt, ambitious changes in the curriculum. Further discussion of governance issues in HE will be invoked as relevant details of the case unfold. V. OPEN SOURCE PROJECT IN A GULF UNIVERSITY Gulf National University (GNU) 4 is a recently established American-style university in the Gulf and has quickly emerged as a leading institution of higher learning in the region. The Business School (B-school) offers degrees in the traditional fields of management along with management information systems (MIS). The MIS program in the B-school emphasizes the application of IT to business processes, and engages in service and research which serve the IT needs of the Gulf region. The integration of open source into the MIS curriculum was the embodiment of a ‘strategic direction’ in the MIS department to go open source on several fronts; research, teaching and administration. However no overarching vision or plan to adopt FOSS existed anywhere else in the university although FOSS exists at the bottom of the software stack and many FOSS academic initiatives were in existence at the time the MIS FOSS project started. The department established an open source interest group for this purpose, made up of faculty, students and administrators. This followed the coalescence of research interests around the concept of open computing among many members of the department’s faculty 5 . In general, the objective of GNU’s open source group was to serve as a knowledge gateway for open source research and practice and its subsequent dissemination in the region. A. ICT Governance at GNU Following a shared governance model of HE, GNU put in place nominal structures of collaborative governance to utilize faculty input into its policy-making including about ICT. Though purely consultative, this model of shared governance, articulated around a set of standing and ad-hoc committees, could have enabled the active participation of faculty into its planning and governance processes. Indeed this worked at the beginning when the university needed faculty resources and input to establish its basic processes of academic governance. But as the university matured and the corporate model of governance took over, faculty role was gradually marginalized especially in matters that were not perceived to be of their 3 Eduforge.org is a virtual collaborative learning and exploratory environment designed for the sharing of ideas, research outcomes, open source educational software, and tools within a community of learners and researchers. They have many links to a range of FOSS initiatives in education. 4 Alias 5 The group coordinator chaired the open source track at a major international conference held in the USA in 2004.

352

direct concern, chief amongst these administrative computing. Moreover academic computing and its corollary the university ICT strategy were to fall prey to the same tide of corporatism in university governance as the university administration created shadow committees mirroring the existing faculty committees to perform strategic ICT planning. The ‘administrative’ committees were responsible to work on basic plans and submit them for examination to the university administration, and hence to the faculty for consultation. It did not take a long time for the faculty committees to cease to function altogether and lose credibility as they were being increasingly marginalized by a university administration that started hammering the motto that ‘faculty belonged into the classroom’ to justify its one-handedness in the management of university affairs. The deterioration of the governance atmosphere in the university reflected at all levels with the disempowerment of faculty trickling down to academic units where Deans and appointed Chairs held absolute power. It is worth mentioning that no formal ICT strategy was in place at the time and that ICT decisions including strategic ones were taken haphazardly and without any significant faculty input. Hence, GNU and the B-School did not have clearly defined strategies for using software either for administrative or academic purposes. Although there had been a clear commitment to proprietary software platforms especially for administrative computing (i.e. ORACLE, SCT Banner), software used for educational purposes seemed to be driven by ad-hoc requests from academic units. The B-School in particular had been toying with both WebCT and Moodle, a proprietary and open source LMS respectively. The university would later reverse its strategy by imposing Blackboard as a mandatory solution for course management. GNU open source project was initiated within these governance premises. B. Open Source Initiatives at GNU The key and initial elements of the project consisted of the implementation of an ‘Open-Lab’ FOSS competence incubator, integrating open source in the MIS curriculum by drawing from a FOSS toolset, hence building open source skills to be later deployed in the university environment and subsequently in the market place. These efforts were in line with the local HE ministry’s and university’s strategies for training technically skilled graduates that are fully prepared for the local marketplace although these strategies were not explicitly stated in terms of FOSS skills. On the curriculum side, Linux and other FOSS productivity tools were introduced within business foundation courses. The MIS department also offered a single shot (special topics) course dedicated to FOSS 6 wherein business-related aspects of open source were examined alongside a lab component primarily based on software demos and tutorials. Moreover the FOSS collaborative model was used to organize student learning. Students were familiarized with virtual collaborative environments as vehicles for student-driven knowledge generation. Department faculty were invited on a weekly basis to partake in the collaborative learning experiment and encouraged to consider implementing it in their own courses. A pedagogical seminar was also held as part of the regular 6

MISXXX--Special Topics in MIS

activities of the university faculty development center where the potential and first results of the use of the FOSS collaborative learning model were presented. The ‘Open Lab’ was installed in the basement of the BSchool building with a Linux server running Novell SUSE 7 , an Apache Web server, and a number of open source applications including OpenOffice, MySQL and IBM Eclipse suite which were used by students to fulfill the course requirements of the ‘Special Topics’ course and other courses where open source modules were introduced. However, no detailed pedagogical plans were developed to integrate FOSS in the curriculum as this was left to the discretion of faculty members. Following the initially successful, though not generalized, introduction of open source modules within existing courses, the thinking evolved into considering a full-fledged open source toolset for MIS students first, which would be generalized to business students later on. This toolset would not be limited to enumerating possible solutions for various course requirements but will dig deep into issues of open source-centered curriculum design. In other words, both the toolset and the FOSS collaborative learning model were to be combined in reshaping the MIS program. Hence, a fullfledged open source infused curriculum both in terms of content and pedagogy was being contemplated. Feasibility studies illustrating the TCO benefit of using open source solutions rather than proprietary ones were elaborated for the restructuring of the existing MIS program as well as the creation of a new graduate offering in collaboration with other academic IT units in the university. This financial component was accompanied with an integrated infrastructure proposal to embed an open source-based learning environment in the BSchool. The general approach was to provide resources to those who were willing to join the FOSS bandwagon and to make of FOSS a dominant reality and a culture of excellence which everybody was willing to belong to. The project was boosted by a university grant and the allocation of laboratory space for the open lab. However this support did not derive from a buyin into the project at the university or college level or from any FOSS strategy implementation which did not exist hitherto. Rather the two main proponents of the project at the time held sway in university and college decision-making for reasons unrelated to the project itself. C. Project Methodology & Outcomes Whilst project outcomes were defined for the MIS department in the beginning, the plan was to roll them over to the rest of the B-School in a second phase and all of GNU at a later stage. The ultimate objective was to make of GNU a showcase for the successful implementation of open source in academe before advocating it outside. The deliverables that were identified at the start were: 1) A portal as a central reference point for FOSS initiatives in GNU; 2) State of the art open computing lab including the open source toolset and assorted training materials; 7 The Linux alliance program was first finalized with SUSE Linux AG from Germany. In the meantime and before implementation started, Novell took over SUSE and reneged on the alliance forcing a costly renegotiation of the alliance program.

353

3) Demonstrable prototypes of some components of the toolset with applications; 4) A lap-top based open source toolset for business students; 5) Integrated program and course development using the toolset and other open source components. In particular, two courses were fully deployed over Moodle.org, an open source LMS, to show the viability of open source LMS in comparison with commercial ones such as WebCT or Blackboard. A workshop was also held for GNU faculty, through the faculty development center, to demo the course implementations on Moodle.org; 6) Workshops and seminars with the participation of industry as well as academic speakers. An open source research and industry forum was held in the spring of 2005 where the findings of this project were presented along with a survey of open source initiatives in the region and elsewhere. The e-government program of Bahrain which was developed on top of an open computing platform was showcased for the occasion; 7) Establishment of a FOSS user group with the active participation of students and faculty all over GNU; 8) Academic alliance programs with FOSS vendors including training, support, and certification for faculty, IT instructors, and students alike; An offshoot of the FOSS project consisted of the development of a portal for collaborative student learning (CLOSE) 8 . The project was designed by students as part of the special topics course and won a regional e-biz award. Hence the overall FOSS project was initially quite successful but started running into major problems at a later stage. Its gradual demise is discussed below: D. Initial Success During the first two years of planning and implementation, the project went full steam and many outcomes were achieved, some even ahead of time. The project was at first a major success in its open lab component. For a fraction of the real cost, the department of MIS at the B-School established a state of the art lab that helped GNU support open source education and training. The hope was that in a second phase, the open lab services would be offered outside of the B-School. Regarding the research and pedagogical aspects of the project, several research works were generated and published in academic and professional journals as well as international conferences. However, a major collaborative piece on “implementing FOSS as pedagogy” which was implemented experimentally in the special topics course described earlier never saw the light because the project quickly lost steam after resources started being withdrawn from the project following the departure of a key collaborator who had marshaled most of the initial resources. Similarly, a major outcome which was the development of an open source education and training portal, planned to be made accessible to GNU students also aborted along with a proposal for a permanent elective course on open computing. An obscure policy on university Web publishing was invoked by the coordinator of academic computing to block any such portal going live. A similar fate 8

CLOSE: Collaborative Learning for Open Source Education

met the student CLOSE project. Furthermore, the project came short in terms of developing a full-fledged open computing program either at the undergraduate or graduate levels. It became clear that such programs would not be adopted as alternatives to existing curricula. To take root, changes of a similar extant would have to clear several ‘governance hurdles’ including the all too-conservative college curriculum committee and the overtly politicized university curriculum committee. The absence of specific policies for FOSS adoption left potential proposals open to arbitrary opposition from many parties. Though the project came short of achieving its overtly optimistic objectives, it could still be considered successful at this initial stage especially with the gradual infusion of open source concepts and tools in the MIS foundation courses which were required courses for all incoming business students. E. Project Roll-Back A series of events coincided not only to stop development and further achievements of the project outcomes but also to roll back the initial achievements to an extent that only individual faculty initiatives to teach open source content within the foundation courses survived. The main reasons for the downturn were the following: 1) The MIS Department Head who championed the open source strategy within the B-School moved to a different institution in a different country and continent altogether. It became clear soon after that the open source project lost its main sponsor who sat in the B-School’s Council of Chairs, hence defended the project’s claim for college resources. The decision to reallocate the open lab to the graduate program was uncontested as it was taken by a body with no representation from the open source project group. This decision was officially motivated by the swelling of student numbers in graduate programs, which were bringing in substantial revenue to the university, and the ensuing need to provide graduate students with better and even privileged access to learning resources. 2) The establishment of a new testing center in the university brought along a renewed requirement to teach computing material that led to the ICDL certification; which is primarily based on proprietary concepts and software 9 . Yet again an undeclared ICT strategy was invoked to bring back proprietary tools into the curriculum. It is worth mentioning that no consultation of any kind took place to establish the testing center and subsequently to require that students be trained in the proprietary skills that were necessary to acquire the related certification. This led to a redesign of the basic introductory IT courses, generally at the expense of open source content. 3) Within coordinated courses, some instructors had throughout expressed dissatisfaction with the inclusion of FOSS in the IT skill set taught to students. Their argument was twofold; first by committing students to 9 Despite its claims of ‘vendor neutrality’ in its descriptions of the skill sets, training materials and test centers are almost predominantly Microsoft oriented.

354

FOSS skills, we were allegedly decreasing their employability given the dominance of the proprietary in the corporate world, and second; there is a lack of good didactical material for the delivery of open source content. However, it was always clear that some were never really impressed with the open source phenomenon and even found it lacking credibility and/or having little chance to forge ahead in the long term. Therefore, they jumped on the first occasion of a shift in the balance of power against the open source tide to help ‘shoot it down.’ 4) The university decided to adopt a standard and proprietary platform for LMS, one based on Blackboard. The decision was made based on a proposal by the administrative IT committee. A last minute push by the faculty IT committee gave a respite to Moodle courses for three years with an injunction to move to Blackboard at the end of that period. Resources were nonetheless withdrawn from Moodle, including server space, bringing into a crisis the fifty or so courses that were hosted on the Moodle server in the B-School and burdening the few faculty volunteers who maintained the Moodle server. The three year period has just come to an end and the plug has been pulled forever on the Moodle experiment despite the recurrent problems with Blackboard in contrast to the greater stability of Moodle which is fast becoming the de facto standard for virtual learning environments [30] 10 . Whenever faced with such argument, the academic computing coordinator and his staff invoked intellectual property issues and other legal risks related to the use of Moodle. Yet again a major commitment of ICT resources for the long term was done in the absence of an explicit strategy and within the framework of a weak governance structure. 5) Proprietary vendors became aware of the threat of open source especially when championed by competent and knowledgeable insiders. As a result, they assailed GNU with attractive offers of academic alliance programs which were very enticing, especially in view of augmenting university graduates marketability. Since the unfolding of the open source project, GNU has entered into strategic alliances with major proprietary vendors covering roughly every computing need of students. Only the laptop initiative developed under the umbrella of IBM ThinkPad University survived until last year yet with a gradual marginalization of its open source component. It is worth mentioning also that an academic alliance with another open source provider which would have given a critical boost to the open source project in GNU fell through because of its lack of competitiveness compared to proprietary ones. This was used opportunistically to undermine the competitive claims of open source. The alliance was sought in this particular case not because of the need to access software that was anyways readily available in the public domain but mainly to provide a mantle of legitimacy to the open source project. The

10 It is operating in nearly 50,000 sites including universities and other types of educational institutions (cf. http://moodle.org/stats/)

vendor lacked a necessary strategic vision to forego profits in the short term for bigger gains in the long term. At present, open source education in GNU is limited to few software demos and a single class lecture within the foundation MIS course. Some students and faculty elect to use popular FOSS tools in system design projects. Web Services and PhP for instance have become central to the development of e-business solutions. However, these remain isolated initiatives and not ones that are central to the delivery of MIS education. On the pedagogical front, there are no remnants of the FOSS model of collaborative learning. Ironically, the massive accreditation literature that the university generates on a continuous basis keeps hammering the theme of student-centered learning without either planning for it or creating the right governance system for it to emerge. Worse yet, whenever a credible experiment came into existence, it was quickly neutralized. VI.

PROBLEMS OF ICT GOVERNANCE IN DEVELOPING COUNTRIES

It is useful to relate the above failures to the ‘developing country’ context of the project and especially its governance aspects: 1) Being in a developing country context, the university or the college lacked any form of significant strategic planning for IT; be it for academic or administrative computing. This led to haphazard decision-making for IT adoption and strategic initiatives. This applies to the inception of the FOSS project itself which was initiated within a policy vacuum. Indeed if a strategic IT plan including provisions for FOSS in university computing and curriculum development were to exist, GNU FOSS project would have survived. Furthermore, in the absence of clear plans and institutional mechanisms to ensure their implementation, the university governance structure would inevitably favor ad-hoc plans and projects of those in power, for as long as they are in power. In sum, most governance structures in developing country contexts are channels for autocratic leaders to extend their rule to reach every corner of the organization. In other words, powerful individuals supersede governance structures and not the other way around. 2) The main sponsors of the FOSS project failed to forge a common vision on open source across the B-School and university-wide. The idea was mainly pushed through the B-School’s IT committee and was since the start misconceived as being yet another IT project. Resources for the project were initially marshaled thanks to the negotiation skills of the MIS department Chair rather than through a strong discourse of legitimacy that should have been aimed at the bulk of faculty in the college and beyond. Better yet, such a discourse of legitimacy should have imprinted governance structures themselves by engendering ICT plans advocating FOSS as a viable alternative if not a preferred course of action for academic computing and IT curriculum development. Vigorous and

355

active leadership supporting a FOSS agenda at the higher level of university leadership is also a must [11]. When faced with a roll-back in resources for the project, the FOSS group could not invoke any such policy and had to yield to the otherwise much more ‘legitimate’ argument of learning requirements of graduate students for instance. In the absence of institutional mechanisms that would guarantee a progressive unfolding of the FOSS project, the intrinsically anti-FOSS governance structure could have been tamed only through a critical mass of proFOSS power players at all levels of the university decision-making structure. 3) The FOSS group which started with a nucleus of faculty members from the MIS department soon expanded to include other members of the faculty especially from the College of Arts and Sciences, including a sizeable number from its Intensive English Program. It is interesting to note that most members were from the social sciences and not from engineering or computer science as would be expected. One possible explanation is the ‘militancy’ side of the project agenda which advocates a vendor-free and open environment for computing where the digital divide is bridged and innovation is taken out of the corporate mould. It is unlikely for similar ‘militant’ agendas to develop in the overtly rationalist environment of computer and engineering sciences. However, it is precisely this perception that open source was being pushed by ‘militant zealots’ in different quarters of the university that probably accelerated its demise. Militancy is essentially antithetical to formal governance. The outcome could have been more positive if this militancy agenda was diluted within more acceptable discourses of curriculum change and through existing channels of academic governance. The FOSS group itself, by its mere existence as a para-academic structure, posed a challenge to official governance. The FOSS group should have concentrated on lobbying for its own acceptation as a legitimate player within the university governance structure instead of situating itself in opposition to it. 4) Proprietary vendors spread fear, uncertainty, and doubt (FUD, [11]) intimidating Academic Computing into discontinuing the open source project. With the consolidation of key proprietary offerings in the HE software market (exp. Blackboard & WebCT; Oracle & PeopleSoft), proprietary vendors leverage is increasing [11]. This was most visible with the Moodle vs. Blackboard battle. This is again a governance issue at stake that materialized through a failure to uphold the necessary control requirements in IT decision-making. It has to do yet again with the developing country context of the university. With the demise of open source, proprietary vendors rushed to forge strategic deals that would ensure their perennial interests. It goes without saying that proprietary vendors wield a lot of influence over university procurement decisions in developing countries, precisely because of the lack of transparency in governance processes [1].

VII. LESSONS FOR DEVELOPING COUNTRIES Though the above developments might sound fatalistic and insinuate the uselessness of progressive efforts to adopt FOSS into a developing country HE context, the author remains adamant that such efforts could be successful if they adopt a governance-aware strategy. The problems encountered during the implementation of a FOSS academic solution in GNU can be used to outline such a successful strategy: - Start at the governance level: It is more important to create sustainable governance structures for ICT planning and implementation than to rush into specific projects within ill-defined structures. In the case at hand, the two project champions should have leveraged their ‘temporary’ power to create FOSS conducive structures within the B-School first and at a wider university level subsequently. - Stress Open Learning: The FOSS model for teaching was advocated more for its collaborative learning potential than its TCO and other technical advantages. However beyond the conventional discourse opposing FOSS to the proprietary, a collaborative model of learning is expected to stir stiff resistance amongst the conservative academic establishment, which is well entrenched within the master delivery model. Yet again, the changes brought about the GNU project seem too radical and too quick. A FOSS-driven model should rather be infused within the general discourse about change towards student-centered learning. As accreditation and quality assurance efforts accelerate, FOSS should be mirrored as central to the changes sought. - Diffuse the battle with administrators: FOSS adoption should not be approached from a software adoption perspective but rather from the standpoint of collaborative learning and reconciling academe with the spirit of openness. There is a lot of confusion out there with regards to the merits of FOSS and its long term viability. However, nobody can contest its open learning capabilities. Administrators can oppose it on technical/financial terms but will be reluctant to do so if it is presented as the core of a new learning model. - Build university-wide capacity and awareness: FOSS initiatives should not necessarily start with MIS and computer science departments by insisting on programming and similar courses. Emphasis should rather be on the use and integration of collaborative learning platforms across the curriculum. In the GNU project, a large number of faculty from the Intensive English Program came on board through the use of Moodle, the open source LMS. - FOSS as Pedagogy: A critical problem that faced faculty adopting the FOSS toolset in the B-School’s MIS department was the lack of a collaborative learning pedagogy. Emphasis should be put as much on developing the rudiments of such pedagogy as on the toolset itself.

356

In essence, all of the above is feasible within a functional HE governance premised on faculty participation and their supreme expert authority on matters of teaching and learning. HEIs in developing countries will not achieve their development function for as long as they remain an addendum to the state bureaucratic system. On the other hand, private universities in the developing world have not amassed a critical mass of resources, either human or financial, to provide any significant alternative. The ICT for development discourse in HE should not be centered on infrastructural requirements, which are still important to fulfill, but highlight the central issue of HE governance based on autonomy, faculty governance, and ethics and social responsibility. REFERENCES [1]

R. J. Abel, “ Best Practices in Open Source in Higher Education Study The State of Open Source Software.” March, Lake Mary, FL, the Alliance for Higher Education Competitiveness, Inc., 2006, pp. 1-47. [2] P. Masson, “Barriers to the adoption of open source: Personal and professional observations” World Campus, PennState University, 2007. Available: http://blog.worldcampus.psu.edu/index.php/2007/04/17/ [3] T. W. Tong. “Free/open source software: Education primer” International Open Source Network, UNDP, 2004. Available: http://www.iosn.net/ education/foss-education-primer/fossPrimerEducation.pdf [4] P.A. David, J.S. Shapiro. “Higher education institutions and the global role of free/libre and open source software.” Report on findings from the FLOSSWORLD survey of developing and transition economies, 2007 Available: http://www.oii.ox.ac.uk/research/FLOSS_HEI_Report.pdf [5] United Nations University. “Free Software in Developing Countries Vital to Future Prosperity and Good Governance: UNU Technology Experts.” Media Advisory, 2006 Available: http://www.unu.edu/media/ archives/2006/files/mre-iist-3-06.pdf [6] V. van Reijswoud, A. de Jager. « Free and Open Source Software for Development,” Polimetrica, Monza: Italy, 2008. [7] K. Gakio “African tertiary institutions connectivity survey (ATICS),” 2006 Report, International Research Development Center, Canada. Available: http://www.aau.org/renu/docs/ATICS2006.pdf. [8] D. Glance, J. Kerr, A. Reid, “Factors affecting the use of open source software in tertiary education institutions”, First Monday,Volume 9,No. 2, Feb 2004. [9] J.A. Lewis “Government open source policies” Center for International Strategic Studies, July 2008. Available: http://www.csis.org/media/csis/ pubs/0807218_government_opensource_policies.pdf [10] S. Bacon, T. Dillon . “ The Potential of open source approaches for education,” Futurelab, 2006, Available: http://www.futurelab.org.uk/ resources/documents/opening_education/Open_Source_report.pdf [11] P.N. Courant, R.J. Griffiths, “Software and collaboration in higher education: a study of open source software,” Available: http://www.ithaka.org/strategic-services/oss/OOSS_Report_FINAL.pdf [12] D.A. Wheeler. “Why Open Source Software / Free Software (OSS/FS, FLOSS, or FOSS)? Look at the Numbers!” 2007. Available: http://www.dwheeler.com/oss_fs_why.html. [13] S. Sahraoui. “Implementation of a New Degree Structure in the Context of Higher Education Reform in Tunisia.” Unpublished Master dissertation, University of Oxford, 2007. [14] J. Maguire, “Open source on the Brink”, in Enterprise Linux IT pp.1-2, Sep 11 2003, Available http://www.Linuxenterprisenews.com/ perl/printer/22278/ [15] G. James, “Electronic Design Automation”, in Electronic Business, Dec, 31 2005, 12, p.24. [16] S.W. Van Rooij, “Perceptions of Open Source versus Commercial Software: is Higher Education Still on the Fence?”, in Journal of Research on Technology in Education, 39(4), 2007, 433-453. [17] S. Yegulpap, “Open Source: When things fall apart”, in InformationWeek, Manhasset, May 2008, issue 1184, p. 20. [18] G. Hein, “Open Source Software: Risks and Rewards” 2004 Available http://net.educause.edu/ir/library/pdf/ECR0405.pdf

[19] S.K. Sowe, L. Angelis, I. Stamelos, Y. Manolopou-los. “Using Repository of Repositories (RoRs) to Study the Growth of F/OSS Projects: A Meta- Analysis Research Approach,” Third International Conference on Open Source Systems. Limerick, Ireland, 2007, pp: 147160. [20] J. Kenny, “ Moodle nudges forward: Can any other learning platform challenge Moodle’s dominance?”, in Guardian Education Supplement, January 8th 2008, p.10. [21] J. Cox, “Universities build Open Source Applications”, in Network World, March 31st, 2008, 25, 13. [22] C.N. Davidson, “We can’t ignore the influence of digital technologies” The Chronicle Review, March 23rd, 2007, B20. [23] A.L. Foster, “Providing online textbooks to the developing world” Chronicle of Higher Education, 54, November 16, 2007. [24] M.A. Peters, (2008) Globalization and the virtues of openness in higher education,” Global e-Journal. Available: http://global-ejournal.org /2008/08/29/globalization-and-the-virtues-of-openness-in-highereducation/ [25] A.H. Moore, “Lens on the Future: Open-Source learning”, in EDUCAUSE Review, Boulder, Sep/Oct 2002, Iss. 2, Vol. 37, p.42. [26] J. Marshall. “Negri, Hardt, distributed governance and open source software,” Journal of Multidisciplinary International Studies, Vol. 3, No. 1, January 2006, pp. 1-25. [27] R.T. Watson, M. Boudreau, P. York, M. Greiner, D. Wynn, “Opening the Classroom”, in Journal of Information Systems Education, Spring, Vol. 19, Iss. 1, 2008, pp. 75-86. [28] N. Bezroukov, “Open Source Software Development as a Special Type of Academic Research” (Critique of Vulgar Raymondism), FirstMonday, 1999, vol. 4(10).

357

ICTD State of the Union: Where have we reached and where are we headed Rabin Patra

Joyojeet Pal

Sergiu Nedevschi

[email protected] Department of EECS University of California Berkeley, CA USA

[email protected] CIS, School of Information University of Washington, Seattle, WA USA

[email protected] International Computer Science Institute, Berkeley, CA USA

Abstract— In this study we examine the history and growth of ICTD since the 1990s. We underline the trends defining this research field and examine the progress in research areas that have come to dominate discussion in ICTD through a thorough literature review of the last decade of ICTD work. In order to answer questions pertaining to the rigor, impact and significance of ICTD, and to compare the expectations and perceived achievements with respect to different development goals, we interview 50 expert ICTD researchers and practitioners. We analyze these results to understand stakeholders' opinions on the past performance of ICTD, both as an academic field and as an area of development practice, and identify defining ideas on the potential directions for the future of ICTD. This study is work in progress and we have continuing research in this area, the subset presented here is rigorous and ready for wider discussion. 1 Index Terms— ICTD, developing world

T

I. INTRODUCTION

HE Scandinavians started it. It may be a lost piece of trivia somewhere, but ICTD as we know it began with the community computer rooms in Scandinavian villages. Starting with the first village (Fjaltring in Denmark), there were several small tele-cottages throughout Sweden, Denmark and Norway by 1985, and it wasn’t until a few years later that Sri Lanka became the first developing country to get its community technology access center [1]. The term ICTD started appearing in academia and industry with minor alphabetical variances such as ICT4D, ICT4B (billions), IT4D and so on by the mid 1990s. The years preceding this were formative in the growth of interest in the subject around the world. The opening up of Eastern European economies coincided with phenomenal periods of growth in China and India. The technology boom in the United States featured a large pool of engineers from various parts of the developing world. The spillover economic effects of this international technology workforce ranged from remittances to home countries to the creation of new small and medium-scale engineering companies [2, 3]. The public discourse of technology as being an engine of macroeconomic growth grew in strength rapidly as the early impacts of globalization in the tech industry manifested themselves first in the West, and soon thereafter in many parts of the developing world. The international faces of this phenomenon

1 This material is based upon work supported in part by the National Science Foundation under Grant No. 0326582.

often were transnational expatriate technologists from the developing world. By the mid 1990s, there was significant buzz on the role of the information technology boom in dramatic global change [4, 5], and a first generation of ICTD literature discussing the specific nature of technology projects in development started a worldwide interest in the field [6-10]. Around this period, there was a dramatic rise in the number of ‘ICTD Projects’ – i.e., technology projects specifically aimed at creating developmental outcomes for their recipients. This trend was partly driven by a slew of research papers and policy documents within international agencies [11-13]. As a result, by the turn of the century, there were thousands of telecenters around the world, funded through various sources [14]. By the early 2000s, engineers became interested in ICTD, not just as a development agenda, but as an area that posed interesting research problems in their own fields of work [1519]. This followed the establishment of numerous academic venues, both as part of existing established forums, and independent venues specifically for the study and discussion of ICTD. Following this early foundational scholarly and practitioner work, more interesting follow-up work has been done on emergent areas of ICTD including digital inequality [20-23], on technology and sectoral development and macroeconomic change [24-26], on technology and urban change [10, 27], and on the ‘potential’ of technology [12, 28, 29]. By the mid 2000s, there was introspective work looking at the performance of the ICTD projects started in the 1990s in India and elsewhere, and also at the causes and outcomes of the interest in technology and development [30-33]. More recently, scholars have looked back at the growth of ICTD through the years by tracing the various stages that ICTD has moved through in this period [34]. Our survey and research here hope to build on such work. As stakeholders in an emerging field of study that is still in the process of defining itself and carving a niche, it is critical that we look back at various points in our progress and review the direction of our growth. Here, we present a fairly comprehensive survey of what has happened in the field, and use the opinion of significant voices from the field to support the literature review, in the hope of shedding some light on the general direction of ICTD work and its validity. We examine

358

the progress in dominant ICTD research areas, we identify the areas that received more attention in previous research, and we compare the expectations with the perceived achievements in each of these areas. We also examine questions concerning the rigor, impact and significance of ICTD, and its appropriateness in serving various development goals. II. APPROACH We begin our study by categorizing some broad domains of ICTD research and practice, and by performing a high-level overview of the progress in each of these domains. We follow this by presenting the results of our interviews with ICTD experts on their opinions about research and practical work in each of these areas. In addition to these discussions, we consider a few recurring themes in our early conversations with interviewed researchers. The first emergent theme was the question of ‘Hope v/s Hype’ in the context of ICTD. This issue, has been explored in the past [35-37], but continues to remains a niggling issue for most practitioners and researchers alike, who face up to it in their own work. A related theme that emerged in our interviews has been that the role of various stakeholders. As ICTD moves from being an experimental area towards mainstream development research and practice, it is inevitable that we face hard questions about what the role of government should be in funding ICTD projects, what kind of ICTD projects are more likely to succeed and so on. Posing these questions to our interviewees, we find an interesting variance of opinion on this issue. Finally we explore issues about the interdisciplinary nature of ICTD – how far does it exist, and has it helped or impeded ICTD. III. METHODOLOGY The research consists of two parts – first, an extensive literature review of the various projects in ICTD over the past decade, and second, a survey of 50 researchers and practitioners in the ICTD space. The two were done simultaneously – i.e. while we had some idea of what the broad literature review would bring up, we had no idea of how similar it would be to the results from our interviews. A. Instrument Design The interview schedule was created over three iterations on the basis of feedback about the interview questions from researchers in our direct contact. The third version of the questionnaire was an online hybrid with a mix of open-ended and close-ended questions, of which the former were created based on categories arrived upon in the first two iterations of the instrument. The final interview was anonymized and took anywhere between 15-75 minutes to complete, depending on the interviewee.

B. Sampling and Recruitment We had a selective sample of respondents from a few major ICTD-related online forums, in addition to other professors and researchers that we contacted to fill out the survey. C. Sample description Our total sample included 50 respondents, who were asked to select their areas of affiliation. We observe a fairly even distribution of these areas, with a slight skew towards education and infrastructure. Overall, a higher number of researchers than practitioners were represented in this survey. TABLE I RESPONDENT DESCRIPTION BY AREAS OF AFFILIATION

Domain Area Healthcare Education Business Agriculture Comm. & network infrastructure User interface design Governance Other

Research

Practice

11 17 8 8 16 11 10 9

3 7 4 3 4 0 1 3

The respondents’ past academic disciplines (regardless of whether they are presently scholars or practitioners) were fairly evenly distributed between science/engineering, social sciences, and hybrid areas such as design and information studies. TABLE II RESPONDENT BY AREAS OF PRIMARY ACADEMIC SPECIALIZATION

Academic Discipline Engineering Information Studies Education Political Science Design City planning Sociology Humanities Media International business and relations ICTD Development Studies Anthropology Agriculture Commerce Physics Environmental Science Total

Number of Respondents 16 7 3 4 2 2 2 2 2 2 2 1 1 1 1 1 1 50

359

D. Caveats The findings are from a fairly small sample because we wanted experts with several years in the field to comment on ICTD issues rather than a broad-based survey of a large number of participants. Over half of our sample is represented by people who have been active in ICTD for over 5 years and are thus familiar with the landscape and the changes in ICTD and are qualified to comment on past and future issues. That said, such a small and selectively sampled survey is prone to bias. This study is a work in progress, and while the results presented here are rigorous and sturdy enough for discussion, there is always scope for increasing the sample size, and finding out through an iterative process what areas might need better coverage in the questions. Finally, one other important caveat here is the regional concentration of respondents. We find a very heavy skew towards respondents currently based in the US and India, but this is explained by the fact that US is somewhat of a hub for researchers and practitioners, and that a large fraction of the ICTD activity is located in India. This also seems to indicate that academics from developing countries are going abroad to study ICTD in American or European Universities. TABLE III LOCATION OF RESPONDENT AND REGIONAL FOCUS

Country

USA India Malaysia Philippines Barbados Brazil Ghana Botswana Chile Colombia Canada Greece Netherlands Nigeria South Africa Spain Sweden Switzerland Uganda Macedonia Nigeria No Specific Region Total

Physical location of Respondent 25 10 2 2 1 1 0 0 0 0 1 1 1 1 1 1 1 1 1 0 0 -50

Location where respondent primarily active in ICTD work 4 20 1 2 1 1 2 1 1 1 0 0 0 0 1 0 0 0 0 1 1 12 50

IV. FINDINGS A. Healthcare The interest in using technology to widen access to healthcare and sanitation pre-dates the ICTD age. While earlier initiatives focused on increasing access to specialists and getting basic diagnosis to remote regions within the developed world [38], recently, the UN Millennium development goals (MDGs) have renewed the focus on ICTD [39, 40]. There are four broad areas where active ICTD projects tackle healthcare challenges. The first has been telemedicine [41], and here the focus has been on using longdistance communication links to expand access to remote rural areas where there are no doctors. For-profit initiatives have been deployed for specific kinds of diagnoses that can be reasonably well managed remotely, such as ophthalmology and dermatology [42]. 2 The second area of ICTD interest in healthcare is information gathering especially for epidemiological research. There have been a number of projects focusing either on general population health surveying, patient health monitoring, or healthcare aid impact assessment [43, 44]. A third focus of ICTD was driven by expatriate doctor communities from developing countries who were interested in contributing back to their home countries through social initiatives. Some ICTD projects have used web 2.0 technologies to connect doctors in the developing world with counterparts and experts in the developed world for remote consulting on specialty care [45]. The fourth area of ICTD in healthcare has concentrated on the use of technology in building low-cost medical diagnostic devices (such as ultrasound, X-ray machines) and sensors. 3 We asked respondents in the study to name the areas of ICTD and healthcare that they felt were of significance looking into the future. The top 5 responses, from a subset of 36 who commented on healthcare, are presented in the following table: TABLE IV AREAS OF FUTURE IMPORTANCE IN HEALTHCARE

Top 5 areas ranked by respondents Medical records Supply-chain management Tele-diagnosis and treatment Collection of epidemiological data User interfaces

Percent Respondents

57.8 50.0 44.7 44.7 28.9

From the interview results, it stands out that supply chain management in healthcare and the design of user interfaces for 2

http://www.clickdiagnostics.com Several of the Microsoft Research Digital Inclusion grantees in 2005 were doing some work in healthcare, at least two were specifically designing low-cost diagnosis devices for developing countries. http://research.microsoft.com/ur/us/fundingopps/RFPs/DigitalInclusion_2005 _RFP_Awards.aspx 3

360

healthcare work were both seen as very important areas for future effort, though both of these are practically absent in our literature review of past work. Medical records, collection of epidemiological data, and tele-diagnosis, all areas with important past work were also viewed as very relevant by respondents. In addition to these, a number of respondents mentioned health education and emergency assistance for locating medical facilities are potentially strong areas for the future. There was surprisingly very little mention of low-cost diagnostic devices. B. Agriculture From the earliest ICTD implementations in developing countries, the apparent incongruity of computers in the rural hinterland has been a key concern for researchers. We see therefore that many early projects tried to increase the relevance of computing in rural areas by providing information on agricultural practices, market prices and government schemes through telecenters [46, 47]. However, persuading rural adults to be regular telecenter users has been a challenge. As a result, besides agriculture, many telecenters have also focused on other services such as online assistance to help small enterprises in villages, for instance, artisans that sell handicrafts on the Internet [48]. There has also been work on turning telecenters into points of purchase in supply chains for rural produce [49]. Other kinds of ICTD projects in the rural agricultural market include the use of cellphones and PDAs in organic certifications for farmers [50], sensor networks in helping water management for rural areas [51], livestock management [52], price information [53] and farming extension and information sharing [54]. With the growing interest in microentrepreneurship, the use of technology to increase access to microfinance has been very prominent in ICTD research. Many projects have attempted to implement technological solutions to assist on-the-ground rural microfinance operations [55, 56]. Work in social sciences has examined whether the ability of technological interventions in increasing the efficiency of rural microfinance operations depends largely on the scale of operations of the organization adopting the technology intervention [57]. A survey of respondents on their opinion about the key areas in agriculture revealed the following: TABLE V AREAS OF FUTURE IMPORTANCE IN AGRICULTURE

Top 5 areas ranked by respondents Best practices and information sharing Market access and information Supply chain management Sensors Access to expert information

Percent Respondents

56.7 51.3 45.9 21.6 18.9

Unlike in the case of healthcare, the responses for

agriculture closely matched past work on agriculture within ICTD, and most respondents were in fairly good agreement about the perceived progress in these areas. What is worth noticing here is that ‘access to expert information’ which had been one of the most important areas of concern in the early days of ICTD takes a lower position, whereas market access and information, despite the apparent challenges in making market information usable for actual transactions, remains a top concern for ICTD experts. C. Education Governments, philanthropic efforts, and private corporations have all found the ICT in education space to be attractive right from the earliest days of ICTD implementations. Many of the early ‘low-cost computing’ projects such as the Hewlett Packard’s 4-4-1 computer, 4 and the Computador Popular were aimed at classroom use [58]. Computer aided learning projects have been among the largest and most prominent project categories within the ICTD space in the last decade, and today, it is arguably true that more poor households have “access to technology” through a child in school than through a kiosk or any other means of computer access [59]. Impact of computers on learning is an area of ICTD that has seen interest from mainstream economics as well [60]. Interface designers have been actively looking at innovative learning methods with computers and a lot of interesting work has been done in shared computing [61], cellphones for game-based learning [62], and networked systems for contextual classroom video instruction [63]. In higher education, distance learning in developing regions has been an early area of interest within ICTD [64, 65]. While a significant chunk of the work with children has been Indiacentric, we find a lot of work in higher education elsewhere, including the use of SMS for question/answers in university lectures [66] in Bangladesh, cellphones for interactive learning [67] in the Philippines, and a one-mouse-per-desk approach for lectures [68] in China. TABLE VI AREAS OF FUTURE IMPORTANCE IN EDUCATION

Top 5 areas ranked by respondents

Percent Respondents

Remote learning Educational games Low cost computing Life-long learning Online content

31.4 25.7 22.8 17.1 14.2

There was far lesser agreement in the education space on the key areas of research among ICTD experts. Two interesting factors emerged in our discussions. First, that lowcost computing, which has attracted a lot of interest in recent years, is seen comparatively as a less important area of research than working with existing devices. This mirrors well 4

http://www.hp.com/e-inclusion/en/project/441_brochure.pdf

361

the general frame of discussion within existing ICTD literature, which has also largely taken the same position. The second important issue is that remote learning, in which we have seen some promising projects start in the last few years, is expected to continue to be an area of focus in the coming years. D. Communications and Infrastructure Among the earliest areas to get engineering scholars interested in ICTD was the challenge of building and deploying novel high quality communication network solutions to connect low income regions with poor legacy infrastructure. Studies suggested that communications backbones could be the first form of infrastructure that would be affordable to deploy in the developing world [15] and that this could in turn be a major driver for wider socio-economic growth [28]. Several projects looked at various inexpensive connectivity solutions both over the short and the long-haul [69], and the use of low-cost WiFi for long-distance connectivity became an important ICTD pursuit, with groups from several parts of the world looking at the scope of this technology in real world deployments [16, 70-73]. Today, long distance connectivity using WiFi has made significant progress showing test throughputs of upto 6 Mbps at almost 400 km distances 5 and enabling functional field deployments for telemedicine applications [42, 74]. Research has also focused on creating appropriate front-end infrastructure to work on low power and unreliable, or intermittently connected networks [75, 76]. Finally, the ‘low-cost computer’ has been one of the most important areas of work in technology infrastructure in ICTD, and probably the area that has generated the greatest amount of industry interest. We found over 50-projects in the past 10 years in this space including many such as Fulong Mini-PC 6, and E-DUC 7, Sirius 8, and SofComp 9 that originated from research labs in the developing world alongside work by big corporations such as Intel’s Classmate 10, and AMD with its PIC 11. TABLE VII

AREAS OF FUTURE IMPORTANCE IN COMMUNICATIONS

Top 5 areas ranked by respondents Wireless/Low cost infrastructure Low cost phones and devices Mobile phones & phone coverage Community radio & TV VoIP

Percent Respondents

41.7 30.6 22.2 22.2 19.4

The results in this category could possibly be seen as underlining the distinction between the interests of researchers and those of the industry. Our sample, despite being open to researchers and industry, had a much higher research component. As a result, we see that low-cost infrastructure, undeniably a major area of concern for ICTD, trumps the low cost phones and devices, the area that shows more activity in our literature review. However, this could also be seen as indicative of what experts feel about the progress of low-cost devices in the real world market. Although we list a large number of low-cost device projects here, only a small fraction of them have actually made any significant impact in the market, and even some of the largest players abruptly left the market. It is also important that most of the research on lowcost devices is on computers, and little or no work exists on making low-cost mobile handsets, though a number of researchers indicate that this is likely to change looking ahead. One potential reason for why work on cellular infrastructure has remained largely unaddressed in ICTD came from the respondents. Several interviewees stated that since worldwide cellular coverage has grown so dramatically over the last few years, cellular coverage is seen as more of an industry problem rather than an academic research area. Furthermore, the closed protocols of dominant cellular technologies are not open to experimentation, and thus harder for researchers to work with. E. Governance E-Governance was a major buzz area in the early days of the ICTD, especially given a general subscription to the view that a lot of the problems of development are because of bad governance [77]. This was especially so in places like India, which had seen sluggish growth in several sectors that were tightly controlled by the state, but in sectors where the state kept itself off, growth came speedily and significantly [26]. Consequently, several projects emerged throughout the late 1990s to increase the use of technology in the processes of governance, first in the developed world and eventually in the developing world [78]. These included projects that migrated a number of state functions online as a way to reduce the manpower cost of governmental transactions [79], enable epayments [80] and e-voting [81], and presumably also reduce the scope for corruption. A number of projects also fundamentally changed certain state functions such as land record maintenance [82]. The status on ICTD in governance thus far has been fairly mixed. Although on one hand some studies have shown generally positive feedback [83, 84], others have raised serious questions about their impacts on the disempowered [85, 86, 87]. TABLE VIII

5

http://radar.oreilly.com/2007/06/wifi-record-range-now-382-km.html http://www.lemote.com/ 7 http://www.e-duc.com/ 8 http://www.fiveriverstech.com/sirius.htm 9 http://www.ncoretech.com/mobilis/index.html 10 http://www.classmatepc.com/ 11 http://50x15.amd.com/en-us/ 6

AREAS OF FUTURE IMPORTANCE IN GOVERNANCE

Top 5 areas ranked by respondents Sharing of public information Digitization of records (land, tax) Improved transparency, corruption reduction

Percent Respondents

47.3 31.5 26.3

362

E-payment and online retailing E-voting

23.6 7.8

The responses from our experts on governance again mirrored fairly closely what we found in the literature review – public information, digitized land and tax records were viewed to be the most relevant research areas. There were also a small number of respondents who felt that e-voting and GISrelated work, both of which have been looked at only to a limited extent, may be among the major areas for ICTD and governance in the near future. F. Design The User Interface (UI) design community has been among the major technical drivers of ICTD research. The basic assumption with UI work in ICTD has been that most technologies prevalent in markets today are designed with high income, educated users in mind, and that re-designing computers and peoples’ interactions with computers could make technology more accessible for the poor and illiterate [88]. Among the earliest ICTD projects – the Simputer [89] had an explicit interface design goal to make the computer a simpler device to use, therefore with few alphabetical keys, and more audio-based interactions. In most of the low-cost computing projects that followed thereafter [90], including most prominently the OLPC, appropriate UI design was a very important component [58]. Significant design work has also been done on building visually enhanced interfaces for illiterate users [91] and speech based systems for agriculture [92] and healthcare [93]. Other design work has looked across domains to redesign existing devices for new application areas – such as innovating with small screens on mobile devices for systematic data collection for healthcare [94, 95], microfinance services for the illiterate [55], and audio-visual based English-language education [96] delivered in local languages. TABLE IX

Voice recognition and synthesis Local language software Translation Accessibility Illiterate-friendly interfaces

G. ICTD Stakeholders One of the interesting questions around ICTD from its earliest days has been the role of various stakeholders in supporting such projects. There were a large number of ICTD projects in the 1990s, a period marked by the increasing privatization of development funding [97] and a massive boost in Corporate Social Responsibility (CSR) funding, which in turn diverted philanthropic funds from companies interested in human development towards ICTD projects [98, 99]. Because ICTD projects are often at the crossroads of business, technology and human development, many initiatives have been supported by large private corporations such as HP, Intel, or Microsoft, especially when there is an intersection of their business interests with ICTD [100]. What is more interesting is that ICTD projects have frequently been couched in a discourse of ‘sustainability’ [101] raising questions on whether these projects should be market driven, and we have seen some research in recent years suggesting that the state should reconsider spending on ICTD projects, specifically telecenters [59]. However, these questions are not easy to answer – the nature of development projects requires that a range of stakeholders work closely together, and ICTD is no different. In looking at how ICTD research can be divided up between various stakeholders, we asked respondents to comment on what they felt were important roles for industrybased research to play. TABLE X AREAS OF FUTURE IMPORTANCE FOR BUSINESS STAKEHOLDERS

AREAS OF FUTURE IMPORTANCE IN DESIGN

Top 5 areas ranked by respondents

limitations, voice recognition cannot be an effective replacement for other technologies – so while speech can work well for small vocabularies and limited UIs, these tasks can also be done equally effectively by inexpensive visual interfaces or numerical keypads. The applications where speech could really be helpful, for example in complex transactions, is also where the technical challenges are greatest.

Percent Respondents

57.1 54.2 20.0 17.1 14.2

In the design area, spreading technology access to populations without English (or other dominant) -language literacy was a major concern for researchers, thus language localization and voice recognition topped the list of future research directions. What is interesting here is that work in both these domain areas is extremely challenging and steps forward have been slow thus far. Due to current technology

Top 5 areas ranked by respondents Microfinance and microcredit Mobile commerce Supply chain management Online commerce Low cost sales devices

Percent Respondents

41.7 41.7 16.7 13.9 13.9

This is an interesting question which needs further examination – on whether markets ought to decide what research agenda is better served by the industry, and where academia should play a role. The surveyed experts seemed to feel that most areas typically involving financial transactions are best left to the industry. We asked respondents which stakeholder does ICTD play for – does it serve a greater development agenda or a greater business agenda – the

363

Potential Impact 100%

Negative No impact No answer Some Significant

80% 60% 40%

Eradication of hunger

Sustainable environment

0%

Gender equality

20%

Impact Thus Far Negative No impact No answer Some Significant

100% 80% 60% 40%

Eradication of hunger

0%

Sustainable environment

20%

Gender equality

To further investigate the gap between what has been expected from ICTD and what ICTD has already achieved, we interviewed respondents on what they felt was the potential of ICTD research to make an impact in achieving various MDGs and on how far work in ICTD over the past few years had

FIGURE II COMPARISON OF RESPONSES ON POTENTIAL AND ACTUAL IMPACT OF ICTD

Poverty alleviation

V. COMMENTARY After analyzing ICTD research and practice through the lens of individual research sub-areas, take step back to think about the bigger issues in the field that appear as a theme throughout the area. The first of these themes is the well-known debate on “hope v/s hype”, on expectations v/s reality that surrounds our field. Our literature review as well as conversations with experts on the matter suggests that the main concerns raised by existing studies were over whether ICTD stakeholders – either implementers or funding partners – made incorrect assumptions about the nature of the expected development outcome of some technology projects. Many reasons are mentioned as being responsible for these effects, including the changing nature of development goals around sustainability, and changing stakeholders with a new breed of technologists entering the development fray [100]. The opinion of ICTD experts further suggests concerns regarding the “hype” aspect: when asked whether they believe ICTD was hyped, 52% of our respondents responded affirmatively, and roughly 30% felt there wasn’t any hype.

Poverty alleviation

Governance

UI design

Energy

Communications

Agriculture

Business

Education

Healthcare

0%

Better governance

20%

Better governance

Partner with non-profits or for-profits Own and implement projects directly Only play regulatory role

40%

Communication technologies

No role

60%

Communication technologies

80%

Healthcare and sanitation

No answer

100%

Healthcare and sanitation

FIGURE I ROLE OF GOVERNMENT IN ICTD

The first thing to note it that most experts felt that ICTD projects generally had potential, but that it had a significant way to go in achieving it fully. This is reflected by the fact that a large number of respondents chose not to answer questions on the perceived impacts of ICTD thus far, but also by the fact that few respondents believed that existing ICTD work has made a significant impact. The areas where most respondents felt there was potential for impact (either some impact or significant impact) were in education, healthcare, communications, and better governance. The area where the overwhelming majority believed the impact to be significant was in communication technologies. Other areas of development such as sustainable environment and eradication of hunger were comparatively seen as outside the scope of ICTD. One remarkable trend we find is that 20% of the respondents’ opinion was that ICTD had an overall negative impact on poverty alleviation, and a small percent felt there was a negative impact on governance as well. While in both categories the proportion of respondents who felt that the impact was positive was either equal or higher, this is nonetheless a real concern that all of us engaged in serious ICTD research need to ask ourselves – whether technologies pose the serious risk of increasing inequalities.

Education for all

We turn now to the respondents’ perceptions of where the government’s role in ICTD ought to be, and as we find, most of the answers are fairly intuitive – regulate communications, directly act in healthcare, education, governance, stay out of design and so on. An interesting finding was that very few of the experts felt that the government should be directly owning ICTD projects in agriculture, most respondents preferring a hybrid role instead. This is a fairly unexpected, because in the earlier days of ICTD, the government’s role in e-agriculture was seen as very important. This could perhaps tie in with the experiences with telecenters and e-agriculture run by state agencies in the past decade, which have often suffered mixed fortunes, usually weak.

achieved that potential. Figure I presents a summary of our results, with development goals ordered (from left to right) by the fraction of interviewees that believed ICTD can make an impact (either “significant impact” or “some impact”) towards that goal. For example, almost 100% of the survey takers believed that ICTD has the potential to make an impact in education for all, while only about 60% believed the same about eradication of hunger.

Education for all

respondents were evenly split – a third believing the development agenda was better served, another third believing that ICTD worked better for business.

364

To comment on the relative importance and research maturity of each of these areas, we also did a quick survey of the publishing record in the past two iterations of the ICTD conference TABLE XI ICTD CONFERENCE PAPERS ARRANGED BY THEMATIC AREAS

Domain Area Agriculture

No. of papers 7

Education

10

Communications (Technical)

12

Communications (Social Sciences)

17

Governance

5

Healthcare

2

Design

15

Stakeholders / Business of ICTD

15

General ICTD

5

Sub-topics Sensors (2) Best practices / information sharing (2) Supply chains (2) Market access / information (1) Literacy (2) Novel input devices (2) User-created content 2 Computer usage models (2) Teacher training (1) Educational games (1) Wireless technologies (3) Mobile phones (2) DTN (2) Disaster prevention and relief (1) Telecommunications (2) Satellites (1) Security (1) Telecenters (general) (8) Low-cost computers (4) Radio (2) Disaster prevention and relief (1) Telecenters (governance) (4) Digitized Records (1) Medical education (2) User interfaces (2) General User-centered design (3) Input devices (3) Illiteracy (2) Voice recognition (2) Accessibility (2) Translation (2) Graphic interfaces (1) ICTD Business models (4) Microfinance (3) Kiosks (2) Models for ICT deployment (2) Supply chains (1) Social entrepreneurs (1) Mobile phones (1) Mobile phone commerce (1) Methodology (3) Gender Empowerment (2)

REFERENCES 1. 2. 3. 4. 5. 6.

Qvortrup, L., The Nordic telecottages : Community teleservice centres for rural regions. Telecommunications Policy, 1989. 13(1): p. 59-68. Saxenian, A.L., Silicon Valley's New Immigrant Entrepreneurs. 1999: Public Policy Institute of CA. Vertovec, S., Transnational Networks and Skilled Labour Migration. COMPAS, Oxford, 2002. Castells, M., Rise of the Network Society. 1996. Negroponte, N., Being Digital. 1995, Cambridge, MA: MIT Press. Bhatnagar, S.C. and M. Odedra, Social implications of computers in developing countries. 1992: New Delhi: Tata McGraw-Hill.

A look at these raises back the issue of the multidisciplinary nature of ICTD. While the diverse nature of our respondent profile is in itself an indicator of the range of scholarly traditions involved in ICTD work, it is not clear how much actual inter-disciplinary work is being done, and if so, what is the exact nature of such collaborations. Within engineering, there is comparatively much greater interest in ICTD from computer scientists, and this may be an outcome of the nature of much ICTD research or that several of the key leading faculty members and senior researchers in ICTD are from a computer science background. We also find that most publishing in ICTD even if interdisciplinary is usually restricted to being either engineering oriented or in the social sciences. Perhaps the most notable exception overall is design, which as a field had already started incorporating ethnography and other methods from anthropology. In fact, 81% of respondents stated that ICTD itself as a field had helped bring multidisciplinary research to greater focus in academia as a whole. Although most experts tended to agree that the multidisciplinary nature of work was a good thing, one concern raised by multiple participants was that of creating scholarly tenure track positions in ICTD. Although information schools have been early in recognizing the importance of ICTD, the case for ICTD is yet to be made for much of social science and engineering disciplines. A final, concerning thought came in reference to the issue of rigor in ICTD research. When asked if the existing ICTD research had been methodologically rigorous, less than a fifth of our respondents said yes, almost half the respondents felt the work thus far hasn’t been rigorous enough. VI. CONCLUSION ICTD has made strides since the 1990s, both the expanding pool of people: researchers, industry and state partners, and high quality new work is evidence of this. What has also worked well for this growing field is that a lot of field projects and research are now closer to maturity, and that we see a lot less work on pilot projects and a lot more work reporting on the progress of initiatives and ideas. A fairly large body of ICTD researchers have now been in the field long enough to think critically and retrospectively on progress in this work. We hope that this document will serve as another step in that introspection.

7.

8.

9. 10. 11.

Howkins, J. and R.L. Valantin, Development and the Information Age: Four Global Scenarios for the Future of Information and Communication Technology. 1997: IDRC (International Development Research Centre). Roche, E.M. and M.J. Blaine, Information technology, development and policy. Information Technology, Development and Policy, EM Roche & MJ Blaine (eds.). Avebury: Aldershot, UK, 1996. Heeks, R., Technology and Developing Countries: Practical Applications, Theoretical Issues. 1995: Frank Cass & Co. Madon, S., Information-Based Global Economy and Socioeconomic Development: The Case of Bangalore. The Information Society, 1997. 13(3): p. 227-244. Hanna, N., Exploiting information technology for development: a case study of India. 1994: The World Bank.

365 12.

13. 14. 15. 16. 17.

18. 19. 20. 21. 22. 23. 24. 25. 26. 27.

28. 29. 30. 31. 32. 33.

34. 35. 36. 37.

Nulens, G. and L. Van Audenhove, An Information Society in Africa?: An Analysis of the Information Society Policy of the World Bank, ITU and ECA. International Communication Gazette, 1999. 61(6): p. 451. Wehn, U. and R. Mansell, Knowledge societies. 1998: Published for and on behalf of the United Nations by Oxford University Press New York. Oestmann, S. and A.C. Dymond, Telecenters - Experiences, Lessons and Trends, in Telecenters: Case Studies and Key Issues. 2001, Commonwealth of Learning: Vancouver. p. 1-19. Brewer, E., et al., The case for technology in developing regions. Computer, 2005. 38(6): p. 25-38. Subramanian, L., et al., Rethinking Wireless for the Developing World. Proc. of the 5th Workshop on Hot Topics in Networks (HotNets), November, 2006. Seth, A., et al., Low-cost communication for rural internet kiosks using mechanical backhaul. Proceedings of the 12th annual international conference on Mobile computing and networking, 2006: p. 334-345. Raman, B. and K. Chebrolu, Experiences in using WiFi for Rural Internet in India. IEEE Communications Magazine, 2007. 45(1): p. 104–110. Jhunjhunwala, A., Unleashing Telecom and Internet in India. India Telecom Conference at the Asia/Pacific Research Center, Stanford University, November, 2000. Hoffman, D.L. and T.P. Novak, Bridging the Racial Divide on the Internet. Science, 1998. 280(5362): p. 390-91. Compaine, B.M., The Digital Divide: Facing a Crisis Or Creating a Myth? 2001: MIT Press. Hendry, J.D., Social inclusion and the information poor. Library Review, 2000. 49(7): p. 331-336. Warschauer, M., Reconceptualizing the Digital Divide. First Monday, 2002. 7(7): p. 0-262. Singh, N., Information Technology as an Engine of Broad-Based Growth in India, in The Future of India and Indian Business, P. Banerjee and F. Richter, Editors. 2002, Macmillan: London. Heeks, R., India's software industry. 1996: Sage Publications Thousand Oaks, Calif. Arora, A., et al., The Indian software services industry. Research Policy, 2001. 30(8): p. 1267-1287. Saxenian, A.L. Bangalore: The Silicon Valley of Asia? in Conference on Indian Economic Prospects: Advancing Policy Reform,. 2000. Stanford, California: Center for Research on Economic Development and Policy Reform. Jhunjhunwala, A., Can Telecom and IT be for the Disadvantaged? Rural Development, 1998. 17(2): p. 321–37. James, J., Low-cost computing and related ways of overcoming the global digital divide. Journal of Information Science, 2001. 27(6): p. 385-392. Heeks, R., Information Systems and Developing Countries: Failure, Success, and Local Improvisations. The Information Society, 2002. 18(2): p. 101-112. Keniston, K., Grassroots ICT Projects in India: Some Preliminary Hypotheses. ASCI Journal of Management, 2002. 31(1). Madon, S., Evaluating the Developmental Impact of e-Governance Initiatives: An Exploratory Framework. EJISDC, 2004. 20(5): p. 113. Donner, J., The Use of Mobile Phones by Microentrepreneurs in Kigali, Rwanda: Changes to Social and Business Networks. Information Technologies and International Development, 2007. 3(2): p. 3-19. Heeks, R., ICT4D 2.0: The Next Phase of Applying ICT for International Development. COMPUTER, 2008: p. 26-33. Thomas, P., Bhoomi, Gyan Ganga, e-governance and the right to information: ICTs and development in India. Telematics and Informatics, 2008. Antin, J. Cultural Assessment for Sustainable Kiosk Projects. in Information and Communication Technologies and Development, 2006. ICTD '06. 2006. Keniston, K., IT for the Masses: Hope or Hype. Massachusetts Institute of Technology (MIT), Cambridge, MA, 2003.

38.

Perednia, D.A., Telemedicine technology and clinical applications. JAMA, the journal of the American Medical Association, 1995. 273(6): p. 483-488. 39. Chandrasekhar, C.P. and J. Ghosh, Information and communication technologies and health in low income countries: the potential and the constraints. Bulletin of the World Health Organization, 2001. 79: p. 850-855. 40. Sahay, S., Special Issue on “IT and Health Care in Developing Countries”. EJISDC, 2001. 5: p. 0-1. 41. Brauchli, K., et al., iPath-a Telemedicine Platform to Support Health Providers in Low Resource Settings. Stud Health Technol Inform, 2005. 114: p. 11-7. 42. Surana, S., et al., Deploying a Rural Wireless Telemedicine System: Experiences in Sustainability. COMPUTER, 2008: p. 4856. 43. Ibrahim, M., et al. Improving Healthcare Delivery in Developing Regions: Coupling Output-based Aid and Mobile Technologies. 2007. University of California Big Ideas 2007 Conference. 44. Chandani, Y. and G. Breton, Contraceptive Security, Information Flow and Local Adaptations: Family Planning in the Philippines and Morocco. EJISDC, 2001. 5(3): p. 1-18. 45. Luk, R., M. Ho, and P.M. Aoki, Asynchronous remote medical consultation for Ghana, in Proceeding of the twenty-sixth annual SIGCHI conference on Human factors in computing systems. 2008, ACM: Florence, Italy. 46. Ramamritham, K., et al., Innovative ICT Tools for Information Provision in Agricultural Extension (December 2005). Information and Communication Technologies and Development, 2006. ICTD'06. International Conference on, 2006: p. 34-38. 47. Kaushik, P.D. and N. Singh, Information Technology and BroadBased Development: Preliminary Lessons from North India. World Development, 2004. 32(4): p. 591-607. 48. Salvador, T., J.W. Sherry, and A.E. Urrutia, Less cyber, more cafe: Enhancing existing small businesses across the digital divide with ICTs. Information Technology for Development, 2005. 11(1): p. 77-95. 49. Kumar, R., eChoupals: A Study on the Financial Sustainability of Village Internet Centers in Rural Madhya Pradesh. Information Technologies and International Development, 2004. 2(1): p. 45-74. 50. Patel, N. and T. Parikh. Designing a Farmer-centric Organic Certification System. in ACM SIGCHI 2008, Workshop on HCI for Community and International Development. 2008. Florence, Italy. 51. Panchard, J., et al. COMMON-Sense Net: Improved Water Management for Resource-Poor Farmers via Sensor Networks in Information and Communication Technologies and Development, 2006. ICTD '06. 2006. 52. Heffernan, C., The Livestock Guru: Demand-led knowledge transfer for poverty alleviation. Information and Communication Technologies and Development, 2006. ICTD'06. International Conference on, 2006: p. 94-100. 53. Veeraraghavan, R., et al., Kiosk Usage Measurement using a Software Logging Tool. Information and Communication Technologies and Development, 2006. ICTD'06. International Conference on, 2006: p. 317-324. 54. Gandhi, R., et al. Digital Green: A Participatory Digital Framework to Deliver Targeted Agricultural Information to Small and Marginal Farmers. in Second International Conference on Information Technologies and Development, Dec 2007, IEEE Conference Proceedings. 2007. Bangalore. 55. Parikh, T., K. Ghosh, and A. Chavan, Design studies for a financial management system for micro-credit groups in rural india. Proceedings of the 2003 conference on Universal usability, 2003: p. 15-22. 56. Sathe, S.K. and U.B. Desai, Cell Phone Based Microcredit Risk Assessment using Fuzzy Clustering. Information and Communication Technologies and Development, 2006. ICTD'06. International Conference on, 2006: p. 233-242. 57. Ratan, A. and M. Gogineni. Cost Realism in Deploying Technology for Development. in Confronting the Challenge of Technology for Development: Experiences from the BRICS. 2008. University of Oxford. 58. Patra, R., Pal, J., Nedevschi, S., Plauche, M., Pawar, U. Usage models of classroom computing in developing regions. in Second

366

59.

60. 61. 62.

63. 64.

65. 66. 67. 68. 69. 70. 71. 72. 73. 74.

75. 76.

77.

78. 79. 80.

International Conference on Information Technologies and Development, Dec 2007, IEEE Conference Proceedings. 2007. Bangalore. Pal, J., Examining e-literacy Using Telecenters as Public Spending. Second International Conference on Information Technologies and Development, Dec 2007, IEEE Conference Proceedings, 2007. 2: p. 59-68. Banerjee, A.V., et al., Remedying Education: Evidence from Two Randomized Experiments in India*. The Quarterly Journal of Economics, 2007. 122(3): p. 1235-1264. Pawar, U.e.a. Multiple mice for retention tasks in disadvantaged schools. in Proceedings of the SIGCHI conference on Human factors in computing systems. 2007. San Jose. Kam, M., et al., Localized iterative design for language learning in underdeveloped regions: the PACE framework. Proceedings of the SIGCHI conference on Human factors in computing systems, 2007: p. 1097-1106. Wang, R., et al., The Digital StudyHall. Computer Science Department, Princeton University, Tech. Rep. TR-723-05, 2005. Feghali, T., M. Martin, and N. Sahyoun, An exploratory case study of a Lebanese e-learning partnership model. World Review of Science, Technology, and Sustainable Development, 2006. 3(2): p. 137-151. Colle, R.D., Building ICT4D capacity in and by African universities.' International Journal of Education and Development using ICT, 2005. 1(1). Islam, Y.M., et al., Mobile Telephone Technology as a Distance Learning Tool. 2005, ICEIS. Librero, F., Uses of the Cell Phone for Education in the Philippines and Mongolia. Distance Education, 2007. 28(2): p. 231-244. Moraveji, N., U. Pawar, and T. Kim, A Mouse on Each Desk: An Inexpensive Classroom Interaction Technique for Remote Teaching. Microsoft Research Working Papers, 2006. Howard, J., C. Simms, and E. Simanis, What Works: n-Logue’s Rural Connectivity Model. A Digital Dividend Study by the World Resources Institute, 2001. Mishra, S.M., et al., Economic Analysis of Networking Technologies for Rural Developing Regions. 1st Workshop on Internet and Network Economics, 2005. Ramanathan, N., et al., Designing Wireless Sensor Networks as a Shared Resource for Sustainable Development. Information and Communication Technologies and Development, 2006. Bhagwat, P., B. Raman, and D. Sanghi, Turning 802.11 inside-out. ACM SIGCOMM Computer Communication Review, 2004. 34(1): p. 33-38. Pentland, A.S., R. Fletcher, and A. Hasson, DakNet: Rethinking Connectivity in Developing Nations. 2004. Zahedi, E., M.A.M. Ali, and M.J. Gangeh, Design of a Web-based wireless mobile teleconsultation system witha remote control camera. Engineering in Medicine and Biology Society, 2000. Proceedings of the 22nd Annual International Conference of the IEEE, 2000. 2. Seth, A., et al., An Architecture for Tetherless Communication. DTN Workshop, 2005. Fall, K., A delay-tolerant network architecture for challenged internets. Proceedings of the 2003 conference on Applications, technologies, architectures, and protocols for computer communications, 2003: p. 27-34. Wei, S.J., Corruption in Economic Development: Beneficial Grease, Minor Annoyance, or Major Obstacle?, in World Bank Policy Research Working Paper No. 2048. 1999, World Bank: Washington DC. Backus, M., E-governance in Developing Countries. IICD Research Brief, 2001. 1. Hasan, S., Introducing E-Government in Bangladesh: Problems and Prospects. International Social Science Review, 2003. 78(3-4): p. 111-126. Kiran, G.R., E-governance services through Telecentres-Role of Human Intermediary and issues of Trust. Information and Communication Technologies and Development, 2006. ICTD'06, Conference Proceedings, 2006: p. 131-142.

81.

Riley, C.G., The changing role of the citizen in the e-governance and e-democracy e-democracy equation. Commonwealth Centre for e-Governance, 2003. 82. De, R. and C. Sen, The Complex Nature of e-Government Projects: A Case Study of Bhoomi, an Initiative in Karnataka, India. Proceedings of Electronic Government, Third International Conference, Zaragoza, Spain, Aug-Sep, 2004: p. 556–557. 83. Okot-Uma, R.W., Electronic Governance: Re-inventing Good Governance. Commonwealth Secretariat, London, 2000. 84. Narayan, G. and A.N. Nerurkar, Value-proposition of egovernance services: Bridging rural-urban digital divide in developing countries. International Journal of Education and Development using ICT, 2006. 2(3). 85. Benjamin, S., Bhuvaneshwari, R, Rajan, P. 'E-Titling' or an "AntiPolitics Machine" necessary to globalize Bangalore? in International Conference on Enhancing Land Registration and Cadastre for Economic Growth in India. 2006. New Delhi: Centre for Science, Development and Media Studies (CSDMS). 86. Vasudevan, R., Changed Governance or Computerized Governance? Computerized Property Transfer Processes in Tamil Nadu, India. Information Technologies and International Development, 2007. 4(1): p. 101-112. 87. Rodrigues-Filho, J. and N.P. Gomes, e-Voting in Brazil: Exacerbating Alienation and the Digital Divide. Proceedings of the 6th European Conference on e-Government. 88. Tongia, R. and E. Subrahmanian, Information and Communications Technology for Development (ICT4D)¿ A Design Challenge? Information and Communication Technologies and Development, 2006. ICTD'06. International Conference on, 2006: p. 243-255. 89. Chandru, V., et al. The Simputer: Radical Simplicity for Universal Access. in Development by Design: ThinkCycle Open Collaborative Conference, Bangalore 2002. 2001. 90. Harvey, F., Computers for the Third World. Scientific American, 2002. 91. Medhi, I., A. Sagar, and K. Toyama. Text-Free User Interfaces for Illiterate and Semi-Literate Users in Information and Communication Technologies and Development, 2006. ICTD '06. International Conference on. 2006. 92. Plauché, M. and M. Prabaker, Tamil market: a spoken dialog system for rural India. Conference on Human Factors in Computing Systems, 2006: p. 1619-1624. 93. Sherwani, J., et al. HealthLine: Speech-based Access to Health Information by Low-literate Users. in Second International Conference on Information Technologies and Development, Dec 2007, IEEE Conference Proceedings. 2007. 94. DeRenzi, B., et al., E-imci: improving pediatric health care in lowincome countries. 2008. 95. Ho, M. Smartphones for Output-Based Aid. in ACM SIGCHI 2008, Workshop on HCI for Community and International Development. 2008. 96. Kam, M., et al., Practical considerations for participatory design with rural school children in underdeveloped regions: early reflections from the field. Interaction Design And Children, 2006: p. 25-32. 97. Atkinson, A.B., New Sources of Development Finance. 2005: Oxford University Press. 98. Jenkins, R., Globalization, Corporate Social Responsibility and poverty. International Affairs, 2005. 81(3): p. 525-540. 99. Fox, T., Corporate Social Resposibility and Development: In quest of an agenda. Development, 2004. 47(4): p. 29-36. 100. Pal, J. Computers and the Promise of Development: Aspiration, Neoliberalism and ‘Technolity’ in India’s ICTD enterprise in Confronting the Challenge of Technology for Development: Experiences from the BRICS. 2008. University of Oxford. 101. Kuriyan, R., K. Toyama, and I. Ray, Integrating Social Development and Financial Sustainability: The Challenges of Rural Computer Kiosks in Kerala. Information and Communication Technologies and Development, 2006. ICTD'06. International Conference on, 2006: p. 121-130.

367

Information Communication Technology and Sustainable Communities in Africa: The Case of the Niger Delta Region of Nigeria. (Feb. 2009) Uduak A. Okon

Abstract— Sustainable development is largely seen as pertaining primarily to environmental issues and grassroots social development. This stereotype misses the reality that sustainable development and the information society are operationally interconnected. The two phases of the World Summit on the Information Society (WSIS) in Geneva (December 2003) and Tunis (November 2005) provide an excellent opportunity to integrate sustainable development principles and practices into the institutions and policy frameworks that are shaping the information society. While the World Summit on Sustainable Development (WSSD) and WSIS Phase I brought these issues to international attention, Southern voices and visions are still notably lacking from the debate As we move into the age of information it is critically important for us to consider the implications of ICT in Sustainable Development and vice-versa. This research undertaking looks at precisely this intersection with its primary focus on sustainable communities. Sustainable Communities development is an evolving discourse. The locally-owned or adapted knowledge of a community is essential for integrated sustainable development, and is becoming a key priority for development practitioners. The popularity of the concept of communities is growing, and so are ideas on how to support these communities with technologies and how these communities might use ICTs to support themselves. These issues are pertinent in assessing the real contribution of ICTs to sustainable development, and therefore merit a closer examination. The paper presents the findings from the study of 9 communities in the Niger Delta region of Nigeria. The aim of the study was to explore how ICTs may contribute to the social sustainability of communities in the Niger Delta Region of Nigeria. The study grounds the understanding of ICT usage among indigenous communities and consumers in the reality of their everyday lives, in order to promote actions for sustainability. Index Terms— Communicative Ecology, Information Communication Technology, Social Change, Sustainable Communities

Manuscript received February 20, 2009. This work is part of a doctoral research. It is funded by a research Grant provided by Akwa Ibom State Government, Nigeria. Uduak Okon is a Doctoral Student at Royal Holloway, University of London and is a member of the ICT4D Collective. (Phone: 01784248786; Mobile: 07776474379 ; Email: [email protected]; [email protected]).

I. INTRODUCTION

D

uring the last decades four themes have emerged as a

response to the collective concerns of world citizens: peace, freedom, development and environment. As a consequence this collective concern and the four themes were manifested in a political process towards sustainability labeled “sustainable development”. The concept of sustainability lacks a widely accepted definition. Sustainable development is a wide concept and has over the years been introduced through many different definitions. The term “sustainable development” was popularized by the World Commission on Environment and Development (WCED), in its 1987 report entitled, “Our Common future”. This report by the Brundtland commission (named after it’s chair Dr. Gro Harlem Brundtland) provided the most commonly cited definition of sustainable development. The Commission wrote: “Humanity has the ability to make development sustainable – to ensure that it meets the needs of the present without compromising the ability of future generation to meet their own needs”. This definition has broad appeal and little specificity, but some combination of development and environment is found in most attempts to describe it. Since the publication of the report, there has been a mass of literature generated in various fields which has resulted in more specific application of the concept, such as sustainable agriculture, sustainable livelihoods, and sustainable transport. Over the years, the interest and scope of the sustainable development debate has grown substantially and this has led to an increasing diversity of interpretations. Mostly, sustainable development is modeled on three pillars used to facilitate the comprehension of the term: the triangle of environmental (conservation), economic (growth), and social (equity) dimensions. According to Selman (1996:31), while such universal aspirations may be helpful, they are insufficient, at the local level, to identify local needs, choose meaningful targets, and most important, “harness the energies of local people and organizations and aid development of the local society and economy to change in ways which are conducive to sustainability” There are newer approaches that argue that the point of departure most be the community. Redclift and Sage (1995)

368

argue that if sustainable development is to mean something, then it must be capable of translating into local action. Liemgruber and Imhof (1998) go even further by suggesting that the true scale for sustainability is the local level, where people interact and communicate and where each individual is affected by everybody’s actions. Woodhouse (2000) offers a similar argument; he states that the concept of sustainability is best understood and evaluated on the basis of a sustainable community. He further states that genuine development enhances the sustainability of the community and this does not necessarily involve economic growth. As the debate on sustainability matures it is becoming increasing accepted that development cannot be sustainable unless it builds on cultural traditions (Nurse, 2004) Traditions, norms and customs are all cultural capital that needs to be preserved and passed to future generations. Sustainable development cannot be a product packaged by the international community and delivered to a local community; it works best when it draws on existing community resources and capacity building efforts (Cooper and Vargas, 2004). It relies heavily on community involvement and commitment, since sustainable development is a way of life. Ideas and projects that are developed locally often have the greatest staying power because the community develops a sense of ownership. Cooper and Vargas also state that sustainable development is both a bottom-up and topdown process. Global commitments made by the nations of the world are necessary but not sufficient to ensure sustainability. Hence, the slogan from earlier Earth Days, “Think Globally, act locally”. Localizing global aspirations is a key challenge for implementing sustainable development strategies and even more so for developing countries where most people live in rural and sometimes remote communities. There is a need to ask local questions and focus on meeting local needs. This paper makes a distinction between external physical dimensions of sustainability and the internal sociocultural dimension. The latter dimension is mainly concerned with individual actions, social structures and the social capital of communities. Contrary to what the proponents that prioritise environmental concerns believe, within developing countries like Nigeria, pressing problems of social exclusion, poverty and unemployment are reducing the attention paid to environmental problems. Because of these problems, societies are less willing to accept the structural changes associated with shifts towards a more environmentally sound patterns of consumption. Thus, for countries like Nigeria, a socio-cultural perspective is integral to all discussions about susutainable development. The aim of this research is to explore from a socio-cultural perspective how ICTs may contribute to and support the sustainable development of communities in the Niger Delta Region of Nigeria. This paper outlines the conceptual framework for the study of ICTs and Sustainable communities and presents a narrative of the findings of the empirical research conducted in Nigeria on the Communicative ecology of Niger Delta communities. This research contributes to the understanding of the interaction of ICT with the communities’

communication ecology with a focus on a (new) definition of sustainability. It adds to the ongoing dialogue among academics, development practitioners, business leaders and government officials, aimed at pinpointing the ways in which ICTs can help make the transition to sustainable development easier, quicker and economically viable. Its focus is on ICTs as means and tools that enable desired changes, since it is these changes, not ICTs that lead to collective action and sustainable development. The presented research is at the base of ICT for development in developing Countries. A thorough understanding of the interaction ICT with the communities’ communication ecology is of primary importance, before the deployment of any ICT solution. Thereby it will be of use for further research and/or development of ICT as part of a solution for indigenous communities. II. CONCEPTUAL FRAMEWORK A. Information Communication Technologies (ICTs) The definition of ICT for this research uses Hamelink’s definition of ICTs: “Information and Communication Technologies (ICTs) encompass all those technologies that enable the handling of information and facilitate different forms of communication among human actors, between human being and electronic systems and among electronic systems” (Hamelink, 1997:3) and Duncombe and Heeks (1999) simplified definition describing ICTs as an 'electronic means of capturing, processing, storing and disseminating information'. This includes the “old” ICTs of radio, television and telephone, and the “new” ICTs of computers and mobile technology and the Internet. Information and Communication Technologies are rapidly consolidating global communication networks and international trade with implications for people in developing countries. Despite this there is a worrying lack of empirical evidence or analysis of the actual experiences and effects of ICTs upon poor people’s economic and social development. The constraints of existing information systems on poor women and men and their intersection with ICTs are also little understood. Little attempt is made by those promoting ICTs for development to assess their impact on the cultural identity, the values, and the state of social equity of the less developed economies. Also lacking are analysis of the social and cultural factors which determine the effective application and use of ICTs by developing countries. Social exclusion in the developing world cannot simply be resolved by technology if consideration of the factors that can ensure respect for cultural values, justice, equity and equality in the distribution of wealth - including information - is absent. Rodgers et al, (1994) argues that by definition, the nature of these elements imposes certain limitations on how ICTs can be appropriated for development. Even if the introduction of ICTs is feasible over a relatively short period of time as some predict, a dialogue between those promoting the technologies and the potential beneficiaries must be the foundation of any

369

development action. In the current information age, the capacity of a society to effectively position itself as a consumer and producer of knowledge is crucial to its social development. ICTs are increasingly playing a crucial role in most communities’ capacities to produce, access, adapt and apply information and thus offer enormous opportunities for facilitating the transfer and acquisition knowledge. They present (atleast theoretically) a promising potential to lead developing countries into the ‘highways’ of development. (Friedman 2006; Castells 2000). Despite the potential role and contribution of ICTs to development, there is still a growing voice in the development field raising serious concerns about the socio-cultural dimension of their application and use (Panos 1998; Wade 2002; Gumucio 2001). Proponents of ICTs (World Bank 2002, UNDP 2001; Pohjola, 2002; Braga, 1998) take an optimistic view and highlight the positive effects of the Internet and other forms of ICTs to create new economic, social and political opportunities for developing countries and the poor, but it also highlighted the need to ask both with regards to individuals and society, especially in third world countries, about the nature and extent of the influence and changes, and about the factors which facilitate or impede these processes, and more importantly how these influences and change may promote actions for sustainability. It has been found more useful to approach the question in terms of different predispositions and experiences, and with different historical, ethnic, linguistic, social, economic or religious backgrounds in different situations, making different use of ICT media (Avgerou, 2001; Walsham, 1993). Taachi et al (2003) further argue that it is important to study the communication needs, communication patterns and network, and the increasingly important impact which developments in communication technology might have on these. O’Farrell (2001) and Heek’s (1999) share the belief that before one can advocate for the development of ICTs among the poor, it is important to understand the existing information systems of the poor, how they interact with more formal information and the best way to strengthen them before intervening with new information sources and means of access sources. Questions such as: who benefits and who loses from the introduction of these technologies; how can ICTs be made useful and meaningful to the developing countries' poor majority who are struggling to meet their basic needs; what are the social and cultural opportunities and risks they present; and how can developing countries meaningfully adopt these technologies while lessening their undesirable social and cultural consequences, are some of the questions that emerge when looking at the potential development impacts of ICTs. These issues are pertinent in assessing the real contribution of ICTs to sustainable Communities development, and therefore merit a closer examination.

According to the organizers of the World Summit on the Information Society (WSIS), “We are in the midst of a revolution, perhaps the greatest that humanity has ever experienced. To benefit the world community, the successful and continued growth of this new dynamic requires global discussion and harmonization in appropriate areas.” Unfortunately, until recently, few discussions have focused on harmonizing the visions of the emerging information society with the principles and priorities articulated by the United Nations’ Millennium Development Goals (MDGs) and the World Summit on Sustainable Development (WSSD). Meaningful discussions about national policy coherence between these processes have been limited by restricted thinking about the information society (IS) and sustainable development (SD). Each has emerged from a different community with a different vocabulary and process for determining national priorities. Information society specialists have primarily been drawn from the fields of telecommunications and economic development. While, sustainable development has been delegated primarily to environmental issues despite the best efforts of its practitioners to articulate a holistic vision of integrated economic, social and environmental decision-making. The lack of interaction between these two policy communities is currently serving to reinforce stereotypes of both fields. Sustainable development is seen as pertaining primarily to environmental issues and grassroots social development, while the information society is perceived as being more relevant to the economic development potential of urban elites. These stereotypes miss the reality that sustainable development and the information society are operationally interconnected. Both terms are increasingly used by civil society and academics to refer to a desired global future that is casting its shadow upon our current time and decisions. For sustainable development to be effective and efficient, it must harness the institutions and tools of the information society. And for the information society to sustain itself, it must pay careful attention to the stocks and flows of resources (material and human) and energy that underpin it. There is now widespread recognition that sustainable development is the responsible way forward and it is enhanced by the recognition that information and communication technologies (ICTs) are key drivers of socioeconomic change and so has to be brought into focus as part of the equation. For example, Radermacher (1998) points out that "Modern information and communication technologies drive the worldwide economic system and the process of globalisation. In this process we see enormous growth worldwide with opportunities for overcoming poverty and promoting human rights but also with major threats to sustainability and to social justice".

B. Information Society and Sustainable development: An emerging Convergence

Only a small amount of literature has attempted to discuss the information society and sustainable development together. Both Felleman (1997) and Pamlin’s (2002) works are good examples, with each offering important and interesting

370

insights into information policy and tools and their role in environmental sustainability. However, more often than not, discourse on these two policy fields has a tendency to provide a very narrow cross-section of issues and perspectives focused primarily on environmental sustainability. To adequately understand the potential for ICTs to enable the achievement of more sustainable development, we must look beyond the direct impacts of the tools themselves to indirect impacts of the broader information society that they enable. A strong area of convergence is local content—the locallyowned or adapted knowledge of a community— which is shown by this paper to be essential for integrated sustainable development. Technology is crucial to sustainable development; local information is fundamental to many sustainable development activities, such as monitoring and raising awareness; local content is pervasive in many sectors of sustainable development. Technology has long been a crucial element in measuring and achieving sustainable development. From satellite imagery, environmental information systems and, today, the use of mobile phones to share health information, there has been a key dependence on ICTs in the creation, dissemination and consumption of relevant, local information. But “even after the first United Nations (UN) World Summit on the Information Society (WSIS) in Geneva 2003, the relationship between issues of the global information systems and of sustainable development is not being discussed adequately. It seems that the interdisciplinary and international research in this field is just beginning (Hilty, Seifert and Treiber 2005). C. Sustainable Communities Sustainable development pursues long-term goals and in trying to explain their endurance, attention to the cultural dimension is essential. The cultural and historical dimensions combine to construct the common social, economic and ecological) pillars of interest of sustainable development. Sustainability must become a primary goal if we want to ensure the long-term health of communities. What is lacking in the debate on sustainability is a strong focus on people. This research does not deal with the all encompassing question of sustainable development in general; rather it is focusing on the internal, social perspective of sustainability, arguing that it is a deeper explanation of sustainability issues. It argues that to explore the meaning of social sustainability, as well as its practical implications, is crucial to the understanding of sustainability issues. The central proposition of this research is that the issue of sustainability should be understood as a social problem, a problem created by, and eventually having its final impact on people themselves. Commoner (1993:23) argues that: “When an environmental issue is pursued to its origins, it reveals an inescapable truth—that the root cause of the crisis is not to be found in how men interact with nature, but in how they interact with each other…”. The human scale (or social dimension) is important for the re-conceptualization of sustainability since the meanings of both environments and

development are both value-laden, involving people in the decision-making and management processes of our society. In other words the inter dependency between development and environment cannot be separated from people’s actions. Manning (1990:291) stresses that “the concept underlying sustainable development is … a human perspective”. Beck (1992:81) argues that environmental problems “are not problems of our surroundings, but – in their origins and through their consequences – are thoroughly social problems, problems of people. Environmental issues and developmental issues are mediated by human beings. According to Giddens’ (1984) theory of structuration, the key to understanding the internal social dimension of sustainability is ‘duality of structure’. His theory argues that individuals and society are one whole: human action creates the structures of society; those structures provide the context for the socialization of humans, and in turn, the human action which will reflect and recreate these structures. This concept of ‘duality of structure’ which consists of individual actions and social structures is the lens through which this research explores ICT use and adoption. Attention is placed on the ability of cultures and relationships, enabled by ICTs, to transform socio-economic patterns. By addressing in more detail how people relate to one another, how shared practices emerge, and how communities evolve, we will be able to understand better when, how, and why such communities use or do not use technologies. Insight into the social capital of communities will also provide better understanding concerning ICT adoption and use by communities than the more traditional analysis of technological requirements, knowledge requirements, and structural aspects of the community. Social capital refers to networked ties of goodwill, mutual support, shared language, shared norms, social trust, and a sense of mutual obligation that people can derive value from. Social capital then is about value gained from being a member of a network. In general, social capital is often seen as the glue that brings and holds communities together (Cohen & Prusak, 2001). Huysman & Wulf (2004a) go even further by saying ‘social capital is considered as a necessary “ingredient” that binds communities over time’. There is no universal development model which leads to sustainable communities; instead development is an essential, multidimensional, dialectic process that can differ from community to community, context to context. Each society and community must attempt to define its own strategy for sustainable development. This research seeks to help Niger Delta citizens determine what constitutes sustainability in their own communities, and how they can act on their new knowledge and how ICTs may facilitate these. III.

CASE STUDY – THE NIGER DELTA REGION

The Niger Delta is the delta of the Niger River in Nigeria and as defined officially by the Nigerian Government, extends over about 70,000 km² and makes up 7.5% of Nigeria’s land

371

mass. Some 20 million people of more than 40 ethnic groups, speaking some 250 dialects live in the Delta. Their livelihoods are primarily based on fishing and farming. Nigeria has become Africa's biggest producer of petroleum, including many oil wells in the Oil Rivers. Some 3 million barrels a day are extracted in the Niger Delta. The region has accounted for more than 80% of Nigeria's export earnings. Much of the natural gas extracted in oil wells in the Delta is immediately burned, or flared, into the air at a rate of approximately 70 million m³ per day. This is equivalent to 40% of African natural gas consumption, and forms the single largest source of greenhouse gas emissions on the planet. The environmental devastation associated with the industry and the lack of distribution of oil wealth have been the source and/or key aggravating factors of numerous environmental movements and inter-ethnic conflicts in the region, including recent guerrilla activity by the Movement for the Emancipation of the Niger Delta (MEND).

benefits from the considerable government revenues from oil and gas sales, as the bulk of revenues have been looted or mismanaged by public office holders at all levels of government. The subsiding levels of a culture of public participation in governance and the prevalence of corruption among public office holders have fuelled fierce competition for political offices with contenders deploying violent mechanisms to deter opponents. The use of state apparatus and armed gangs by politicians has been on the increase across the country, and with it the harassment and abuses of the rights of citizens. The majority of citizens have thus been alienated from the political processes, while those that speak out are sometimes targeted for intimidation and, sometimes, even outright elimination. The increase in militia activities in the Niger Delta is a reflection of disenchantment and local desperation in the struggle for survival, as well as a quest to seek attention of the world to their plight in the hands of trans-national oil and gas corporations and the Nigerian government. Criminal gangs are also increasingly exploiting the breakdown of communal order to unleash mayhem on citizens. However, the increasing global appreciation of the problem is an opportunity for concerted action. That is why it is critically important to the region to get on the global information highway and for ICT media to reflect local content to enable to people advocate for their rights as citizens. Community mobilization and collective action is essential not just to counter looting of revenues allocated to the different tiers of government, but also to demand for and defend their democracy. IV.

Fig. 1: Map of Nigeria Map of Nigeria numerically showing states of the Niger Delta region depicted in red: 1. Abia, 2. Akwa Ibom, 3. Bayelsa, 4. Cross River, 5. Delta, 6. Edo, 7. Imo, 8. Ondo, 9. Rivers

A. Community Related Challenges The Niger Delta remains largely underdeveloped although there are areas where expatriates working in oil companies like ExxonMobil, Shell, Chevron etc. live, which look like a communities in well developed countries. However, amidst growing exports and increasing revenues to national, state and local governments, poverty levels in the country have been increasing while social infrastructures collapse. Dependence on oil and gas revenues has been a major factor in the inability of the state to create an enabling environment for sustainable development with the result that citizens are disenchanted while violence has escalated in the country. The increasing violence in the Niger Delta complicates a social situation characterized by massive poverty and environmental degradation. Corruption has robbed communities of potential

METHODOLOGICAL APPROACH

The methodological approach is based on Ethnography and Participatory Research. The research methodology incorporated Ethnographic approach into the Participatory Research framework. Taachi et al. (2002) have used this kind of approach successfully for a research designed to develop a transferable methodology for the evaluation of community multimedia centers. It was developed to focus on actual practices of use and interactions with technologies in the wider context of people’s lives and social and cultural structures (Tacchi and Slater, 2003) Ethnography is used to guide the research process and make sense of the complete range of social relationships and processes. The research will involve individuals and groups researching their own sociocultural settings and experiences. They will reflect on their values, shared realities, collective meanings, needs and goals. A major factor considered in designing the methodology is the evolving paradigm in development research. Most notable is the ‘things’ versus ‘people’ debate (Korten 1995; Chambers, 2003). Previously research in ICT for development had been focused on ‘things’ i.e. infrastructure, connectivity, hardware and surveys of ICT use in developing countries etc. (Hoffman,1985; Flamm,1987; OECD 1988) More recently

372

there has been a shift in research focus from ‘things’ to ‘people’. In an evolving paradigm of development there is a new high ground, a paradigm of people as people (Kling, 2000; Heeks 2001; Madon, 1997). The research is people-centered and the methodological design reflected this emphasis. The approach is open, evolving, participatory, employing diverse methods, interactions seek to enable local people and it is a bottom-up approach. Central to these methods were participant observation, keeping of field notes, in-depth interviews, focus group discussions and group interviews. The PAR strategy adopted here aimed to produce knowledge and action directly through research and to empower people on a deeper level through the process of constructing and using their own knowledge. The study is an integrated three stage process of social investigation, education and action designed to support those with less power in their community settings. The field work for the research is in three key stages that reflects the overall Participative Research Framework (see Table 1). The findings for the first 2 phases of the research are presented in this paper. TABLE 1 RESEARCH FRAMEWORK

PHASES

Purpose

METHODS

Phase 1 Preliminary Study

Redefining ‘Community’ and ‘Sustainability’

Focus Group Discussions Interviews

Phase 2 Ethnographic Study

Understandin g the Communicative Ecology

Participant observation Field Notes In-depth interviews Diaries

Phase 3 Participatory Action Research

Engaging communities in critical dialogue

Focus Group Discussions Workshops

A. Sampling The Niger Delta consists of 9 states, and the study was conducted in 3 states, Rivers, Bayelsa and Akwa Ibom states. These 3 states were predetermined and selected based on the diversity of their cultures, ethnicity, differing levels of

development and population to effectively represent the Niger Delta as a whole. During phase 1 of the study, 9 communities were studied. Three in each state and four focus groups was conducted in each state. Each group had similar membership, and so two of these constituted of women participants (young women and older women) and 2 of men participants (young men and older men). In total there were therefore be 12 focus groups. Three focus groups were conducted in more urban communities and 9 conducted in rural communities. This was to enable a comparative analysis of rural versus urban constructions of meanings of ‘community’ and ‘sustainability’. Participants were drawn from various backgrounds ranging from students, professionals and civil servants to artisans and market women. Group size ranged between 6 -12 people. 104 semi structured interviews were also conducted in the nine communities. In phase 2 of the field work, the sampling size was smaller to allow for a more in-depth and rigorous exercise. The ethnographic approach meant longer periods of time was spent in each community. The aim of the study was to understand the communicative ecologies of the target communities by investigating the local information flows, social networks and structures and how these may affect people’s perception and use of ICTs. It was conducted in Akwa Ibom and Bayelsa states; four communities were studied, using participant observation, field notes, in-depth interviews and diaries. Phase 2 spanned four months. The researcher immersed herself in each community for a month. V.

FINDINGS

The findings presented in this section are the results from the first 2 phases of the empirical study. A. Redefining ‘Community’ and ‘Sustainability’ Developing a clear picture of what is meant by Community sustainability from a group of people who are generally not reflective is not an easy undertaking. The study set out to answer three key questions about Niger Delta Communities (1) What kind of community is it? (2)What is to be sustained? (3)And How will it be sustained? The definition of community has been contested within sociology from Tönnies onwards (Tönnies 1957). For developing countries with a strong cultural heritage ‘community’ means something different from the western construct of community. By understanding what ‘community means to local people at a local level, it should be possible to develop sustainable strategies that are meaningful and germane to people in those communities. Key elements that are integral to healthy communities like Identity, ownership, participation, cohesion were given special focus. The key findings from the preliminary study were that the parameters that define communities for the Niger Delta people are different from those of western communities. Geography and ethnicity played a strong role in defining who belonged to which community, although that was not always the case.

373

There were cases of people who resided in particular communities who were considered outsiders and were not involved in community activities because they belonged to a different ethic group or are not originally from there. In some cases people living outside a geographic region have a stronger identity with their communities of origin even though they don’t live there. Language is a crucial element in belonging to the community, if you had a different language you were not considered a community member in the rural communities, Culture, Social Norms and Values, Shared leadership and Lineage make community bonds much stronger. Their ‘emic’ definition of a sustainable community also differed from the western construct of sustainable communities. The major reasons for this are that community is viewed differently and the priorities of the people are based on their socio-economic situation and their cultural heritage. Also the struggle for survival supersedes all other development needs of the people. Their construct of a sustainable community is: • A place where everyone (who shares the same heritage, language, values and ethnicity) has an equal opportunity to participate and contribute. • A place where citizens shared strong community values, where their relationships, community leadership structure and Indigenous customs and culture are preserved for future generations. • A community where there is use of ICT with local content and language, and where people (especially the youths) have the skills and capacities to use ICTs effectively. • One where there is access to information and education. • A community where there are income generating activities and employment opportunities. • A community where there are functioning schools and health centers. • And people are able to provide for the daily sustenance of their families. So the sustainable development priorities of these communities differ significantly from the accepted models for sustainable communities. These communities also have social structures that required participation from representative groups like the youths, women and men associations. The traditional leadership structure is very vital in the defining community laws and future development agendas. There was a strong sense of ownership and identity among community members which are crucial elements in building healthy communities. With a better understanding of what ‘community’ and ‘sustainability’ meant the people, the next phase of the research was to study the communicative ecologies of these communities in order to be able to explore how ICTs maybe deployed to meet the sustainable development needs of these communities. B.

The Communicative Ecology

An analytical framework for the analysis of the communicative ecology with eleven key concepts grouped under three categories was developed: These categories are Community practice; Community Information and Communications systems and; ICT Use and Impact. The classification of the concepts under the three categories was based on: causal relationships, contextual factors or intervening conditions, and actions and consequence of the previous two. See Fig. 2 below.

Concepts

Categories

• Communal life • Social Networks and Connections • Structures affecting individual actions • Development priorities • Access to Information • Information sharing mechanisms • Constraints to information Access • Information needs • ICT Awareness • Local appropriation of ICT • Individual’s exploitation of ICT • ICT impact

Community Practice

Community Information and Communication

ICT Use and Impact

Fig.2. Analytical Framework VI.

COMMUNITY PRACTICE

A. Communal life Life in these communities is shaped by three major aspects: the conventional face-to-face interaction, villagers’ interest in participating in communal issues and the ways and channels for information exchange. A well-built social interweave where everybody knows one another was observed. This network of connections creates the rules of reciprocity, epitomized by membership in community organizations which is a platform where people exchange experiences, stories and every kind of information that is of villagers’ interest. Communal life is also dictated by the leadership and

374

traditional values of the community. Societal norms are set by the village head, although most have been handed down by a previous generation. There is a very strong respect for their community laws and they believe that obeying these laws is important to the development of their communities. The communal life in these communities supports active community involvement and mobilization for actions that may lead to the development of their community. The sense of responsibility and identity the people have is very strong. B. Social networks and Connections Communities can be understood as ongoing systems of interactions, a set of social identifications and interactions. Social networks build and support the development of community capacity and individual agency critical to social development. The conceptualizaion here, refers to peoples connections with other people and the community. The idea is to focus on how the structure of ties affects individuals and their relationships. The shape of a social network helps determine a network's usefulness to its individuals. In the Niger delta it was observed that the people have very close connections to family and have strong family ties. The implication of this is that their actions are greatly influenced by their interaction with family members. Outside of the family, the church, community groups and associations provide avenues for social interactions. Types of information exchanged with their social network are about community events, happenings in the community, community infrastructural developments, business, financial challenges and politics, personal and family issues. The young men have the widest network of connections within the community and they are key information carriers within the community. This is because it is acceptable and even expected for them to circulate and interact more. They also act as vigilante for the community. A sense of belonging and the concrete experience of social networks and the relationships of trust that are involved in these connections bring significant benefits. However, the sense of attachment and quality of social networks varies between different people. It could be argued that for sustainability to take place in a community we should be focusing on enhancing the quality of social networks etc. rather than the creation or strengthening of ‘community’. This is the line taken by writers such as (Stacey, 1969). Exploring this concept adds to the context creation of these communities.

C. Structures affecting individual actions This concept explores what is responsible for individual actions in the community, this adds to the causal conditions for people’s action which may prove useful in understanding how these may facilitate actions for sustainability. Respondents expressed the desire to uphold the family reputation within the community, so they were constantly

careful in their actions to protect this family reputation. Another factor that affected individual action was the desire by the people to ensure community support in times of need and to avoid punishment and enforcement of community laws. There are societal expectations in these communities that guide the choices people make; they are expected to enhance the good norms of the society and to serve future generations. Membership in community fraternities, associations also has an influence on how people conduct themselves in this community. For example there are community laws that penalize women for certain offences that do not apply to men. For example insubordination and adultery. There are also certain community development meetings that women cannot participate in, like meetings to resolve inter community conflicts. D. Development priorities The kind of development people want plays a role in what they would commit to in terms of sustainability. Development that does not meet the people’s needs may not get the required support. Preserving indigenous knowledge and values is very important to the sustainability of these communities. Majority of the respondents felt that their local self government and community leadership structure was a positive development that if preserved would benefit future generations because it had proven effective in running their communities. The observation and interviews revealed that they value their traditions and communal norms very highly. As one respondent put it “our traditions and values are what makes us unique, that is our identity” These traditions have been passed down to the younger generation by inculcating it in them when they are young. Community associations also reinforce these traditions. Social, educational and economic development was highlighted as the development most desired in their communities. That is the way people interact with each other for the advancement of their communities. Education was also highly desired. In a rural community like Mbiaya Uruan where poverty is high, many children could not get further than a primary education and because of this they grow up not having many opportunities and many of them migrate to the city centers where there are better opportunities. The youths prioritized human development, capacity building and technological advancement as most important for sustainability. Inconsistent power supply hinders their use of ICT and unless there was an attempt by the government to provide consistent power supply, they did not see how they could maximize ICT for their benefits.

VII. COMMUNITY INFORMATION AND COMMUNICATIONS SYSTEMS The concepts under this category are comprised of contextual, mediating factors that affect access to information in the communities. Within communities, people also live and

375

participate in what we could call an “information ecosystem.” This does not just refer to the information that a community presently has available, but all the human and technological elements through which people become informed. The “media,” of course, plays a critical role in the information ecosystem, but they are hardly all of it. Numerous actors, both individual and institutional, contribute to the information ecosystem by creating, storing, disseminating, interpreting, and consuming information. Understanding how and why a community accesses information is critical in exploring the role ICTs play in meeting the community’s developmental goals. A. Information sharing mechanisms Interactions and communication is a vital part of Niger Delta communities. Most communication in the community is done through mobile phones and face to face interactions. In the rural sites communities share and disseminate information using town criers or announcers and community meetings. One of the key ways of receiving information is through a town crier, who disseminates community information about road repairs, meetings and new laws to community members. The information passed by the town crier is by far the most important information of all to the people. Information is also received through heads of family. Information from neighboring communities is passed on by ‘Okada’ (commercial motorcycles) riders in the community. Within these communities there are no public transport systems. There are intercity buses that drop people off at the entrance to the communities, but ‘Okada’ riders are the main means of transportation within the communities. Because of this they have access to all sorts of information which they pick up from the motor parks in town and from the passengers they carry, so they have become key information carriers for their community. See Fig. 4 below.

Okada

Fig. 4 Information Flows Community leaders filter certain developmental information that comes in and transmit information they consider useful to community members. This goes to show the kind of influence that the leaders have on the community’s access to

information. Observations and the interviews revealed that over 80% of these leaders are illiterate and totally out of touch with what is happening on the global scene, yet the educated ones in the community allow these leaders to preside over them. People with mobile phones exchange more information than those without. Local and state information is received from Radio and Television. Radio and town criers are the most constant means of receiving information. B. Constraints to information Access Gender is one of the factors that hinder access to information, the widening technological gap between the sexes was observed and it is reinforcing traditional forms of power dynamics and hierarchies between men and women. A large proportion of women in the region don’t have the capacities to use ICTs. ICT solutions need to mainstream gender perspectives into any initiatives to ensure equity. The regions integration and participation in the global information highway is also constrained by factors such as the high cost of access, high Internet access costs, low bandwidth, poor ICT infrastructure, Inconsistent power supply and often unreliable communication facilities. Traditional Leadership can also be a constraint, the discretion of the leaders affects information access, as stated earlier they filter the information coming into the community and if it is deemed not suitable, it is withheld from the people. In oral cultures, the collective memory and importance placed on the elders to store information creates a strong system for information flow (Slim & Thompson 1993). There is a need for mediation between the traditional and emerging information systems when considering the socio-cultural and economic leap that will be required for societies, accustomed to receiving information orally from a known and trusted source, to new digital, text based information sources. Lack of education was another barrier observed, the illiterate ones in the community where not given certain information because it is believed it would be of no use to them. For example information from the Niger Delta Development Commission that could benefit the whole community is told to only the leaders and the educated ones. So there are people that are totally excluded from happenings within the community, the deployment of ICT could further isolate this socially excluded groups. C. Information needs A number of factors affect how different members of the community may understand or use information, such as gender, economic status, literacy, etc. Each community has it’s own way of communicating and finding out about what is going on in their area and outside. The information needs identified by students and workers were: Latest discovery in science and technology, Events around the world, Entertainment information i.e. music, creative arts, Local and state news, Health news and information; Social events; Sport news; Business information i.e stock news; Job opportunities and Information on community development. Traders and

376

The investigations revealed that a combination of old ICTs (Radio and Television) and new ones like mobile phones, computers and internet (very limited) were used in the Niger delta region of Nigeria. The impact and use of these ICTs was explored in this category to identify the factors that influence the adoption and use of ICTs and identify tangible economic and social benefits arising from having access to and using ICTs. A. ICT Awareness Computers are viewed as a symbol of modernity. A lot of the older people were quite intimidated by the prospect of using one. People under the age of forty said ICTs are important for their individual lives and their communities. Some young people saw it as a tool that helps learning; some others view it as a means of getting beneficial information. One respondent said “it can give everybody information about the events happening in different communities and how they overcame such problems so the community can learn from it” They also thought it can be useful in building capacities and personal growth and can also provide an opportunity for wealth creation. Some cited the fact that many people now earn an income from running calling centers, so could also be income generating. These calling centers are usually little tents that are constructed by the road side where people make local and national calls from mobile phones for as little as N25 (1p) per minute, and international calls for N75 (3p) per minute. These calling places are all over the place on almost all street corners. Profits are marginal and competition fierce, but people eke out constant income from offering these services. B. Individual’s exploitation of ICT There were no fixed land lines in any of the communities studied, so all calls are made with mobile phones. The graphs below show ICT usage in two of the communities studied. Mobile phones and Radio were the most affordable and most used. In more rural communities like Mbiaya Uruan radio is a major source of Information. There were many people interviewed who owned televisions and some times because of power outage, would get to watch it only once a week, and in cases where the outage is due to faults, they would go for months without watching TV. Radio has become a primary source for getting information on local, state, national and international news and events. Radio is considered cheaper to run because it is battery operated, no power supply needed, there is a huge supply of radios and low level batteries imported from china that is readily available in most corner shops. Phones are also affordable, portable and mobile, although not as cheap as radio but definitely highly

ICT Media Use in Anua 30 Television 25 Mobile Phones No. of People

VIII. ICT USE AND IMPACT

desired. Internet use is very limited as stated earlier, because cyber cafés are located outside the community and travel costs is included with airtime costs. Most people have never used the internet and can’t use computers but have the perception that it is very costly and don’t even try.

Radio

20

Mobile Phones

Radio

Internet

15 Non-internet computing Internet

10

Non-internet computing Television

5 0 1 Media Used

ICT Media Use in Mbiaya Uruan 25 Television 20 No. of P eople

market women are interested in product prices. The elders and leadership were more interested in information that has to do with community development and politics.

Radio

Radio Mobile Phones

15 Mobile Phones

Internet

10 5

Non-Internet Computing

Non-Internet Computing Television

Internet

0 1 Media Used

Fig. 3 ICT use C. ICT Impact This concept explores the interaction between Community Information Systems and ICT Tools. The study showed that ICT further reinforces the social structures of these communities. The greatest area where ICT is affecting community systems is in strengthening social networks, social interactions have been greatly enhanced by using mobile phones, it is getting more people involved in the happenings in the community. ICT use is also strengthening the sense of identity people have with their communities, there are indigenes that work and live outside of their communities and radion and mobile phones are keeping them connected to happenings in their communities, because as revealed from the findings of the preliminary study Niger Delta citizens feel a sense of identity with their communities of origin, and are committed to the development of their indigenous communities. In this way ICT is helping to strengthen these bonds. ICT use is also impacting the development priorities of community members, because it connects them to the rest of the world. This connection has positive benefit for the communities. They are becoming more aware of global issues

377

like HIV/AIDS, environmental degradation and human rights and have mobilized themselves to address some of these issues. ICTs have also had an impact on the security of citizens in these communities. The Nigerian Government is very ineffective in protecting their citizens from crime. So citizens have been forced to find ways to fill this gap by defending and protecting their own communities and ICT is playing a crucial role in that. Mobile phones are being used effectively by vigilantes for security. The following is a summary of areas where ICT is impacting communities in the Niger Delta • Community mobilization • Community watch • Circulates information faster • Connects people together, Fosters good relationships • Income generation The study shows that ICT is being integrated into the social fabric of the community, especially mobile phones. Computers and the Internet are not as yet having any real impact in the communities. There is still a long way to go before ICT can begin to meet the information needs of these communities, because they do not have input on content, this is designed by others who may not be sensitive to needs of the people. These initial positive development of ICT use shows that there is great potential for ICT, if appropriately deployed to meet community information needs. IX. IMPLICATIONS OF ICT IMPACT ON SUSTAINABILITY Incorporating sustainable development at the local level is one the biggest challenges facing the movement towards sustainability, how to take global principles and make them concrete locally. Communities which are culturally rooted, locally produced and technologically adapted are being rapidly eroded. Vast literature frame the sustainability discourse largely in terms of an environmental agenda preoccupied with ‘green’ issues. Literature has largely neglected the community development as sustainable and livable places which adapt their unique cultural identities and specific historical heritage to contemporary needs. Analyzing the cultural environment and information ecosystem is a method of understanding the complexity of a given community, its culture, its fragility, resilience and it’s reaction to change. The factors relevant to sustainability of communities identified from the research are: • Awareness about sustainable development; • Information availability and thus transparency; • Public participation in governance; • Empowerment of citizens, especially women; • Fostering of cultural diversity; • Building capacity • Social cohesion

• Preservation of Indigenous cultures • Job creation and skills acquisition The empirical data provided possible directions for exploring how ICTs may be employed to contribute to sustainable development for the Niger Delta people. ICT can play a role in facilitating sustainability by: 1. Strengthening community systems that enable cohesion and collective action 2. Providing appropriate information on sustainable development practices through: • Public awareness campaigns and social reorientation on development issues; • Education; • Provide access to information, 3. Electronically documenting indigenous knowledge and practices for now and future generations. X.

CONCLUSION

The study provided a better understanding of how the use of technology affects relationships within the community and how individuals use technology to develop their relations with others in community. The study also showed the importance of mobile phones and radio in changing the lives of the people. Mobile phones is by far the most common communication technology to effect tangible positive change in these communities (market and trading information, emergency and security communications, strengthening kinship relations and social interactions) and is the backbone of ICTs. At present there is a vast unmet demand for radio broadcasting with local content in these communities. A combination of mobile technology and rural broadcasting will enable information and communication services reach more people than any other medium. Overall, the study has provided a better understanding of the use of technology in community life and presented opportunities of how ICTs may be deployed to help these communities towards sustainability. REFERENCES [1] [2]

[3] [4] [5] [6] [7] [8]

Avgerou, C. (2001), The Significance of Context in Information Systems and Organisational Change. Information Systems Journal 11:43-63. Barton, C. & Bear, M. (1999) Information and Communication Technologies: Are they the key to viable business development services for micro and small enterprises? Report for USAID as part of the Microenterprises Best Practices Project. March 1999 by Development Alternatives Inc, MD, USA Beck, U. (1992) Risk Society: Towards a New Modernity. New Delhi: Sage. (Translated from the German Risikogesellschaft published in 1986 Bell, S. (2003). Measuring Sustainability: Learning by Doing. London: Earthscan Publications Ltd. Braga, C. P. (1998). Inclusion or Exclusion, Information for Development (InfoDev), The World Bank, http://www.unesco.org/courier/1998_12/uk/dossier/txt21.htm Castells, M. (Eds.) (2000) The information Age: Economy, Society and Culture, Oxford: Blackwell, 3 volumes Cohen, D., and Prusak, L. (2001). In good company: How social capital makes organizations work. Boston: Harvard Business School Press. Chambers, R. (2003) Whose Reality Counts? Putting the First Last. Intermediate Technology Publications, London.

378 [9] [10] [11]

[12] [13] [14] [15] [16] [17] [18]

[19] [20]

[21] [22] [23] [24] [25]

[26] [27] [28]

[29] [30] [31] [32] [33]

Commoner, B. (1993). Population, Development and the Environment: Trends and Key Issues in the Developed Countries. In Population, Environment, and Development. New York: United Nations. Cooper PJ. and Vargas CM, (2004) Implementing Sustainable Development: From Global Policy To Local Action. Rowman and Littlefield, Lanham, MD Duncombe R., and Heeks R. (1999) Information, ICTs and Small Enterprise: Findings from Botswana. Manchester: University of Manchester, Institute for Development Policy and Management, Working Paper 7. Felleman, J. (1997). Deep Information: The Role of Information Policy in Environmental Sustainability. Greenwich, CT: Ablex Publishing Corporation Flamm K. (1987) Targeting the Computer, New York, the Brookings Institute Friedman, T.L. (Eds.) (2006) The World is Flat: The Globalized World in the Twenty-First Century, UK: Penguin Giddens, A (1984) The Constitution of Society: An Outline of the Theory of Structuration. Polity, Cambridge. Gumucio DA. (2001), Making Waves: Stories of Participatory Communication for Social Change, Rockefeller Foundation. Hamelink C.J (1997) New Information and communication Technologies: Social development and Social change, Discussion paper No.6, Geneva: UNRISD. Heeks R (1999) Information and Communication Technologies, Poverty and Development. Development Informatics: Working paper Series. Paper No 5, June 1995 Institute of Development Policy and Management. Published on the www: http://www.man.ac.uk/idpm/idpm Heeks, R. (2001) What Did Giddens and Latour Ever Do For Us? Academic Writings on Information Systems and Development', Information Technology in Developing Countries , 0, Vol.11(1) Hilty, L. M., Seifert, E. K., Treiber R. (2005) Information Systems for Sustainable Development, Swiss Federal Labs for Materials Testing & Research, Switzerland; Wuppertal Institute, Germany; Hochschule Niederrhein, Germany. Hoffman, K. (1985) Microelectronics, International Competition and Development Strategies: The unavoidable issues- Editor’s introduction. World Development 13(3): 263-272 Huysman, M., and Wulf, V. eds. (2004). Social capital and information technology. Cambridge, MA: MIT Press. Kling, R. (2000) Learning about information technologies and social change: the contribution of social informatics, The Information Society , Vol. 16 No.3 Korten D.C (1995) When Corporations Rule the World, Kumarian Press. Liemgruber W. and Imhof G. (1998) Remote Alpine Valleys and the Problem of Sustainability. In Andersson, L & Blom, T (Eds) Sustainability and Development - On the Future of Small Society in a Dynamic Economy. Universit of karlstard research report 98:8 Madon, S. (1997) The Information-Based Global Economy and SocioEconomic Development: The Case of Bangalore, The Information Society, 13 (3) Manning E.W (1990 Conservation strategies: Providing the vision for sustainable development/ E.W. Manning Environment Canada, [Ottawa, Ont.] : Alternatives 16, 24–29 Nurse K, 2004, Culture as the Fourth Pillar of Sustainable Development. www.fao.org/sard/common/ecg/2700/en/Cultureas4thPillarSD.pdf. Assessed 15/01/09 O’ Farrell, C. (2001) Information Flows in Rural and Urban Communities: Access, Processes and People. IRDD, The University of Reading . Pamlin, D, ed. (2002). Sustainability at the Speed of Light. Solna, Sweden: WWF Sweden, [Online] Available: http://www.panda.org/downloads/general/ict_sustainability.pdf. Panos (1998), The Internet and Poverty. Panos Media Briefing, No. 28, Panos Institute, London. Pohjola, M. (June 2002). The New Economy: Facts, Impacts and Policies, in Information Economics and Policy. Radermacher F.J (1998) Intelligenz - Kognition - Bewußtsein: Systemtheoretische Überlegungen, technische Möglichkeiten, philosophische Fragen. In: Interdisziplinäre Beiträge zur Kommunikation und zum Mensch-Technik-Verhältnis (C. Stadelhofer, ed.), Band 6, S. 146-193, Kleine Verlag GmbH, Bielefeld (1998)

[34] Redclift M. and Sage, C (Eds) (1995) Strategies for Sustainable Development – Local Agendas for the South. John Wiley & Sons, Chichester [35] Rodgers G. et al. Social Exclusion: Rhetoric, Reality, Responses, International Institute for Labour Studies, Geneva, A contribution to the World Summit for Social Development, 1995. [36] Selman, P. 1996. Local Sustainability: Managing and Planning Ecologically Sound Places. New York: St.Martin’s Press. [37] Slater, D., Tacchi, J and Lewis, P. 2002. Ethnographic Monitoring and Evaluation of Community Multimedia Centres: A study of Kothmale Community Radio and Internet Project, Sri Lanka. London: DFID. [38] Slim, H. & Thompson, P. (1993) Listening for a Change: Oral History and Development. London. Panos Publications. [39] Stacey, M. 1969. The Myth of Community Studies, British Journal of Sociology, 29, 134-147 [40] Taachi, J. and Slater, D 2003. Modernity under construction: Comparative ethnographies of internet. Amsterdams Sociologisch Tijdschrift 30, No. 1-2 Special Issue Digital contact. pp 205-222 [41] Tacchi, J , Slater, D., and Lewis, P. 2002. Ethnographic Monitoring and Evaluation of Community Multimedia Centres: A study of Kothmale Community Radio and Internet Project, Sri Lanka. London: DFID. [42] Tacchi, J, Slater,D and Hearn, G. 2003 Ethnographic Action Research: a User’s Handbook (New Delhi:UNESCO, http://cirac.qut.edu.au/ictpr/downloads/handbook.pdf [43] Tönnies, F. 1957. [Gemeinschaft und Gellenschaft] Community and Society. East Lansing, MI, Michigan State University Press. [44] UNDP (2001), Making new technologies work for human development, Human Development Report 2001, UNDP, Oxford. [45] Walsham, G. (1993) The Emergence of Interpretivism in IS Research Information Systems Research 6(4): 376-394. [46] Wade, R. (2002) Bridging the Digital Divide: New Route to Development or New Form of Dependency? In Global Governance, Vol.8, No. 4 Oct. Dec.2002 [47] Woodhouse, P (2000) Environmental Degradation and Sustainability. In Allen, R &Thomas, A (Eds) Poverty and Development – Into the 21st Century. Oxford University Press, Oxford [48] World Commission on Environment and Development. (1987). Our Common Future. The Bruntland Commission, UNESCO. [49] WSIS (World Summit on the Information Society) Civil Society Plenary (2003) Civil Society Declaration to the World Summit on the Information Society .

379

Integrating Health Information Systems in Sierra Leone Johan Sæbø, Edem Kwame Kossi, Romain Tohouri Golly-Kobrissa, Ola Titlestad, Jørn Braa

This paper presents an ongoing project in Sierra Leone to integrate health information systems at district and national level through a novel approach. Employing solar-powered low-energy computers running Linux, a wide consortium of local and international actors have tried to counter the severe problems of electricity supply breakdowns and computer viruses. The paper discusses the experiences from this effort, as well as the integration process itself, and the corresponding capacity building strategies. The findings so far suggest that alternative technologies, namely solar power and open source software, can be fruitful to apply in such infrastructural settings as Sierra Leone presents. Furthermore, the technical solution to an intermediary step towards integration shows some promising results. Index Terms—Sierra Leone, health information systems, integration, solar power

I. INTRODUCTION

I

N this article we describe and discuss a project to develop an integrated Health Information System (HIS) in Sierra Leone, which has been going on since early 2007. The article focuses on the key challenges facing the project; 1) Building agreement on the need for an integrated approach to HIS in order to solve the prevailing fragmented situation of multiple vertical reporting systems which have no coordination and shared data standards – and building a consortium to carry out the task. 2) Handling the extremely poor infrastructure in Sierra Leone by establishing a “new” solar powered computer infrastructure. 3) Developing a free and open source database application which may handle the problem of interoperability with existing systems and include and manage data from the various health programs and data sources. 4) Capacity development; how to establish a training scheme and a support structure sufficient for rendering the new computerised system sustainable. Ravaged by a prolonged civil war, which was declared over on 18 January 2002, Sierra Leone consistently scores extremely low on human development indicators. The public .

health system, suffering from a huge loss of both personnel and infrastructure during the war, is slowly rebuilding the capacity to improve the service provision across the country. This effort is supported by the government and the international community through many agencies such as UNAIDS, WHO, UNICEF, GTZ etc. and aims at achieving health millennium development goals. The rapid growth of various health initiatives has created a situation of fragmented information systems, common also in other developing countries [1, 2]. An initiative to counter this fragmentation by integrating and strengthening the HIS together with using novel ICT solutions to cope with the extreme infrastructural challenges has been initiated and piloted over the last year. This initiative is supported by the Health Metrics Network (see http://www.who.int/healthmetrics/) together with other agencies such as the World Bank and UNAIDS. In order to address the fragmentation, an integrated district based data warehouse approach has been followed. As a first step, data reported from the health facilities to the district using various health program specific reporting formats and systems are all being captured in one database application, the District Health Information Software (DHIS version 2, from now referred to as DHIS). Unified processing and analysis of these previously disparate information flows have then made it possible to better assess the quality of data and discrepancies between the different data reporting formats. As a result overlaps, gaps and inconsistent definitions of data variables between the different reporting formats have been identified. As a result of this, a revision of the data collection tools within an integrated framework is being planned. Furthermore, the district based database application supports unified data management, data disaggregating (e.g. make it possible to “follow” the data from the national level down to the health facility where it is collected), data analysis, programmatic reporting, changes in health units organisation, and local system integration. A number of paper based reporting formats and routines are currently in place in Sierra Leone Some of these are also computerized, such as HIV/AIDS reporting, which consists of paper forms reported from the facilities, and a database application called Country Response Information System (CRIS, from UNAIDS) located in the districts, where the HIV/AIDS reporting forms are captured. One of the goals is to

380

establish interoperability and data exchange between DHIS and CRIS and other software. In order to achieve this integration, many other critical issues such as infrastructure that the system will rely on and human capacity need to be addressed. The national power supply system is extremely poor, as many other national infrastructures. The main power source of the existing computerized HIS are generators. As the government is not able to provide fuel regularly to run generators, this reliance on diesel contributes to undermining the system. Human capacity both in terms of data analysis and information use and IT skills is another key issue that has to be tackled. Many information/IT officers (called M&E officers; Monitoring and evaluation) at district level have received on-the-job IT training, but because of power shortage, computer troubles (viruses, damaged and old computers) and absence of follow-up, the training endeavors were not capitalized on and they are still unskilled or uncomfortable in IT. The rest of the article will present the case of Sierra Leone by focusing on the challenges described above. Some background to HIS and integration, as well as the methods applied, will be described first. II. BACKGROUND: HIS AND INTEGRATION Fragmentation and poor quality and use of data are major problems with health information systems in African countries. Integration of information systems is often perceived as a technical task involving primarily incompatible software and infrastructures. While this is an important part of the picture, fragmentation and poor coordination between organizations, and, as in the case of health, between providers of different services, together with political and social aspects more generally, are as important [3]. This fragmentation leads to gaps and overlaps, and incompatible definitions in the data that is being collected, registered and reported. Experiences from other African countries, as well as from Sierra Leone, show that without shared data standards, data exchange cannot take place. Integration is therefore first of all about data standardization. However, while the practical system focus of integration is on standardization, the overall integration will need to involve and enroll as many as possible of the various actors in the health system in the process. Over the last decade the Health Information System Programme (HISP, www.hisp.info) has addressed these problems. The DHIS software was successfully used for integrating data and health services in first South Africa, and later in many other countries [4]. Integration of data and interoperability between information systems as well as increased use of information are key issues in the HISP approach. The health information infrastructure may be regarded as heterogeneous networks of actors (Latour 1987), which may

be more or less integrated, or aligned. Communication and interaction between the components of the network may be regarded as going through gateways. Integration can then be perceived as software exchanging data through gateways, but this can also mean integration between paper based and computer based components. An integrated HIS allows data to be collected and analysed at one point and to be shared among the modules or parts of the system. This makes the data analysis more meaningful as all data is analysed together. It also entails knowledge integration as well as solving challenges of political and economical nature [5], [3]. This process has shown to be non-trivial [6], especially in developing countries [3]. The fragmented HIS leads to poor quality health information which can be rendered useless. Integrated HIS is therefore relevant, but it does not solve all the problems of HIS in developing countries. III. METHODOLOGY The research enveloping the project in Sierra Leone has been carried out along the lines of Action Research (AR). AR is a form of participative research where the researcher takes part in the change processes in an organization, actively trying to improve some stated problem [7], [8]. While it was initially not used much in IS research, and had a lot in common with anthropology and social studies, it has over the last two decades increasingly become accepted as a way to generate very relevant research findings on technology in its human context [9]. While we acknowledge the critique often raised against using AR in the field of IS (or any field, for that matter), we also claim that we, through our active participation, gained knowledge we would not easily come by taking a more passive role. Baskerville and Wood-Harper provides an overview over common critiques of AR. However, they conclude that the traps found are not unique for AR as a specific method, but are likely to trouble a researcher using any method in social science [9]: “A number of problems confront the action researcher such as lack of impartiality, lack of discipline, confusion with consulting and its context-bound nature. However, these problems confront researchers using alternative methods as well. The difficulty with action research may be a matter of degree, and the easy loss of scientific rigour” (ibid, 144) The authors have been an active part in the consortium described in this paper. Through this role, we have been able to engage in a close collaboration with users at different levels. Our findings originate in analysis of our work, including software development and customization, database development, implementation of pilot sites in Sierra Leone, training of all monitoring and evaluation officers in the country over 6 weeks, on-site training in the pilot districts, and discussions with district personnel to clarify data flows,

381

reporting, work practices, and the like. The scientific rigour, to quote Baskerville and Wood-Harper, has been upheld by applying qualitative research methods in the daily work, such

as interviews and observationary techniques, and carefully recording the findings.

IV. PREVIOUS SITUATION Table 1 summarizes the strengths and weaknesses of the previous HIS, and Figure 1 depicts the overall fragmented HIS in Sierra Leone. At facility level, data is captured on multiple paper forms and registers. This is sent to the districts, where the data is entered into various software applications (CRIS, Excel spreadsheets, and other single-

program (only catering one health program) software). This fragmentation is replicated at the national level, with various standards for data storage. The result is poor accessibility of data, overlap between the many databases, and hence less use of data for processing and analysis.

Table 1 Strengths and weaknesses of previous HIS Strengths Weakness • Poor IT skills of data management teams at all levels • Clear and understood reporting procedure • Lack of power supply • Lack of resource to fuel regularly the generators • CRIS used in all districts • Lack of IT support and recurrent viruses problems • Computers in all districts • multiple and not well coordinated tools and formats for data reporting from facility to district, with overlaps and gaps, are in use; • Information awareness and • Multiple computer based tools in use, also with overlaps (same data captured computer usage very good in in different systems), and they are not linked one of the districts visited; • Data analysis and local reporting carried out in districts by re-capturing data in multiple spreadsheets designed excel spreadsheets and used for particular program • Data aggregated to district totals before reporting to national; making quality activities, up-to-date graphs on check of data difficult program activities displayed on • No feedback on reports from national to district; the wall. • Reporting from hospitals very poor (e.g. incomplete, poor coverage); morbidity and admissions/discharges reported, but not time spent by patients in hospital (laying days/bed-days), resources utilisation therefore not possible to deduct • Anti RetroViral therapy (ARV) patients managed by not-optimal paper record system; very difficult to provide outcome and cohort indicator reports based on current system

382

National: Fragmented reporting; gaps & overlaps Data sources not linked ICS

Other data sources

SU

Exce

SU

ICS

Facility survey

(hospitals- poor)

Exce

Exce

Data

District Fragmented Data management

CRIS

Exce

Exce

Exce

CRIS

Data

Exce Exce

Data Summary reports Compiled monthly reports Data

Summary report

Facility Multiple & registers

IC

Mor idit .

Hospital

AR

Summary report

ARV patient Paper

PMTCT

Othe

In & out patients Records / registers

Fig. 1. The previous Health Information System in. Sierra Leone was characterized by fragmentation and overlap between data collected by the different actors

V. PROPOSED SOLUTION From the current fragmented HIS described above, the main aim was to provide meaningful and relevant information for decision making and to diminish the workload of staff who is collecting and reporting the data. The key concepts used to address this issue are “integration” and sustainability. The suggested way forward depicted in figure 2 is to use DHIS to integrate the various data flows and data sources and thereby to provide an integrated framework for M&E and data management. The approach is to design data entry interfaces that are similar to existing ones in order to enter all data in

DHIS. But at the same time, data can be imported from existing applications into DHIS. So instead of entering overlapping data at district level in many databases, now data will be entered in one application (DHIS) and then exported in appropriate format to other applications. This model brings together essential data and indicator sets from all relevant sources disintegrated at the level of the reporting unit. Revision, harmonization and standardization of data and indicator sets and tools for registration, collection and reporting of data are part of this process and particular attention must be given to the infrastructure that support the system as well as capacity building.

383

Figure 2: Vision Integrated & Linked data Sources All information available through “one point” / Data repository

Dissemination & sharing; Reporting, indicators M&E, GIS/ Maps web-dissemination

Census data

DHIS

CRIS

Moved to & Managed by DHIS Facility

Gateway: Data extract & integrate

surveys Excel

Health & Demogr. surveys

Other data sources

Gateway: Data extract & integrate

Epi-Info ICS

Vital statistics

Other Services statistics

National level 1. Export data Captured by DHIS

District level Dissemination in districts; Reporting, indicators M&E, GIS/ Maps

DHIS

Facility level

VI. INTERVENTIONS A. The consortium : Once a decision was made to improve the HIS in Sierra Leone based on the findings of the initial assessment, the Ministry of Health (MoH), the Health Metrics Network (HMN) and HISP had organized several meetings during which many questions regarding project aims were addressed. In doing so, some key actors were identified and enrolled in an alliance of complementary and experienced actors to run the project: Ministry of Health of Sierra Leone: is represented at all levels of the project purposefully to make its participation and ownership very strong – as manager, user and beneficiary. Its experience in health management in the local context and its good understanding of what is needed both as a supplier and consumer of health information was crucial. Health programs: it is useful to look at health programs within the MoH as separate entities, as they are key actors of the integration process. They have their own HIS, which is the main reason for the current fragmentation. Some have computerized HIS and some not. They are autonomous vis-àvis the MoH. Examples: Mother & Child Health, EPI (immunization), HIV/AIDS, Tuberculosis program Health Metrics Network –HMN : is a global agency under the WHO and aims at strengthening developing countries’ HIS. They have developed a HIS framework and tools for

CRIS

Data capture

Feedback

Gateway: Data extract & integrate 2. Import data Captured by ICS Epi-info etc

Epi-Info ICS

Other Services statistics

Data capture

Feedback reports: M&E, indicators, Supervision District meetings

assessing country HIS (see http://www.who.int/healthmetrics/). In this project, HMN plays the role of executive sponsor and project manager. Health Information System Program – HISP: is a global network in HIS development, open source software, ICT for Development and research and education. It comprises several actors from a number of countries and has been successfully implementing sustainable and integrated HIS in developing countries. The software named District Health Information Software (DHIS) is developed within the HISP network. HISP started in South Africa 1995 and has been instrumental in developing the HIS there. Since 2000 HISP has developed into a global South-South-North collaborative network with nodes in South Africa, India, Tanzania, Nigeria, Mali, Vietnam and Norway (see http://hisp.info). ICT Development Center – IDC Sarl : is an international IT company based in Mali, specializing in open source and eHealth solutions for Africa. It offers eHealth consultancies, eHealth software development such as telemedicine applications and has a strong experience in HIS and IT project implementation in developing countries. IDC is also member of HISP and has contributed in the design and implementation in Sierra Leone. The task in Sierra Leone is to provide regional support and help enabling the local core team assume full ability in running the system. Inveneo: is a US IT company specialized in open source and low power computing and has deployed low power infrastructures in many developing countries. Inveneo’s strategy is to build alliances with local IT companies that can

384

be empowered in order to provide support for the low power competency and responsibility, two pluses indicate substantial equipment and avoid expensive and problematic traveling competency and involvement, while one plus indicates some from US. involvement only. This only relates to the roles in Sierra The attracting characteristics of this consortium lay in the Leone, specifically). This overlapping contributes a lot in the diversity and the synergy of the actors enrolled. As depicted in common understanding of the problem in its context. table 2, their expertise cover all fields requested and are overlapping (Three pluses indicate that this is the core : Table 2: Expertise of the consortium members Actors Politic Policy Health HIS Open Hardware making source +++ +++ +++ ++ MoH + +++ + Health programs +++ +++ + +++ HMN ++ ++ +++ +++ +++ HISP +++ +++ +++ + IDC +++ +++ Inveneo • The use of Ubuntu Linux OS makes the system virus free and unlikely to corrupt database files. System start or restart erases any changes in system files. B. Infrastructure This reduces significantly the system maintenance To face the challenging infrastructure context of Sierra need. Leone, a decision was made to deploy Inveneo low power hardware running Linux operating system. In doing so, we use • The server runs also Ubuntu Linux and hosts the DHIS one stone to kill two birds: respond durably to the lack of application. DHIS is then accessed remotely from power, and counter computer virus. other computers regardless its operating system. The Inveneo hardware is an open hardware design (all plan and software are published through the Inveneo website) that In each site one Inveneo desktop, one server and a mixed – operates using open sources software. wired and wireless – local area network(LAN) were deployed. Each site in the project was provided with an Inveneo All equipment deployed are based on low power. The 12v computing station and an Inveneo hub server. These power is provided by a car battery which can supply the computers are designed for challenging environments and system for 8 to 12 hours of working when it is charged and have the following capabilities: can be charged by either solar panel or the existing generator when this is turned on.

Figure 3 Inveneo server

• Very low power consumption (total power consumption of 18 watts at peak operation including LCD monitor) allowing the units to operate efficiently on 12 volts DC, with renewable power from solar panel, generators, etc. • Solid state design (fanless, driveless) eliminates fans, hard disks, or moving parts of any kind, improves durability in hot, humid and/or dusty environments, making it ideal for use in Sierra Leone.

The LAN allows M&E officers to smoothly and seamlessly access DHIS both from the existing MS Windows computer and from the new Inveneo desktop. Existing computers can be used only when the generator is running since they are not low power based. But the Inveneo desktop and laptops can access DHIS at anytime when the battery is charged. Thus both M&E officers as well as the district medical officer are able access DHIS over a LAN even in times where no external power source is available. The server comprises two hard drives, synchronized. In case of a problem, the currently running (now faulty) hard drive can be replaced by the second one as an emergency solution. Inveneo unsuccessfully sought an eligible local partner that could be trained to be responsible for the in-country support. Nevertheless, the emerging local core team for the project described here came to constitute this role during the pilot implementation phase.

385

C. Software Development The software application used for the new integrated HIS solution is called DHIS, and is an open source software application developed in the global HISP software development community. Briefly the DHIS can be described as a tool for data collection and visualization of statistical data, tailored to support integrated health information management activities. It is a flexible tool which allows for extensive customization to meet local needs for HIS standardization of collection forms, reports and data analysis. The process of customizing the DHIS to the Sierra Leone context has been ongoing since July 2007 and has gone through many feedback cycles between users and stakeholders in the field in Sierra Leone, the IDC team responsible for the local customization, and the developers in Norway. During the more intense periods the online communication (chat, e-mail lists, issue tracking, source code repository) between Sierra Leone and Norway has been so active that new builds (versions) of the software have been released on a weekly basis with rapid response to user feedback. Furthermore, the innovative designs of feedback reports in Sierra Leone, created in collaboration with HMN, local M&E officers and the IDC/HISP have pushed and inspired the global developers to come up with a better and more generic report solution that later has benefited users also in other countries. While the DHIS software is based on web technologies and can be used in networked settings, it also supports stand alone installations in locations with no internet. This adaptability to a varying infrastructure has been especially important in Sierra Leone as there have been local networks in each district set up around an Inveneo server, but at the same time no network connectivity between most of the districts and the national server has been present. DHIS has been installed on a district Inveneo server enabling multiple desktops in the same building to access and use the application simultaneously. This enables sharing of the district data warehouse and access to all data to all program managers, M&E officers and other users, as well as a secure server environment where the data are protected on a virus free Ubuntu Linux system. On a monthly basis each district exports its data to an external XML file and sends it to the national level either using a dialup connection or using a flash-drive and physical transport. To address the problems of the previous situation of the HIS, a new computerized system was introduced following best practices from HMN and HISP, and implemented using

the DHIS software. Key principles of DHIS design strategy are; 1) a modular data approach focusing on data item as the atomic unit rather than the larger data collection forms, 2) disaggregation of data enabling drill-down to health facilities from all levels, and 3) to collect and process data across health programs on an integrated data warehouse. All these principles, although arising from many years of implementation in various other African and Asian countries, were highly relevant to the Sierra Leone context. A major problem in the previous solution that had to be addressed in the DHIS was the duplication of data reporting from the health facilities, as the same data items was collected in multiple forms. At the same time it was an important requirement from the users that the computerised form in DHIS would look exactly like the paper forms they were familiar with. The software was customised so that all data items collected in the various paper forms were defined as individual elements in the database, but without duplication. In the DHIS these data elements are the atomic units and the data collection forms are just visual representations of a given collection of such elements, as illustrated in figure 4. While the user interface for data collection maintained the look of the fragmented paper system, the database behind was organised so that no duplication took place. E.g. if a data element was shared by two collection forms, then after registering its value in one form the value would automatically appear in the second form without a need to enter it again. Another benefit of such a modular approach was that all data elements are stored in the same integrated data warehouse and available for processing and correlation independent of any collection form. It was also possible to define calculated indicators based on these data elements to analyse coverage and rates as opposed to the “raw” numbers. E.g. BCG coverage (rate of infants given the BCG vaccine against tuberculosis) could be calculated by dividing the monthly reported data on BCG doses given, by the population estimate under 1 year for the same area and thereby provide the M&E manager with information comparable across time and space. A set of feedback reports were designed to provide districts and chiefdom administrations with a summary report of last month’s data, which combined a few selected key data elements and indicators and used charts to visualise trends over time, comparisons across areas, as well as an overview of the reporting completeness.

386

Figure 4 Overview of integrated data warehouse

D. Capacity building In each district, two M&E officers are working to collect, prepare, report, and analyse aggregated data. During six weeks in May-June 2008, extensive training was given for all M&E officers. The training was given as a three-module course, each module taught one week. Given the authors earlier experiences with training similar positions in the use of the same software in other countries, we opted for a one-week introductory course to the use of computers in general, followed by one week of training in the use of DHIS and one week in use of health information. In the end then, only 1/3 of the training was actually in the use of the software in question. Two weeks extra training were seen as necessary to increase the level of post-training use. First of all, few M&E officers actually had any experience with computers at all. So one week was dedicated to lift everyone to the same basic level of computer literacy. Second, sustainability can only be achieved if the HIS actually serves a need. In a context of poor information understanding and use, the collection and analysis of data will not be institutionalised. Therefore, the third week was dedicated to public health administration using primary health care data, and challenges and implications of a supporting HIS. Data entered during the week of DHIS

training was analysed, interpreted, and presented, and actions to improve the situation were discussed. A clear distinction between the three modules were not drawn, on the contrary, efforts were made to integrate them, for example by using epidemiological data in excel spreadsheet training. But on the whole, the capacity building was centred on the three following components: 1) Using computers, to get the most of DHIS training 2) DHIS operation, including regular tasks and basic database maintenance 3) HIS and data use, to create understanding and motivation. Using data from DHIS and analytical and presentation tools taught in the first week. This approach had a few benefits that justify the relatively long time spent by M&E Officers away from office. First of all, the basic computer skills taught in the first week were essential to both understand how to use DHIS, and also to use Microsoft Excel, Powerpoint, and Word, which are common to analyse and present data once entered and processed in DHIS. Furthermore, it enabled a faster progress during week 2 and 3, which meant that a few, inexperienced M&E officers would not slow down those already familiar with computers. Secondly, special topics relevant for the context could be examined, with the students, in depth. The training schedule was revised daily based on the progress so far, and what the students wanted to learn. Motivation and inclusion played a

387

major role in forming the curriculum. For instance, while most of the districts are not connected to the internet, we found time for a session on browsers, free email hosts, search engines, and how to connect with mobile broadband through one of the local mobile network providers. Another example is the issue on computer viruses and possible solutions to the problem, which were explained in depth by, so to say, popular demand. Thirdly, and quite importantly, the extensive training was used as an arena for user feedback to the system implementers. Especially the third week focused on plenary discussions on topics such as factors influencing the health status of the community, the validity and use of data forms, and data quality. In addition, requirements to DHIS and report designs were voiced during the training of the software. The training was a dialectical process, in which both trainers and students learned from each other and converged on a common understanding of the challenges ahead and the way forward for the project. A few examples can highlight the mutual learning experienced. One form, for filling in among other things stocks of antigens at the facilities, was fraught with errors. Using the validation rules we had set up in DHIS, we could catch these errors after entering the data into the software. Running these validation checks, on one occasion only 5 of the 21 M&E officers found no errors in the form they had just filled in. The errors had two causes; one was wrong calculation of stock balance, another was a wrong interpretation of the data that was to be entered. It turned out that the M&E officers, and most likely also the nurses filling in the forms at the facilities, were not fully aware of what should be recorded; vials or doses. For example, BCG vaccines come in vials of 20 doses each. If you don’t use all 20 doses at once, you are supposed to discard the rest, because rapidly they lose potency. So the data elements of Start Balance, Received, and End Balance should be dividable by 20. Distributed and Discarded, however, could be different, but should add up to vials of 20 doses too. Counting the various data elements differently was quite common, however, and several M&E officers did not understand the correct procedures when the errors showed up in the software. A quite lively discussion erupted, only to calm down when a representative from the EPI program at national level clarified correct use. Then, for another data element, where the header was “Doses discarded due to”, and the subheader was “Expiry date”, many forms actually contained the expiry date of the various antigens in stock, rather than amount of doses discarded due to them being too old. This revealed both a lack of understanding of the forms at facility level, and also a possible ambiguous design of the form itself. A similar example comes from the inconsistent closing dates for the data recording books at the facilities. For one form, which was supposed to cover one month, we would find that some facilities reported as early as the 25th day of the month. Since some facilities were far away from the district

center, they closed the books and did the aggregation for the district health team a few days earlier. Other facilities just filled in the form when they had time, and the exact day could vary from month to month. Thus some forms could contain data from 23 days, some for 34 days, etc. The M&E officers being trained knew about these varying practices, although they did not initially see this as a problem affecting data inconsistencies. Lastly, efforts have been made to include the new technology applied to the capacity building activities. This has been done by focusing on strong on-site support by local partners who have been given extensive training in the Inveneo technology. VII. DISCUSSION: Our discussion builds on an ongoing project, and it would be too early to draw many conclusions at this time regarding sustainability and the long-term implications of the applied approaches. Nevertheless, the project can already show to interesting findings regarding the four topics presented in depth earlier; the consortium, infrastructure and software solutions, and capacity building. A. The consortium As we have seen above, the previous HIS was extremely fragmented, but the solution proposed was not to solely provide a new system that aimed at integrating everything. The focus was not merely on the application but on the whole information infrastructure, which comprises both technical and non-technical components. The diagnosis of the HIS went in depth and revealed that existing infrastructure and applications as well as human capacity were weak and needed to be addressed with equal attention. Being mindful of users and collective participation, the strategy was to start with a prototype that would contain data from several existing data sources such as EPI, facility survey and population estimates. The prototype was then a useful concrete example of what and how the proposed integrated HIS could be like. It acted as a boundary object and enabled actors participation that gave more input to the project [5] [10]. On the one hand, by showing this concrete example, first, major actors such as HMN and MoH got a good understanding of the potential and benefits of such system. Second, it helped health programs that were reluctant to collaborate within the newly proposed framework to realize that everybody will get more output from the new system by gathering all different sources in one data warehouse, while at the same time pursuing their own trajectory. For health programs that did not have any computerized HIS, it was a great opportunity to move forward. On the other hand, those who are feeding the system with data immediately saw that i) at district level, the capability of the system to avoid overlapping data by displaying automatically the value of one

388

overlapping data element already typed for a given facility and a given period will considerably reduce their workload, since overlapping rate among two forms can be as high as 50 %. The system’s capability of networking was also attracting for them. ii) at the grass root level, health care providers who are also those who are producing data had understood that the proposed solution can trigger a process that will lead to harmonization of reporting tools which in turn will diminish the 17 overlapping forms they have to fill at the end of each month. The alignment of interest of actors at all levels [11] made a large alliance of actors that was strong enough to get a momentum in the project and got them very involved and active. During several months, they all worked together through a process of improvement, feedback, and feedforward between developers, implementers, managers and users. The heterogeneous and complementary nature of the consortium was useful in this regard. At the early stage of the project, an important attention was given to the infrastructure issue. The team wanted a sustainable and flexible solutions that suit the context of Sierra Leone – lack of power supply, no resource to fuel generators, viruses to name a few. Then came the idea of bringing Inveno experts with their low power computers that can work with solar panels and car batteries and based on Linux and open source applications. In the field, IDC experts were trained by Inveneo because by then, they were the only able to have a good understanding of the technical system. They are supposed to continue the training of the core team as well as organizing a local company to provide ongoing technical support. The overlapping expertise or the continuous expertise domain of the consortium helped keeping in place the expertise of one member even when he left. The fact that IDC has a good background in hardware and open source helped to acquire knowledge from Inveneo and give it back to the local core team on the basis of learning by doing. The HISP-IDC team is engaged in a long term support of the project and local expertise building. B. Infrastructure The new infrastructure ensures data safety, and since its deployment, users have not experienced loss of data as they used to previously. Even when their existing computers are infected by viruses, the server as well as the Linux desktops still work. So now they have power to run computers and computers to practice what they have learned during the intensive training without spending more money for generator fuel. One month after deployment in four piloting district and at the MoH, one sever stopped working and the diagnosis has revealed that the motherboard was off because of overheating, while it was supposed to be designed to overcome heat. The problem was reported to Inveneo who sent a new motherboard and it was an opportunity for them to improve the design of their computers. All computers were then upgraded and since then, similar problems have not been reported. By “black-

boxing” DHIS on the server without screen, keyboard and mouse, the equipment is kept out of reach of local “IT experts” in districts who used to format hard drives as the standard solution for any trouble shooting. The approach chosen in Sierra Leone has been to go around the problem of the current infrastructural problems of no power supply by establishing a new technical infrastructure based on solar power and low-power computers. Technically, this is an innovative and wise solution. However, it will only work if the consortium succeeds in establishing an effective technical support structure, including training and maintenance. C. Software solution As far as the integrated data warehouse is concerned, the prototype strategy helped users at all levels to grasp the system and give more input to its improvement. The DHIS is continuously being improved through these feedback processes involving all the stakeholders; computer users, information users, implementers, hardware builders. Users in the field are aware of the flexibility of the DHIS and see this as an opportunity to make the work as easy as possible for them. For instance, both users and project managers acknowledged that it is time to improve feedback and reporting system from district M&E to the health programs by shifting from a manually generated report to one “push button” report generation. This was the highest priority local requirement for the DHIS from the users. But at the developer’s level, the credo is to always make the globally used software as flexible and generic as possible and keep the adaptation to the local context for the customization phase. This mechanism and philosophy improved the software a lot at the global level as feedback and experiences from the field in Sierra Leone were highly relevant to other countries as well. The win-win alliance between local needs and DHIS development at global level was enabled by the prototyping strategy at local level and the need and philosophy of developing DHIS as a flexible standard; that is, flexible both in terms of use and change. For the new software solution to be accepted among the local users it was important that the DHIS was customized so that the electronic data collection tools looked exactly like the paper forms. The modular data element design enabled collection forms to be generated simply by combining data elements and without disturbing the carefully designed integrated data warehouse underneath. The way already registered values would be automatically filled in the electronic forms when opening a second form using the same data element was highly appreciated by the health workers that had previously wasted a lot of their time on manually filling the exactly same data in various multiple forms. Modularizing data in fine-grained data elements as opposed

389

to the traditional collection forms has several advantages. This metadata model enabled collection forms similar to the paper versions to be generated, which was an important requirement to get buy-in among the end users, simply by combining data elements, and without disturbing the carefully designed integrated data warehouse underneath. Furthermore, in an environment with changing requirements and a range of stakeholders with different needs, the modular design is flexible to change as standardization takes place at the data element level. New collection forms, indicators or report formats are just new compositions of data elements which can easily be defined in the user interface without changing the underlying integrated data warehouse. This flexibility has been important in the early phases of implementation when the formats of feedback reports and indicator definitions went through many feedback cycles among the users, the HMN specialists and the developers. The introduction of feedback reports has been an important step in the DHIS implementation process and a maturing stage towards a revision of the entire HIS reporting system. The integrated data warehouse at district level has provided the local M&E officers access to their data across health programs, and the tools for processing and analyzing data have helped to illuminate gaps and problems with the existing data collection forms. Feedback reports, in stark contrast to typical data collection forms, have focused on data analysis and use and not upward reporting, and have given local M&E officers a quick overview of the situation from month to month. These reports have enabled local monitoring of key performance indicators of the services provided, and identified gaps in reporting from the chiefdoms and health facilities. In stead of presenting data in fragmented and overcomplicated forms, these reports have used just a fraction of the data collected, extracting only the most relevant data from the health programs, and used charts and simple tables to visualize the information. D. Capacity building The extensive training provided some benefits, which were briefly mentioned above. These were: 1. A week-long general computer-course enabled a higher quality of the HIS-specific training. 2. Time was given for student-led discussions on relevant topics, which clarified both for the student body and the trainers various issues of relevance 3. The training was used to get feedback on the software and HIS from the students As mentioned earlier, the extensive training was given to develop a solid base both for using the software developed, and for creating the demand that could make it sustainable, by fostering a culture for using information. While it is too early to make any conclusions regarding sustainability, the preparatory week of general computer knowledge had the

effect that the main aim of the training, namely correct use of DHIS, was attained to an acceptable level. In addition, it gave the trainers and students some time to get to know each other, which proved crucial when health information issues were discussed at a later stage. In this case, the benefits of spending more time with the students became apparent during the third week. The examples from the immunization forms show how the training facilitated peer learning. Discussion amongst the M&E officers established a common understanding on important issues like standards and routines. It also provided the trainers with an understanding of the situation “on the ground”, and what could be done to counter any challenges. Lastly, since the trainers were also the authors of this paper, the points about discussions and user-feedback highlights the potential power of action research. Being part of the consortium and taking an active role in the process to improve the current situation, in this case as trainers for the district M&E officers, the authors gained knowledge about the issues raised in the previous paragraphs. As the training unfolded, and over time the students became more secure on each other and the trainers, the discussions flourished and provided an extremely rich arena for learning about the Sierra Leone health system. VIII. CONCLUSION The HIS in Sierra Leone, presented in this article, has been developed in an evolutionary participative manner, and it is still in the making. Earlier approaches to develop HIS in Sierra Leone have been to focus on partial solutions for one organization or health program, or, in some occasions, to implement new reporting formats including several of the health programs. A major problem has been that fragmentation has rather increased than decreased. In the latter cases where all-encompassing reporting formats have been introduced, various groups of users and health programs have not been satisfied, and they have continued to use the old reporting formats in addition to the new ones, and thereby worsening the situation. The particular approach chosen by the ongoing project in Sierra Leone has been different from earlier attempts in that it is trying to incorporate all the various requirements for reporting and use of data in a step-wise approach. By taking the existing data as a point of departure, users have been able to join all the various reports being collated at district level within one database framework. In this way they have also for the first time been able to assess the quality of the data they are collecting – through being able to access and analyze the data – and they have been able to compare the data across the fragmented reporting structure. Through this integrated access to their own data, they have been enabled to identify the

390

problems with the current reporting formats; inconsistent data definitions between formats, missing data and overlaps. This learning process is now triggering, and feeding into, a much needed revision and overhaul of the entire reporting system, planned for early 2009. A principal difference between the chosen approach and the previous approaches is the allowed and promoted - flexibility to change the system as users and involved health programs learn concretely about current shortcomings and potentials for improvements. Ideally, of course, revisions of reporting formats should have taken place up-front, as part of the initiation of the project. Despite the high level profile of the project (Ministry of Health, HMN/WHO), however, the various health programs and donor agencies could not agree on shared revisions and new standards for data reporting. By having demonstrated how the reported data could be integrated, and the usefulness of such an approach, the flexible software and general approach have served as an effective means in negotiating new data standards, unifying the actors, and thereby building the consortium pursuing an integrated approach. The evolutionary approach followed is relying upon, and feeding into, the ongoing formal and informal on-the–job training that is carried out as part of the capacity developing activities. By basing the training on real data and the use of the system in practice, the training sessions become a key participatory input for the improvement of the system, while at the same time enhancing the learning process. Introducing radically new technical infrastructure and type of computers, as the Inveneo technology, represent a considerable capacity development challenge. The best way to approach this challenge is to make the technical part of the HIS integral to the training. Contextualizing the training by basing it on the problems in the district, including the hardware parts, is the best way to approach also this part of the sustainability challenge.

391

1. 2.

3.

4.

5.

6. 7. 8. 9.

10. 11.

Braa, J., et al., Developing Health Information Systems in Developing Countries. The "Flexible Standards" Strategy. MIS Quarterly, 2007. 31. Sæbø, J., J. Braa, and O. Chandna. A Flexible Approach to Integrating Health Information Systems: The case of Data Warehouse as Integrator in Botswana. in IFIP WG 9.4. 2007. São Paulo. Sahay, S., E. Monteiro, and M. Aanestad, Configurable politics: trying to integrate health information systems in developing countries. Special Issue of JAIS on eInfrastructure, 2007. Braa, J., E. Monteiro, and S. Sahay, Networks of Action: Sustainable Health Information Systems Across Developing Countries. MIS Quarterly, 2004. 28(3): p. 337-362. Puri, S.K., Integrating Scientific with Indigenous Knowledge: Constructing Knowledge Alliances for Land Management in India. MIS Quarterly, 2007. 31(2): p. 355-379. Hanseth, O., et al., Reflexive Standardization: side effects and complexity in standard making. MIS Quarterly, 2006. 31(2): p. 355-379. Checkland, P. and S. Holwell, Action Research: Its Nature and Validity. Systemic Practice and Action Research, 1998. 11(1): p. 9- 21. Avison, D., et al., Action Research. Communications of the ACM, 1999. 42(1): p. 94-97. Baskerville, R.L. and A.T. Wood-Harper, A Critical Perspective on Action Research as a Method for Information Systems Research in Qualitative Research in Information Systems, M.D. Myers and D. Avison, Editors, Sage Publications. p. 129-145 (2002). Bowker, G.C. and S.L. Star, Sorting Things Out: Classification and its Consequences. 1999, Cambridge, MA: MIT Press. Timmermans, S. and M. Berg, Standardization in Action: Achieving Local Universality through Medical Protocols. Social Studies of Science, 1997. 27: p. 273-305.

392

Mobile Telephony Access and Usage in Africa Chabossou, A., Stork, C., Stork, M., Zahonogo. Z.

Abstract—This paper uses data from nationally representative household survey conducted in 17 African countries to analyse mobile adoption and usage. This paper shows that countries differ in their levels of ICT adoption and usage and also in factors that influence adoption and usage. Income and education vastly enhances mobile adoption but gender, age and membership in social networks have little impact. Income is the main explanatory variable for usage. In terms of mobile expenditure the study also finds linkages to fixed-line, work and public phone usages. These linkages need however be explored in more detail in future. Mobile expenditure is inelastic with respect to income, ie the share of mobile expenditure of individual income increases less than 1% for each 1% increase in income. This indicates that people with higher income spend a smaller proportion of their income on mobile expenditure compared to those with less income. The study provides tools to identify policy intervention to improve ICT take up and usage and define universal service obligations based on income and monthly usage costs. It help to put a number to what can be expected from lower access and usage costs in terms of market volume and number of new subscribers. Linking this to other economic data such as national household income and expenditure surveys and GDP calculation would allow to forecast the economic and social impact of policy interventions. Concrete recommendations are being made for policy interventions and regulatory measures to decrease access and usage costs.

in developing economies and find that they are playing the same crucial role that fixed telephony played in developed economies in the 1970s and 1980s. Their study covered 38 developing countries for the period 1996 to 2003 and provides evidence of the strong impact of mobile in developing countries. TABLE I ICT DENSITIES IN 2007 (SOURCE: ITU DATABASE 2008)

Internet users per 100 inhabitants Sub Saharan Africa Low Income Lower Middle Income Upper Middle Income High Income Countries Benin Botswana Burkina Faso 1 Cameroon 2 Côte d'Ivoire 3 Ethiopia Ghana Kenya Mozambique 4 Namibia Nigeria Rwanda 5 Senegal South Africa Tanzania Uganda Zambia

Index Terms—Economics, Information technology, Mobile communication, Developing nations

T

I. INTRODUCTION

The information, telecommunication and broadcasting sector is increasingly integrated into the day to day activities of businesses and lives of households and individuals worldwide. Numerous studies have shown that ICTs contribute to economic growth, employment and social inclusion. It is therefore important that policymakers monitor progress towards access and usage of ICTs. A paper by Roller and Waverman (2001) suggests that the spread of modern fixed-line networks in OECD countries was responsible for one third of output growth between 1970 and 1990. The importance of ICTs for economic growth are further supported by studies from Jalava & Pohjola (2002), Oliner & Sichel (2000), Pohjola (2001), Niininen (2001), Sichel (1997), Jorgenson et al (2005), OECD (2003) and UNCTAD (2006). Waverman et al (2005) investigate the role of mobile phones

6.47

Main Mobile (fixed) cellular telephone telephone lines per subscribers 100 per 100 inhabitants inhabitants 3.17 26.77

6.04 15.69

5.87 15.23

24.06 50.99

31.48

21.01

86.18

54.78

43.85

112.42

1.66 4.25 0.59 2.23 1.63 0.35 2.77 7.99 0.93 4.87 6.75 1.08 6.62 8.16 0.99 6.48 4.19

1.22 7.28 0.70 0.79 1.41 1.06 1.60 0.71 0.33 6.66 1.07 0.24 2.17 9.56 0.58 0.53 0.77

20.98 75.84 10.90 24.45 36.60 1.45 32.39 30.48 15.42 38.58 27.28 6.53 33.31 87.08 20.40 13.58 22.14

Country level data reveals a global digital divide between continents and countries but also within countries 1

Internet and main fixed line data from 2006 Internet and main fixed line data from 2006 3 Internet and main fixed line data from 2006 4 Main fixed line data from 2006 5 Internet data from 2006 2

393

(Baliamoune-Lutz, 2003). Table 1 displays the difference between low, lower-middle, upper-middle and high-income countries for Internet, fixed-line and mobile density. It also display the same information for countries covered by this studies and the sub-Saharan Africa (SSA) averages. The SSA average for Internet users and fixed-line users is lower than the low income average. It can also be seen from Table 1 that several countries from this 16-country sample are below the low income and SSA averages. Countries have a much lower ICT density compared to the low-income average. Table 2 and Table 3 document the digital divide within countries for access to fixed telephony in urban areas compared to rural ones and knowledge about and usage of the Internet across disposable income quartiles. TABLE 2 HOUSEHOLDS WITH WORKING FIXED-LINE PHONES

National Rwanda Uganda Tanzania Mozambique Cameroon Kenya Ghana Nigeria* Benin Burkina Faso Côte d'Ivoire Ethiopia Botswana Senegal Namibia South Africa Zambia*

0.10% 0.30% 0.90% 1.70% 1.80% 2.30% 2.60% 2.70% 4.60% 4.70% 4.80% 7.60% 11.00% 11.70% 17.40% 18.20% 2.44%

Major Urban 1.20% 1.60% 2.80% 7.30% 3.50% 11.40% 6.30% 9.70% 15.50% 18.30% 10.80% 46.00% 7.80% 21.50% 34.10% 38.90% 5.20%

Other Urban 0.00% 2.00% 2.50% 1.30% 3.50% 0.90% 4.60% 5.70% 6.80% 16.70% 7.10% 22.70% 17.00% 19.20% 29.60% 18.50% 7.55%

Rural 0.00% 0.00% 0.20% 0.00% 0.00% 1.40% 0.30% 1.20% 0.60% 0.30% 1.30% 2.90% 8.70% 0.60% 7.90% 2.10% 0.00%

Bagchi & Udo (2007) are that African countries are not experiencing the benefits from ICT that are being experienced by OECD nations and lag severely behind even global averages in ICT adoption in particular in investment-intensive infrastructure such as fixed-line telephony and broadband Internet access. 6 McCormick & Onjala (2007) list the following reasons for the low level of ICT access and usage in Africa: • • • •

weak telecommunications infrastructure generally low level of economic activity irregular or non-existent electricity supplies; and lack of human resource capacity, lack of skills and brain drain.

The results from the researchICTafrica.net household survey, however, shows that the monthly cost of telecommunication services is the main reason behind low 6

levels of access and usage. TABLE 3 ICT KNOWLEDGE AND USE, AND MOBILE ACCESS ACROSS DISPOSABLE INCOME CATEGORIES

Benin Botswana Burkina Faso Cameroon Côte d'Ivoire Ethiopia Ghana Kenya Mozambique Namibia Nigeria Rwanda Senegal South Africa Tanzania Uganda Zambia

16+ knowing what the Internet is

16+ using the Internet

lower three dispo sable inco me quarti le 17% 14% 6%

33% 52% 17%

lower three dispo sable inco me quarti le 6% 2% 2%

16% 19% 10%

Owning a mobile phone or active SIM card lower top three dispo dispo sable sable inco inco me me quarti quarti le le 17% 63% 53% 83% 19% 50%

35% 12% 6% 26% 25% 2% 16% 35% 4% 37% 42% 6% 4% 39%

54% 31% 18% 27% 52% 10% 60% 52% 13% 71% 75% 16% 15% 60%

10% 4% 0% 5% 9% 1% 4% 10% 1% 9% 7% 1% 1% 1%

25% 14% 3% 8% 32% 3% 25% 22% 4% 14% 38% 4% 7% 13%

28% 33% 1% 53% 42% 20% 37% 74% 4% 29% 54% 14% 12% 36%

top dispo sable inco me quarti le

top dispo sable inco me quarti le

74% 63% 11% 79% 79% 49% 86% 93% 26% 77% 84% 46% 46% 84%

The great potential of ICTs as catalysts of social and economic development is clearly recognized in continental discourses of agencies such as NEPAD, economic regional association strategies and in national policies. ICT dissemination and adoption in developing countries facilitate the achievement of major development goals in the areas of health, education, governance and others. People in developing countries need easier and cheaper access to ICTs. They need enhanced ICT skills to better employ these technologies in their homes, schools and jobs. Policies and strategies that have been adopted however have not been successfully in addressing the question of how African countries can catch up with global levels of ICT access and usage and how the poor can join the newly developing information societies. For this it is important to understand the factors that lead to adoption of ICTs by households and individuals. It is further of importance to understand what factors other than income impact on usage once households or individuals have adopted ICTs. II. DATA AND DESCRIPTIVE STATISTICS The data that is being used for this study stems from

See also Sciadas (2003)

394

surveys conducted by Research ICT Africa (RIA) 7 in 17 African countries during the end of 2007 and the beginning of 2008 8 . The data is nationally 9 representative on a household level and for individuals 16 years of age or older. The survey was stratified into metropolitan, other urban and rural areas. Enumerator areas (EAs) were sampled for each stratum using probability proportional to size (pps). The individual to be selected for each household was randomly selected from all household members and visitors that stayed at the home on the night the household was visited and that were 16 years of age or older. The RIA questionnaire was divided in three sections. The first part, the household rooster, collected information about all household members. The second part collected household related information. The head of the household or someone that manages the household answered part one and two. The third part, the individual section, was answered by a randomly selected individual 16 years of age or older that slept in the house the night of the interview and included household members and visitors. TABLE 4 RIA 2007/2008 HOUSEHOLD SURVEY SAMPLE

Benin Botswana Burkina Faso Cameroon Côte d'Ivoire Ethiopia Ghana Kenya Mozambique Namibia Nigeria Rwanda Senegal South Africa Tanzania Uganda Zambia Total

Major Urban 432 348 416 490 502 1,173 473 472 562 311 895 415 432 779 634 436 405 9,175

Other Urban 336 241 329 347 312 631 324 557 312 294 1,012 333 312 465 393 347 212 6,757

Rural

Total

333 229 332 398 298 551 295 432 257 280 844 330 337 527 463 344 264 6,514

1,101 818 1,077 1,235 1,112 2,355 1,092 1,461 1,131 885 2,751 1,078 1,081 1,771 1,490 1,127 881 22,446

A household constitutes of a person or group of persons, irrespective of whether related or not, who normally live 7

Research ICT Africa (RIA) is a network of universities and research institutions from 19 African countries hosted at The EDGE institute, Johannesburg, South Africa. For more information see www.researchICTafrica.net. 8 These are Benin, Botswana, Burkina Faso, Cameroun, Cote d’Ivoire, Ethiopia, Ghana, Kenya, Mozambique, Namibia, Nigeria, Rwanda, Senegal, South Africa, Tanzania, Uganda and Zambia. The data for Zambia is not included in this analysis yet since the data had not been weighted by the time of submission of this article. 9 The data for Zambia and Nigeria are national extrapolations but not nationally representative. In Zambia the selection probabilities had to be estimated since enumerator ID information was not unique. In Nigeria the sample was drawn only from 6 of 36 provinces.

together in the same housing unit or group of housing units, have common cooking arrangements and share financial resources. Maids, guards and baby-sitters that stay with a family are part of the household. A head of a household is a person who economically supports or manages the household or for reasons of age or respect, is considered as head by members of the household or declares himself or herself to be head of a household. III. STATED AND REVEALED PREFERENCES The RIA household survey collected information from mobile users about their monthly expenditure but also from non-users about their willingness and ability (WTP) to spend on mobile telephony. The first is known as revealed and the latter as stated preferences. WTP measures are widely used to provide information to policy makers regarding the economic value of non-market, non-pecuniary or environmental goods. For stated preferences, no behavioural changes can be observed; the individuals only state that they intend to behave in that particular fashion (Adamowicz et al, 1994). The payment ladder method with exponentially distributed amounts was used to infer about the willingness and ability to pay for mobile services of non-users. The payment card was first used by Mitchell & Carson (1984). The payment ladder is a kind of payment card which lists WTP values from low to high. Enumerators read the values to the respondent, starting at the top of the list and moving down. They asked until the first three values were answered with no. If the respondents were almost certain about their willingness and ability to pay a monetary value then a tick (√) was placed in the space next to that amount. If the respondents were not sure about an amount then it was simply left blank. If the respondents were almost certain that an amount was too high then a cross (x) was placed next to the amount. Three crosses (NOs) were required to complete the question to assure that no intransitivities occurred. The highest amount the respondent was willing and able to pay and the lowest amount the respondent was not willing and able to pay was captured. The difference between these two values is the range of uncertainty (Bann, 1999). Having two points increases the chance of eliciting the respondents’ actual willingness to pay. The payment ladder avoids starting point bias and reduces the number of outliners (Bateman, 2002). To avoid range and centring biases 10 the payment ladder was based on exponential value increment, as suggested by Row et al (1996). This approach has three mayor advantages. First, the scale is consistent with the hypothesis of increasing measurement error with increasing value. For instance a person might value a cup of coffee in a café at $2,00 plus minus $0,20, but a vacation at $2.000 plus minus $200. Second, according to Row et al (1996) there exists evidence that measurement errors in values obtained in contingent valuation studies have a log-normal distribution and that a logarithmic transformation of the WTP data addresses the increasing measurement error in hypothetical 10

See Venkatachalam (2004) and Row et al (1996).

395

WTP data. Therefore the exponential payment ladder can be seen consistent with the error distribution of WTP values. The third argument Row et al (1996) put forward refers to the concept of “just noticeable” differences and they uses the example of light sources to explain this. A source of light has a brightness B. The differences between two of such sources is “just noticeable” if the difference can be detected in 75% of the time. When bringing these sources in a sequence arranged in order of increasing brightness B1, B2, …, Bn,so that each source is “just noticeable” to the preceding one, than the relationship to each other is given by Weber’s law: B

,

(1)

where k is a positive constant. Hence the “just noticeable” differences are increasing proportionally to the sequence of sources and can also been written by exponential function: (2) The value of k can be seen as a percentage increase between adjacent scale values. The survey used 28 values, starting with a zero (not interested at all) and ending with an empty cell for amounts higher than listed. Equation (2) was used to compute the 2nd amounts to 27th amount. K was selected so that (1 + k)n-1 equals the highest value on the list. 11 IV. MOBILE ADOPTION / ACCESS This section analyses factors that contribute to the probability of an individual to adopting mobile telephony, ie either owning a mobile phone or active SIM card. The dependant variable is dichotomous and using linear regression models would not be appropriated. 12 Probit models are being used instead, assuming normal distributed error terms. The assumption about the error term is arbitrary and its validity cannot be tested. 13 A model can only be identified once an assumption is imposed on the mean and variance of the error term. This implies that coefficients cannot be interpreted directly since they are influenced by the necessary assumptions. Changing the assumptions about the error term changes the values of the coefficients uniformly (Long, 2006), but it does not affect the probability of an event occurring 14 . It affects the spread of the distribution but not the proportion of the distribution above or below the threshold (Long, 2006). The decision to adopt mobile telephony is the result of several factors which cannot be observed and which will be different for each individual. What can be observed is whether an individual has a mobile phone or active SIM card (Mobile=1) or not (Mobile=0). One individual might neither be able to afford nor be interested in a mobile phone while

another might just be close to getting one and still saving money towards it. For both individuals it can only be observed that they do not have a mobile phone ie Mobile=0. The process leading to that decision is unobserved and referred to a latent variable. Probit or Logit models tie observable (independent variables) to the latent variable though contributions to the probability of the latent variable taking a value above or below a threshold that would lead to the observable outcome, adoption or not. Many factors are considered in the economic literature as influencing the latent variable, 15 such as socio-economic factors (social networks, income, prices, household size, education, age and gender) and environmental factors (infrastructures, enumerator area, country’s characteristics). Piccoli et al (2001) established that the education level of an individual influences ICTs adoption and usage, through influencing an individual’s capability to use technology. The theory of technology use also points to age as a factor that influences when and how an individual uses technologies. Some studies indicate that age is a key factor in the use of technology, with younger people tending to exhibit higher use levels. Consequently, these individuals tend to use ICTs technologies more than older users. Gender and technology studies have found that men and women adopt and use technology differently (Gefen & Straub, 1997; Venkatesh & Morris, 2000). Men’s decisions to use technology are more strongly influenced by their perception of usefulness, while women’s decisions are based more on perceptions of the technology’s ease of use (Venkatesh & Morris, 2000). Further, men and women may view the same mode of communication differently. 16 Table 5 in the appendix. displays the Probit model configuration and Table 6 in the appendix details the results for each country. Table 6 provides summary results without reporting coefficients, showing for each of the independent variables whether their impact on the probability of mobile adoption is significant and whether it is positive or negative. 17 The results suggest that the variables used to analyse mobile phone adoption contribute significantly to explaining the probability of adoption. Income increases the probability of mobile phone adoption as expected. Only for Ghana and Rwanda it wasn’t significant. Substituting individual income with disposable income increases the pseudo R2 in both cases and let to income being significant with positive coefficient. Disposable income has the advantage over individual income that it takes into account the buying power of dependents that do not have own income. It does not generally lead to better results though. Income being a barrier to adoption has two consequences. First, people will increasingly adopt mobile telephony should their income increase. Secondly, they might

11

For example with n = 24 cells which have to be computed and $1000 the highest value, the equation would be given by 1000 = (1 + 0,35)23 12 See for example Maddala (1983) and Gourieroux (1986) 13 For a more detailed discussion on this assumption see Long (1997). 14 See Long (2007) pp 49-50 for a proof.

15 See for example : Kwapong (2007) Fabiani al (2005), Bagchi & Udo (2007), Kabbar & Crump (2006) 16 See Gefen & Straub (1997); Ono (2003) 17 The detailed results can be found in the appendix.

396

be adopting it at current income if access (handset) prices and usage cost would come down. Age influences for some countries positively and for others negatively the probability of mobile adoption. Age is likely to impact positively the probability to adopt mobile up to certain level and then decreases. Using age2 instead of age in subsequent research could get to the bottom of this. 18 The gender variable is mostly insignificant as expected (11 countries). Only in Senegal, Tanzania and Burkina Faso does being a woman decrease the probability of mobile telephony adoption. In South Africa and Mozambique it even increases the probability significantly. The result from the Probit models confirm the assumptions about education as well as the rural areas. Membership in social networks contributed positively to the probability of mobile adoption in 7 out of 16 countries. For the remaining countries it was insignificant. Randomly selected individuals 16 years from each household without mobile phone or active SIM card were asked for their willingness and ability to pay for a handset and what they think a handset would cost them. These average figures are being shown in Table 7 in the appendix. The difference between average willingness to pay and average expected costs confirms income as the main adoption barrier. Table 7 also indicates the number of new mobile phone users operators could expect should they offer mobiles for US$20, US$15 and US$10 respectively. 20US$ handsets would attract about 3 million new customers in Kenya and Ivory Coast, for example. Ethiopia could expect 2,6 million more users if the handset price would drop to US$10 and provided that SIM cards would be available. Ethiopia is the country with the widest gap between willingness to pay and expected cost and the country with the lowest mobile teledensity (see Table 1). The handset price is also known to fluctuate with the availability of SIM cards. V. INCOME ELASTICITY / E-USAGE Perl (1983), Taylor (1994) and Taylor (2002) developed telecommunication demand models that can be addressed price and income elasticities for public phones, mobile phones and local, national and international fixed-line call as well as cross price elasticities. 19 Universal Service Obligations (USO) in this context have been thoroughly discussed in the literature. 20 Garbacz & Thompson (1997, 2002, 2003) find that price elasticities for landline services in the USA approach zero in recent years and that subsidies therefore have little or no impact on universal service. Crandall & Waverman (2000) Eriksson et al (1998) and Hausman et al (1993) support these findings. Wallsten (2001) finds that greatly reduced international settlement rates for telecom traffic between the United States and the rest of the world had the effect of reduced prices and 18

See Tegegne (1999). See for example Kaserman et al (1990) and Parsons (1998) 20 See Gasmi et al (2000), Panzar (2000), Rosston & Wimmer (2000), Valletti (2000) 19

increased traffic in developing countries due to relatively high price elasticities. However, Garbacz & Thompson (2005) point out that international calls are less of relevance for poor people in developing countries compared to local and national calls, where price elasticities are low. They also find that the mobile monthly price elasticity exceeds fixed-line price elasticity in developing countries by a substantial amount and suggest that wireline phones are substitutes in the mobile market while mobile phones are not substitutes in the wireline market but complements. This is to some extent contradicted by Esselaar & Stork (2005) who find studying usage data for nine African countries that mobile telephony is a substitute for fixed-line telephony across all income groups. Clarke & Wallsten (2002) argue that universal service for the poor in developing countries is generally ineffective. Modelling income elasticities could establish alternative approaches to current universal service definitions. Rather then specifying coverage or geographic spread, universal service obligations (USOs) could be in terms of costs for monthly usage packages that take into account existing income barriers. Lee (1999), in estimating access demand for mobile networks, shows that per capita GDP and the size of the existing fixed-line network increase the probability of people subscribing to mobile telephone services. This implies that the mobile telephone network is interdependent with the fixed-line network. Demand, as understood by economists, is defined as the amount of goods or services people are willing to buy at a certain price (ceteribus paribus - other things including income equal). In this sense, demand is a concept affected by buying power. A person may have needs but not demand du to income limitations. With insufficient income, demand can be zero or low even if the need is urgent. Demand or buying power for a goods or services arise from the consumers’ preferences for specific goods or services. Thus defining a good or service, and studying how the consumer orders his/her preferences for such a good or service in relation to other available goods and services is crucial. Demand arises from a previous knowledge of the good and a subjective evaluation of its benefits and costs. Communications services differ from other goods or services since some require a subscription prior to usage such as post paid mobile and fixed line services. Even for prepaid mobile services a distinction can be made between access (cost of a handset) and usage costs. Subscription is necessary to use the Internet as well in most cases (notable exception is the Internet café). In the case of telephony the access service provides utility to a customer in terms of granting the ability to make and receive calls, even though it is uncertain whether the ability will be exercised. This is the so-called option value (Hee Lee, 2006). For the purpose of analyzing demand income elasticity for mobile telecommunication services we assume that the potential usage basket of a consumer consists of calls (q) and a composite good (x) that represents all other goods and services. We further assume that the consumer’s decision on

397

individual demand for calls is made by maximizing a utility function within a given budget constraint. In addition to the factors of calls (q) and a composite good (x) affecting his/her utility, income (y) can be considered. Moreover, if call externality is incorporated in the utility of an individual, the number of existing subscribers who are making calls to and receiving calls from the individual affects the demand for calls. Thus, after including the size of the network (N) in order to reflect call externality, the utility function of a customer (U) is represented in expression (3):

logarithm of the left hand side and the right hand side: (8) Where and the socio-demographic factors set. The transformed demand functions for mobile phone services are then expanded by socio-economic factors and postulated to be of following form: (9)

(3) At the same time, the budget constraint considering access charge and call charge is given by ,

(4)

Where r is the price of access to the telecommunications the price of a call, y is income of the consumer and system, p the price of the composite good x. For the utility function in (3), the number of calls represents direct benefits of using the mobile, while the number of subscribers represents indirect benefits through network size. Given the budget constraint in (4), and assuming the utility function is strictly quasi-concave and differentiable, using Lagrangian techniques the demand functions for calls and the composite good, as shown below in expressions (5) and (6), can be derived from the optimal solution of the utility function: (5) (6) These functions differ from conventional demand functions in two aspects: The budget constraint is y-r, rather than y, which reflects the conditionality of having to purchase access to telecommunication services, and the demand functions depend upon the number of subscribers, which reflects the access and usage externalities. Perl (1983), Taylor (1994, 2002) and Taylor and Kridel (1990) developed telecommunication demand models that are now widely used in the literature. The model has the following form: ,

(7)

Where q denotes use of the telephone network, p denotes the price of access and use, y denotes income, and u denotes a random error term. Network externalities as well as now socio-demographic factors can be included in this model as well. 21 Equation (7) can be transformed by taking the natural 21 Haque (2007) suggests ways of testing for the functional form of the demand function for telecommunication services. This would be beyond the scope of this paper and will be addressed with a separate paper.

m = monthly mobile expenditure in US$, converted using nominal end of 2007 exchange rates (source IMF) Y= natural logarithm of individual income in US$, converted using nominal end of 2007 exchange rates (source IMF). The income is the sum of any salary or wage, selfemployment income, property income, income from agricultural produce, pension, transfer income and scholarships. A = age in years PP = dichotomous public phone variable: used public phone in the last three months =1, otherwise 0 WP = dichotomous work phone variable: access to a phone at work =1, otherwise 0 G = dichotomous gender variable, female=1, male=0 F = dichotomous fixed-line variable: fixed-line access at home =1, otherwise 0 SN= Dichotomous variable: social network (church groups, sports clubs etc.) membership= 1, otherwise 0 R = Dichotomous variable: rural= 1, urban (other urban and major urban) =0 The price variable used in the analysis is represented by the amount that an individual paid for access and usage of mobiles. However since country results are being computed individually, there will not be a change for any of the responses, the cost of the usage basket is the same for everyone in a country. This lets the price variable become a constant. Table 8 in the appendix presents the results from robust regressions round for each country using weighted data. Income is in each case significant and positively related to mobile expenditure. The gender variable is insignificant for 10 countries which is somewhat surprising given that the average income of women is lower compared to that of men for all countries but Mozambique. However income is already taking care off in the equation. This means that apart of being at different income levels women spend the same on mobile phone as men. For Botswana, Cameroon, Mozambique and Namibia the gender variables was significant and negative, which means that women spend less. Nigeria had the only positive significant coefficient. 22 22

These finding can be explored further by using mean rank comparisons for the ratio mobile expenditure / disposable income.

398

Having a working fixed-line phone at home is linked to higher mobile expenditure in 8 out of 16 countries. This can be explained by two factors. First, households with fixed-line phones are generally wealthier which might not directly be reflected in the individual income of the responded (dependents). Second, having a fixed-line phone at home provides more opportunity to talk, such as calling home. These results confirm the existence of a relationship between mobile and fixed-line networks. To contribute to the complementarity and substitutability discussion further the fixed-line expenditure would need to be investigated for any affect through the number of mobile phone users in the household as well. The survey data also can be used to explain usage pattern in more detail. Public phone usage has only been significant for 6 out of 16 countries. For Benin, Cameroon and South Africa it has a negative affect on mobile phone expenditure and for Burkina Faso, Rwanda and Senegal a positive one. This too needs to be analyzed in more detail. Public phones might predominantly be used by lower income groups or might actually substitute mobile phone expenditure. More country specific research is required to come to definite conclusions. Equally for the impact of access to a work phone or private use. Fore five countries the work phone variable shows as positive and significant. While having access to a work phone is a good indicator for income one might have also have suspected a negative impact on mobile usage expenditure since the work-phone could be used to substitute for mobile calls. Age has a significant and negative impact in six countries where that variable is significant. For those countries a youth user’s effect can be established. The main objective is to estimate demand elasticities. A distinction is made between income and price elasticities. Income elasticity describes the magnitude and also therefore the responsiveness of a demand change for a good in respect to a income change. Graphically expressed, the income elasticity is –ceteris paribus- the slope of the demand curve, dependent on the income.

eY =

ΔQ /Q ΔY /Y

Where: Q = quantity demanded; Y = income and, ey = income elasticity. The first derivative of the regression equation for income yields the coefficient for Y, which is the income elasticity, that is, the percentage change in mobile expenditure at a 1% change in income. 23 The estimations results show that income variable is statically significant and positive for all 17 countries. 23 See Wooldrige (2006) pages 46 and 705, Hej et all (2004) page 296.

Figure 1: Income elasticity: percentage change in mobile expenditure for 1% change in income.

The figure above shows the percentage change in mobile expenditure for a 1% income change. For all countries the mobile expenditure is inelastic with regard to income. Nigeria has the highest income elasticity and Mozambique the lowest. Since these are national figures based on cross-section data it implies that people with higher incomes spend a smaller proportion of their income on mobile telephony compared to low income households. Figure 2 confirms this. Income elasticity seems to be linked to relative cost of mobile usage, penetration and coverage. Those factors need to be explored in more detail in subsequent studies utilising supply side data and qualitative research.

399

price elasticity will be relatively low (much less than -1). But, as prices fall, in a much cheaper service environment, she argues that people will start using the phone for many nonessential purposes; ranging from relationship maintenance and simple conveniences. Added to this, pricing innovations that enable poorer people to purchase small denomination pre-paid calling cards to make phone calls it is intuitive that the number of calls made will increase significantly. Milne (2006) argues that when this happens, the price elasticity will rapidly rise towards -1 or even more than -1. One of the biggest problems in estimating price elasticity of demand (PED) for telecom services in developing country situations is the lack of accurate usage data. Unlike in developed countries where usage is easily obtainable from monthly bills for post-paid connections, the vast majority in developing countries, in the case of the current study, between 78% and 99% have no billing records (for prepaid shares see Table 9). Many use public phones, phones at work or other people’s phones which equally does not generate records.

Figure 2: Share of mobile expenditure per month in terms of individual income.

VI. PRICE ELASTICITY / E-USAGE Price elasticity describes the magnitude and also therefore the responsiveness of a demand change for a good in respect to a price change. Graphically expressed, the price elasticity is –ceteris paribus- the slope of the demand curve, dependent on the price.

eY =

ΔQ /Q ΔP / P

Where: Q= quantity demanded; P = price of the good and ep = price elasticity. The price elasticity of demand is in almost all cases, except for inferior goods, negative, if the price increases the demand decreases. To simplify the notation, demand elasticities are in general expressed in absolute value and therefore expressed as positive (Frank & Bernanke, 2003). Milne (2006) provides a very intuitive argument for the changing structure of price elasticity in a hypothetical case. She argues that when phones become available for the first time in a low-income community and usage charges or tariffs are relatively high (in terms of income) people will make limited number of essential calls. In such a scenario, even if the prices falls by a small amount the number of such calls; for instance in an emergency or substituting a telephone call for a bus ride to a city, will not increase by much. In other words

Figure 3: If calls were cheaper what would you do? (source: RIA 2007/2008 household survey).

400

could be reached with US$1, US$2 and US$5 Average Revenue per Users (ARPU) business models. VII. CONCLUSION

Figure 4: Average reaction to cost of using your phone came down by half or going up double (source: RIA 2007/2008 household survey).

Another difficulty is that cross section data, like the one that is being analysed here, is not suitable to calculate price elasticities since it only captures data for one snapshot in time. Changes in consumer behaviour due to price changes can only be analysed using time series data. Cross section data only allows a rough glimpse at price elaticities by using contingent valuation methods. Respondents with mobile phone or active SIM card were asked how the would react to price decreases and increases. The answers to these questions are being displayed in Figure 3 and Figure 4. Figure 3 shows that the vast majority of respondents would make more calls if prices would come down. Senegal and Ethiopia are the two countries with the highest share of respondents that would use the saved money for something else. Figure 4 shows how respondents state that they would react to a doubling and halving of call charges. What can be seen is the response is asymmetrical, ie respondents would react to price increases stronger than they would do to price decreases. However, approximating price elasticity from these two contingent valuations results is strictly speaking not possible. Respondents state their preferences and do not reveal it (observed behaviour reacting to price changes). Table presents data on the untapped market, those 16 years or older that do not have a mobile or active SIM card at present. It shows the results for the number of people that

This paper shows that countries differ in their levels of ICT adoption and usage and also in the factors that influence adoption and usage. Income and education vastly enhances mobile adoption but gender, age and membership in social networks have little impact. Income is the main explanatory variable for usage. In terms of mobile expenditure the study also finds linkages to fixed-line, work and public phone usages. These linkages need however be explored in more detail in future. Mobile expenditure proofs to be inelastic with regard to income, ie as income increases mobile expenditure increases to a lesser extent indicating its importance in individual budgets. Key policy interventions would be regulatory measure to increase access and usage of mobile services by promoting network investment. The current super profits being enjoyed by operators across the continent as a result of prices that are amongst the highest in the world need to be moderated. In most markets there is not effective competition requiring regulatory interventions to ensure cost based wholesale and retail pricing. Increasing Coverage: Licence fees for infrastructure investment should cover the administrative and usage costs of national resources only. The current high cost of licences in most countries used to generated funds for state coffers translate in high prices for consumers as operators recover their licence costs. Countries need to look at incentives to encourage networks rollout not disincentives. Wider Access: Access prices can be reduced by exempting telecommunication equipment, in particular mobile handsets and services from import duties and additional taxes. Allow more Usage: Policy measures to increase the competition within the industry are the best mechanism to reduce usage costs. In a competitive environment the operator can choose to compete on price or on service quality. Regulatory measure can improve competition on price by creating price transparency. Price Transparency: Contract and prepaid tariffs vary across operators to an extent that a product to product comparison is impossible for consumers. Non- transparent pricing, including rapidly changing promotions and misleading media campaigns, make it difficult for end-users to make informed decisions. This prevents operators from having to compete on prices and leads to higher average prices. Establishing price transparency on the basis of published monthly usage baskets will make the cost to the end-user transparent and force operators to compete on price or service quality. Cost-based Interconnection rates: Above cost interconnection charges are often used by dominant operators to restrict small operator from gaining market share. High interconnection charges make off-net calls expensive and punishes consumer or small operator. Cost based interconnection rates (or benchmarked on cost base) increase the competition between operators and hence lead to lower

401

usage cost. Licences for Low ARPU business models: Regulators should award additional licences to applicants that are able to offer low ARPU business models, preferably below US$2. The correct policy and regulatory measures can boost mobile access and usage. Avoiding sector and service specific levies and additional taxes will result in lower recovery costs and therefore lower prices. Lower prices would allow millions of people currently unable to afford to access mobile services to do so and those currently restricting their usage as a result of the high cost of services to use services more extensively and effectively. This is likely to produce greater profits for companies overall and Governments will as a consequence have even more money in their coffers due to taxes on greater operator revenues. VIII. REFERENCES [1]

[2] [3] [4]

[5]

[6] [7] [8]

[9]

[10] [11] [12] [13] [14] [15]

Adamowicz W. & Louviere J. & Williams (1994). Combining Revealed and Stated Preference Methods for Valuing Environmental Amenities. Journal of Environmental Economics and Management, Vol. 26, pp.271292. Bagchi, K. and G. Udo (2007). Empirically testing factors that drive ICT adoption in Africa and OECD set of Nations, Issues in Information Systems, Vol. 8, No.2, pp. 45-52. Baliamoune-Lutz, M. (2003). An analysis of the determinants and effects of ICT diffusion in developing countries, Information Technology for development, Vol.10, pp.151-169. Bann C. (1999). A Contingent Valuation of the Mangroves of Benut. Forestry, Department, Peninsular Malaysia, Danish Cooperation for Environment and Development and the Johor State. www.soc.uoc.gr/calendar/2000EAERE/papers/PDF/A6-Bann.pdf Last Appeal: 10.07.07 Bateman I., Carson R., Day B., Hanemann M., Hanley N., Hett T., Jones-Lee M., Loomes G., Mourato S., Ozdemiroglu E., Pearce D., Sugden R., Swanson J. (2002). Economic Valuation with Stated Preference Techniques – A Manual. Edward Elgar, Cheltham, UK, ISBN: 1-84064-919-4. Clarke, G., & Wallsten, S. (2002). Universal(ly bad) service: Providing infrastructure services to rural and poor urban consumers. World Bank Policy Research Working Paper, 2868, 1–54. Crandall, R. & Waverman, L. (2000). Who pays for universal service? When telephone subsidies become transparent. Washington, DC: Brookings Institution Press. Eriksson, R., Kaserman, D., & Mayo, J. (1998). Targeted and untargeted subsidy schemes: Evidence from postdivestiture efforts to promote universal telephone service. Journal of Law and Economics, 41, 477– 502. Esselaar, S., and Stork, C. (2005). Mobile Cellular Telephone: Fixedline Substitution in Sub-Saharan Africa, The Southern African Journal of Information and Communication, Issue 6, pages 64-73, ISSN 16072235. Fabiani, S., F. Schivardi and S. Trento (2005), ICT adoption in Italian manufacturing: firm-level evidence, Industrial and Cooperate Change, pp.1-25. Frank R. & Bernanke B. (2003), “Principles of Economic”. McGrawHill, Boston Garbacz, C., & Thompson, H. G., Jr. (1997). Assessing the impact of FCC lifeline and link-up programs on telephone penetration. Journal of Regulatory Economics, 11, 67–78. Garbacz, C., & Thompson, H. G., Jr. (2002). Estimating telephone demand with state decennial census data from 1970–1990. Journal of Regulatory Economics, 21, 317–329. Garbacz, C., & Thompson, H. G., Jr. (2003). Estimating telephone demand with state decennial census data from 1970–1990: Update with 2000 data. Journal of Regulatory Economics, 24, 373–378. Garbacz, C., & Thompson, H. G., Jr. (2005). Universal telecommunications service: A world perspective. Information Economics and Policy, 17, 495–512.

[16] Garbacz, C., & Thompson, H. G., Jr. (2007). Demand for telecommunication services in developing countries. Telecommunications Policy 31 (2007) 276–289. [17] Gasmi, F., Laffont, J. J., & Sharkey, W. (2000). Competition, universal service and telecommunications policy in developing countries. Information Economics and Policy, 12, 221–248. [18] Gefen, D. & Straub, D. (1997). Gender difference in the perception and use of e-mail: an extension to the technology acceptance model. MIS Quarterly, 21, 389– 400. [19] Gourieroux, C. (1986). Econométrie des variables qualitatives, Economica. [20] Haque, O., M. (2007). Income Elasticity and Economic Development, Springer, ISBN 0-387-24292-9. [21] Hausman, J., Tardiff, T., & Belinfante, A. (1993). The effects of the breakup of AT&T on telephone penetration in the United States. American Economic Review (American Economic Association Papers and Proceedings), 83, 178–184. [22] Heij, C., de Boer, P., Franses, P., Kloek, T. and van Dijk, H. (2004). Economteric Methods with Applications in Business and Economics, Oxford University Press, ISBN 0-19-926801-0. [23] Jalava, J. and M. Pohjola (2002). Economic growth in New Economy: Evidence from advanced economies, Information Economics and Policy, Vol.14, pp. 189-210 [24] Jorgenson D, M.S.Ho and K .Stiroh (2005). Productivity: Information Technology and the American Growth Ressurgence. London; Cambridge: The MIT Press. [25] Kabbar, E.F. and Crump B.J. (2006). The factors that influence Adoption of ICT by recent refugee immigrants to New Zealand, Informing Science Journal, Vol.9, pp.111-120. [26] Kaserman, D. L., Mayo, J. W., & Flynn, J. E. (1990). Cross subsidization in telecommunications: Beyond the universal service fairy tale. Journal of Regulatory Economics, 2, 231–249. [27] Kelly, T. (2004). Changing ICT ranking of African Nations. The Southern African journal of Information and Communication, Issue No. 5, 2004. [28] Kwapong, O.A.T.F (2007). Problems of policy formulation and implementation: The case of ICT use in rural women’s empowerment in Ghana, International Journal of Education and development using Information and Communication technology, Vol.3, No. 2, pp. 68-88. [29] Long, S. (1997). Regression Models for Categorical and Limited Dependent Variables, Advanced Quantitative Techniques in the Social Sience Series, Sage Publications, ISBN 0-8039-7374-8. [30] Long, S. & Freese, J. (2006). Regression Models for Categorical Dependent Variables Using Stata, Stata Press, ISBN1-59718-011-4. [31] Maddala, G. S. (1983). Limited-dependent and Qualitative Variables in Econometrics, Cambridge University Press. [32] McCormick, D. & J. Onjala (2007). Methodology for Value Chain Analysis in ICT Industry Frameworks for the Study of Africa, Institute for Development Studies University of Nairobi [33] Milne, C. (2006). “Telecoms demand: measures for improving demand in developing countries – a toolkit for action”. Discussion Paper WDR 0603, Main Report (World Dialogue on Regulation). [34] Mitchell R. & Carson R. (1984). A Contingent Valuation Estimate of National Freshwater Benefits – Technical Report to the US Environmental Protection Agency. Resources for Future, Washington D.C. [35] Niininen, P. (2001). Computers and economic growth in Finland, in Information Technology, Productivity, and Economic Growth: International Evidence and implications for economic Development, Matti Pohjola, ed., oxford University Press. [36] OECD (2003). ICT and Economic growth. Evidence from OECD countries, industries and firms. Paris (France). [37] Oliner, S.D. and D.E. Sichel (2000). The resurgence of growth in the late 1990s: Is Information technology the Story?, Journal of Economic Perspectives, Vol.14, pp. 3-22. [38] Ono, H. (2003). Gender and the Internet. Social Sciences Quarterly, 84, 111–121. [39] Orbicom (2003). L’observatoire de la fracture numérique... et au-delà. Conseil national de la recherche du Canada, Québec (Canada). [40] Panzar, J. C. (2000). A methodology for measuring the costs of universal service obligations. Information Economics and Policy, 12, 211–220. [41] Parsons, S. G. (1998). Cross-subsidization in telecommunications. Journal of Regulatory Economics, 13, 157–182. [42] Perl, L. J. (1983). Residential demand for telephone service. White Plains, NY: National Economic Research Associates.

402 [43] Piccoli, G., Ahmed, R. & Ives, B. (2001). Web based virtual learning environments: a research framework and a preliminary assessment of effectiveness in basic IT skills training. MIS Quarterly, 25, 401–426. [44] Pohjola, M. (2001). Information technology and economic growth: A cross-country Analysis, in: Information Technology, Productivity, and Economic Growth: International Evidence and implications for economic Development, Matti Pohjola, ed., oxford University Press. [45] Roller, L. H. & Waverman, L. (2001). Telecommunications Infrastructure and Economic Development: A simultaneous approach. The American Economic Review, Vol. 91, No 4, 909-923. [46] Rosston, G. L., & Wimmer, B. R. (2000). The state of universal service. Information Economics and Policy, 12, 261–283. [47] Row R. & Schulze W. & Breffle W. (1996). A Test for Payment Card Biases. Journal of Environmental Economics and Management, Vol. 31, pp. 178-185. [48] Sichel, D. (1997). The Computer revolution- An Economic Perspective, Washington D.C.: Brookings Institutions Press. [49] Taylor, L. D. (1994). Telecommunications demand in theory and practice. Dordrecht: Kluwer Academic Press. [50] Taylor, L. D. (2002). Customer demand analysis. In M. Cave, S. Majumdar, & I. Vogelsang (Eds.), Handbook of telecommunications economics (pp. 97–142). Amsterdam: Elsevier Science. [51] Taylor, L.D. and Kridel, D.J. (1990). “Residential Demand for Access to the Telephone Network,” in Telecommunications Demand Modeling, ed. by A. de Fontenay, M.H. Shugard, and D.S. Sibley, North Holland Publishing Co., Amsterdam. [52] Tegegne G.E., (1999). “Willingness to pay for environmental protection: an application of Contingent Valuation Method (CMV) in Sekota District, Nothern Ethiopia”. Ethiopian Journal of Agricultural Economics Vol 3: 123-130 [53] UNCTAD (2006). “Rapport 2006 sur l’Economie de l’information”, UNCTAD/SDTE/ECB/2006. [54] Valletti, T. M. (2000). Introduction: Symposium on universal service obligation and competition. Information Economics and Policy, 12, 205– 210. [55] Venkatachalam L. (2004). The Contingent Valuation Method - A Review. Environmental Impact Assessment Review, Vol. 24, pp. 89– 124. [56] Venkatesh, V. & Morris, M.G. (2000). Why don’t men ever stop to ask for directions? Gender, social influence, and their role in technology acceptance and usage behavior. MIS Quarterly, 24, 115–139. [57] Wallsten, S. (2001). Telecommunications investment and traffic in developing countries: The effects of international settlement rate reforms. Journal of Regulatory Economics, 20, 307–323. [58] Waverman, L., Meschi, M. & Fuss, M. (2005). The impact of telecoms on economic growth in developing countries. In Africa: The Impact of Mobile Phones: Moving the Debate Forward, The Vodafone Policy paper series, No.2. [59] Wooldrige, J. (2006). Introductory Econometrics A modern Approach, 4th edition, South-Western Cengage Leraning, Canada, ISBN 13 : 9780-324-58548-3.

403

IX. APPENDIX TABLE 5: PROBIT MODEL SPECIFICATION

Variable Dependent Variable

Variable Name Mobile (M)

Independent Variables

Income (I) Gender (G)

Type of Variable Dichotomous variable: Individual has a mobile or active SIM card = 1, 0 otherwise Continuous variable: Individual income in US$ based on end of year nominal exchange rates (source IMF) Dichotomous variable: female = 1, male =0

Expected sign Na

Comment

positive

Income is assumed to be the main barrier to adoption The assumption is that there is no gender difference when it comes to mobile adoption Technology being a domain for the younger generation Education should enable individuals to use mobile phones, while also being an indicator for income, model would drop variables if multicollinearity persists

Not significant

Age (A)

In years

Negative

Tertiary (T)

Dichotomous variable: highest education being tertiary = 1, otherwise 0 Dichotomous variable: highest education being secondary = 1, otherwise 0 Dichotomous variable: highest education being primary, remedial or traditional =1, otherwise 0; Dichotomous variable: highest education being vocational =1, otherwise 0 Dichotomous variable: rural = 1, urban =0

Positive

Dichotomous variable: social network (as church groups, sports clubs etc.) membership = 1, otherwise 0 Captures various factors that either are constant for a country such as the price for mobile or fixed telephony or that serve as reference for dichotomous variables such as the educational variables and rural-urban location.

Positive

Secondary (S) Primary (P) Vocational (V) Rural (R)

Social Network (SN) Constant

Negative

Negative

Mobile network coverage is less in rural areas compared to urban (major urban and other urban) ones and less adoption is expected therefore Membership in social networks might increase communication need Captures no education, but also urban

TABLE 6: NATIONALLY REPRESENTATIVE PROBIT RESULTS FORE EACH COUNTRY USING SAMPLING WEIGHTS (SOURCE: RIA 2007/2008 HOUSEHOLD SURVEY)

Country

Pseudo R2

Prob > chi2

Benin Botswana Burkina Faso Cameroon Côte d'Ivoire Ethiopia Ghana

0.396 0.16 0.253 0.258 0.292 0.452 0.158

0.00 0.00 0.00 0.00 0.00 0.00 0.00

Positive Coefficients + = significant at 0.1 level, ++ = significant at 0.05 level, +++ = significant at 0.01 level Negative Coefficients - = significant at 0.1 level, -- = significant at 0.05 level, --= significant at 0.01 level I A G T S P V R SN Constant +++ ++ No +++ +++ +++ +++ --no --+ No No ++ +++ No Skip -+++ No +++ ----+++ +++ +++ Skip --No No +++ ++ No +++ +++ +++ ++ --++ --+++ -No +++ +++ No +++ --+++ No +++ No No +++ ++ No +++ --No --No No +++ +++ +++ ++ --++ No

404

Kenya Mozambique Namibia Nigeria* Rwanda Senegal South Africa Tanzania Uganda Zambia*

0.258 0.214 0.224 0.172 0.244 0.215 0.171 0.172 0.357 0.2894

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

+++ +++ +++ +++ No ++ +++ +++ +++ +++

No No No -No No --+ + No

No +++ No No No --+++ No No

+++ +++ +++ +++ Skip +++ +++ Skip +++ Skip

+++ +++ +++ +++ +++ +++ +++ +++ ++ +++

.++ +++ No No +++ ++ No +++ No +++

+++ No No + Skip ++ No + +++ +++

No ----No ----------Ski p

++ No No ++ No No No +++ No +++

----No No --No No -------

TABLE 7 WILLINGNESS AND ABILITY TO PAY

Average willingness and ability to pay for a mobile handset in US$ 7.45 19.14 8.92 15.11 29.70 6.06 14.02 16.98 4.00 24.64 5.65 19.55 17.44 10.89 17.42 7.45

Benin Botswana Burkina Faso Cameroon Côte d’Ivoire

Ethiopia Ghana Kenya Mozambique Namibia Nigeria* Senegal South Africa Tanzania Zambia* Benin

Average expected cost of a mobile handset US$ 11.44 27.38 12.84 22.16 30.06 64.19 23.15 26.68 23.2 25.12 12.57 2543.0% 32.41 17.3 22.43 11.44

National New users at 20 US$ for an handset 124,972 119,014 427,032 864,053 3,057,420 1,436,628 1,283,271 2,857,406 56,457 162,992 356,907 1,336,691 2,549,812 1,422,927 682,864 124,972

New users at 15 US$ for an handset 487,176 196,496 1,242,397 1,728,316 3,539,351 1,637,668 1,469,652 4,160,498 79,895 192,395 1,004,573 2,169,548 3,251,782 2,102,510 1,061,607 487,176

New users at 10 US$ for an handset 677,715 228,203 1,451,446 1,865,876 3,914,283 2,644,673 1,841,837 5,658,430 287,147 232,584 2,527,884 2,301,775 3,991,768 3,272,065 1,598,555 677,715

TABLE 8: MOBILE EXPENDITURE - ROBUST REGRESSION RESULTS FOR EACH COUNTRY USING SAMPLING WEIGHTS (SOURCE: RIA 2007/2008 HOUSEHOLD SURVEY)

Country

R2

Prob >F

Benin Botswana Burkina Faso Cameroon Côte d'Ivoire Ethiopia Ghana Kenya Mozambique Namibia Nigeria Rwanda Senegal South Africa Tanzania Uganda Zambia

0.399 0.298 0.206 0.22 0.196 0.353 0.145 0.251 0.189 0.449 0.458 0.339 0.266 0.477 0.127 0.292 0.378

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Only significant coefficients are being displayed: * = significant at 0.1 level, ** = significant at 0.05 level, *** = significant at 0.01 level Y A G WP PP F R SN 0.35*** 1.07*** -0.27*** 0.46*** -0.26** 0.33*** 0.28* 0.35*** 0.27*** -0.32** -0.22* 0.54* -0.4* 0.36*** -0.02* 0.45** 0.37* 0.34* 0.29*** -0.01** -0.47*** 0.23* 0.47*** -0.01** 0.33*** -0.02*** 0.44*** 0.47*** 0.11** -0.03*** -0.72*** 1.04*** 0.43*** -0.23* 0.37** 0.71*** 0.24*** -0.41* 0.24*** 0.55*** 0.66*** 0.43*** 0.247* 0.47*** -0.02*** 0.2** -0.2*** 0.3*** 0.33*** 0.84** 0.39 *** 0.48** -0.24* 0.43** 0.52*** -0.01** 0.69***

405

TABLE 9: MOBILE PENETRATION (SOURCE: RIA 2007/2008 HOUSEHOLD SURVEY)

Benin Botswana Burkina Faso Cameroon Côte d'Ivoire Ethiopia Ghana Kenya Mozambique Namibia Nigeria* Rwanda Senegal South Africa Tanzania Uganda Zambia*

Monthly average mobile expenditu re in US$ 8.33 10.18 5.84 7.14 12.52 3.81 10.44 10.41 6.26 11.41 10.88 6.02 11.00 15.88 7.44 5.75 10.55

Current market in US$ million 11.38 6.67 10.77 21.29 63.13 5.29 78.23 112.11 30.47 7.14 686.54 3.13 27.54 320.49 30.79 16.81 25.96

16+ with mobile phone or active SIM 1,365,851 30.2% 654,737 59.5% 1,844,701 27.2% 2,979,597 36.5% 5,042,524 41.8% 1,387,910 3.2% 7,491,378 59.8% 10,772,696 52.0% 4,865,758 25.7% 625,707 49.3% 63,101,014 77.3% 520,259 9.9% 2,502,300 39.8% 20,185,135 62.1% 4,138,338 21.5% 2,924,095 20.7% 2,459,961 45.5%

16+ with more than one SIM card 496,917 61,670 380,945 240,473 762,295 8,379 832,341 2,796,971 143,404 39,090 12,265,752 16,170 125,251 2,200,647 602,730 526,378 110,683

Number of duplicated SIM Cards 1,173,454 129,323 943,819 600,756 1,741,585 17,282 1,969,676 5,932,015 286,808 83,807 26,381,512 32,340 291,243 4,845,907 1,301,997 1,097,654 253,279

Total Number of active SIM cards 2,047,486 722,390 2,408,616 3,440,472 6,037,870 1,410,159 8,691,409 13,984,190 5,012,287 670,424 77,954,949 536,429 2,705,744 22,938,052 5,070,790 3,505,813 2,605,368

Share of Prepaid users 95.96% 99.28% 96.69% 88.04% 91.77% 88.31% 99.83% 98.89% 98.89% 89.95% 99.33% 94.77% 99.70% 78.64% 96.95% 97.84% 99.7%

TABLE 10: MOBILE WILLINGNESS AND ABILITY TO PAY (SOURCE: RIA 2007/2008 HOUSEHOLD SURVEY)

Benin Botswana Burkina Faso Cameroon Côte d'Ivoire Ethiopia Ghana Kenya Mozambique Namibia Nigeria* Rwanda Senegal South Africa Tanzania Uganda Zambia*

16+ without mobile phone or active SIM card 69.8% 3,162,099 40.5% 446,140

Number of 16+ without a mobile phone or active SIM willing and able to spend monthly:

Average monthly WTP in US$

1US$ or more 2,797,101 311,446

2 US$ or more 1,402,507 199,511

5 US$ or more 161,217 66,192

2.94 4.28

Monthly untapped market in US$ million 8.26 1.47

72.8% 63.5% 58.2% 96.8% 40.2% 48.0% 74.3% 50.7% 22.7% 90.1% 60.2% 37.9% 78.5% 79.3%

4,929,897

4,371,694

1,875,892

430,952

3.13

5,177,393 7,033,592 42,497,353 5,036,815 9,941,748 14,078,222 644,056 18,541,687 4,735,492 3,779,221 12,331,758 15,066,652 11,174,801

3,452,460 4,485,498 10,231,145 2,849,435 5,866,299 1,407,840 275,364 7,989,151

1,855,275 3,645,855 3,104,395 1,953,135 5,235,785 1,199,765 247,254 6,234,941

550,724 1,677,528 74,428 984,279 1,245,083 376,037 71,171 5,128,000

3.75 6.86 1.53 9.34 3.30 2.96 4.88 6.09

13.71 13.14 31.44 25.68 38.40 25.69 6.70 1.35 65.25

3,428,481 7,604,512 5,560,959

1,294,681 5,551,777 4,750,935

502,730 2,209,625 1,064,087

3.28 4.34 2.61

11.33 36.27 21.42

54%

2,944,732

1,933,833

1,310,576

329,922

3.45

8.2

406

Numeric Paper Forms for NGOs Gursharan Singh, Leah Findlater, Kentaro Toyama, Scott Helmer, Rikin Gandhi, Ravin Balakrishnan

Abstract—Non-governmental organizations (NGOs) working in disadvantaged communities have a variety of data-collection and analysis needs, for example, for performing surveys or monitoring programs. Because much of this data collection occurs in environments with insufficient IT support and infrastructure, and among populations not always comfortable with technology, paper forms rather than electronic methods remain the predominant means for data collection. We consider the design of machine-readable paper forms for NGOs. We first examine the unique needs of NGOs that interact with underprivileged populations through interviews with eleven organizations and an in-depth investigation of one NGO’s specific form-filling requirements. These explorations led to a focus on numeric forms – forms with questions requiring responses largely constrained to numbers. We then present an experiment which evaluates how a variety of formats for numeric data would fare with users from backgrounds similar to those who might fill out such forms. Our goal was to balance the tradeoff between ease-of-use among our intended population and machine readability. Combining the results of the experiment with an analysis of machine-readability from a technical perspective, we propose the best numeric input methods for different NGO form filling requirements.

Index Terms— machine-readable forms, paper forms, input methods, ICT for development I. INTRODUCTION Monitoring, evaluation, measurement, and selfassessment are among several critical tasks for non-profit organizations working in global development. Knowing the nature of one’s impact, ideally with accompanying quantitative information, allows for self-correction, reports to sponsors and potential donors, and external influence. Most program assessment requires data collection in some form as a first step. Data collection can be tedious and expensive (in labor, time, and financial cost), and it is thus desirable to extract as much value from the effort as Manuscript received September 22, 2008. Revised February 20, 2009. G. Singh was with Microsoft Research India, Bangalore, Karnataka, India (email: [email protected]). L. Findlater is with University of British Columbia, Vancouver, BC, Canada (email: [email protected]). K. Toyama is with Microsoft Research India, Bangalore, Karnataka, India (email: [email protected]). S. Helmer is with University of British Columbia, Vancouver, BC, Canada (email: [email protected]). R. Gandhi is with Microsoft Research India, Bangalore, Karnataka, India (email: [email protected]). R. Balakrishnan is with University of Toronto, Ontario, Canada (email: [email protected]).

possible. In this regard, information technologies frequently play an important role in the storage, analysis, and display of painstakingly collected data. Microfinance institutions benefit from back-end databases that store client data [22]; healthcare institutions have need to maintain medical records; and NGOs in general can benefit from longitudinal data collected over the life of their programs. Organizations have also considered the use of information technology for the task of collecting the data itself. Microfinance accounting information systems [28] and healthcare information on PDAs [2] are two significant examples. The expectation is that these efforts minimize transcription of data from paper forms and allow for more rapid analysis. There are, however, many who question the value of electronic means for data collection in poor environments. One cost-benefit analysis of mobile devices used by microfinance institutions to interact with their clients suggests that the benefits of electronic technology for data collection do not always outweigh the costs [9]. Others point out that electronic mechanisms are distrusted by populations who are used to physical evidence of transactions [23]. Paper, on the other hand, is a ubiquitous, low-cost, and well-understood medium. Even in the developed world, paper forms remain widely used for the purposes of gathering information, despite ongoing advances in digital technology. However, while it may be easy to collect data on paper, transferring that data into a format suitable for subsequent computer based storage and analysis remains a difficult problem that is currently typically resolved only by tedious manual data transcription. In this paper, we first study how paper forms are used for data collection in the context of non-profit nongovernmental organizations (NGOs) that seek to monitor their own programs. We performed interviews with eleven development-focused organizations involved in healthcare, microfinance, education, and agriculture. We then probed deeper with one of the organizations to better understand their data-collection pipeline. This investigation resulted in a proposal for data collection that uses paper for the “frontend” collection tasks, and subsequently uses a combination of scanner and PC to digitize the data. Rather than attempt to handle the myriad of possible paper forms right away, we instead focused on an important subset – forms used for collecting numeric input, which was found to be frequently used by NGOs. Although many additional factors besides input type will determine the value of such a system in a real environment, our study provides a first step towards building an effective system. We conducted a study of how

407

people with varying educational backgrounds responded to form-filling tasks using several different “input methods” for recording numbers. Finally we analyzed the literature on machine-readable forms, and gauged how each of the input methods would fare for subsequent automatic digitization of the data. II. RELATED WORK We review four categories of research on which our work builds. First, a number of electronic devices and software tools have been designed to complement paper in information work. Second, some user studies have compared digital and paper-based methods of data entry. Third, some research has focused on using paper as an interface for systems that interact with underserved populations. And, finally, existing work on machine-readable forms will be reviewed in a separate section. A. Complementing Paper The importance of paper as a medium to capture data is recognized, even in technology-heavy developed-world environments: “Rather than pursue the ideal of the paperless systems, we should work toward a future in which paper and electronic document tools work in concert and organizational processes make optimal use of both” [25]. And, indeed, many tools have addressed this need. Early research by Johnson et al. [13] showed how paper documents could be stored, indexed, distributed and processed by capturing them as images with scanners. Wellner’s [29] classic Digital Desk system used a camera and a projector to record input and project documents. Guimbretière’s PADD [12] attempts to bridge the gap between the paper and the digital world. Built on top of PADD, Liao et al.’s [17] Papiercraft implemented a gesture interface for natural editing. In a lighter-weight approach, the Anoto pen [1] uses paper printed with unique marks and a small camera in the pen to synchronize changes on the paper to a digital version of the document. Arai et al. [3] show how a camera-assisted pen can capture words on paper by hyper-linking them to other content. Koike et al. [15] used projected Venn diagrams to record input. Although these systems combine the strengths of paper and electronics, their benefits come with additional costs, which is a concern for NGOs. B. Digital versus Paper Some researchers have compared paper and digital systems for data entry, and overall tend slightly in favor of paper. Galliher et al. [8] found that people are more likely to complete paper forms than digital ones. They cite technical difficulties, as well as stolen or lost devices. On the other hand, errors of omission were more common on paper. Shelby-James et al. [26] disprove an often-made claim that electronic data capture is more accurate than paper-based methods. They found error rates with handheld computers were over sixty times that for paper-based data entry.

C. Data Collection for the Developing World In spite of the trend to place digital technology directly in front of underserved populations, a few researchers have focused on using paper instead. Parikh et al. [24] systematize paper-based data collection by using paper user interfaces and automated forms processing for microfinance institutions. Mackay et al. [20, 21] propose using paper on top of tablet PC displays. Parikh et al.’s [23] CAM system demonstrates use of a camera-equipped mobile phone to read bar codes printed on paper forms. Bar codes cause the phone to issue audio annotations, which instruct the user to transfer data on the paper form into the mobile phone. DeRenzi et al. [6] use a PDA based system to significantly increase adherence to medical protocols for pediatric health care. These systems all incur a device cost for the person doing the data collection. Ellison et al. [7] stress the importance of participatory monitoring and evaluation for a small scale NGO. Chand et al.’s [5] Jadoo, is a paper-only exploration of how structured content on paper can be made easy to deal with even for people with little formal education. Our early investigations agree with this work on the value of paper, and its likely continued use by NGOs. Our research builds on the above work and focuses on the novel problem of designing paper forms which should, on the one hand, allow easy machine readability, and on the other hand, be easy to fill in for people with diverse educational backgrounds. III. NGOS AND FORMS Over the last several years, we have both formally and informally interviewed eleven organizations – five NGOs, four microfinance institutions, and two clinics – in India with respect to their data-collection needs and experiences with both digital and paper-based forms. These organizations varied significantly in size and in their use of forms (from requiring a handful of forms filled in per month, to hundreds of thousands per year), but all had in common that they had data-collection needs and continued to utilize paper forms for this purpose, despite isolated trials with electronic data collection. Below, we outline what we learned from these organizations, and go into some detail about the datacollection system for one organization that typifies the NGO experience with forms. A. Overall NGO Experiences with Forms The need for data collection is widespread among NGOs, and means for doing it efficiently and subsequently being able to analyze the results easily are, in the words of one NGO, “the need of the hour” [14]. Despite the prevalence of paper forms, however, NGOs do not seem completely satisfied with their systems (or lack thereof) for incorporating collected data in their routines.

408

The goals of most data collection fall into one of several categories. For surveys, the intent is simply to understand the state of a population as it exists. For monitoring and evaluation of projects, the objective is accountability, both in terms of whether NGO staff and beneficiaries are performing tasks they have committed to, or whether the intended outcomes of a project are being accomplished. For baseline surveys, the collected data helps identify various parameters and tune efforts before work on a particular project is started by an NGO, for example, the majority occupation of the population in the area. Forms can be filled in by different people with different backgrounds, but for the most part, there seem to be two classes of people filling in forms: those who are paid by the NGO to perform data collection and those who are direct beneficiaries of the NGO’s programs. The former category consists of people who are almost always literate, and typically have completed secondary school; some have advanced degrees, as well. The latter range wider in their backgrounds, but frequently have not had much formal education. Anyone asked to fill in a form will likely be literate, but there are cases where the form-fillers can just barely sign their name and write numerical digits. Furthermore, their handwriting may not be very good. Forms involve recording of numerical data (e.g., age, price, income, dates, durations), textual data (e.g., name, free-text comments), and binary or multiple-choice questions. It is worth noting that a significant fraction of the questions asked on forms appear to be reducible to numeric information, with multiple-choice answers. The key exceptions are names and free-form comments. Also, in some cases, photographs were part of the data collected. Overall, the repeated concerns expressed by NGOs regarding data collection were: cost, time, training, data accuracy and consistency, storage, and means of data analysis. Among those NGOs who had experimented with electronic systems, these issues were again highlighted, in addition to difficulties with infrastructure and maintenance. We discuss each of these concerns briefly below.

of data collection. In another case, a bicycle courier was paid 15 cents per village to collect sheaves of paper records and bring them back to the NGO office.

Cost: NGOs are run on tight budgets, and many expressed concerns even about the cost of paper and printing. (Some of the NGOs we spoke with explicitly requested a ream of “white paper” as a meaningful gift in exchange for their time.) Digital equipment, of course, can be prohibitively expensive, and even those that can afford the capital costs have difficulty with costs of maintaining technology. One interesting point here is that centralized technology in an NGO office is easier to justify than devices that must scale in proportion to the number of respondents (or equivalently, data-collection staff). The cost of staff hours for data collection also adds up, although these costs match labor costs of the local area, and are typically very low by developed-country standards for hourly wages. Among the NGOs we consulted, data collecting staff were paid as little as US$4 for a 10-hour day

Data storage and analysis: Almost all of the NGOs we spoke with understood the value of digitizing data once it was collected. Many employed data-entry staff whose sole job was to take paper forms and convert them into digital format; others outsourced these tasks to transcription services; in some organizations volunteers handled this task. These methods are all costly – for some organizations, the costliest part of the data pipeline – or irregular, and depressingly often, we were shown stacks of paper forms that had yet to be processed; in some cases, they had been gathering dust for years. Analysis of data most frequently involved the use of spreadsheet software as well as tables in word-processing software filled in by hand. Two microfinance institutions used custom software, which produced fixed reports once the data was entered.

Infrastructure and maintenance: Among the organizations that had either experience with or expressed interest in information technology (IT) systems for data collection and processing, all cited challenges with power and maintenance issues. Fig. 1 shows a person filling out a form with light from a flashlight, a frequently encountered situation.

Fig. 1. A participant fills out a form during a power outage.

Training, data accuracy, and consistency: Data-collection tasks often require some training, and systems involving digital devices only add to this need. Training is a time- and management-intensive activity that NGOs invest heavily in. Ensuring data accuracy and consistency of routine surveys is another headache for many NGOs. Staff, who themselves are not paid particularly well, may have little motivation to perform data-collection tasks with care. In many cases, the data-collection staff itself needs to be monitored, to ensure that data is being correctly collected according to specification. Even dedicated staff, however, often embed form-filling tasks into what can otherwise be busy daily schedules. Thus, data accuracy and consistency can suffer.

409

Other issues: One thoughtful NGO head mentioned that there would be some value in systems that helped organizations develop good questionnaires for data collection. He felt that a lot of data collected by NGOs was done without a clear understanding of what information was most useful for evaluation purposes [14]. B. One Particular NGO’s Experience GREEN Foundation is a small NGO based in Bangalore, India, whose mission is to spread sustainable agricultural practices among rural farmers. With an annual budget of about US$100,000, they have 15 field staff tasked with covering an operational area of approximately 100 villages. Recently, they began a project that involves group sessions with farmers, where facilitators mediate discussion based on video content [11]. As part of the monitoring and evaluation of this project, both facilitators and farmers are requested to fill in paper forms. The program has been running for over a year, in each of 12 pilot villages, with typically three sessions per week. The first author visited these sites several times over a period of two months to better understand how the forms were used. Their case is particularly interesting for us, because it provides information about the two broad classes of formfillers: (1) two groups of paid staff of the NGO, all of whom are literate (senior staff with ten or more years of formal education, and junior staff, with at least eight years of education), and (2) smallholder farmers, who earned no more than $2 a day, and rarely had more than six years of education; many were all but illiterate.

videos, their interests, suggestions, and so on. The junior staff then summarizes feedback from every session onto a form, shown in Fig. 2. These per-session forms are analyzed by the senior staff weekly, and finally aggregated onto a monthly form. These forms are currently filled in plain text in Kannada (the local language in the region). The filled forms are then sent to the regional office. The data is entered into a database by a human transcriber and is then analyzed to spot trends and results in the villages of operation. We observed that the form-filling sessions in villages tend to become interactive “classroom sessions” (Fig. 3) leading to vital exchange of information between the NGO staff and the participants. Forms filled by the farmers convey what they want to be taught, and forms filled by on-field staff help document their staff progress. Overall, our key findings were that: (1) data collection and form-filling are important activities for many NGOs; (2) cost and ease-of-use are major concerns, often preventing technology-heavy systems; and (3) digitized data is desired, but digitizing data is the bottleneck for data-collection efforts. These findings confirm findings from earlier work and additionally identify an important problem faced by many NGOs.

Fig. 3. The verandah of a house used for meetings and data collection.

IV. PROPOSED DATA COLLECTION PROCESS A. Proposed System

Fig. 2. A GREEN form filled by the junior staff.

After every video session, the farmers attending the session are asked questions related to the practices shown in

Based on our findings from NGO interactions, we believe that the following combination of paper forms and hardware can solve many of the data-collection challenges outlined above, while making a good tradeoff among the constraints. For hardware, one PC, a scanner, and a printer is required. We then envision that specially designed paper forms will be printed and used in the field to collect data. Once completed, the forms are scanned (a scanner with an autofeeding mechanism would be ideal), and special software digitizes the filled-in content. To do this, some software innovation is required to allow those creating the forms to easily design customized, machine-readable forms. In addition, there must be software that can robustly digitize any content that is scanned. Due to the limitations of technology for digitizing freeform handwriting, the expectation is that any given form

410

will be implemented using multiple-choice or numeric responses as much as possible, because these responses are comparably easy to digitize. Free-form text will be given space within blank rectangles on the form, which can be scanned and stored as images (and not converted to electronic text). The proposed system keeps costs low. No per-staff device is required, and all of the equipment is available as massmarket off-the-shelf hardware, which helps to keep costs low and alleviates technical maintenance needs. Since NGOs that collect data would like to have it in digital form eventually, the willingness to invest and maintain at least one PC is assumed (many will already have a PC). Printers and scanners add costs that are small compared with the PC itself, and paper is the only additional ongoing cost. The system would maintain all of the advantages of paper, namely its low cost, a well-understood “interface”, the lack of need for power or maintenance, and robustness in the field. Finally, the proposed system addresses the data-entry bottleneck by providing an automated means of digitizing much, if not all, of the data. This proposal still leaves us with the following challenge: on the one hand, existing techniques for machine reading of hand-marked forms is reliable only when the forms are designed and filled out in a particular way (e.g., “bubbles” on standardized test forms) or when digits are neatly written (e.g., post-office automated digit-reading systems for sorting mail); on the other hand, the groups that we anticipate will fill in these forms are less familiar with standardized forms, and many have borderline penmanship. This is a non-trivial challenge that requires both engineering and interface design. In the remainder of this paper, we consider how best to handle numeric input on paper forms with the goal of achieving a reasonable tradeoff between ease of user comprehension, user accuracy, and machine readability. V. NUMERIC INPUT METHODS We considered 10 different methods for entering numbers on paper forms, shown in Table 1. These were various versions of numeric entry, as commonly found in forms, and ranging from those that were specifically designed for machine readability (e.g., bubbles), to those that were handwritten. Eight were marking-based input methods, where the user marked the desired number either by filling in a bubble, circling the number, drawing a checkmark inside a box, or ticking the number (Table 1a-h). The marking-based methods were tested in both coded and uncoded versions, as described below. We also tested 2 handwritten methods (Table 1i-j): digit per box, where each digit was written in a separate box, and digits in one box, where all digits were written in a single box. The un-coded marking methods provided a row for each digit, so, for example, a 2-digit number would require two rows of input (Table 1b shows the number “58” entered in

un-coded Bubbles). Coded marking methods, in contrast, only provided a single row for all digits (see Table 1a). The advantage of the un-coded methods is that they allow for a complete set of numbers: for example, with 4 rows users can enter any number from 1 to 1000. The coded marking reduces the set of numbers that can be input: for example, “58” and “85” would appear the same when entered as coded input and numbers with repeating digits, such as “22”, cannot be input at all. However, the advantage of coded marking is that it greatly reduces the visual complexity of the form and saves physical space, which in turn saves paper. Coded marking could be useful for situations where the numeric data is nominal rather than scalar. If the form is used to record attendance at village meetings, for example, each individual may have an identification code assigned to them. Since these are nominal values they could be assigned with the goal of entering them as coded input. TABLE I NUMERIC INPUT METHODS CONSIDERED. a.

Coded Bubbles

b. Un-coded Bubbles

c.

Coded Circles

d. Un-coded Circles

e.

Coded Checkbox

f.

Un-coded Checkbox

g.

Coded Ticks

h.

Un-coded Ticks

i.

Digit per box

j.

Digits in one box

VI. TECHNICAL ISSUES WITH MACHINE READABILITY OF NUMERIC INPUT Designing paper forms for automatic processing is a mature field in the developed world, with successful applications in mail sorting using postal codes (e.g., [27]), academic marking systems, and systems for processing

411

applications (e.g., [10]). Moreover, there are numerous commercial software packages available that require little more than a PC and a document scanner. The process itself involves two stages. The first is form registration, which aligns the document so that values can be extracted. The second, and more challenging task is extracting the values from particular locations in the document. These values are extracted using either optical mark recognition (OMR) or intelligent character recognition (ICR), depending on the input method. Processing forms in the developing world, however, adds numerous challenges. The cost and ease of deployment may increase since many commercial applications may lack support for local languages. Another challenge, based on our NGO interviews, is that since forms may be filled out in less than ideal conditions, they may be wrinkled or dirty, which can affect both the form registration and value extraction. Commercial software packages also generally assume that the end users (the individuals filling out the forms) are literate and have a certain level of education. These users are expected to have some level of competency in filling out, which is not always the case in the developing world. All of these challenges need to be weighed against the ease of use for different input methods. The un-coded and coded checkbox and bubble input methods could be processed using OMR. In general, this method of input is nearly 100% accurate, assuming the user has followed the instructions [4]. This rate is reported for systems requiring specialized scanners and forms, which is not applicable for most NGO use; systems likely to be used by NGOs will have less accuracy. In addition, in the developing world, users will likely be less familiar with such forms, thus resulting in more user errors for OMR processing (for more detail on errors see Results section, below). Moreover, if the forms are degraded because they have not been filled out or stored in antiseptic conditions, less specialized systems may be more likely to mistake smudges, wrinkles and dirt for marks. The circling and ticking input methods we tested would also fall under OMR, but these types of marks are not generally supported in commercial software. The reason for this is possibly that the recognition accuracy is poor given the variability in the location and form of the marks, especially when contrasted to checkboxes or bubbles. The digits in one box and digit per box input methods, on the other hand, require ICR processing, which is less accurate. ICR technology leans upon advances in machine learning that have allowed systems to learn how to discriminate characters from samples of handwriting, and is still an active area of research [18]. Currently, even the most accurate ICR systems are only capable of 98% accuracy [19], and this is the case for digits written by a more literate population than our target users. In the case of forms requiring recognition of characters of partly non-alphabetic systems, the accuracy is much worse. For example, for

Kannada, an Indian language of interest to the NGOs we surveyed, the recognition rate can be less than 90% [16] and as mentioned earlier may not be supported by commercial software. It should also be stressed that accuracy rates in ICR are stated for sanitized data sampled from a completely different distribution of people. It is certain that illiteracy, poor handwriting, and form degradation will reduce the accuracy and this adds uncertainty as to whether the technology is applicable in this setting. VII. EXPERIMENT Our goal in designing paper forms is to achieve a reasonable balance between ease of user input and machine readability. We thus also conducted a controlled study to test user performance, accuracy and preference with respect to the 10 different input methods. A. Participants We recruited 40 participants who ranged in age from 1750 years (M = 26.2). Their formal education varied from four years of schooling to undergraduate university level. Most of them spoke at least three of the following languages: Hindi, Telugu, Kannada, Sinhala, Tamil, Bengali, Konkani and English. Every participant could write in at least one language, although many had not held a writing implement since finishing school. A very basic vision test (reading a series of numbers aloud) was used to screen participants before starting the experiment. The participants came from disparate professional backgrounds: auto-rickshaw (three-wheeled mini-cab) drivers, farmers, restaurant workers, security guards, housekeeping personnel, cab drivers, army soldiers, and machine operators. Participants’ literacy levels, ages, genders, and occupations were recorded. Participants were compensated with a small gift for their time. B. Task The task was to enter 20 numbers on a paper form, with five each of the following: 1-digit numbers, 2-digit numbers, 4-digit numbers and 8-digit numbers. Numbers to be entered were displayed on the forms themselves. A sample form for the digit per box input method is shown in Fig. 4. The numbers included in the tasks were randomly generated with no digit being duplicated in a number (for the sake of the coded input methods) and every participant saw the same set of numbers (though not in the same order). No participant entered the same set of numbers for more than one type of input method. C. Experimental Design A single factor within-participant design was used: each participant completed the experimental task with 10 forms, one for each input method. Presentation order was counterbalanced using a Latin square design and participants were randomly assigned to an order.

412

Finally, at the end of the study a short interview was conducted to compare the experience of using different input methods.

Fig. 5. A participant in a village filling a form on floor – locally known as the ‘free desk’.

VIII. RESULTS

Fig. 4. Sample digit per box form. The numbers above each set of boxes are the numbers to be entered for the experimental task.

D. Measures We measured speed and accuracy, and asked participants to provide preference feedback. Speed was measured as the time to complete the entire task for each input method and was measured manually with a stopwatch. Errors were counted after the forms were complete. A mark on a number that logically should not be there, or would obviously make a machine register the input incorrectly was counted as an error. Subjective feedback was collected for each input method using 5-point Likert scales on difficulty, confusion, and perceived speed. E. Procedure Experimental sessions took 45–80 minutes per participant, as they were all given as much time needed to complete or quit the task. Most of the participants were from villages in north and south Karnataka and we chose a quiet, isolated environment in which to conduct the sessions in each of these villages (Fig. 5). The remaining sessions were conducted in an office environment, where participants were called in. We ensured that there was sufficient lighting in both locations. Participants were first given a background questionnaire to collect demographic information. Then, for each of the 10 input methods, participants were given time to examine the form and ask questions about the task. Once they were comfortable with the nature of the task, they were asked to begin. Times were recorded for each of the 4 subtasks, namely, 1-digit numbers, 2-digit numbers, 4-digit numbers and 8-digit numbers. After each input method, a questionnaire was used to collect subjective feedback.

Participants were divided into two educational groups: those with up to middle-school education (7 years or less, M = 6.1), and those with more (M = 11.9). A 2x10x10 (educational group x input method x presentation order) repeated measures (RM) ANOVA on the main dependent variable of speed showed no significant main or interaction effects of presentation order, so we simplify our analysis by only examining the effects of education group and input method (2x10 RM ANOVAs). All pair-wise comparisons were protected against Type I error using a Bonferroni adjustment. Where df is not an integer, a GreenhouseGeisser adjustment for non-spherical data has been applied. One outlier in each of the low and high education groups were excluded because they were more than two standard deviations away from the mean on the dependent variable of speed. Thus, we report on data from 38 participants (10 in the low education block and 28 in the high education block). A. Speed Average time to complete the forms with each input method is shown in Fig. 6. Participants in the lower education group took 31.8 minutes on average to complete all forms, while the higher education group took only 21.1 minutes on average. Note that we ran the RM ANOVA on a log transform of the speed data since the original speed data violated the homogeneity of variance assumption (significant Levene’s test). Both the education level of participants and the input method significantly affected the time it took to enter numbers into the forms (main effect of education block: F1,36 = 23.3, p < .001, η2 = .393; main effect of input method: F4.99,180 = 58.8, p < .001, η2 = .620). More interestingly, some input methods were relatively better for the higher education group than for the lower education group (interaction effect between input method

413

and education: F4.99,180 = 2.37, p = .041, η2 = .062). To understand which input methods were better within each group of participants, we performed pair-wise comparisons and summarize the significant results (p < .05) as follows.

problematic for machine readability, as discussed in Section VI. We counted three types of errors: (1) when the number of digits in an entered number was large (four or eight) participants often found it hard to match the positions of digits and rows in un-coded versions (Fig. 8a) and marked two numbers in the same row; (2) ill-formed numbers and numbers flowing out of boxes were also problematic (Fig. 8b-c); (3) sometimes it was difficult to distinguish which number had been marked on the control (Fig. 8d).

Fig. 6. Task completion time by education and input method. (N = 38)

1) The handwritten input methods were generally fast, and this result was clearest with the higher education group. These two methods were not found to be any different from each other in terms of speed. For the higher education group both handwritten methods were faster than all other methods. The lower education group, on the other hand, was not as relatively fast with digit per box: it was not found to be different from several of the coded input methods (circles, checkbox and ticks). 2) The bubble input methods were the slowest for the higher education group, but did not have as large a negative impact on the lower education group. For the higher education group, both un-coded and coded bubbles were slower than all other input methods with only one exception (coded bubbles and coded checkbox were no different from each other). For the lower education group, however, the bubble methods were not found to be significantly slower than any of the other 6 types of marking-based input methods. 3) Coded input methods were also generally faster than their un-coded counterparts, suggesting that the added complexity of having several lines of input instead of one increased the difficulty of the task. For all coded versus un-coded comparisons of ticking, checkboxes and circling in both education groups, the coded versions were faster with one exception (coded ticking for the higher education group was not faster than uncoded ticking). B. Accuracy Error rates were uniformly low; on average, less than 1 error was made per form (Fig. 7). Fig. 8 shows examples of entries that were counted as errors and a variety of additional anomalies (not counted as errors) that could be

Fig. 7. Average error rate by education and input method. (N = 38)

There was a significant main effect of input method on the number of errors (F2.38,85.5 = 3.02, p = .046, η2 = .077), but using a Bonferroni adjustment indicated that no pairwise comparisons were significant. No main or interaction effects of education group were observed. C. Subjective Measures Difficulty: Participants generally found all methods easy to use: on a 5-point scale from easy to difficult the average overall rating was 2.0 (SD = 0.8). Some methods were felt to be more difficult (a 2-way RM ANOVA showed a main effect of input method: F5.12,184 = 3.94, p = .002, η2 = .099). Pair-wise comparisons show that the digits in one box method was less difficult than un-coded bubbles and uncoded ticking. No other significant main or interaction effects were found. Confusion: Participants generally claimed not to find the input methods to be confusing: the average rating on a 5point scale was 4.2 (SD = 1.8), where 5 represented “not at all confusing”. A 2-way RM ANOVA revealed that some input methods were more confusing than others, but pairwise comparisons showed that the only statistically significant comparison was that numbers in a box was less confusing than un-coded ticking (p < .013). Though not statistically significant, digits in one box received the best rating from participants (M = 4.8, SD = 0.6). No other significant main or interaction effects were found. Perceived speed: In general, perceived speed matched actual speed. There was a main effect of input method on perceived speed (F6.22,224 = 5.17, p < .001, η2 = .126). Pair-

414

a. Numbers filled in same row in an un-coded input method

b. Ill-formed numbers

c. Numbers crossing boundaries

d. Ambiguous mark

e. Impatiently filled bubbles

f. Digits indexed on the number

g. Darkening on the number in bubble

h. Overlapping circles

Checkmark extending out of box

j. Double ticks

k. Long, overlapping ticks Fig. 8. Examples of errors (a-d) and other anomalies.

l. Light numbers

wise comparisons showed that both the handwritten types (digit per box and digits in one box) were perceived to be faster than coded bubbles (p < .05). Digits in one box was perceived to be the fastest, since participants also felt it was faster than two of the un-coded input methods (bubbles and checkbox). Participants in the lower education group rated the tasks as slower than those in the higher education group, a result that matched actual speed (main effect of education block on perceived speed: F1,36 = 7.29, p = .010, η2 = .168). No significant interaction effect was observed between input method and education group. D. Qualitative Observations We made a number of observations that, though not directly the focus of the study, illustrate practical issues that may be encountered with paper-based forms in this context. Several of the less educated participants expressed great enthusiasm with the study. They savored holding a pencil since they had some formal education but their jobs never required them to write or read anything. Other participants were more apprehensive about holding a pencil because they did not know how to write, but they agreed after being encouraged by earlier participants. A few participants without computer backgrounds felt a natural inclination to add leading zeros where there were unused “slots”. For example, in a preliminary investigation prior to the formal tests, they marked the month of February in the date field as “02”, filling both the input columns, instead of only “2” and leaving a column blank. Some circumstances of rural life affect the form-filling task in unexpected ways. In one case, we experienced a power outage during a preliminary field experiment (again prior to our formal trials). One of the participants, who was over 50 years of age, found it particularly difficult to continue, although his vision was fine in bright light.

One participant commented that she depended on her husband for filling out textual information since she did not know how to write text. However, she was comfortable with numbers, and felt that if forms only required numeric entry, she could do it herself. E. Summary of Results In terms of speed, degree of formal education impacted the effectiveness of individual input methods. In particular, both un-coded and coded bubbles were the slowest input methods for the higher education group but were no slower than any other input method for the lower education group. We had anticipated that the handwritten methods (digits in one box and digit per box) would be relatively faster for the higher education participants than the lower education ones, but both groups of participants, it turned out, were faster with these than with most other input methods. Coded input methods were also generally faster than un-coded methods and resulted in fewer errors. As expected, higher education participants completed the task more quickly than lower education participants. Note that since the lower education group had one third as many participants as the higher education group, there was less statistical power to detect differences among the input methods; this could explain why some pair-wise comparisons were significant for the higher education block but not for the lower education block. IX. DISCUSSION Combining our understanding of the state of the art for machine-readability of numeric forms with the user study, we find that we can make the following recommendations for typical NGO conditions. At the highest level of granularity, if all of speed, machine accuracy and user accuracy are desirable characteristics, coded checkboxes seem to be the best fit

415

irrespective of education level of users. If the users are literate, digits in one box might be a better choice as it was the fastest method for the more highly educated group, with low error rate (zero for our participants), and if carefully written these digits have reasonable machine readability characteristics. The relative performance differences between the higher education and lower education groups suggest that there may be a tradeoff between the benefits of structure and simplicity for different levels of education. The nonhandwritten methods may be useful for low literacy participants since they do not require the same writing proficiency as the techniques that require handwritten numerals (digits in one box and digit per box). Our lower education participants had on average 6 years of schooling. In our experience from interviewing NGOs, we found that some users of their forms may have even less education, which could magnify the relative performance differences we found based on education. This highlights the need to consider the educational background of the target user population. Although coded methods were in general quicker to fill than their un-coded counterparts, and also resulted in fewer errors by participants, they are not necessarily suitable for all types of forms. In particular, coded methods cannot handle numbers with repeating digits. As such, if a form has some fields that require entry of such numbers, it might be best for the sake of consistency to stick to an un-coded method throughout the entire form. Interestingly, participants rated handwritten numbers best overall. Both digits in one box and digit per box methods were reported to be less confusing and easy to fill. In contrast, the bubble methods were not particularly favored, and were the slowest. Taken as a whole, our results indicate that the proposed system of using structured paper forms with subsequent automated scanning is likely viable. However, the results also caution that there are nuanced tradeoffs that need to be made when choosing the type of input method to use. The fastest and most preferred techniques (handwritten numerals) are generally the least accurately recognized by a machine, although the digit per box method is arguably a reasonable compromise in this regard since the box provides some structure that eases the task of the recognition algorithm. X. CONCLUSION AND FUTURE WORK We have explored the problem space of data collection by NGOs in underserved communities where resources and technological literacy are typically scarce. In particular, we focused on how a particular class of paper forms – those that require only numeric data entry –might be best structured to support subsequent automated data transfer to enable computerized storage and processing. The main contributions of our work are: (1) an interview-based

analysis of challenges faced by NGOs around data collection and form filling, (2) a proposal for a semi-automated system that uses paper as the interface for data collection, and a scanner and PC for digitization, (3) an empirical study with 40 participants of how 10 different number-based input methods fare among people with both lower and higher educational backgrounds, and (4) an analysis of how easily the forms filled using each of these input methods can be automatically digitized. Our main findings are that users tend to prefer and perform best with techniques that require handwritten numbers. However, these are not the most easily recognized by a machine. If multiple factors such as speed and accuracy of entry, and machine readability are taken into account, coded checkboxes are likely the best option, although choosing a coded method limits the type of numbers that can be captured (e.g., numbers with repeating digits are not easily handled by the coded techniques we considered). Another key finding is that the educational background of the user can significantly impact performance with different input methods, and as such the demographics of the target user population should be carefully considered when making decisions as to the type of input method to use. In our study, participants used the same input method across each form. Given the tradeoffs we found with respect to coded and un-coded methods, it would be interesting to probe further into how people might perform with forms that have a mixture of input methods, each optimized for the types of questions being asked on the form. It is unclear if having more than one input method on a form would be overly confusing to users, particularly those with limited levels of literacy. Finally, there are many other challenges that will need to be addressed in building an effective system to collect and process data using paper forms as a front-end, and automated input recognition as a back-end. Different types of data can pose different problems. Our work with only numeric data is a first step in this direction. It would be interesting to actually build a system that handles the processing of forms in real use, and to study its performance in the field. We intend to explore this in the near future with a partner NGO. REFERENCES 1. Anoto, http://www.anoto.com 2. AED Satellife, http://www.healthnet.org 3. Arai, T., Aust D., and Hudson S.E. (1997). PaperLink: a technique for hyperlinking from real paper to electronic content. Proc. CHI ’97, p. 327-334. 4. Bergeron, Bryan P. (1998, Aug.). Optical mark recognition. Postgraduate Medicine. 104(2). 5. Chand, A., Dey, A.K. (2006). Jadoo: a paper user interface for users unfamiliar with computers. Proc. CHI 2006. p. 1625-1630. 6. DeRenzi, B., Lesh, N., Parikh, T., Sims, C., Mitchell, M., Maokola, W., Chemba, M., Hamisi, Y.,

416

Schellenberg, D., Borriello, G. (2008). e-IMCI: Improving Pediatric Health Care in Low-Income communities. Proc. CHI 2008. p. 753-762. 7. Ellison, M. Firm foundations: Improving NGO data collection, http://www.id21.org/id21ext/s7bns1g1.html 8. Galliher, J.M., Stewart, T.V., Pathak, P.K., Werner, J.J., Dickinson, L.M., Hickner, J.M. (2008). Data Collection Outcomes Comparing Paper Forms With PDA Forms in an Office-Based Patient Survey. Annals of family medicine. 6. p. 154-160. 9. Gogineni, M., Ratan, A.L. Evaluating the viability of a mobile phone-based, SMS/GPRS-enabled, client data collection channel for urban microfinance. Microsoft Research India Technical Report. 10. Gravic, http://www.gravic.com/remark/remarkcustomers.html 11. Green Foundation, http://www.greenconserve.com 12. Guimbretière, F. (2003). Paper Augmented Digital Documents. Proc. UIST 2003. p. 51-60. 13. Johnson, W., Jellinek, H., Klotz, L., Rao, R., Card, S.K. (1993). Bridging the paper and electronic worlds: the paper user interface. Proc. CHI ‘93. p. 507-512. 14. Khan, Aqueel. ASKIndia, private communication, 2008. 15. Koike, H., Sato Y., Kobayashi Y., Tobita H., and Kobayashi M. (2000). Interactive textbook and interactive Venn diagram: natural and intuitive interfaces on augmented desk system. Proc. CHI 2000. p. 121-128. 16. Kunte, S.R., Samual, R.D.S. (2007). An OCR System for Printed Kannada Text Using Two - Stage Multi-network Classification Approach Employing Wavelet Features. Proc. ICCIMA 2007. p. 349-353. 17. Liao, C., Guimbretière, F., Hinckley, Hollan, J. (2008). Papiercraft: A gesture-based command system for interactive paper. ACM Transactions on ComputerHuman Interaction. 14(4). p. 1-27. 18. Liu, C., Fujisawa, H. (2008). Classification and Learning Methods for Character Recognition: Advances and

Remaining Problems. Studies in Computational Intelligence, SpringerLink. 19. LeCun, Y., Bengio, Y., Bottou, L.and Haffner, P. (1998). Gradient-Based Learning Applied to Document Recognition. Proc. IEEE. 86(12). p. 2279-2324. 20. Mackay, W.E., Pagani D.S., Faber L., Inwood B., Launiainen P., Brenta L., and Pouzol V. (1995). Ariel: augmenting paper engineering drawings. Proc. CHI ‘95. p. 421-422. 21. Mackay, W.E., Pothier G., Letondal C., Bøegh K., and Sorensen H.E. (2002). The missing link: augmenting biology laboratory notebooks. Proc. UIST 2002. p. 4150. 22. Mifos, http://www.mifos.org. 23. Parikh, T.S, Javid, P., Sasikumar K., Dhosh, K., Toyama, K. (2006). Mobile phones and paper documents: evaluating a new approach for capturing microfinance data in rural India. Proc. CHI 2006. p. 551560. 24. Parikh, T., Aditya, V.P.S, Vellayutham, M. (2003). Automated Forms Processing and Paper User Interfaces for Data Collection from Village Microfinance Groups. Unpublished draft, http://people.ischool.berkeley.edu/~parikh/papers/itira20 03.pdf 25. Sellen, A.J., Harper, R.H.R. (2001). The Myth of paperless office. MIT Press. 26. Shelby-James, T.M., Abernethy, A.P., McAlindon, A., Currow, D.C. (2007). Handheld computers for data entry: High tech has its problems too. Trials. 8(5). http://trialsjournal.com/content/8/1/5 27. Siemens, http://www.industry.siemens.com/postal-automation/ 28. SKS Microfinance http://www.sksindia.com/technology.htm 29. Wellner, P. (1993). Interacting with paper on the DigitalDesk. Communications of the ACM. 36(7). p. 8796.

417

Rajnikant’s Laptop: Computers and Development in Popular Indian Cinema Joyojeet Pal parental consent and wisdom would typically be highlighted in the couple’s struggle. Abstract: The role of technology in development has seen an exceptional spike in interest in the past decade in academia and industry. India has been a key location in the study and practice of technology and development because of the centrality of the technology industry in its growth over the recent two decades. Examining the public discourse on technology in areas where computers are not ubiquitously used or available, we find that cinema is important in constructing how people in Indian villages and even cities perceive computers. Turning back to the films themselves, we find a strong aspirational discourse in the ways in which computers and technology-users are portrayed on Indian popular cinema. These in turn relate to how technology and development have been portrayed in cinematic discourse over generations, in cinema from around the world, from the earliest days of silent film. We find from interviews with users of technology in rural India that these images create strong ideas of value and legitimacy derived from using technology. Reflecting on this, we find that study of public discourse has been a critical missing piece of ICTD studies. A study of media imagery such as the one presented here is a small step in that discussion.

K

I. PROLOGUE

ARTHIK is a hormonal rich brat, Sakhti is a hardworking medical student from a poor family. The class-crossed coupledecide to marry, against the wishes of both sets of parents, and thus the plot thickens in Alai Payuthey, a 2000 blockbuster hit from south Indian filmmaker Mani Rathnam. This kind of face-off between youthful love and parental opposition to marriage, a veritable obsession of Indian cinema, is perhaps one of the oldest and most successful themes. Back in the old days, Karthik would probably have been reduced to begging office to office for a job wearing a tie, finding in his useless paper degrees a metaphorical foil for the oppressive market economy [1] and thereafter turn to a life of dubious ethical distinction [2, 3]. Sakhti meanwhile would probably sit home sacrificing square meals and running a bare-bones household with a sanctimonious smile [4]. A number of themes, including the dependence of the man on the system to earn an honest living, the helplessness of the woman outside of the home domain, and the importance of

Manuscript received February 20, 2009. Joyojeet Pal is a research associate at the Center for Information and Society at the University of Washington. Email: [email protected]

Instead, in Alai Payuthey, the couple turns to a new direction for its salvation – technology. Karthik starts a computer software company with his friends, eventually winning an outsourced contract from the US that fixes for good their financial troubles, rubbing in the process a few parental noses in the dirt on the gold-paved streets of south Indian cities. Karthik fails most stereotypes of cinematic occupational characterization. He is no idealistic teacher, nor upstanding cop, nor charismatic businessman. He is at best a lovable nerd, traveling on a motorbike to work daily with a laptop strapped to his back instead of a holster. In effect, he is the archetype of exactly what he isn’t in the movie – the perfect candidate for an arranged marriage. Before resigning myself to academia, I tried a hand at scriptwriting in India. The process for selling a script typically involves going to a movie producer for a ‘reading.’ The process is somewhat ad-hoc, and depending on one’s familiarity with the producer, it can involve a detailed description of every element of the screen or quite simply a very broad outline. In my case, I had to give very specific descriptions of each character, scenario, and outcome. While my stories were not entirely outlandish, producers would point out how certain characters needed tweaking, or deleting entirely, because the “public will not accept it.” While failing to sell scripts, I had a comparatively more successful doublelife as a researcher working with children in villages that had recently got computers. I had noticed over and over, that children and adults alike who had never seen computers before in their lives had built ideas about technology from popular cinema. At the end of one particularly disappointing script reading when all my characters and plotlines were deemed unsaleable, I saw a Tamil film, Vyapari [5]. The script on that one probably sold easier, the plot revolved around a protagonist driven by his admiration for Bill Gates which leads him to use technology to clone himself to make twice as much money. II. INTRODUCTION In the 1990s, information and communications technology had an extraordinary impact worldwide both in terms of economic growth and in social change through access. The concurrence of this technological ‘revolution’ with the changing economic and political conditions in various postsocialist nations created a renewed interest in the area of technology and development not just within academia [6], but within business and management strategy circles [7] and also

418

within a rapidly expanding international development practice. A new wave of scholarly research publishing on ICTD (Information and Communications Technology for Development) started in the late 1990s, with much work on digital inequality issues [8-11], on technology and sectoral development and macroeconomic change [12-14], on technology and urban change [15, 16] and a very sizable body of work on the ‘potential’ of technology [17-19]. A look through the history of research in ICTD shows that an inordinately large portion of research and practice in this field has been focused on India. The reasons for this are multifold, but the perceived role of the technology sector in India’s ‘leapfrog’, and the investment of several high-tech companies’ philanthropies as well as international agencies in the rural technology projects have both played important roles in this concentration. In the early years, much of the ICTD literature in India and the South Asian region in general, looked more into the role technology had in facilitating a leapfrog either for individuals, or for the economy as a whole; consequently, the tone of much work in this period was comparatively more enthusiastic. By the mid 2000s, much introspective work looked at the performance of ICTD projects in India and elsewhere of the 1990s, and the causes and outcomes of some of the interest in technology and development [20-23]. While a lot of such work was empirically grounded, most critical discussion in ICTD focused on the supply side – i.e. why certain projects or project-types gained centerstage and how they performed, while comparatively a lot less effort went into the demandside, i.e. determining why ICTD projects have come to find apparent or stated enthusiasm among users populations, and in what ways is this enthusiasm manifested in public discourse. We begin a discussion in the latter through an examination of representations of technology in popular Indian cinema, especially in the south of the country, where the technology sector has had a comparatively larger economic and social impact. III. RELATED WORK There is little if any specific work within the relatively new ICTD field on media representation of computers and its impacts on technology adoption. There are however a number of important areas of related work that are foundationally useful to this paper. A. ICTD Theory ICTD is still at a fairly young stage for theory-building work, and it is here that the multi-disciplinary nature of ICTD has served as a mixed-bag rather than as a clear advantage. Social theory work in ICTD benefited in its early days from Manuel Castells, a sociologist and planning theorist who formed some of the early foundations of ICTD [24, 25], and much work in the ‘information society’ studies [26-28]; and more recently (and importantly for this paper) from the

perspective of gender studies [29, 30]. The theory building process has also benefited from works that have taken a multidisciplinary approach, looking at ICTs from the perspective of one or a few specific existing bodies of social theory including innovation diffusion [31], social inclusion [32], technology artifacts [33], and stakeholder theory [34]. Another interesting area of theory work has been that of technology in developing regions, though not necessarily specifically about projects aimed at development, such work on transnationalism and technology usage of people from developing regions [35, 36] and ICT-related research oriented specifically towards technology use in identity formation and articulation [37-39]. B. Technology and Cinema Ideas of technology and society have dated back over a century to the fanciful silent short film Trip to the Moon [40], and the representation of technology in cinema has straddled the line between science fiction, and what may be, and technology and social readjustment. The latter, especially the idea of technology as transformative, came to centerstage in the 1920s and 30s, around the period and often theme of rapid industrialization. Scholars have been interested in two important and fairly consistent themes along technology and society in cinema – the first, the dystopian ideas of technology and urban living – either in the present or future as seen in Fritz Lang’s futuristic epic Metropolis set in a fractured 21st century or Charles Chaplin’s Modern Times [41], set in a dehumanized 20th century factory floor. These ideas of technology as mystical, all pervasive, and potentially harmful have been a consistent theme of cinema and literature throughout the decades of vast technological change around the world [42, 43]. The second early theme had a more positive view of technology, primarily industrial production, and bears its origins in the early Soviet cinema highlighting technology as a nationalist enterprise, a means of social and economic development [44] and later emulated several other nationalistic cinemas, and this connection of cinema with modernity has been an important theme of cinema around the world [45]. Both these themes have been prominent in Indian cinema, and the two manifestations of these have been in the struggle between traditionalism and modernism, one seen in other third world cinema as well [46, 47], and of course the use of technology in nation-building as the second theme [48]. So while technology in a more general sense was a consistent theme of cinema, computers themselves did not appear until well into the 1950s, and even then they tended to be fanciful clunky large panels (reasonably similar to what computers really looked like!) used in science fiction cinema. The first films to feature computers were science newsreels in the immediate postwar period, and in 1951, the first film with a rudimentary computer was When Worlds Collide featuring a computer in a war to keep earth safe from a renegade asteroid [49]. Several such films followed. The rising popularity of television in the US spurred the

419

appearance of computers on screen, starting with the 1962 scifi show The Jetsons [50], in scattered episodes of the spy caper serial The Avengers [51], and finally in 1966 with the hugely popular sci-fi series Star Trek [52]. This fantasy archetype is important, since we see this early imagination of computers reprise itself in contemporary Indian cinema. The next paradigm featured a cinematic imagination of computers guided by the popular conceptions of artificial intelligence – as typified by the man v/s machine face-off in 2001: A Space Odyssey [53]. But the individual user of computers remained by and large the scientist, as indeed was true for the world at large, until in the late 1970s and early 1980s, when kids in regular schools started getting access to computers. This triggered a spate of ‘hacker’ films around teenagers and youngsters using computers starting with the Canadian Hide and Seek about a kid who sets off a nuclear emergency by hacking into a mainframe [54]. But the ubiquitization of computers on screen, where computers became everyday items, happened only around the mid-1990s, with a huge spike in technology-related blockbuster cinema made for popular consumption in Hollywood, around the time of the Silicon Valley economic boom, [55] and the widespread permeation of the Internet. 1 As we discuss the representation of technology in Indian cinema, we find many of these themes repeated, in Indian cinema – but one striking difference. While the computer user as ‘hacker’ is a consistent theme in Hollywood (and used selectively in India, as we shall see), the computer user as ‘aspirational’ is almost entirely missing. C. Cinema and Public Discourse in India In exploring why technology in cinema is important to our understanding of ICTD, we turn briefly to the importance of cinema in public discourse in India. The political importance and development discourse of cinema gained early recognition among the Bolsheviks following Lenin’s own affirmation of cinema as a tool of propaganda [56]. Much Indian cinema under the British was censored and limited to non-nationalistic themes, but the postcolonial period threw open narrative floodgates. Jawaharlal Nehru, the first prime minister, recognized the importance of cinema in the nation-building exercise, commissioning Italian director Roberto Rosselini to make a film on India in 1950, thereafter supporting films like Mother India and several of Raj Kapoor’s early postindependence films [57]. In this period, a lot of the screenwriting in the Bombay film industry was done by a number of socialist secular writers who were formerly active in the freedom movement, and subsequently incorporated several nation-building themes into their cinema. But it was further south in the Madras film industry that cinema as a political weapon was actively imagined by politicians and filmmakers alike [58]. The Dravida Munnetra Kazhagam (DMK) as a political party was quick to use films as a 1

This is somewhat comparable to the appearance of computers and computer users in India, which became much more common a decade later when home computers achieved reasonable (though far from ubiquitous) permeation in urban India.

propaganda tool for spreading its message, and began a trend of what is now synonymous with several south Indian cinematic cultures, of practically replacing temple idols with cinematic stars larger than life, and usually extremely political [59]. 2 This north-south divide in the manifestation of technology in public discourse exists, especially in looking at the on-screen use of technology by ‘politically active’ stars and examining how that relates to their own political motivations. Much work has been done on the cinematic culture of India, and its representation of modernity [60-62], and more recently on television in the post-liberalization India [63]. Significant commentary on media and public discourse has focused on the influence of cinematic and television representations on public imagination including on economic and social decision-making [64-68]. Interestingly, while much work has focused on the protagonists and their characterizations as elements of modernity [69-71], little has focused on the actual artifacts themselves, despite as we will find, much continuity among these symbols of modernity. Finally, this article draws at its heart a bountiful literature in critical theory and film [72, 73] including in cinematic traditions of the developing world [74]. IV. METHODOLOGY This paper is primarily a proposition based on readings of theoretical work in media studies and ICTD, and an expansive content analysis of a large number of films. 87 films are cited here, of which a subset of 43 films form the core of this argument. Each of this latter group is an Indian film that has from the late 1990s or early 2000s some reference or characterization relating to technology and society. Given the limitation of space, I refer primarily to the overall characterization and reference to technology rather than a analysis of specific details within the script such as dialogue. In addition to the observations from films, excerpts from interviews that were part of an earlier interview-based study on conceptions of technology and computers in rural India are used here. The excerpts used here are largely those that have some mention of cinema, or reflected ideas that relate directly to the themes emergent from our examination of films. These interviews are in fact the motivator for this study, as it emerged in this earlier study that cinematic descriptions of computers and computer-users were very important in idea formations about the benefits of technology among populations that had little or no prior experience of direct use of computers. V. ANALYSIS From the early days of Indian cinema, modernity has had a tenuous track record. Consequently, technology and its representations have often been caught in an awkward contested space between traditional and modern values, often implying the schism between rural and urban landscapes. One 2 This is further interesting by the fact that DMK was an explicitly atheist party.

420

of the earliest films to straddle this space, Dharti Mata [75], literally meaning Mother Earth, had two friends, one who goes abroad to study and returns an urbane engineer home to find mineral wealth on his ancestral soil, and his buddy, who stays back and decides to stay true to his rural roots and farm the same earth on which the former wants to drill. The clash between the protagonists ends with the village-bound son-ofthe-soil winning over, an ending that would reprise itself repeatedly over the years. Fast forward to the new millennium, and Kandukodain Kandukondain [76], the Indian remake of Sense and Sensibility . Here, the protagonist is a woman, and lives in her ancestral village, but when the going gets tough, she leaves for the city, ends up happily ever after as a software engineer. Indeed Kandukondain may not be the archetype for all films of the generation, but as a popular blockbuster, the film featured several of the key themes that we highlight here on the changing view of modernity, especially because of the reversed gender roles in the film. A. ICTs and gender empowerment During field research in rural Karnataka in 2005, at a village where the local government had just installed computers in a primary school, an interesting insight on computers and the workplace came from a 21-year old teacher at a village computer center. Both her parents were illiterate casual laborers. “I want to move out of the village. I am looking for a job with computers because my parents will let me move to Udupi or even Bangalore if the work is in computers. For any other job, they won’t let me leave the village.” Geetha, Computer Teacher, Rural Udupi, Karnataka About a year later, an interesting corollary to Geetha’s statement came from a taxi driver interviewee, in the neighboring state of Tamil Nadu, a father of two girls in their 20s. “Both my daughters work in Chennai in computers. In the early days, we would never let our (referring to the Thevar caste) women travel to the city to work, but if they work for computers that is good. There are good facilities with only ladies housing, and many other families from our village have sent their daughters to work in Chennai now.” Selvaraghavan, Taxi Driver Coimbatore District, Tamil Nadu There is already a growing corpus of work on female technology workers in Indian call centers which probes this issue more closely, [29], and the local movie theatre as a starting point into some of the changing social conceptions of women’s work. If we think back to Indian cinema and think back to the typical occupations of women on screen … well, reasonable

women don’t have jobs. And those that do, typically do so because of the failure of some critical male provider. Thus, the dead, drunk or incapacitated rural husband or father gives way to the woman who works the field, at the risk of lascivious attention of the agrarian landlords or plantation managers [77, 78]. When the female ends up working in the city, if she’s generally less educated or desperate enough, and will likely end up as a casual laborer, in which event she is at risk from the lecherous building contractor [79, 80]. Now these informal sector jobs tend to go to the ‘mother figures’ in movies, an obvious consequence of these positions cutting a harder case for visual glamour. The young heroine in a film tends to work more often in organized sector. Here too, many of the same problems exist. There are screen roles where women portray alternatives to the mother figure such as nurses or doctors [81-83] or caring social workers or teachers [84-87], reasonably respectable jobs for women, and not macho enough for the male protagonists. The female character who fails to secure any of the above, spills over into the organized employment sector. The villainy of the rural landlord and the urban building contractor are now reprised in the lascivious white-collar bourgeois rogue harassing female co workers [88-90] and in the rare cases, the daring self-starter saleswoman [91-93] or a spoilt heiress boss [94, 95]. In most of these cases, a male character, usually the hero, offsets the perils of the woman’s tryst with the man’s domain of the economy. Of course, the exceptions exist – the woman embraces the perilous world with vengeance or connivance, the former when she takes on the role of Hindu goddess Durga as a police woman [96, 97] or avenging angel [98, 99]. What remains degenerates quickly in the caste hierarchy of professions from the mildly uncomfortable bar dancer or performer [100, 101], to the circumstantial prostitute [102104], and finally to the campy gangster’s moll [101, 105, 106] Kandukondain Kandukondain [76], took a gentle step in a different direction. The female protagonist in the film, a Brahmin girl, Sowmya is reduced to penury after being hoodwinked out of her ancestral village home by evil relatives. She proceeds to move from the village to the city, where she rectifies her family’s situation by learning to use computers and getting a job as a software engineer. The actress not only lifts her entire family out of poverty, but emerges as the most valuable technologist in her firm and almost flies out to the US to work as an engineer before better sense prevails and she marries her wooing admirer. What is uniquely compelling about the characterization of Sowmya is that she remains ‘traditional’ to the end, but straddles the modernity space effortlessly. She dresses conservatively, except for dream sequences with song and dance, serves her family dutifully, and acquiesces to an arranged marriage. When she does start her software job, her workplace does not demand flashy hackers, instead, her interactions with her engineer boss highlight the importance of urban traditional middle-class work ethos. Despite her success, she does not turn into a slick skirt-wearing executive, and in doing so, kills two birds with one stone. She remains the agreeable screen heroine and makes it to poster child for ‘appropriate ICTD’ –

421

a woman who wins her way out of poverty, and manages stay culturally rather untainted.

daughters married off. In the 2000s, that role has been taken over by the US-returned (or not) software engineer. 3

While there had probably never been an Indian film with a female character playing an engineer save for the oddball automobile mechanic’s daughter, the female software engineer and outsourcing center workers hit the stands in a huge way around this time [107-113], including films with a reversal of roles – where the female lead plays an accomplished technologist of some form, and the male lead is portrayed as professionally subservient [114, 115]. One film stands out as an interesting continuation to Kandukondain. In the Telugu film Anand [116], the protagonist Roopa takes Sowmya’s position a step further. Here, she not only supports herself through a software job, but sheds much of Sowmya’s demure qualities in reinforcing her independent identity and challenging traditional behavior roles expected from her by suitor and in-laws alike.

The classic ‘catch groom’ of south Indian cinema was Arvind Swamy, who in the blockbuster hit Roja [121] played a computer engineer whose arranged marriage is the early plot of the film. Over time, two broad strands of software engineer-related marriage scenarios have emerged – the first in which the engineer is the middle-class hero [122-124] and the second is the counterpoint – in which parents are shown hankering after a groom who is a software engineer or NRI, and the hero in this case is usually a son-of-the-soil type [76, 125, 126], and while the endgame of these films is often an ode to the anti-hero, the focus on the software engineer as essential to middle class aspiration is nonetheless highlighted. What is also interesting is to look at the contrast between the engineer/technocrat protagonist in cinema around industrial themes with the engineer in a technology film. The factory films frequently had a foil – typically industrialist antagonist pitted against the engineer [127], and worker, rather is a the hero [128]. A “good” engineer typically aligns with the worker instead of the industrialist, breaking class barriers [94, 129]. The flatness of the technology sector is reinforced through two factors – first, the “proletariat” is missing in the software company, unlike for instance the range of films about both factories with workers and service sector offices where the ‘peon’ symbolized the underclass [130-132] and second, the boss himself (not necessarily herself!) being sharing class with the employees [133, 134]. My reason for excluding gender is that women and workplace already come loaded with a large literature as emphasized in the previous section, thus the grounds for evaluating “flatness” do not exist in the same vein – women never played engineers or evil industrialists (heiresses, though) so the terms for historical comparison do not apply. A woman’s professional ascendance in the a technology job is in itself shown as having a range of power dynamics and it would therefore be confounding to analyse the gender depiction based on flatness alone.

An even more interesting addition to the technology and gender empowerment idea comes in the film Swagatam [117]. In it, the male protagonist is a demanding customer at an arranged marriage matchmaking bureau. The manager of this bureau (coincidentally the female lead) has an online candidate repository, and tells the male lead to use the computer to filter through his requirements. She kindly reminds the hero that a woman’s greatest trauma is being rejected at the arranged marriage meeting, and that technology should be used effectively to circumvent this problem, and thereby empower women. Given that marriage is practically an ever-present theme in Indian cinema, it is not surprising that several other films have used technology as a go-between for arranged or other marriages [113, 118]. Although the computer as a cupid connector has been a fairly strong theme through most Indian cinema, the female software engineer has been less so. Regional cinema such as Bhojpuri, Bangla, and Oriya largely cater to rural audiences (or what in trade parlance is referred to as ‘B circuits’), and still feature a lot fewer women in professional positions. In Hindi cinema, on the other hand, female characters who used computers in Hindi films, have tended to be more westernized or upper class, such as the student who does computer science in college as a qualification, before her eventual marriage, like Madhuri Dixit in Hum Aapke Hain Koun [119] or the ‘smartypants’ Preity Zinta in Koi Mil Gaya [120], characters significantly different than the Sowmyas or Roopas, where the appropriation of technology into the middle class domain has been less of a concern. In our discussion of aspiration, we return to this issue. B. Computers as Aspiration, Engineer as Hero There are few windows more insightful into the aspirational environment in India than the marriage market. Till date, the highest catches in the dowry market in rural India go to holders of government jobs – which offer among other things – stability, an invaluable element in the rain-dependent agrarian landscape. But films have long portrayed doctors and ‘big officers’ as good catches for families trying to get

In some films, the less than desirable character turns to computers as a means of social acceptability and in others [135, 136], the drive is primarily economic, often explicitly as a means of getting jobs in the US [134, 137]. We find among male software engineers a distinction between the north and south Indian stars. In Hindi films, the engineer stereotype is typically applied to the ‘cool youngster’ hacker characters, usually teen idols, [138-141], very comparable to the Hollywood depictions of computer teenager users from the 1980s and 1990s whereas most of the south Indian depictions are not necessarily hackers who do cool things with 3 In the last year South Indian actresses, Renuka Menon, Gayathri Raghuram, and Kanika Subramanian have all left the film industry to marry software engineers in the US. In the past, Swarnamalya, Ravali, Jyothirmayi, Maheshwari the common stereotype/joke was of actresses leaving the film industry to marry doctors abroad, which is interesting and a potential topic for a larger study on ‘legitimacy’ of actresses.

422

computers, but rather people who have technology-related jobs. This distinction is doubly interesting since it not only underlines the differences between the viewing cultures of popular Hindi and regional cinemas, a difference that has been articulated in Ashish Rajadhyaksha’s examination of the ‘westernizing trend’ of Hindi cinema [62], an upper class urbane narrative, in short. In contrast, south Indian cinema, while the fantasy sequences such as songs etc. have roughly the same ‘modernity’ of the Hindi counterparts, the narrative themes are much closer to the wider middle-class audience appeal. Thus, the persistence of software engineer characters is seen not just in the ‘urban’ south Indian cinema (meaning urban in both setting and primarily appeal) but also in the ‘mass’ cinema. C. ‘Mass Films’ and the Computer as Benevolent Superstar “Computers can be used to fight evil. We can do anything with a computer” Shivraj, 5th grader Devanahalli, Karnataka “Computer can save us. When neighbouring country is attacking, this is known to our scientists by tracking it on the computers.” Udhaykumar, 5th grader, Coimbatore, Tamil Nadu In this section, we turn to the difference between the aspriational cinema, which we cover in the previous two sections, and the ‘mass films’ which not only highlight an interestingly different view of technology and modernity, but also bear uncanny resemblance to some of the Hollywood characterizations of the past. By ‘mass films’, we refer here to the films that are primarily intended towards urban lowerclass and rural viewers. The term, derogatory as it may seem, is quite commonly used by filmmakers and analysts alike to refer to ‘people’s’ cinema, and several major film stars have come to be associated with this movement. While these can also refer to low-budget movies made for rural audiences, a fairly large share of ‘mass films’ are actually big budget productions usually featuring a movie star with a significant, often frenzied fan following. Such actors typically associated with ‘mass’ cinema tend to have reliably larger-than-life characterizations and public images [69]. The dialogue in ‘mass films’ is typically political and populist, almost propagandist, and can probably be traced back to the cinema of M.G.Ramachandran, who never drank or smoked on screen, committed no act of villainy, remained utterly attractive to women, helpful to every needy person, and entirely unbeatable in any physical or intellectual contest. His ‘mass star’ successors are somewhat less pious in contrast, but fairly comparable in their abilities to commit superhuman acts, ricochet bullets, and most important of all, save mankind. Fans of ‘mass stars’ frequently refer to their cherished stars

with honorifics such as ‘Dear Leader’ or ‘Elder Brother,’ several of them end up in politics, and the opening few weeks of a film by a mass star can almost certainly expect to run to full houses, not to mention the celebrations and prayers that accompany any release. Several stars, especially in the south, have landed in the ‘mass star’ image trap including Mohanlal and Mammootty from Kerala Rajkumar in Karnataka, Balakrishna, Nagarjuna, and Chiranjeevi in Andhra Pradesh, and Vijaykanth and Rajnikanth in Tamil Nadu, icons with possibly the most frenzied fan following of all. The use of technology by these mass stars is particularly interesting, because their audiences are often not the middle class aspirants that watch many of the urban-based films listed above. Interestingly then, we find the ideas of technology use in such mass star movies somewhat simplistic, but in line with the overall theme of what a mass star does – save the world. Typically a mass star is a perfect son-of-the soil, thus almost always speaking in the vernacular through a film (unlike in Hindi films, for instance, where English is often interspersed with the Hindi dialogue) excelling and promoting local arts and so on. In a sense, the mass stars’ use of computers is comparable to their use of English. Mass stars generally use English only occasionally, laconically, and with much emphasis, usually when challenged by an English speaking adversary or intransigent love interest. In the same vein, a ‘mass star’ may use a computer to underline just that he can, even if it is entirely irrelevant to the plot. As a son of the soil, he is able not just to play the game by the rules of the hinterland, but can match up to modernity [142]. In Sivaji - the Boss 4 [143], by some estimates the top grossing India films of all time, Rajnikanth, arguably India’s most maniacally followed star, plays a software engineer-cumsocial worker bent on rectifying the ills of the Tamil world. In his quest, Sivaji uses his voice-recognition-enabled laptop to control his vigilante operations and outmaneuver his rivals. Similarly the use of computers to maintain and control databases of villains has been used effectively by megastars Vijaykanth [144], Chiranjeevi [145], and Ajithkumar [146, 147]. The idea of computers as a tool to assist heroes in cleaning out society has been used to much effect in recent times, mostly in films featuring online websites for ‘submitting un-rectified crimes’ for appropriate appraisal and corrective action by the hero on the other end of the internet connection [148, 149] The quote from Shivraj at the start of this was not an isolated one. We were surprised at how often in our interviews in rural India, the same answer was repeated to us over and over “Computers can do anything” sometimes ranging from children’s fantasies to adults with no direct experience with 4

Boss here stands for Bachelor of Social Science

423

computers allocating human attributes to the machines, “Computers can teach us English” in clear seriousness. It is far fetched to ascribe such notions specifically to films, but it’s worth looking briefly at some of the omnipotent deeds of computers. Perhaps a compelling analogy would be that computers can do anything, just like Rajnikanth. It is not surprising in this light to know that movie star Mammootty was selected as the Brand Ambassador for the rural computing initiative in Kerala, Akshaya. If we look back before the computer-using days of the mass stars, shades of fantasy and omnipotence among computers in India film were akin to western cinema of the 1960s, especially in early Indian science fiction [150, 151]. Computers and their ability for visual magic broke box office records with ‘Miss World’ Aishwarya Rai’s first screen appearance in Jeans [152]. Here, her brother creates a visual double of Aishwarya to convince twin brothers that they were both in love with two different girls, a reasonable exercise in balancing technological savvy with the need for romance in scripts. Jeans and the hugely popular MTV video of Michael Jackson’s single ‘Black or White’ [153] featuring faces morphed into one another, or simply the glut of Photoshop engineers in India, also helped create one of the most enduring computer tricks in Indian films, still a common feature of ‘mass films’ – the ability of a machine to take a scanned picture of a child lost several years ago, and morph out of it a perfectly accurate image of the adult version, auto-adjusting for sartorial grace and the few extra facial pounds on the likes of Vijaykanth. [154, 155]. CONCLUSIONS Is there intentionality in these portrayals? My opening the article with my personal sob story of failing to sell my scripts was meant to imply at least partly that there is at least some level of production-level engineering of characterizations. I would argue that that question impossible to answer given the recursive loop on whether society influences media or viceversa. The idea for this research emerged as after confounding outcomes in other research that indicated a mismatch between peoples’ stated interest in technology and their actual use of ICTD projects in rural India. People were very excited about computers and the possibility of their own access to them, but unclear on how technology could be practically useful in their lives. Such ideas about technology were further seen to not just influence researchers’ estimation of what the likely adoption for certain ICTD projects may be, given the apparent enthusiasm about technology, but could also influence the populations’ own propensities to invest in ICTD projects, without a necessarily clear idea of the value of such technology. For serious scholars of ICTD, closing this phenomenon off as the ‘buzz’ of technology is not adequate, and an examination of the discourse encompassing this buzz is necessary. As ICTD matures as a strong body of independent literature, it will be vital to incorporate learnings from existing

bodies of thought in development studies. Here, I aim to very gently open the windows into the concerns of critical theory and discourse analysis. This analysis of computers and cinema in India brings to light a number of interesting factors about how filmmakers have chosen to portray technology in Indian cinema, and it is particularly interesting to look at where these portrayals are comparable to other cinematic traditions, and where, like in the aspirational characterizations, it is unique and deeply related to the prevalent discourse of technology being a means of social and economic leapfrog. What is also remarkably telling is the much greater prevalence of technology-related themes and characterizations in south Indian films than in north Indian cinema, given the disproportionately higher concentration of technology-related industries and opportunities in the major south Indian metropolises. These cities like Hyderabad, Chennai, and Bangalore saw tremendous demographic, landscape, and social transformations in the late 1990s, the same time that computers as movie characters spiked up the charts. Many of the aspirational characters may well draw their original inspirations from real-life heroes, often engineering graduates from the hinterland who moved to the cities, and saw remarkable growths in their own incomes and social statuses, often moving abroad, enjoying opportunities that would have been much more challenging to aspire to a generation back. Most importantly, many of these graduates were women. What the cinematic analysis shows us is indeed a reflection of a larger perception of what counted as the ‘India Rising’ metaphor for the urban middle classes and the rural aspiring middle classes, which incidentally, largely comprise the regular film-going population. It is the aspirations of these classes that we see portrayed in a subsection the South Indian cinema of today. It is thus decidedly interesting to compare the output geared at these aspirational markets with the poorer hinterland markets of the mass films which take on a simplistic form, carrying forth in selective filters the transplant of middle class imagination of technology down to the economic and social chain. In conclusion, we turn briefly to the iconoclastic proposition of popular Indian cinema being proactively used to impact development. While the producers may indeed have been sitting by with whips, were screenwriters of the films we see above thinking of issues of technology and development, and of impacting society, or is what we see here a sheer reflection of popular psyche. Irrespective of that, as demonstrated, research has already shown that certain types of entertainment media experiences in India have had positive social outcomes, especially in gender-related issues like female child protection and domestic abuse prevention. This could well be dangerous knowledge for the ICTD community, a significant part of which is comprised of scholar-activists. Indeed some of us may stand up and ask if cinema can and should be proactively be used to impact technology uptake. It is important that we think of these questions since we are after all in development and for many of us, all research implies the responsibility of considering real world intervention

424

capability. When we do start thinking in that direction, perhaps a social scientist in the room can stand up and point out the parallels of such ideas with Leninistic cinematic propaganda, and ask if technology intake ought to be promoted at all. After all, asking questions is what we do, decisive action is Rajnikanth’s domain.

19.

20.

ACKNOWLEDGEMENTS Thanks to Eric Brewer for bankrolling my whimsical work, and to Divya Ramachandra, Kentaro Toyama and Jenna Burrell for their interest in this particular research. REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.

13. 14. 15.

16.

17. 18.

Roy, B., Naukari. 1954, Bimal Roy Productions: India. Chandra, N., Ankush. 1986: India. Bapaiah, K., Aaj Ka Daur. 1985: India. Muthuraman, S., Aarilirunthu Arubathu Varai. 1979. Chidambaram, S., Vyapari. 2007: India. Castells, M., Rise of the Network Society. 1996. Negroponte, N., Being Digital. 1995, Cambridge, MA: MIT Press. Hoffman, D.L. and T.P. Novak, Bridging the Racial Divide on the Internet. Science, 1998. 280(5362): p. 390-91. Compaine, B.M., The Digital Divide: Facing a Crisis Or Creating a Myth? 2001: MIT Press. Hendry, J.D., Social inclusion and the information poor. Library Review, 2000. 49(7): p. 331-336. Warschauer, M., Reconceptualizing the Digital Divide. First Monday, 2002. 7(7): p. 0-262. Singh, N., Information Technology as an Engine of Broad-Based Growth in India, in The Future of India and Indian Business, P. Banerjee and F. Richter, Editors. 2002, Macmillan: London. Heeks, R., India's software industry. 1996: Sage Publications Thousand Oaks, Calif. Arora, A., et al., The Indian software services industry. Research Policy, 2001. 30(8): p. 12671287. Madon, S., Information-Based Global Economy and Socioeconomic Development: The Case of Bangalore. The Information Society, 1997. 13(3): p. 227-244. Saxenian, A.L. Bangalore: The Silicon Valley of Asia? in Conference on Indian Economic Prospects: Advancing Policy Reform,. 2000. Stanford, California: Center for Research on Economic Development and Policy Reform. Jhunjhunwala, A., Can Telecom and IT be for the Disadvantaged? Rural Development, 1998. 17(2): p. 321–37. James, J., Low-cost computing and related ways of overcoming the global digital divide. Journal of Information Science, 2001. 27(6): p. 385-392.

21. 22. 23.

24. 25. 26.

27. 28. 29.

30.

31.

32. 33.

34. 35.

Nulens, G. and L. Van Audenhove, An Information Society in Africa?: An Analysis of the Information Society Policy of the World Bank, ITU and ECA. International Communication Gazette, 1999. 61(6): p. 451. Heeks, R., Information Systems and Developing Countries: Failure, Success, and Local Improvisations. The Information Society, 2002. 18(2): p. 101-112. Keniston, K., Grassroots ICT Projects in India: Some Preliminary Hypotheses. ASCI Journal of Management, 2002. 31(1). Madon, S., Evaluating the Developmental Impact of e-Governance Initiatives: An Exploratory Framework. EJISDC, 2004. 20(5): p. 1-13. Donner, J., The Use of Mobile Phones by Microentrepreneurs in Kigali, Rwanda: Changes to Social and Business Networks. Information Technologies and International Development, 2007. 3(2): p. 3-19. Castells, M., The Informational City: Information Technology, Economic Restructuring, and the Urban-regional Process. 1991: Blackwell Publishers. Castells, M., Information technology, globalization and social development. 1999, Geneva: UNRISD. Roche, E.M. and M.J. Blaine, Information technology, development and policy. Information Technology, Development and Policy, EM Roche & MJ Blaine (eds.). Avebury: Aldershot, UK, 1996. Mansell, R., Knowledge Societies: Information Technology for Sustainable Development. 1998: Oxford Univ Press. Webster, F., Theories of the Information Society. 2002: Routledge. Radhakrishnan, S., Examining the “Global” Indian Middle Class: Gender and Culture in the Silicon Valley/Bangalore Circuit. Journal of Intercultural Studies, 2008. 29(1): p. 7-20. Kuriyan, R. and K.R. Kitner. Constructing Class Boundaries: Gender and Shared Computing. in Second International Conference on Information Technologies and Development, Dec 2007, IEEE Conference Proceedings. 2007. Bangalore. Roman, R., Diffusion of Innovations as a Theoretical Framework for Telecenters. Information Technologies and International Development, 2003. 1(2): p. 53-66. Warschauer, M., Technology And Social Inclusion: Rethinking the Digital Divide. 2004: MIT Press. Sein, M.K. and G. Harindranath, Conceptualizing the ICT Artifact: Toward Understanding the Role of ICT in National Development. The Information Society, 2004. 20(1): p. 15-24. Bailur, S., Using Stakeholder Theory to Analyze Telecenter Projects. Information Technologies and International Development, 2006. 3(3): p. 61-80. Burrell, J. and K. Anderson, "I have great desires to look beyond my world:" Trajectories of Information

425

36. 37. 38.

39. 40. 41. 42. 43.

44. 45. 46. 47.

48. 49. 50. 51. 52. 53. 54. 55. 56. 57.

and Communication Technology use among Ghanaians. New Media and Society, 2008. 10(2): p. 203-224. Morton, H., Islanders in Space: Tongans Online. Small Worlds, Global Lives: Islands and Migration, 1999: p. 235–254. Franklin, M.I., I Define My Own Identity: Pacific Articulations of'Race'and'Culture'on the Internet. Ethnicities, 2003. 3(4): p. 465. Rai, A.S., India On-line: Electronic Bulletin Boards and the Construction of a Diasporic Hindu Identity. Diaspora: A Journal of Transnational Studies, 1995. 4(1): p. 31-58. Bastian, M.L., Nationalism in a Virtual Space: Immigrant Nigerians on the Internet. West Africa Review, 1999. 1(1). Méliès, G., Le Voyage dans la Lune. 1902: France. Chaplin, C., Modern Times. 1936, United Artists: USA. Atkinson, P. Technology and the Vision of Utopia. in Imagining the Future: Utopia, Dystopia and Science Fiction. 2005. Monash University, Melbourne. Bendle, M. Zarathustra’s Revenge: The Sordid Utopia of Contemporary Science Fiction Films. in Imagining the Future: Utopia, Dystopia and Science Fiction. 2005. Monash University, Melbourne. Youngblood, D.J., Movies for the Masses: Popular Cinema and Soviet Society in the 1920s. 1992: Cambridge University Press. Singer, B., Melodrama and Modernity: Early Sensational Cinema and Its Contexts. 2001, New York: Columbia University Press. Akudinobi, J., Tradition/Modernity and the Discourse of African Cinema. 1995, IRIS. Lopez, A.M., " Train of Shadows": Early Cinema and Modernity in Latin America. Multiculturalism, Postcoloniality, and Transnational Media, 2003. 40(1): p. 48-78. Mukherjee, R., Hum Hindustani. 1960. Maté, R., When Worlds Collide. 1951, Paramount: USA. Zaslove, M. and C. Nichols, The Jetsons. 1962, Hanna-Barbera Productions: USA. Newman, S., The Avengers. 1961, American Broadcasting Company (ABC): USA. Roddenberry, G., Star Trek. 1966, Paramount: USA. Kubrick, S., 2001: A Space Odyssey. 1968, MGM / Warner: USA. Bonnière, R., Hide and Seek. 1977, Public Broadcasting System (PBS): Canada. Faden, E., The Cyberfilm: Hollywood and Computer Technology. Strategies, 2001. 14(1). Kepley Jr, V., Soviet Cinema and State Control: Lenin's Nationalization Decree Reconsidered. Journal of Film and Video, 1990. 42(2): p. 3-14. Mishra, V., Bollywood Cinema: Temples of Desire. 2002: Routledge.

58. 59. 60. 61. 62.

63.

64.

65. 66.

67.

68. 69. 70. 71. 72. 73. 74. 75. 76.

Dickey, S., The Politics of Adulation: Cinema and the Production of Politicians in South India. The Journal of Asian Studies, 1993. 52(2): p. 340-372. Hardgrave, R.L., Politics and the Film in Tamilnadu: The Stars and the DMK. Asian Survey, 1973. 13(3): p. 288-305. Dwyer, R. and D. Patel, Cinema India: The Visual Culture of Hindi Film. 2002: Rutgers University Press. Appadurai, A. and C. Breckenridge, Public modernity in India. Consuming Modernity: Public Culture in a South Asian World, 1995: p. 1–20. Rajadhyaksha, A., The Bollywoodization of the Indian cinema: cultural nationalism in a global arena. Inter-Asia Cultural Studies, 2003. 4(1): p. 2539. Mankekar, P., Screening Culture, Viewing Politics: An Ethnography of Television, Womanhood, and Nation in Postcolonial India. 1999: Duke University Press. MacPherson, Y., Images and icons: harnessing the power of the media to reduce sex-selective abortion in India. Gender & Development, 2007. 15(3): p. 413-423. Fernandes, L., Nationalizingthe global': media images, cultural politics and the middle class in India. Media, Culture & Society, 2000. 22(5): p. 611. Jensen, R.T., E. Oster, and R. National Bureau of Economic, The Power of TV: Cable Television and Women's Status in India. 2007: National Bureau of Economic Research Cambridge, Mass., USA. Holbert, R.L., D.V. Shah, and N. Kwak, Political Implications of Prime-Time Drama and Sitcom Use: Genres of Representation and Opinions Concerning Women's Rights. The Journal of Communication, 2003. 53(1): p. 45-60. Dolnick, S., Keep Movie Heroes Alcohol-Free, Bollywood Urged, in Washington Post. 2008: Washington, DC. p. C 07. Pandian, M.S.S., The Image Trap: MG Ramachandran in Film and Politics. 1992: Sage. Jacob, P., From co-star to deity: Popular representations of Jayalalitha Jayaram. Women: A Cultural Review, 1997. 8(3): p. 327-337. Sharma, A., Blood, Sweat and Tears: Amitabh Bachchan, Urban Demi-God. You Tarzan: Masculinity, Movies and Men, 1993: p. 167-180. Lehmann, C. and L.S. Starks, Spectacular Shakespeare: Critical Theory and Popular Cinema. 2002: Fairleigh Dickinson Univ Press. Bordwell, D., Making Meaning: Inference and Rhetoric in the Interpretation of Cinema. 1989: Harvard University Press. Gabriel, T.H., Towards a Critical Theory of Third World Films. Film Theory: Critical Concepts in Media and Cultural Studies, 2003. Bose, N., Dharti Mata. 1938, New Theatres: India. Menon, R., Kandukondain Kandukondain. 2000.

426

77. 78. 79. 80. 81. 82. 83. 84. 85. 86. 87. 88. 89. 90. 91. 92. 93. 94. 95. 96. 97. 98. 99. 100. 101. 102. 103. 104. 105. 106. 107. 108. 109. 110. 111. 112. 113. 114. 115. 116. 117.

Khan, M., Mother India. 1957: India. Roy, B., Madhumati. 1958, Bimal Roy Productions: India. Chopra, Y., Deewar. 1975: India. Krishnamurthy, R., Thee. 1981: India. Majumdar, P., Aarti. 1962, Rajshri Productions: India. Sahu, K., Dil Apna Aur Preet Parai. 1960, Mohan Pictures: India. Balachander, K., Manathil Uruthi Vendum. 1987: India. Chopra, Y., Daag. 1973: India. Padmarajan, P., Koodevide. 1983: India. Kapoor, R., Shree 420. 1955, RK Films: India. Kapdi, M., Aaitya Bilavar Nagoba. 1979: India. Prakash Rao, K.S., Secretary. 1976: India. Balachander, K., Avargal. 1977: India. Chopra, B.R., Pati Patni Aur Woh. 1978, B.R. Films: India. Rawail, H.S., Patanga. 1949, Varma Films: India. Neelakantan, P., Raman Thediya Seethai. 1972, AVM Productions. Paranjpe, S., Chashme Buddoor. 1981, PLA Productions: India. Kanwar, R., Laadla. 1994: India. Mirza, A., Raju Ban Gaya Gentleman. 1992, Sippy Films: India. Rama Rao, T., Andhaa Kanoon. 1983: India. Dixit, C.P., Fakira. 1976, Eros Entertainment: India. Bhogal, A., Zakhmi Aurat. 1988, Manta Movies: India. Chopra, B.R., Insaf Ka Tarazu. 1980, B.R. Films: India. Sippy, R., Shakti. 1983: India. Hussain, N., Yaadon Ki Baaraat. 1973, United Producers: India. Balachander, K., Arangetram. 1973: India. Sasi, I.V., Avalude Ravakul. 1980: India. Samanta, S., Amar Prem. 1971, Shakti Films: India. Ramachandran, M.G., Ulagam Sutrum Valiban. 1973, Emgeeeaar Pictures: India. Anand, V., Jewel Thief. 1967, Navketan Films: India. Bhadran, Olympian Anthony Adam. 1999: India. Madan, Pellaina Kothalo. 2004, AA Naluguru Films: India. Selvaraghavan, K., Aadavari Matalaku Ardhalu Verule. 2007, Sri Saideva Productions: India. Yelleti, H., Ankit, Pallavi and Friends. 2008: India. Goud, R.K., Jodi No. 1. 2003, RK Movies: India. Shankar, S., Kunjikkoonan. 2002: India. Prabhakar, T., Neethone Vuntanu. 2002, Lakshmi Art Pictures: India. Vamsy, Avunu Valliddaru Ista Paddaru. 2002, Anandi Arts: India. Selvaraghavan, 7G Rainbow Colony. 2005, Ayngaran: India. Kammula, S., Anand. 2004: India. Dasarath, Swagatam. 2008, Aditya Ram Movies: India.

118. 119. 120. 121. 122. 123. 124. 125. 126. 127. 128. 129. 130. 131. 132. 133. 134. 135. 136. 137. 138. 139. 140. 141. 142. 143. 144. 145. 146. 147. 148. 149. 150. 151. 152. 153. 154. 155.

Kathir, Kadhalar Dhinam. 1999, A.M. Rathnam: India. Barjatya, S., Hum Aapke Hain Koun? 1994, Rajshri Productions: India. Roshan, R., Koi...Mil Gaya. 2003, Film Kraft: India. Rathnam, M., Roja. 1992, Hansa Pictures: India. Kranthikumar, Kanden Seethaiyai. 2001: India. Vasanth, Satham Podathey. 2007: India. Renjith, Nandanam. 2002: India. Narayana, K.L., Rakhi. 2006, Sri Durga Arts: India. Selva, V., Youth. 2002, Ayngaran International: India. Chopra, Y., Aadmi aur Insaan. 1969: India. Chopra, R., Mazdoor. 1983: India. Chopra, Y., Kala Patthar. 1979: India. Hingorani, A., Dil Bhi Tera Hum Bhi Tere. 1960. Anthikad, S., Nadodikattu. 1987. Chatterjee, B., Chhoti Si Baat. 1975: India. Menon, R., Kandukondain Kandukondain. 2000: India. Menon, G., Minnale. 2001: India. Bhargavan, Thiru Ranga. 2007: India. Ravishankar, Varushamellaam Vasantham. 2002: India. Rasool, Unnai Paartha Naal Muthal. 2004, Anandi Arts: India. Singh, A., Raqeeb. 2007: India. Ghosh, K., Fida. 2004: India. Roshan, R., Krrish. 2006, Film Kraft: India. Anand, S., Bachna Ae Haseeno. 2008, Yash Raj Films: India. Sivamani, Law and Order. 2002, Jyotirmayi Pictures: India. Shankar, S., Sivaji. 2007, AVM Productions: India. Murugadoss, A.R., Ramana. 2002, Oscar Films: India. Vinayak, V., Tagore. 2003, Oscar Movies: India. Vasu, P., Paramasivan. 2006: India. Vishnuvardhan, Billa. 2007, Sri Keerthi Creations: India. Shankar, S., Anniyan. 2005, Lakshmi Ganapathi Films: India. Jayaraj, R., 4 The People. 2004: India. Parvez, S., Jumbish. 1986, Ghalib Studio: India. Srinivasa Rao, S., Aditya 369. 1991, Sridevi Arts: India. Shankar, S., Jeans. 1998, Sri Surya Films: India. Landis, J., Black or White. 1991, Michael Jackson: USA. Perarasu, Dharmapuri. 2006, A.M. Rathnam: India. p. 150 min. Vidhyadharan, Vaitheeswaran. 2008, Annamalai Films: India.

427

Regulatory Independence and Wireless Market Development: A Comparative Analysis of Two African Nations Annemijn F. van Gorp & Carleen F. Maitland Abstract— This study analyzes the nature of regulatory independence and its influence on wireless market development in Tanzania and Botswana. The study finds that the level of regulatory independence is associated with improved market conditions. The research has implications for theories of regulation and market development in low income countries. In particular the Tanzania case suggests that the independence of regulation can have secondary effects such as diversity of technologies and faster transitions to advanced technologies, while the reversal of independence in Botswana highlights the need for greater insights into the under-theorized dynamic nature of regulatory independence. Index Terms— regulatory independence, Botswana, Tanzania, mobile market development

T

I. INTRODUCTION

elecommunications infrastructure has long been seen as a catalyst for economic development [1, 2]. However, its growth is sometimes hindered by a variety of factors including its past as a publicly provided good, regulation, and characteristics inherent to network technologies. As network technologies, telecommunication systems require investments that are susceptible to expropriation and hostage-taking [3, 4], as exemplified in many cases by the government’s power to specify a variety of conditions including geographic market definition, coverage obligations, universal service obligations and even prices. Consequently, the level of investment in, and hence the widespread availability of telecommunications services, including fixed, mobile and Internet access is likely to be influenced by regulatory quality that reduces uncertainty and insulates the sector from politically driven actions. During the past two decades many countries, high and low income alike, have sought to improve their regulatory quality by establishing autonomous national regulatory authorities (NRAs). The level of independence of these authorities from Manuscript received September 22, 2008. This work was supported in part by a grant of the Africana Research Center at the Pennsylvania State University. A. F. van Gorp is with the Ted Rogers School of Information Technology Management at Ryerson University, 350 Victoria Street, Toronto, ON M4V1N1 Canada (phone: +1-416-979-5000 ext. 2482; e-mail: [email protected]). C.F. Maitland is with the College of Information Sciences & Technology at the Pennsylvania State University, PA 16802 USA (email: [email protected]).

ministries and other government departments is a key feature of that regulatory quality [5]. Further, empirical studies have found that independence has significant effects on both market and regulatory performance [e.g. 4, 5]. However, many of these studies are carried out either exclusively in relatively high income countries or in global studies covering a broad range of institutional environments. Consequently, they often contain two implicit assumptions that may not apply in low income countries, which in turn raise interesting questions. The first assumption is that ministries and other government departments are able to wield power, from which independent NRAs are insulated. However, in some low income countries ministries and other government departments are not well-functioning entities and are unable to wield power. In these environments, is regulatory independence still important and if so how? The second assumption is that gains in independence are sustained. In low income countries where power shifts and political upheavals create generally less stable political environments, assumptions about the linear nature of progress on independence may not apply [6]. Under what conditions are reversals most likely? What are the implications for market development? Further, whereas earlier research on regulatory quality and market development focused primarily on fixed telecommunications infrastructure, more recent research that includes mobile market development suggests that it is less sensitive to regulatory quality [7, 8]. These findings suggest that in low income countries, where mobile markets often play a more significant role in terms of providing access, regulatory quality may play a much less significant role. Hence, in these markets does regulatory independence influence mobile and wireless market development, and if so, how? Also, is the nature of this influence similar to that found in fixed networks? In answering these questions this research provides further nuance to theories of regulation and market development. In particular, it sheds light on the relationship between regulatory independence and its stability and wireless market development in low income countries. The analysis employs a comparative case approach of national regulatory authorities (NRAs) of two African countries, namely Tanzania and Botswana. While being among the poorest in the world, these nations are recognized as having two of the better regulatory environments in the region [9]. Data were

428

collected through in-depth interviews and document analysis. Comparative case analyses generate insights into the nature of independence and stability as well as their implications for particular characteristics of the wireless market development. The paper is structured as follows. Section 2 provides background on regulatory independence and its effects. Section 3 describes our research method and data collection techniques and is followed by Section 4 in which the two country cases are presented. Section 5 provides a brief cross-case analysis and in section 6 discussion and conclusions are presented. II. REGULATORY INDEPENDENCE AND ITS EFFECTS While research has found that regulatory independence has significant effects on telecommunications market development, to date it is unclear the extent to which these findings apply in lower income countries. In particular, given the differences in the state of institutional development in some lower income countries and the potential challenges of maintaining institutional gains, it is unclear how independence will influence market development if at all. These questions of influence are particularly salient for the wireless market both because it has experienced such rapid growth in low income countries and because evidence suggests the influence of regulatory quality is lower in this sector. These issues are discussed in turn below. A. Regulatory Quality For years scholars have argued for the necessity of regulatory independence both for proper functioning of the regulator and to enhance both regulatory and market performance [6, 10]. Regulatory performance is the extent to which regulations fulfill their immediate objective, such as lowering interconnection prices, collecting universal service revenues or lowering entry barrier for new entrants. These outcomes are expected to create conditions which in turn improve overall sector performance by, for example, lowering consumer prices, extending networks, and improving service quality. Regulatory independence has been characterized by institutional arrangements that foster clarity of roles of the regulator, and accountability and transparency in the process of regulatory decisions [6, 10]. The delegation of authority and need for independence is driven in part by the desire to establish credibility but also in some cases to insulate the regulatory policies from future politically-driven changes (Gilardi 2007). Further, independence is frequently established at the time the autonomous regulator is developed, a frequent driver of which is privatization and liberalization of monopoly public telecommunications operators (PTOs).To this extent, in a study of PTO ownership in Europe, Bauer [11] found that levels of independence were highest for those nations that were slow to privatize their PTO. Thus, he concludes the level of independence may be driven in part by the level of state ownership in the PTO. Independence is a multidimensional construct, with both formal and informal components, each with several dimensions. While it is often conceptualized as one

component of general regulatory quality (as will be discussed further in the following section), research has suggested it has a unique contribution to market development [4, 5], although findings are mixed [e.g. 11]. Studies of the relationship between independence and market development have employed both qualitative and quantitative methods, with the latter becoming increasingly prevalent along with data availability and relying predominantly on econometric analyses of the formal components of independence. Measures of formal independence vary in their complexity, ranging from simple dummy variables to highly complex measures [e.g. 4, 5, 11]. The most recent and extensive of these is the EURI-I index, which is based on eleven general measures 1 of formal regulatory independence including: 1. multi-sector, 2. multi-member, 3. funding, 4. reporting, 5. shared roles, 6. legislative appointment, 7. fixed terms, 8. renewable terms, 9. staff, 10. budget, and 11. experience [5]. While many of these variables are self-explanatory, several deserve further discussion. The multi-sector measure is based on the presumption that a multi-sector regulator will achieve a higher level of independence from any one ministry, with similar reasoning for multi-member regulators (1) versus single member (0). Reporting indicates whether regulators are required to report only to the Minister (0), to both the Minister and the legislature (.5) or only the legislature (1). The shared roles measure indicates whether or not a regulator shares its duties with the government (0) or is granted exclusive powers (1). Finally, experience is measured as whether or not the NRA has been in existence for at least two years. While the EURI-I index provides a complex measure of independence it is developed only for the European context. Hence, it is unclear the extent to which their results that regulatory independence reduces interconnection rates only when the public telecommunications operator is partially owned by the state, are generalizable to lower income contexts. In addition, a question remains to what extent this operationalization of regulatory independence relates to regulatory issues other than interconnection. That effects of regulatory quality might be different in lower income countries are suggested by the research of Gutierrez [4] and Wallsten [12]. Gutierrez [4] found that when separating low versus high income Latin American countries the influence of regulatory quality on market development is indeed different. In particular, regulatory quality has less of an impact on market development in the low income countries. The author proposes this may be result of the greater difficulty in implementing a positive regulatory environment in low income countries or that their institutional changes were more recent and therefore have yet to produce changes in the market. Conversely, the variables of competition and privatization have a greater effect in lower income countries, which may be attributed to the greater room for improvement. Similarly, in a study of the impact of 1 Here we exclude the study-specific measure of interconnect powers, which measures the extent to which the regulator has powers over interconnection. This measure is useful to studies of interconnection but may not be appropriate for studies of other regulatory issues.

429

an independent regulator on market performance in Africa and Latin America, Wallsten [12] found the regulatory variable on its own is insignificant in explaining teledensity. Hence, while research concerned with the implications of regulatory quality is becoming more nuanced, with independence, competition, and privatization having distinct contributions to market development, it is unclear the extent to which the direction of these developments are relevant for studying independence is low income countries. The following section discusses the broader context of independence and its implications for regulatory independence in low income contexts. B. The Context of Independence While at some level challenges to independence are universal, in different contexts they are likely to vary in their degree. For instance, while nearly all government departments face resource constraints, in low income countries the lack of resources of the regulator and society in general raise special issues for independence. Examples include the inability to pay competitive wages to regulatory staff, resulting in high levels of employee turnover as well as the general scarcity of qualified personnel resulting in the necessity of hiring staff with potential conflicts of interest (e.g. are former Ministry or PTO employees). This occurred, for example, in Sri Lanka where the former Managing Director of Sri Lanka Telecom (SLT) was named Director General of the regulatory authority [13, 14]. This lack of resources may not only affect the regulator itself, but also government departments that serve as partners and possibly opponents of the regulator as well. For example, also in Sri Lanka, Balasooriya et al. [13] found that the Fair Trading Commission, which by law was to serve as the competition authority for the country, was “almost inoperative despite its investigative and quasi-judicial power assigned by its Act” (p. 388). Such circumstances suggest that measures of a regulator’s independence that focus solely on the regulator may not suffice. Earlier studies on the effects of governance or regulatory quality on telecommunications market development focused on the broader governmental environment. For example, studies examined the effect of the credibility of the policy regime [15, 16] through use of the POLCON index [15], a measure of veto points within a government that in turn constrain any one political actor from changing government policy. The studies found that over a wide range of high and low income countries those with lower likelihood of arbitrary policy changes had higher growth in fixed teledensity. Examining the relationship between general regulatory quality exclusively in the telecommunications realm is the work of Gutierrez and Berg [17] and Gutierrez [4]. In these two studies, the measure of regulatory quality is developed from a dichotomous measure of of the the presence (1) or absence (0) of a regulatory framework, to one in which regulatory development is measured as an index. The index includes (1) the separation of operations and regulatory activities, although not necessarily the existence of a separate

regulator, (2) the degree of freedom from political and industry interference, clarity of regulatory functions, accountability and transparency, and (3) the legal basis of the creation of the regulatory body. These studies suggest that independence will be influenced by a variety of contextual factors. In particular, factors that influence the ability of governmental departments to fulfill their role in terms of providing checks and balances will impact independence, positively or negatively. For example, as described by Jain [18] and Samarajiva [19] in India and Sri Lanka respectively, the judiciary played an important role in mediating the relationship between the PTO, ministry and regulator. Whereas in the Indian case the judiciary challenged the position of the regulator, in Sri Lanka it was supported. Further, in both cases the judiciary fulfilled its role in providing checks and balances. However, does this imply that in other environments where ministries, judiciaries and other administrative bodies are unable to wield power, that independence is still important? C. Changes in Regulatory Independence A second assumption implicit in much of the research on the effect of regulatory quality and independence on market development is that gains in quality and independence are maintained. As noted by Stern [6, p. 69], “Even when formally independent regulatory agencies have been set up, as in Argentina or Hungary, the question remains as to how far their independence is (a) genuine and (b) sustainable.” Despite the possibility of reversal little attention has been paid to its implications. This is not to say, however, that the dynamic nature of regulation is not considered. Indeed, several econometric studies use time series data that capture changes in regulatory quality and independence over time. For example, in the EURI-I index, which measures independence in the 15 original EU member states between 1998 and 2003, regulatory institutions declined in two countries, remained constant in four countries, and improved in nine countries [5]. However, because declines are far fewer than improvements their implications are not obvious from the broader analysis. It is also important to note that even where the changes in regulatory quality and independence are positive, the effects may not be. Gutierrez [4], in a study of 22 Latin American countries during the 1980-1997 period, examines changes in regulatory quality and suggests that effects on market development may be nonlinear. In particular, he proposes that at low levels of regulation, further enhancement of the regulatory framework at first increases telecom market performance but then over time its impact is slowly diminished. The possibility of nonlinear effects of institutional improvements begs the question as to how declines in institutional quality affect market performance. However, prior to understanding the effects of declines, more information is required about their fundamental nature (i.e. in what areas of regulatory quality are declines most common) and in what conditions are they most likely to occur.

430

D. Regulatory Quality and Wireless Market Development The dynamic nature of regulatory independence raises a second issue which is its differential impact on wireless market development. In particular, Andonova [8] finds that whereas an increase in the institutional quality positively affects the number of Internet hosts, there is no effect for cellular penetration. Further, the static effects of regulatory quality were less for cellular penetration than for fixed line services. Andonova [8] proposes these findings are due to wireless infrastructure being less of a sunk investment, one that can be redeployed and thus less prone to expropriation and hostage taking. While this may be true in theory, the rapid pace of technological change may hinder the redeployment of wireless infrastructure. Other reasons for the lack of influence of regulatory quality on wireless services are possibly the lower level of regulation, greater prevalence of private ownership and the high level of demand. These factors suggest that a closer examination of the relationship between wireless market development and regulatory quality is required. Earlier studies of regulatory quality [e.g. 12, 17, 20] actually used wireless market development as a predictor of fixed line market growth. Clearly greater understanding of the influence of regulatory quality on wireless market growth (both fixed and mobile/cellular) is required. Also, whereas studies have focused primarily on market growth in terms of access and/or users as well as efficiency of the sector, the relationship between regulation and the wireless market given the less central role of regulation may require a more nuanced understanding. Further, given the rapid growth of wireless markets worldwide, it may be time to look at higher level service criteria than merely access. Thus, this research seeks to provide further nuance to theories of the role of independence in telecommunications market development. Research on regulatory independence in the 80’s and 90’s was primarily normative, given the limited experience of regulators. Next, qualitative case study research began to emerge and provide insights into independence as a contribution to regulatory quality. As experience with independence grew and data became available econometric studies began to provide systematic evidence of its effects. Here, we seek to contribute to research based on in-depth case studies that can subsequently inform more systematic analyses. In particular, we seek to provide insights into the nature of independence, how and why it changes and how it affects wireless market development in low income countries. These insights may subsequently generate greater nuance in both qualitative and econometric studies. III. METHODS The study takes a comparative, explanatory case study approach [21]. Tanzania and Botswana have been selected as cases for their similarities as well as differences. They are both members of the Southern African Development Community, which generates a common approach to

regulatory governance, albeit a general one. Further, both countries have in recent years experienced relatively strong telecommunications regulatory governance, with Botswana being recognized by the ITU [22] and Tanzania by the 2006 ICT Investment Summit [9]. Furthermore, both countries have recently implemented a converged licensing framework as two of the first countries in the continent. The implementation of such licensing frameworks has the potential to significantly affect wireless market development, as indeed has been the case in Tanzania [23]. In addition to these similarities there are differences as well. First, the two countries come to the strong regulatory positions via different routes, with Botswana having a past characterized by political stability and strong governance throughout the government, as compared to Tanzania that has had a history of political and economic turmoil. Second, most recently while Tanzania’s regulator has maintained its independence, Botswana has decreased [9]. This difference enables a comparative analysis of regulatory quality as well as its link with wireless market development. These dynamics in regulatory governance in both countries, historically and most recently, facilitate comparisons that can identify their underlying factors. Data were collected through 46 face to face interviews conducted in Tanzania and Botswana during September –November 2006, with managers at incumbent and mobile operators and Internet Service Providers (ISPs), as well as with policy makers from ministries responsible for telecommunications and regulators from the national regulatory authorities. Additional data were collected through policy and document analysis from 2006 until present. Data were analyzed through a combined deductive and inductive analysis. The goal of the deductive analysis was to characterize the degree of independence according to the EURI-I index as developed by Edwards and Waverman [5] and changes in policies and market development, while the inductive analysis was used to identify possible underlying factors influencing these phenomena, with a particular focus on the dynamics of regulatory independence (i.e. sustainability of gains in independence), and the role of, and relation between, Minister, judiciary, and other administrative bodies with the regulatory authorities. IV. REGULATORY INDEPENDENCE AND WIRELESS MARKET DEVELOPMENT IN TANZANIA AND BOTSWANA In the following sections the development of regulatory bodies and markets for wireless services are discussed. Both cases start with an overview of regulatory and wireless market developments as related to the implementation of converged licensing frameworks, followed by a more detailed analysis of these countries’ evolving regulatory independence. A. Tanzania Case

431 2

Regulation and Market Development in Tanzania In 1993 Tanzania liberalized its telecommunications sector by splitting its state owned enterprise, the Tanzania Posts and Telecommunications Corporation, into three separate entities, namely the Tanzania Posts Corporation, the Tanzania Telecommunications Company Limited (TTCL), and the Tanzania Communication Commission (TCC). The latter became responsible for regulating the telecommunications market, whereas Tanzania’s ministry responsible for communications - currently known as the Ministry of Infrastructure Development - was responsible for developing higher level policies. Established in 1993, TCC was among the first 30 autonomous regulatory authorities in the world [5, 23]. Nearly simultaneous to this early liberalization process, in 1993 three mobile operators were licensed [24], with two more were added in 2000. These later entrants, Vodacom Tanzania and Celtel, are now market leaders in voice services provision3. The addition of these two new providers spurred growth: While until 2000 subscriber numbers for mobile telephony remained low, since 2000 the number of telephony users has grown exponentially 4 . Nevertheless, adoption numbers still remained low as compared to many other countries. Hence, steps to further liberalize the sector and increase competition were taken. One of the first steps towards full liberalization of the market was the partial privatization of incumbent fixed line provider TTCL in February, 2001. Two companies, one of which was the mobile operator Celtel International, obtained 35% of the shares, the Government of Tanzania kept 36%5, and the remaining 29% went to local and international financial institutions as well as TTCL employees. At the time of the privatization, TTCL was granted a four year exclusivity period for fixed line telephony provision, until 2005. Next, as part of Tanzania’s liberalization strategy, regulator TCC merged with the broadcasting and postal regulators into TCRA, the Tanzania Communication Regulatory Authority, established through the new telecommunications Act in 2003. After TTCL’s exclusivity period ended, in early 2005, TCRA introduced a range of new regulations to further stimulate competition; the most famous one being the introduction of a converged (technology and service neutral) licensing framework. This meant that all operators as of then were eligible to provide both fixed and mobile services: firms

once relegated to particular service categories (e.g. fixed, mobile, value added, etc.) could obtain licenses without restriction to the types of services offered. Consequently, all of Tanzania’s large operators now have a network facility license, network service license, and application service license, which in some instances are complemented by a content services license, alongside frequency spectrum licenses for wireless services provision. The new licensing framework resulted in market entry and an expansion of the range of (wireless) technologies [23]. By May 2006 four new service providers had frequency assigned, and started rolling out mobile services including 3G through CDMA2000 [23]. In addition, one recent (2000) entrant from the Internet access market has expanded its services to broadband and voice services provision, bringing the total number of mobile providers in Tanzania from 5 to 10 [23]. Further, the number of internet service providers rose from 11 to 236. New technologies are reflected in the use of CDMA (due to scarcity of GSM frequency bands), as well as upgrades of mobile networks to third generation technologies (both UMTS and CDMA). Also, most operators indicate an interest in the potential for WiMax rollout, which is globally still in its infancy7. Regulatory Independence in Tanzania: The Roles of the NRA and Minister The growth of Tanzania’s market has been significantly influenced by regulation, the most recent one being the introduction of a converged licensing framework. As already reported by Van Gorp and Maitland [23], Tanzania’s regulatory authority TCRA is characterized by a number of factors that have enabled it to introduce such competition stimulating regulations. Its regulatory independence is one of the key features, as will be discussed in more detail next. The independence of TCRA can be analyzed first by the EURI-I indicators suggested by Edwards & Waverman [5]. Among these factors, those that suggest the TCRA is independent include (1) multi-sector jurisdiction (telecom, postal and broadcasting); (2) multi-member board control; (2) fixed term appointments of board members; (3) appointments of board members are renewable only once; and (4) experience – with 13 years TCRA is one of the older regulatory agencies in the world. However, one indicator does not necessarily imply independence: while TCRA receives its funding directly through licensing fees and levies

2

For a more detailed overview of market developments and Tanzania’s liberalization strategies see 23. Van Gorp, A.F. and C.F. Maitland, Regulatory Innovations in Tanzania: The Role of Administrative Capabilities and Regulatory Governance. Info: The Journal of Policy, Regulation and Strategy for Telecommunications, Information and Media, 2009. 11(1): p. 64-77. 3 By August 2007 market shares for voice providers are: 1. Vodacom with 51% market share; 2. Celtel with 26% market share; 3. MIC Tanzania-Tigo with 13% market share; 4. Zantel with 7% market share; and 5. TTCL with 3% market share. See http://www.tcra.go.tz/publications/telecom.html. Last accessed August 6, 2007. 4 For a detailed overview of number of subscribers over time, see also http://www.tcra.go.tz/publications/telecom.html . Last accessed August 6, 2007. 5 See http://www.ttcl.co.tz/about_history.asp Last accessed August 6, 2007

6 Quote taken from East African Business Week, “TZ Okays More Cellular Firms”, 22 May 2006. See http://www.busiweek.com/index.php?option=com_content&task=view&id= 1582&Itemid=39 Last accessed August 11, 2007. 7 Even though much progress has been made; few challenges have surfaced. Due to the high demand, regulator TCRA was forced to temporarily stop frequency applications for spectrum that includes the common bands for GSM, CDMA, and WiMax, and was forced to review the band plan in 2006. Subsequently, in 2007, plans were initiated to comprehensively survey Tanzania’s ICT infrastructure, with an eye towards determining the extent to which further opening of the market in the future is desirable, and to determine the optimum number of operators in the sector, as well as the degree of (lack of) competition in different segments of the market. See also Last http://www.cellular-news.com/story/31356.php?source=newsletter accessed May 30, 2008.

432

by industry, Parliament and Minister play a role in approval of TCRA’s budget. Nevertheless, as per the EURI-I index, there are a number of indicators that would actually suggest ‘dependence’. For example, Tanzania’s staff number of 97 is far below that in Europe, however for African standards is relatively high [9]. Additionally, TCRA primarily reports to the Minister while the legislature is hardly involved. Parliament does receive TCRA’s budget and annual report from the Minister, but no specific approval powers etc. are stated in the TCRA Act of 2003. The board members of TCRA are also not appointed by Parliament, but through a committee established by the Minister. Further, as per the TCRA Act of 2003, the Minister has the power to engage in many regulatory activities. For example, TCRA is not allowed to award or cancel licenses with an exclusivity period, universal service obligations or any license for a term of more than five years without consulting the Minister. These factors implying certain degrees of dependence vs. independence have also changed throughout the years. In favor of independence is first TCC’s evolution from a single sector regulator into a converged regulator responsible for postal, telecom and broadcasting. Second, in some areas the Minister’s powers and influence in the regulatory process have decreased. Third, during the time that the regulator was purely in control of telecom, under the name TCC, board members could renew their terms an unspecified number of times, whereas currently more stringent rules apply where board members can only renew their appointment once. However, regardless of developments in the direction of more independence for the regulator, the Minister is still awarded an extensive formal role. Hence, the question remains: why is TCRA referred to as an “independent” regulator? Particularly interesting is the fact that the incumbent TTCL remains partially state-owned, which in many countries has led to government protection and delayed implementation of liberalization due to incumbent-initiated court proceedings. First, in Tanzania, little evidence exists of incumbent-initiated court proceedings. In interviews with managers at TTCL’s competitors, including mobile operators and ISPs, attempts at delaying the implementation of liberalization mechanisms through court proceedings were not mentioned, and stands in stark contrast with a country like South Africa [see e.g. 9]. There exist two potential underlying reasons for the lack of judiciary involvement. First, Tanzania is known to have a fairly weak judicial system, which has restrained market development in many sectors due to the lack of (threat of) enforcement, leaving government unchecked by law [25, 26]. This may have refrained TTCL from taking action. A second reason for limited action from the side of TTCL might lie in the problems with management control, which first changed from government to Celtel during TTCL’s partial privatization in 2001. Even though in the years after Celtel extended its shareholdership to 65% shares while government remained 35%, problems in management control led to the government taking back management control. In

July 2007 Sasktel International of Canada started a three year contract to take over management control8. Further, Tanzania’s broader institutional endowment is known for a highly dominant Executive (including President and Ministers) in national policy making processes. While Parliament has powers of scrutiny, due to the large majority of one party in Parliament, at the end the Legislature thus cannot easily hold the Executive accountable [27]. This might explain the limited role of Parliament in telecommunications matters; besides enacting the TCRA Act of 2003, Parliament’s single formal role in relation to TCRA regards to how it may provide funds to TCRA, and that it shall receive TCRA’s annual report from the Minister. However, this does not explain the limited role the ministry has played. The limited attempts at protecting the incumbent may partially be due to the ministry’s lack of resources. Currently, the Ministry of Infrastructure Development employs roughly 100 people responsible for departments spanning from postal to transportation, but as of 2006 only 6 staff are assigned to telecommunications and ICT matters [23]. Consequently, the ministry realistically does not have a strong influence on ICT and telecommunications policy or regulation. As one manager from the private sector put it, “the regulator is trying to compensate for lack of policy from the Ministry’s side”. Furthermore, the manager mentions that the Minister has an overseeing role, but that in practice TCRA is “independent”. Another manager says “frankly speaking, the ministry doesn’t have enough staff. They should actually give more input”. This relative lack of involvement of the ministry in the telecommunications sector might be further reflected in the lack of engagement of operators with the ministry (as opposed to the regulator), as is explained by regulatory affairs managers at two mobile operators. B. Botswana Case Regulation and Wireless Market Development in Botswana The case of Botswana shows a very different history of regulatory quality and wireless market development than Tanzania. Botswana, due to its population of 1.8 million, has a very small market with, besides an incumbent fixed line operator, until very recently only two mobile operators. While seemingly a small number of mobile operators, at the time of licensing in 1996 when market liberalization started and the Botswana Telecommunications Authority (BTA) was established, many thought that only one mobile operator would be viable due to the small population size. By 2007 already 1.2 million people used mobile phones [28]. But nevertheless, it was felt that competition in the Botswana telecommunications sector developed “unevenly across different regions of the country and at different levels” [29, p. 3], as outside of cities and major districts, provision of telecommunications services often remained limited or even non-existent. A new service neutral licensing framework was 8

See e.g. http://wirelessfederation.com/news/sasktel-takes-over-at-ttcl/ Last accessed February 18, 2009.

433

developed in hopes of “this imbalance [to] be corrected” [29, p. 4]. On June 20, 2006 the Minister of Communications, Science and Technology announced the intent to further liberalize the telecommunications sector through five separate changes to license conditions, as well as having the incumbent rebalance tariffs and allowing new entrants to tender for service neutral national licenses [30]. The first five measures were accomplished by March, 2007 through the introduction of the “Service Neutral Licensing Framework in the Era of Convergence” [31]. Prior to the change in the licensing framework, the market was categorized into “Fixed, Cellular, Internet Service Providers (ISP), Satellite and Data etc.” Market segments determined as “non-competitive” were restricted in terms of the number of players within the segment. In Botswana these were Fixed and Cellular, whereas ISPs, data service providers and paging services providers have been determined to be competitive9. Hence, only one fixed and two cellular providers were licensed, while a significantly larger number of ISPs and data providers obtained licenses. The new service neutral licensing framework required the existing fixed and mobile operators (BTC, Mascom Wireless and Orange Botswana) to obtain so-called “public telecommunications operators” (PTO) licenses. Under this license, any one of them is eligible to provide both cellular and fixed services. In addition, all PTO licensees have become eligible to operate the international gateway and to “self-provide” – a relief to mobile operators that previously were required to use incumbent BTC’s backbone. The three PTOs however are still not eligible to provide value-added Internet services. Hence, BTC continues to use its subsidiary Botsnet for Internet services provision [9]. The introduction of a service neutral license nevertheless was a surprise to some market players. As explained by regulatory managers at incumbent BTC, it was expected that a third mobile operator was to be licensed instead of a fully revamped licensing framework to be implemented. BTC applauded this decision, as BTC was “not sure” if it could have applied for a mobile license otherwise – as two managers explain. Interestingly however, three regulatory staff at regulatory authority BTA explain that BTA actually recommended the Minister to license a 2nd fixed and 3rd mobile operator after stakeholder consultation processes. Nevertheless, the Minister had the authority to either accept or refuse BTA’s recommendations. Further, as a manager at regulator BTA explains, the incumbent BTC lobbied the government – concerned that under BTA’s recommended plan BTC could not have a mobile license. Hence, according to a BTA employee, the Minister decided to convert the three major operators’ licenses into one service neutral license. Thus, while throughout the years the growth in ICT connectivity in Botswana has been greater than expected, the introduction of the new licensing framework has not 9

See http://www.bta.org.bw/licensing.html Last accessed March 30, 2008.

generated a significant new impulse to market entry as it has done in Tanzania. Moreover, even though the PTO market segment shall only in 2009 be considered for further liberalization and market entry [31], as explained at the Ministry of Communications, Science & Technology, “The current regime means there is no intent to get more than three main voice operators”. The future licensing of new national network providers will depend on the Minister, who is in charge of awarding licenses. Since April 2008 BTC has started its mobile operations under the brand name Be Mobile. Managers from industry explain that indeed more competition is felt due to the regulatory changes. Perhaps as a result of this, more advanced services are introduced: Mascom Wireless has launched a 3G and 3.5G HSDPA network10, while Orange has introduced the Blackberry in February 2008 on its GSM/GPRS/EDGE network11. And, even though the new regulations had ISPs remain dependent on infrastructure provision by third parties [9], as of July 2008 value added network services (VANS) providers (e.g. ISPs) may tender for Fixed Wireless Access Spectrum in a number of frequency bands [32]. Regulatory Independence in Botswana Similar to the Tanzanian case, Botswana’s regulator BTA scores well on a number of variables of Edwards and Waverman’s [5] EURI-I index. BTA scores well on (1) multi-member board control, with 5 non-executive members; (2) fixed term appointments of board members (4 years); and (3) experience – with 12 years of experience, Botswana has significant regulatory experience. In addition, BTA has both telecommunications and broadcasting departments, and as such may be perceived as a multi-sector regulator. However, there are separate boards for both sectors. There are a number of factors related to independence as identified by Edwards and Waverman (2006) that BTA does not score very well on, however. First, similar to Tanzania, the staff number of BTA is low compared to for example European regulatory authorities, at 70 staff in 2006. Nevertheless, for African standards it is very high; particularly when taking into account the low population size of Botswana (1.8 million) [9]. A second factor that BTA does not score very well on is the lack of involvement of the legislature. Botswana’s Parliament (National Assembly) does not have a dedicated communications portfolio committee, and although the Parliament has passed the Telecommunications Act of 1996 as well as the Amendment Act of 2004, Parliament does not have significant formal powers in relation to BTA. It is only eligible to (1) appropriate money for the Authority’s fund; and (2) receive an annual report and auditing account within 30 days after the Minister’s reception of both [33]. This also implies that there are two other factors not in favor of independence: there is no 10 See http://www.cellular-news.com/story/33201.php. Last accessed September 20, 2008. 11 See http://www.orange.co.bw/press_room/news_page.php?newsID=11 “Orange and RIM Introduce BlackBerry for the first time in

Botswana”, February 14, 2008. Last accessed September 20, 2008.

434

legislative appointment of board members, and BTA reports only to the Minister. Third, BTA’s board members may renew their appointments an unspecified number of times. Finally, the shared roles of BTA and Minister, and BTA’s budget, seemingly are indicative of a limited degree of independence. The Minister has the power to interfere in a number of regulatory areas. First, as per the Telecommunications Amendment Act of 2004, the Minister has to approve all licensing of fixed line and cellular telephone service and may set licensing fees. Additionally, the Minister has the “power to make regulations” [33, B101], and thus may interfere in many areas of regulation. Thus, as much as six indicators have been identified that could imply a low level of independence. Further, some issues are indicative of the dynamics, including reversal, of independence. Especially in its early days, BTA was perceived as a model regulator, and moreover, governance in Botswana in general has been of a high standard. BTA’s exemplary regulatory governance and independence is for example reported in a 2001 report by the ITU which states that “the Botswana experience also offers a number of world models. Among these are that BTA has achieved a high level of independence as measured by the lack of influence from the government in implementing its mandate. Its virtually unfettered authority to license operators and self-financing operation may also develop as a world model. BTA further provides good models of strong legal processes in carrying out its regulatory mandate.” [22]. Additionally, with regard to licensing, the ITU (2001) report even states that “BTA is one of the rare regulatory bodies that has been given almost complete freedom to decide which services are to be licensed, how many licenses should be granted for each service and which operators are to be awarded a license” [22, p. 27]. Not only the ITU has noticed Botswana’s good governance; throughout the region regulators refer to Botswana’s strong governance. For example, a former regulator from South Africa perceives BTA to be a “model regulator”, which by industry managers and regulators across the Southern African continent is seconded [9]. Botswana’s impressive regulatory governance is furthermore acknowledged in previous research. McCormick [34], suggests in her article with the revealing title “Telecommunications reform in Botswana: a policy model for African states” that Botswana has been able to develop a model of policy and regulatory governance known by significant transparency in decision making. However, the amendments to the Telecommunications Act of 2004 have led to BTA’s degree of dependence becoming point of debate. The Telecommunications (Amendment) Act of 2004 substitutes a number of sections of the 1996 Telecommunications Act that give more ‘power’ to the Minister. In particular, these changes enable the Minister to 1. determine the use of surplus funds that accrue to the Authority, 2. make regulations, on the recommendation of the Board, 3. set licensing fees, and 4. approve all decisions on the licensing of fixed line and cellular telephone service [33]. Thus, a number of decision making powers formerly under the authority of the BTA Board have been transferred back to the Minister. Further, the Minister’s control over BTA’s

budget has been observed immediately: in 2006, BTA provided to the treasury Pula 10 million, and in 2007 Pula 2.5 million (respectively about USD $1.5 million and USD $375.000) [28]. Additionally, the government can now take part of the profits of BTA. A manager at BTA suggests that the Minister taking back power is a trend observed in more southern African countries, including Lesotho, South Africa, and Namibia [9]. Indeed, that these developments represent a transfer of power from the regulator back to the Ministry is confirmed by people in the industry and at the regulator as well. However, while theoretically the Minister does have more power, reality might be slightly different due to underlying resource issues. As one of the Directors at BTA states, “even though the Minister legally has more power […] at present that is not an issue”. Further, another Director at BTA indicates, “there is a lot of consultation between the ministry and BTA – BTA has a lot to say. Liberalization was initiated by BTA. The ministry relies a lot on BTA because it is better resourced”. The Director furthermore continues: “The Ministry is really under-resourced. Most work is carried out by BTA. A policy direction should come out, which BTA would then have to implement.” This however does not always happen. As the Director continues, “a problem […] is that BTA basically made the national plan. This is not desirable for checks and balances.” Nevertheless, regardless of these issues, generally speaking Botswana is still perceived by many people, including those directly involved in Botswana’s private sector, as having a very good regulator. V. CROSS CASE ANALYSIS The cases of Tanzania and Botswana bring to bear two key aspects related to independence: the relation between Minister and regulatory authority as well as sustainability of regulatory independence. First, according to the independence indicators by Edwards & Waverman’s EURI-I index [5], Tanzania currently scores better. The difference however primarily lies in terms of board members of the regulatory authority of Botswana being renewable an unlimited number of times. Further, in Botswana, the Minister has significant control over the budget of the regulatory authority. However, in both countries there are shared roles between the Minister and NRA, and in both countries the Minister theoretically has significant room to implement regulations. The case of Tanzania however showed that even though this formally might be the case, it does not necessarily stand in the way of independence in day to day operations. Due to limited resources at the ministry, Tanzania’s regulatory authority has enjoyed full freedom to design regulations, and even has played a major role in drafting policies; which typically is a Ministerial responsibility. While over the years in Tanzania formal independence has increased, in Botswana it has actually decreased. The Minister’s decision to ignore BTA’s recommendation to tender for a third mobile operator but instead to allow the

435

incumbent fixed line operator to start offering mobile services illustrates this point. Thus, the cases provide evidence of independence being a dynamic phenomenon. Under what conditions are reversals most likely? Comparing Botswana and Tanzania one can conclude, although tentatively given the limited number of cases, that reversals are more likely in countries where the government owns a larger stake in the incumbent, and thus has incentives to protect the incumbent. Finally, the cases provided evidence that suggests that regulation has an important role to play in determining the number of carriers and hence competition, which has been shown to have positive market effects in many countries. Tanzania’s implementation of its converged licensing framework with an open approach to market entry led to market entrance of 4 new operators and additionally an impetus to innovation (i.e. the offering of more advanced technologies). Botswana’s approach led to significantly less impetus, even though competition of course was stimulated with there being a third mobile operator now. VI. DISCUSSION AND CONCLUSIONS This study analyzed the nature of regulatory independence and its influence on wireless market development in Botswana and Tanzania, and as such, the study has implications for theories of regulatory independence in low income countries. First, we find that traditional notions and measures of independence may under- or overestimate a regulatory authority’s status, particularly when the broader institutional context is not taken into account. When other administrative bodies, such as the ministry, legislature or judiciary, are weak, independence of an NRA may generate greater freedoms than suggested by their legal status. As reflected in the case of Tanzania, to date these freedoms have generated positive outcomes for market development. However, this freedom exists with minimal checks and balances, which are typically a requirement for stable political systems. This finding suggests that independence of the regulator should not be assessed solely by characteristics of the regulator itself. Measures must further consider: what is the regulator independent from? This may be interpreted as a call to return to studies that examined the broader construct of regulatory quality and not just independence. Indeed, some measures of regulatory quality such as the POLCON index, which measures veto points in a government, reflect the existence of checks and balances. However, the checks and balances related to the independence of the regulator are more subtle. As suggested by Samarajiva [19, p. 81] who described the increasing credibility and independence of the regulator in “The incumbent operator appealed against one of the interconnection decisions, and for the first time the appeal went to the courts, not to the political and administrative authorities as in the past.” In this case the incumbent may have had the right to appeal to the courts all along, but never needed to.

Indeed, Bauer [11] suggests that complaint records and court decisions may be an adequate measure of one component of independence, as long as they are free of a systematic bias. While specific case records may be difficult to obtain and bias even more difficult to assess, as the above quote suggests, a measure that reflects simply the involvement of the courts may adequately differentiate different levels of independence. Also, the case data, together with other instances of reversals, suggest there are a variety of reasons for these reversals, which may be more or less strategic. The case of Botswana reflects a strategic reversal in that it enabled the ministry to create market conditions favorable to the incumbent. However, for example the regulator in Sweden, which was recently identified by the EU as needing greater independence due to a court ruling that resulted in reduced powers for the regulator 12 , reversals can occur for other reasons as well. This together with the previously mentioned declines in independence reported in the EURI-I index (see section 2.3), indicate that independence reversal is an issue common to both high and low income regions. Hence, a more nuanced understanding of the dynamic aspects of independence, as well as the day to day practices in case of shared roles between Minister and regulatory authority, as well as the causes and market effects of reversals is required. Further, insight into reversals in regulatory independence may also contribute to understanding reversals in the more general realm of regulatory quality [see e.g. 35]. Finally, the research findings reported here and elsewhere suggest that regulatory independence has different effects on fixed versus wireless market development. Indeed, the lower levels of regulation in these markets and their rapid growth across all nations, independent of institutional arrangements, has led some researchers to question the role of regulation altogether. First, these questions may be more salient in high income countries where the diversity of checks and balances is in general greater. Second, if indeed the effects of regulation on traditional measures of market development such as teledensity and efficiency are universal, it may be that new measures of market development are needed. The case of Tanzania suggests that measures such as the diversity of technologies and speed of transition to advanced technologies, both likely important to consumer satisfaction and to enhancing the true benefits of wireless technologies, may be fruitful areas for future research. REFERENCES [1] [2]

Saunders, R.J., J.J. Warford, and B. Wellenius, Telecommunications and economic development. 1983, Baltimore and London: Johns Hopkins University Press for the World Bank. 395. Bonciu, F. and H. Williams, The contribution of the mobile industry to economic development in Romania. Info, 2006. 8(5): p. 37-55.

12 See “Commission opens three new cases on independence and effectiveness of telecoms regulators in Latvia, Lithuania and Sweden”, IP/08/1343, September 18, 2008. Retrieved from http://europa.eu/rapid/pressReleasesAction.do?reference=IP/08/1343&type =HTML&aged=0&language=EN&guiLanguage=fr. Last accessed September 22, 2008.

436 [3] [4] [5] [6] [7] [8] [9]

[10] [11]

[12] [13]

[14] [15] [16] [17] [18] [19] [20] [21] [22] [23]

[24] [25] [26] [27] [28] [29]

Williamson, O.E., Credible commitments: Using hostages to support exchange. American Economic Review, 1983. 73(4): p. 519-540. Gutierrez, L.H., The Effect of Endogenous Regulation on Telecommunications Expansion and Efficiency in Latin America. Journal of Regulatory Economics, 2003. 23(3): p. 257-286. Edwards, G. and L. Waverman, The Effects of Public Ownership and Regulatory Independence on Regulatory Outcomes. Journal of Regulatory Economics, 2006. 29(1): p. 23-67. Stern, J., What makes an independent regulator independent? Business Strategy Review, 1997. 8(2): p. 67-74. Varoudakis, A. and C.M. Rossotto, Regulatory reform and performance in telecommunications: Unrealized potential in the MENA countries. Telecommunications Policy, 2004. 28: p. 59-78. Andanova, V., Mobile phones, the Internet and the institutional environment Telecommunications Policy, 2006. 30(1): p. 29-45. Van Gorp, A.F., Increasing Regulatory Capacity: The Role of the Region in Shaping National ICT Policy in Southern Africa. PhD Dissertation, the College of Information Sciences & Technology, the Pennsylvania State University, University Park, PA, USA. 2008, p. 207. Melody, W.H., On the meaning and importance of 'independence' in telecom reform. Telecommunications Policy, 1997. 21(3): p. 195-199. Bauer, J.M., The coexistence of regulation, state ownership and competition in infrastructure industries: Quagmire or Institutional Innovation?, in Quello Center of Telecommunication Management and Law Working Paper Series. 2003, Michigan State University: East Lansing, Michigan. Wallsten, S.J., An Econometric Analysis of Telecom Competition, Privatization, and Regulation in Africa and Latin America. The Journal of Industrial Economics, 2001. 49(1): p. 1-19. Balasooriya, A., Q. Alam, and K. Coghill, The effectiveness of the telecommunications regulatory regime: The case of Sri Lanka Telecom. Public Administration and Development, 2006. 26: p. 383-393. Samarajiva, R., The Role of Competition in Institutional Reform of Telecommunications: Lessons from Sri Lanka. Telecommunications Policy, 2000. 24(8-9): p. 699-717. Henisz, W.J. and B.A. Zelner, The institutional environment for telecommunications investment. Journal of Economics and Management Strategy, 2001. 10(1): p. 123-147. Henisz, W.J., The institutional environment for infrastructure investment. Industrial and Corporate Change, 2002. 11(2): p. 355-389. Gutierrez, L.H. and S. Berg, Telecommunications Liberalization and Regulatory Governance: Lessons from Latin America. Telecommunications Policy, 2000. 24(10-11): p. 865-884. Jain, R., Interconnection regulation in India: Lessons for developing countries. Telecommunications Policy, 2006. 30: p. 183-200. Samarajiva, R., Regulating in an Imperfect World: Building Independence through Legitimacy. Info, 2001. 3(5): p. 363-368. Gutierrez, L.H., Regulatory Governance in the Latin American Telecommunications Sector. Utilities Policy, 2003. 11: p. 225-240. Yin, R.K., Case Study Research: Design and Methods. Third ed. 2003, Thousand Oaks, CA: Sage Publications. ITU, Effective Regulation Case Study: Botswana 2001. 2001: Geneva, Switzerland. p. 48. Van Gorp, A.F. and C.F. Maitland, Regulatory Innovations in Tanzania: The Role of Administrative Capabilities and Regulatory Governance. Info: The Journal of Policy, Regulation and Strategy for Telecommunications, Information and Media, 2009. 11(1): p. 64-77. Moshiro, S., Licensing in the Era of Liberalization and Convergence. 2005, ITU. Mwaikusa, J.T., The Limits of Judicial Enterprise: Judicial Powers in the Process of Political Change in Tanzania. Journal of African Law, 1996. 40(2): p. 243-255. Gloppen, S., The Accountability Function of the Courts in Tanzania and Zambia. Democratization, 2003. 10(4): p. 112-136. Lawson, A. and L. Rakner, Understanding Patterns of Accountability in Tanzania. 2005, Oxford Policy Management, Chr. Michelsen Institute and Research on Poverty Alleviation (REPOA): Oxford, UK. BTA, Annual Report 2007. 2007, BTA: Gaborone, Botswana. Pheko, T.G., Speech by the Chief Executive, Mr T.G. Pheko, at the Award Ceremony of the Service Neutral License to Mascom Wireless

[30]

[31] [32]

[33] [34] [35]

Botswana, at the BTA Offices. 13 june 2007. 2007: Gaborone, Botswana. Venson-Moitoi, P., Press Statement June 20, 2006: Further Liberalization of the Telecommunications Sector. 2006: Gaborone, Botswana. Retrieved September 20, 2007 from http://www.bta.org.bw/Liberalisation/Press%20Statement%20-%20L iberalisation%203.pdf. BTA, Service-Neutral Licensing Framework in the Era of Convergence. 2007, BTA: Gabarone, Botswana. p. 7. BTA, Public Notice Radio Spectrum Licensing Tender No. BTA BTA/PT/02/2008.2009: Public Invitation to Tender for Fixed Wireless Access Spectrum in the Frequency Bands 1785 – 1805 MHz, 2500 – 2690 MHz and 3400 – 3600 MHz. 2008: Gaborone, Botswana: http://www.bta.org.bw/tenders/2008/Spectrum%20public%20announ cement%20(3).pdf. BotswanaGovernmentGazette, Telecommunications (Amendment) Bill, 2004. Bill No. 15 of 2004. 2004: Gaborone, Botswana. p. 3. McCormick, P.K., Telecommunications Reform in Botswana: A Policy Model for African States. Telecommunications Policy, 2001. 25(6): p. 409-420. Kaufmann, D., A. Kraay, and M. Mastruzzi, Governance Matters VII: Aggregate and Individual Governance Indicators 1996-2007, in Policy Research Working Paper 4654. 2008, The World Bank Development Research Group Macroeconomics and Growth Team & World Bank Institute Global Governance Program: Washington, D.C.

437

Social Enterprises: A Vocational Entrepreneurship Framework for Street Youth Paul Javid12, Kentaro Toyama1, Manna Biswas3 Microsoft Research India1, University of California Berkeley2, Praajak3 Abstract – It is estimated that there are 100 million “street children” in the world [1]. Street children typically live independent of families in makeshift living arrangements, and survive on little but their wits and the camaraderie of small gangs. To better understand the lives of street children, we conducted 150 days of ethnographic investigations in and around Calcutta, with an emphasis on street children who live in train stations. Our hope was to understand their typical challenges, and to see whether they could be addressed with information and communication technology. Among other things, this study led to the identification of an unreported knowledge gap among the children. Namely because their world is limited to geographical “islands” within a couple-block radius of train stations, they were entirely unaware of the of microentrepreneurial possibilities beyond their islands. In the second half of this paper, we describe a preliminary trial with an NGO to expose street children to entrepreneurial possibilities using mediated video instruction. Such a system shows promise, if combined with ongoing social support: some children were both quick to understand and try their hand at new professions, and were to continue their professions for at least two months with ongoing NGO support.

E

I. INTRODUCTION

stimates suggest that there are approximately 100 million street children or street youth who live and work on the roads and squares of cities all over the world [1]. These terms connote a wide range of children who are either homeless or who work on the streets but sleep at home (typically slum community). Many remain in contact with their families, are male, and work on the streets to augment their household income. Many others have separated from or were abandoned their families due to war, poverty, family alcoholism, or abuse, and other factors [1].

India has the largest population of the world’s street children [2]. At least eighteen million children live or work on the streets of urban India, doing work as porters at railway terminals, informal tour guides, tea sellers, and so forth [3]. Calcutta, for example, may have as many as 500,000 street children, who typically speak the language of their state (e.g., Bengali in Calcutta), as well as a local dialect of their home community. They have little or no formal education, and live in a larger context of urban poverty. Calcutta has a population of over 14 million people, and a density of 24,760 people per

square kilometer [4]. About 4 million live in one of about 5000 slum communities, and 42% earn between US$132 and $372 dollars per year. These slum communities have an overall literacy rate of 28.5%, and only 3.9% of males and 1.3% of females have graduated from secondary school [5]. Few have access to opportunities, social networks, or marketable skills that would provide them with a steady stream of reasonable income; the majority find irregular income streams from the informal service industry or smallscale industries such as tailoring, carpentry, blacksmithing, or manufacturing [6]. Indian street children are routinely detained illegally, beaten, tortured and sometimes even killed by police, based on the general population’s mistrust of street children, and the corruption of the police force [7] [8] [9]. In this paper, we describe the three phases of research we conducted to provide a technology-based system that provides older street children with information about available incomegenerating opportunities: First, we conducted an ethnographic exploration of street children in and around Calcutta. Second, we tried a couple of short-term interventions to understand the impact they would have on youths who live in train stations. These early investigations led to the novel understanding that street children have an extremely limited view of the job opportunities available to them, due to a kind of tunnel vision about the world around them. Third, we then implemented and observed a preliminary trial of a new system called Social Enterprises, which informs street children of a diverse set of micro-entrepreneurship opportunities. II. AN ETHNOGRAPHY OF CALCUTTA STREET CHILDREN In an effort to understand the life of street children, we began with an eight-month-long ethnographic investigation of street children in and around Calcutta, beginning October 2007. Our goal was both to get a background sense for the life of street children in this region, as well as to identify opportunities for ICT (information and communication technology) interventions. Methodology The first author moved to Calcutta to live near street children for nine months. He stayed in an apartment near Park Street, which is a busy urban street of food stalls and street youth, with whom he had daily interaction. On a more formal basis, he spent over 150 days in the field either in Calcutta or at Malda train station, about 8 hours by train from Calcutta, spending between 2 and 8 hours a day conducting structured and unstructured interviews with local

438

street vendors and micro-entrepreneurs (N=50), street children who lived either on city streets (20) or in train stations (10), and leaders (8 non-profit directors, a dozen non-profit midlevel managers, and 2 primary school principals who work with street children) and field staff (6) from a local nongovernmental organization (NGO) that specialized in working with street children. Observations of the routines and behaviors of these groups were also made throughout the period. Among the street children, we cast a wide net, but focused primarily on subjects who did not claim to possess any formal job skills, expressed interest in wanting a more stable career, had received no formal schooling, were considered too old to re-enter a traditional public education track, and had found alternative means to earn a living and survive. The NGO, Praajak, has presence in four train stations in West Bengal, one of which is Malda. Praajak works with marginalized children and adolescents in west Bengal. Its aim is to create a gender- just environment of care and protection around these children so that they grow up as more “responsible, productive and law-abiding citizens.” Praajak's programs include education, health and social rehabilitation work. Praajak works mainly with younger boys. The interviews focused on general aspects of the lives of street children that are lost in tables of global statistics. Apart from information about their age, gender, and general well-being, specific questions were asked about life as they knew it: What were the children’s relationships to their parents and relatives? Where do they consider home? Why did they leave home? Who are their friends, and how did they meet? Who are the people they can’t trust and why? How do they earn money? How much money do they earn? What strategies do they employ to survive on the streets? Do they have career aspirations? Do they understand what sorts of careers are available? What are their plans for the future? Did they perceive any inhibitors to their plans? Key Findings Life Overall There are two categories of street children (ages 8 to 20) who live in urban cities like Calcutta: (1) slum dwellers, and (2) street dwellers [10]. The largest group of street children is migrant street children who are first or second generation migrants: either they themselves are migrants, or they have migrated with their families [11]. In Calcutta, some of these migrants come from Bangladesh, while a smaller percentage comes from surrounding rural village areas. Most of those who have migrated with their families live in one of 5,000 slum communities in Calcutta (group 1), while those who have migrated alone find shelter on the streets (group 2). While one might assume that children migrating alone is uncommon, worldwide roughly 48 percent of refugees are children [1]. These two groups of street children have significantly different lifestyles, and sources of income [12]. Those in the first group typically share some form of family income from their father and/or mother, while they themselves

are employed as beggars. Such beggars typically wear “costumes” to disguise themselves as homeless and in need of food, and roam the streets with their mothers begging for money. At times such families might ask for consumable food items, but the large majority of food donations is later re-sold and liquidates for cash. For such families, begging is not the main source of income, but such extra income is used by the family to purchase extra consumer goods such as a radio. Those in the second group do not have a primary residence, and therefore find shelter on the streets in the city. This group typically earns less money than those in group 1, through “wholesale vending.” They will purchase small food consumables, such as packets of gum, from wholesaler street vendors (commonly available in cities like Calcutta), bring the food items to high-traffic areas, and resell those items to passerby clientele. For the children at the train stations, a majority of the children come from families who live in villages within a 12-hour train journey. They have been living in train stations from anywhere between 3 months and 8 years. All of the trainstation street youth are self-employed. They have two typical sources of income: (1) the buying and reselling of various goods, such as water bottles or various small food items, at train stations, and (2) service-related jobs such as sweeping and cleaning of train stations. Children might board a train after it stops at the station, sweep, and ask for recompense from the passengers. Typically the children earn between US$0.75 and $1.00 per day for such services. They do not have a concept of savings. They spend a majority of their money on food and entertainment, and what is left over, is usually given to other members of their gang (about which more is below). Entertainment might include the purchase of alcoholic beverages, or watching movies at trainstation “movie theaters”: Some groups will set up TVs and show movies to other youth for a price. The Malda trainstation youth frequented such locations approximately twice a month. While the Malda train station youth had a more stable sleeping situation through space provided by Praajak, most train station youth are nomadic, moving from one station to another, spending a few weeks at any one location. Gangs and Hierarchy Street children as a whole, self-organize into gangs, in order to increase individual productivity and share livelihood goods. [13] Gangs are territorial, and control particular streets or alleys, preventing those outside the gang from doing business in their geographical area. Gangs also provide their members with lines of trust: i.e., the names and locations of street vendors who will sell and/or purchase items from the gang. It also is a form of insurance and protection; not only does the gang protect its members in terms of helping in dire cases with life necessities such as food, but it will train its members to avoid the local police, and if caught, how to react appropriately to avoid being thrown into jail. The gang therefore, is a major broker between any of its members and the outside world: providing information about markets and their vendors, information about police, and information about

439

other children and gangs [14]. While such street gangs are not typically contentious, fights between children are commonplace. Initiation into a group requires into a group requires finding a leader and proving your worth through sexual favors and financial payments [13]. It should be noted that the nature of these gangs is different from the street gangs such as those in the United States or Brazil, who manage their territory with heavy-duty weapons, and whose income sources are typically illegal [15]. In contrast, street children in India live a carefree lifestyle, and are not felt to be dangerous or intimidating by surrounding populations. The gangs at the train station show further structure and organization. There are three tiers within the gangs, classified roughly by age. Leaders, generally between 16 and 21 years old, are responsible for the overall well-being, safety and material well-being of the group. Below leaders, there are the laborers, between 9 and 16 years old, who do much of the day-to-day work of the gang and earn approximately $.50 to $1 per day. While there is no formal payment structure between the laborers and the leaders, there is a general understanding that they need to pay the leaders a percentage of their extra income that is not used as food, typically less than $1 per week. Below the laborers, there are peons who are the youngest children in the group, usually under 10 years old. While peons will probably bring in an equivalent amount of money into the group as laborers, they are not as selfreliant, and typically accompany laborers on cleaning/selling trips when boarding trains. Entrepreneurship Street children have an informal understanding of markets and pricing. They have a solid understanding of the different areas of the city where lower, middle, and upper class city dwellers live, which areas of the city to buy the cheapest good and which areas of the city goods can be resold with the highest profit margins [16]. Some items, such as food perishables, are bought from wholesalers in large markets, and resold to individual customers on the streets, while other items are bought from wholesalers and are resold to street vendors. Such children serve as middlemen in the formal marketplace, commonly supplying goods to local retailers on a per-need basis. At times, after mutual trust has been built between the street child and the street vendor, the vendor might call upon him to source certain goods on his behalf, and be paid a small amount for such services. Train station youth are entrepreneurs. While one might readily assume that given the no-skill, no management, no savings lifestyle of these youth, that they are not “business savvy,” the reality is the contrary. Train station youth are incredibly good at assessing consumer demand for different types of products in different locations, and at different times of the year. For instance, during Diwali (the festival of lights), which is celebrated in India in October, youth understand that the demand for candles and other light-giving instruments will go up, and therefore they go into the business of buying and reselling candles. At times they might even

travel to train stations in areas of India where Diwali celebrations are more concentrated. Train station youth also understand the concept of pricing structures and profit maximization. When they realize that there is a demand for a particular good, say bottles of water, rather than just source one type of water bottle and sell only that, they understand that if they have different kinds of water bottles they might be able to build a price structure which increases overall profit. For instance, they might be able to buy larger water bottles for only a 20% increase in cost, but be able to sell the water bottle for double the price of the smaller ones. Street Smarts Train station youth also have developed acute survival skills that allow them to continue their lifestyle on the train stations [15]. They know the rotations of the police officers, which police officers to avoid at which particular times, which police officers would ask for a bribe, and which ones might be a little more forgiving. They know the times of arrivals and departures of trains within a six train station radius (stops between train stations can be anywhere within 3 – 50 km), and know which trains will likely be carrying more passengers at which times and days. All of this information is pertinent to both maximizing profit and keeping as much of it as they can (without having to pay bribes to local police officers). Finally, train station youth understand that they are not “liked” by all – including the management of train stations. Whenever management comes to a train station to inspect its “cleanliness” all of the train station youth vacate its premises for the day. This partially is to avoid being forcefully thrown out by the local police, but it also is to insure a healthy relationship between themselves and local police officers. Technology Usage Street children at times use pay telephones to keep in touch with their clients. Pay telephones in India cost $0.20 per minute for a local call. There are two types of clients that are typically contacted through with a telephone (1) helperclients, and (2) street vendors. Helper-clients refer to those people in the city who have built some sort of relationship with the street child and are involved in a rehabilitation program of some sort with them. The first author received approximately 2 unsolicited calls from street children per week throughout his stay in Calcutta. Such calls either were to ask for money or for moral support. Street children might also occasionally call street vendors asking them if they require any type of good that they could source for them. Rohit, for instance, on occasion would call his eye glasses street vendor client to see if he would like him to go to the market and buy glasses for him. This call was necessary rather than an in person conversation, because the cost for Rohit to make the call was less than if he would have taken a local bus to meet the local street vendor retailer. Train station children have no interaction whatsoever with technology beyond the electronic billboards available at some train station in India. Microcosmic Perspectives

440

Typical lives of older street youth might look like this: Bablu, 19, is a migrant to Calcutta, whose family lives in Varanasi (a 13-hour train ride from Calcutta). He passed 6th grade and then dropped out of school to earn income for the family. Has strong ties with his family and has family “emotional” support, but decided to move to Calcutta in search of better job opportunities. Upon arriving in Calcutta, given his lack of education, he was unable to find formal salaried employment. Since Bablu comes from Varanasi which is a major tourist destination, he has a smattering of English knowledge, and therefore, earns his main source of income as an informal tour guide in Calcutta. He asks passerby clientele if they would like further information about a particular area of Calcutta or if they would like to be guided to a particular destination. After services rendered he asks for payment of approximately US$1. Bablu claims to make approximately $5 per week. His younger friend, Chotu is 16-years old and a runaway youth, whose family occupies a nearby slum in Calcutta. His father is a fruit re-seller; he buys fruit from a wholesaler and resells it in nearby neighborhoods on a hand-pulled cart. His mother is a house-worker. Chotu has never attended school, and is addicted to cocaine. His main source of income is through buying and reselling packets of gum. He buys a 12-pack of gum from nearby street vendors and hawks single packs around the city. He can make a profit of US$0.10 per packet of gum sold, selling an average of 3 packets of gum per day, or $0.30. While train station youth might be incredibly perceptive of their own local environment, beyond a mile of the train station, the youth seem to have almost no knowledge whatsoever of the business opportunities that might be available to them. If particular types of food, for instance, are only of demand inside the city next to the train station, but not inside of the train station, train station youth would likely have no idea that such opportunities or demand existed. Additionally, their vocations and acquired skills are strictly limited to what they can directly see and experience at various train stations within their immediate reach. III.

ACTION RESEARCH

In addition to the ethnographic explorations, we also conducted small-scale action research to understand the consequences of two kinds of interventions: micro-loans and provision of video-based information. Experiment 1: Loans Some street youth appear to be eager to start businesses on their own, and they actively seek out people who can help them with start-up capital. Of course, many of these children are con artists. Street children, out of necessity, learn how to turn all kinds of “aid” to their short-term advantage in ways which may not necessarily be the intent of the benefactor. Some youth, however, persevere in their requests for help with legitimate jobs and seem genuinely interested in trying an alternate lifestyle. Microcredit, of course, has an established

history and is intended to help the poor start their own businesses, and so one question is what would happen if small loans were made available to street children. In an attempt to understand how such efforts might end up, we provided a loan to two youths that allowed them to start two different businesses: Raja as a shoe shiner, and Stephen as a bicycle rickshaw puller (names have been altered). These were businesses which seemed easy enough for someone to pick up quickly, and which could be undertaken with capital investments of less than US$150. To avoid straight monetary transactions, the youths were provided not with cash, but with the tools of their trade as they felt necessary. Raja thus received a shoe-shining box, creams for shining, shoebrushes, and Stephen received a cart for his bicycle. A vocal agreement was made such that about 20% of revenue would be repaid to the lender at the end of every week. Expected repayment from the bicycle rickshaw would be about $4 / week, and from the shoe-shining box, $2 / week. Raja began his new shoe-shining business within one week, in a high-traffic location with plenty of middle-class passersby. After two weeks in the business, however, Raja reported that shoe-shining was not very profitable. On average, he would earn approximately US$1.50 a day – which he acknowledged was approximately double his previous earnings from various illicit activities, but he was unsatisfied with the amount of work needed to make that extra income. He was working approximately 10 hours per day, every day of the week. Additionally, the local police, who were familiar with Raja as a delinquent, pestered him about licenses and so forth (strictly speaking, one is legally required to have a business license, but this is enforced only at the pleasure of local police officers, who often extort money from those without licenses), and on one occasion, beat his hands to render his shoe-shining work more difficult. Within weeks, Raja had sold his shoeshining equipment (as the author discovered from the acquaintance to whom Raja had made the sale), and disappeared. Stephen’s story was similar. Within weeks of being provided the cart, he, too vanished, presumably after liquidation. There are two key learnings from this experiment: First, transitions to more legitimate activities for street youth can be challenging. It is difficult for one to make a change in lifestyle from one based on gang life and at one’s pleasure, to one that requires daily routine and endurance, even if the result is increased income. Second, from the point of view of running successful interventions, it will probably help to carefully screen potential candidates for entrepreneurship and to provide ongoing social support once they begin their businesses. Experiment 2: Video-Based Information Among information and communication technologies, video is undoubtedly the richest medium for recording and demonstrating activity, at least among those that are available to the mass market. Street children, who are semi-literate at best, find it easy to understand. Video captures the reality of

441

the situation; it is very easy for a viewer to grasp the full context of what is being recorded. In contrast, PCs are expensive and complex, and audio devices and mobile phones are either feature-poor or portable enough to become easy targets for theft. We thus also experimented with producing vocational training video content and tested approaches to video screening and mediation. The first video experiment was intended to better understand their receptivity towards learning about vocational business skills through mediatedvideo. First, one local entrepreneur in Calcutta was identified, an “iron wala,” whose job consisted of sourcing clothing from surrounding neighborhoods, ironing the clothing, and returning the clothing to its owners. The first author, with help from a local community member who spoke both English and Bengali (the language of the iron “wala”) conducted an interview with the ironing wala. The video focused the skills required to run the business (i.e. how to iron, how to source clients, where to save daily earnings, etc.) The author filmed his day-to-day operations of the business including: collecting clothes, ironing clothes, collecting payment, and returning clothes. In total the entire interview and demonstration lasted 13 minutes. The author then selected a local street child, Rahul, of age 18, who had not attended school and whose father was a fruit re-seller in middle class neighborhoods. The first author brought Rahul over to his residence, and on a laptop, mediated the entire video of the local ironing wala, pausing the video every couple of minutes to insure understanding and knowledge retention. Rahul was then asked to evaluate the value of the video by answering the following questions: 1) had you seen an irnoning wala before? 2) did this video help you learn the trade of running an ironing business? 3) if not, would this video help you learn those skills? 4) what else would help you learn the skills to learning ironing and run an ironing business? There were two main observations based on the feedback session with Rahul: First, while Rahul had seen ironing wala’s before, he stated that this video had “good information,” that helped him better understand the business. Rahul stated that numeric information such as the number of clients served per day or the charges for different types of clothing were particularly useful. Second, the video was motivating for Rahul. When asked if he would like to start an ironing business, he seemed skeptical, but thought that if he was going to, that he would have wanted to see this video first for additional information. Key learnings from the feedback session include (1) the importance of video in displaying information about local businesses and (2) information contained in the video can be motivating for street children. The severe constraints of street children make it difficult to imagine technology interventions that would be meaningful for them that don’t devolve into empty or temporary aid. We believe, however, that several classes of projects are possible: „ Monitoring or check-in systems for lifestyle changes (such as a new job). „ Educational tools to support vocational training.

„ Information dissemination about opportunities and local demand.

local

business

„ Social networking tools to connect local youth with mentors and/or entrepreneurs „ Audio-based building.

feedback

systems

for

community

We felt that relatively quick impact could be had by providing information about the kinds of professions that were open to them outside of the small communities they were familiar with. This reasoning, in combination, with our positive experiences with video suggested an intervention project we call Social Enterprises. IV. SOCIAL ENTERPRISES The long-term goal of Social Enterprises is to provide a vocational awareness, training, mentoring and monitoring model to help street children gain self-employment and financial independence. In the short term, we focused on providing vocational awareness via mediated video instruction for relevant, generically available jobs. By “relevant, generically available jobs”, we mean those that have the following criteria: a local business that could provide at least $1.50 of net income per day, initial capital investment below $200, job-skill requirements that can be learned in less than one week, equipment and material needs that can be met locally, and unsaturated market demand in surrounding localities and cities. In its current form, Social Enterprises includes (1) a videocontent production, (2) mediated instruction for dissemination, and (3) a mentoring framework involving a local NGO, established entrepreneurs, and train-station youth. Video-Content Production One of the major components of SE involves the production of video content. Videos produced are informational and instructional in nature. All SE videos involve a formal interview with a local entrepreneur and an informal demonstration of the nature of the business. The first part of the production of the video content involves choosing appropriate entrepreneurs. Once a local entrepreneur is located, an NGO field staff, who is trained in the basics of video recording, sets up a time and day to conduct the interview with the entrepreneur. During the interview the content producer asks the local entrepreneur a series of shortanswer questions aimed at giving the viewer a general feel for the tasks required to perform the particular business venture, and the average revenue and profit of the business. Questions include: What is your occupation? How long have you had this occupation? How did you start in this business? How did you learn the skills for this business? How did you find the location for your business? How many hours / week do you work? What is your average daily income? Etc. During the demonstration portion of the video, the local entrepreneur will demonstrate the activities involved in the business (i.e., if it is a food business, they will show food preparation). The videos

442

are captured using inexpensive, MiniDV camcorders, and tripods. Inexpensive wireless microphones are used to increase audio quality – especially given the loud background noise on major streets in Kolkata. Videos range from 2 min. to 4 min. in length. Video is then edited and stored on a PC. SE has one technician who has basic computer literacy, and is trained on the basics of video post-production. The technician will minimally edit the video, insuring accuracy and clarity of the content before it is store locally. If the content is deemed unusable by the technician, the content is re-recorded. After the post-production process, the video is stored in a local database along with title information and additional metadata. Videos are then given by hand to local NGO partners who work with SE.

where they legally, and socially would not be disturbed. Once the business has been identified, the location for the new venture found, and the materials sourced, the final step before starting the venture requires street youth to learn, first-hand, through day-to-day observation of a local entrepreneur in the same business. This process will likely continue for a day to three days depending on the complexity of the business. After these couple of days, the youth is ready to start his venture, and in collaboration with the local NGO, begins their operations. During the next couple of months, the NGO will monitor the progress of the venture, its revenue and profitability, while ensuring that the youth is actively pursuing the venture on a day-to-day basis, at least 10 hours per day, except on Sundays. Monitoring and evaluation by NGO staff continued at a pace of once a week.

Mediated Instruction The principal means of distributing videos from the SE database to the train station youth is by physically carrying the DVDs to the local NGO who then hands the DVDs to local field staff. A member of the NGO field staff who is trusted by the train-station youth is then assigned as a mediator. The mediation sessions are organized by Praajak. Praajak will inform train station youth from surrounding train stations to come and attend an event at their site location. Between 5 – 100 youth have been present at a given mediation session. Mediation sessions have a dual purpose: (1) to inform train station youth of various types of vocations they could get involved in, and (2) to filter likely candidates for the SE system. During a single mediation session *all* of the videos are shown (at the moment of the last mediation session, only 27 videos had been generated), which is a total of approximately 1.5 hours of video. The mediator’s role is to help the client audience believe that this in fact is a viable business opportunity, and that if they desire, one that can be replicated in a nearby locality. Given the short length of the videos, the mediator generally waits until the end of the video to have discussion about the video and resolve remaining questions, evaluate knowledge gained, etc. Mentoring Framework The mentoring framework includes all aspects necessary to assist an SE candidate proceed through the necessary stages of exploring, identifying, learning, understanding, implementing, and continuing a new business venture. The first step to the mentoring framework involves identifying likely candidates for the Social Enterprises system. These decisions, we leave to the NGO, who has the background and experience to identify those candidates who are sincerely looking for employment. Such a filter is necessary to avoid working with and financing clients who do not have the emotional or psychological ability to successfully start and manage a small start-up enterprise. The local NGO would then work with the street youth to acquire the materials to set up the start-up venture, and find a specific location within that locality (i.e. a street corner),

V. PRELIMINARY TRIAL Social Enterprises was implemented by the first author in conjunction with Praajak over a two-month period. During this time, our goal was to identify at least three candidates for self-employed entrepreneurship, and to observe (1) how they handled the transition, (2) what benefits and challenges emerged, and (3) whether the overall system was worthwhile to continue. Methodology Over a two month time period, two mediation sessions were organized by Praajak. They informed train-station youth from Malda and surrounding train stations to come and attend an event at their site location. A small projector projected the video on a wall in Praajak’s office where the train station youth were asked to gather. The projector was connected to speakers to amplify the sound. Between 5 and 100 youth attended each of the sessions. These sessions serve a dual purpose: (1) to inform train station youth of the various vocations, and (2) to filter likely candidates for further investment and mentorship. During a single mediation session all of the 27 videos we had recorded were shown, which totaled approximately one and a half hours of video. The mediators were instructed to provoke discussion, answer questions, and otherwise convince the audience that the vocations displayed were in fact viable business opportunities in their own locality. Given the short length of the videos, pauses generally occurred between videos. After the video was shown the mediator probed the audience for interest in the business. Some questions asked included “Do you like this business?” “Would you be interested in starting this business?” “What types of information were useful in this video?” Given the large audience, candidates had to be screened before they would be able to proceed to the next phase of the SE system. In rough order of importance, candidates were chosen based on the following criteria: age, trustworthiness, work ethic, business experience, and vision. Candidates had to be at least fifteen years old; this is due to the fact that younger children might still have the possibility to re-enter the formal school environment, as encouraged by Praajak.

443

Ultimately, three youth who indicated interest in the businesses shown in the video were chosen for the next phases, which include an introduction to an established entrepreneur in the chosen lines of business, capital investment by the NGO to start the business, and once-a-week monitoring by NGO staff during the two-month course of our trial. All capital costs were recorded. During the monitoring sessions, revenue information from the new businesses was requested (averages over the week), along with information related to the overall operations of the business. Gathered data included average number of hours worked per day, number of days worked per week, average number of customers who visit and/or purchase from the business each day. Additionally all information related to procuring products for the business was recorded including how much money was spent on which different items for the business to startup, how often did which different items need to be repurchased and how much did they cost. Results Overall, things went well for the three youths. There were fluctuations in their personal commitment to the business, but the regular check-ins by the NGO made a substantial difference. The youths found the new lifestyle a significant adjustment, and they needed continued encouragement by the NGO. They also reported that the videos were a primary source of motivation for them to start their businesses. For the youth, the videos not only demonstrated additional business opportunities, but showcased the lifestyle of an entrepreneur as one who performs the types of tasks they were already used to conducting in their day-to-day lives. Finally, youth reported that the financial assistance from Praajak was another primary reason why they felt they were willing to take the risk of abandoning their previous train station lifestyle, including the guaranteed income and their social circle of friends, moving to a new area and running a business. Value of Video The advantages of video as a medium were further highlighted. Entrepreneurs explained the nature of their business, many of which are not that difficult to start or manage, their income (and the fact that they were able to provide for their families), that it is local (the video visibly shows the streets of Kolkata and speech in the local language), and so forth. In this way, the train-station youth were able to relate to the local entrepreneur, and gain intuition for whether or not the vocation was appropriate for them. Further indications of the value of video were shown in the overall enthusiasm of the youth to start such businesses after only having seen a couple-minute video. Over half of the train station youth expressed their willingness to further explore the possibility of learning more about and/or perhaps starting one such business. Increased Income The greatest incentive for the train station youth to attempt their new businesses was the prospect of increased average

income. While previously train station youth earned between $.50 and $1 per day, after starting and running their own business ventures, average daily income increased to approximately $1.50 per day, an increase of 50% to 100% per day. However, their overall revenue increased by approximately 1500%, which meant youth were managing a much larger amount of money. At the train stations, a large percentage (approximately 90%) of daily revenue was profit, because the goods they were procuring were in many cases free (such as water bottles left behind by passengers). However, in their new ventures, average daily revenue was approximately $12. Most of this money however, was used to recuperate costs, hence only approximately a $1.50 per day of revenue was actually profit. It also should be noted that while the average daily income for the businesses is greater, the income is much more variable. On certain days, for instance, a new business might not have any revenue whatsoever (due for instance to monsoon weather conditions). Such days can be extremely frustrating for the youth. Given that their previous earnings were much more predictable (when they wanted to make money, although little, they likely could find opportunities at the train station), when low-revenue days occur in their businesses, their first reaction is to liquidate assets and return to their previous income earning profession. The average cost of setting up a business and purchasing its first inventory order was $180 dollars. This included various necessary one-time purchase items, such as a stand where the items would stationed and sold on a street corner that accounted for approximately 30% of the overall cost of the business. Youth Reactions Train station youth remained positive throughout the entire process of learning new business opportunities, setting up their own businesses and running their own business. However, youth also expressed equal discomfort with various aspects of their new lifestyle. Two particular areas of discomfort for the youth included change in working and time management lifestyles and change in social scenery. Before beginning their own businesses, train station youth enjoyed the freedom of an “eat what you kill” working methodology: they worked when they wanted, how they wanted, and with whom they wanted. The longer hours they worked, the more money they earned. After the transition, Praajak required the youth to run their businesses 6 days per week and attend to their businesses at least 10 hours per day. This standardized the lifestyle of the youth, with regular sleeping hours, regular attendance to the location of their business and attention to customer needs and demands. Additionally, they no longer had as much time to enjoy the flippant nature of their social lives on the train station. The work was also more detailed oriented: youth had to keep track of their current inventory, and when necessary, work with Praajak to restock supply. These two changes were the primary discomforts for the train station youth who participated in the SE system. On the other hand, youth who were involved felt increased motivation and began to grow a vision for their future. One

444

youth stated “This business opportunity has changed the way that I am planning for my future.” This youth was referring to the increased understanding that he had gained of financial planning as a result of running his business for only two weeks. Overall, youth seemed happy to have become selected investees of Praajak, as they knew that they had been filtered, and chosen to participate in this pilot project. Furthermore, the families of the train station youth expressed gratefulness to Praajak for having chosen their son for this program, and saw the business as a means to long-term sustainable livelihood. Parents, therefore, played a large role in encouraging their children to push themselves and continuing their business ventures. System Adjustments Through the two-month preliminary trial, various areas of potential improvement have been identified for the SE system. One possibility is with the variety in the video content. Currently all of the videos showcase urban entrepreneurs. However, given the heavy support from parents for the program, many of whom live in rural areas, there could be value in producing video content on village-level entrepreneurship opportunities. Such content would be used to encouraging youth to return home and pursue village businesses in their own home communities. Through our trial, it became clear that there is little need for heavy day-to-day involvement from local entrepreneurs: local entrepreneurs do not need to take on a mentorship role with the train station youth in order for them to begin and operate their businesses. Rather, the street youth are savvy enough that with just basic observation they can pick up the vocational and business skills by watching entrepreneurs in action. However, mentorship, monitoring and evaluation remain necessary to motivate the train station youth through typical start-up pains (variable revenue, change in lifestyle, etc). Increased parental involvement and encouragement might be a viable way to sustain these tasks. Overall Conclusions Overall the results of the SE preliminary trial are positive: three youth with no prior experience running or operating a business successfully started, managed, and led their own business venture. Video played a crucial role in informing youth of the various opportunities available and encouraging them of the likely possibility of higher profitability. Support from Praajak staff in video mediation further enhanced video credibility. Furthermore, ongoing mentoring and monitoring from Praajak staff encouraged youth to push through tough business times. Motivation levels of the train station youth were variable throughout the process. While initially there was extreme enthusiasm to start and run a business of their choice, over time, this enthusiasm varied depending largely on the profitability of the venture. Additionally, while the system continues to work well, it is heavily reliant on intervention from local NGO field staff in all stages, including video production, mediation, candidate

screening, business bootstrapping and ongoing support and evaluation. NGO staff also played a large role in accounting procedures and budging for the business opportunities. Such interventions need to be replaced with some other form of monitoring and evaluation. Furthermore, train station youth find the social adjustments difficult; moving to day-to-day hourly operations, spending a majority of their time away from their other train station friends, and at times living and sleeping at new locations. This requires further intervention on behalf of the NGO. In the future, parents of the children in collaboration with local village-level mentors could replace the NGO’s role. One issue is that the profit variability of the businesses can be discouraging. While on average, income is greater, on bad days, it is less. For youth who are used to living hand-tomouth, these instances can be frightening. Overall the SE system seems to work well but remains at a small scale, and still requires continuous monitoring and encouragement of the newly self-employed youth. It is important to note that the train station youth and Praajak have a long history of trust. At scale, developing this level of trust or identifying other organizations with such trust will be difficult. VI. RELATED WORK Vocational Education for Development In recent years there has been increased demand in developing countries, such as India, for public and private involvement in providing skill-based educational programs as a means to provide the poor with skills to lift themselves out of poverty. The government of India launched the urban Self Employment Program (USEP) and The Urban Wage Employment Program (UWEP) in 1997 to assist individual urban poor beneficiaries for setting up self-employment ventures. The program involves participants in the acquisition of vocational and entrepreneurial skills. Additionally, because it is a government subsidized program, participants only need to pay $50 for six months of training. Approximately 150,000 people in India have been trained through this program [17]. The Department of Rural Development in India set up the Rural Development and Self-Employment Training (RUDSETI) that provides skill training to the rural poor [17]. This program has benefited over 150,000 trainees, of which approximately 66% have successfully established entrepreneurial ventures, such as repairmen of two wheeler vehicles, watch and mobile repair, etc [17]. Corporate Social Responsibility initiatives such as that of the Byrraju Foundation provides training services in areas such as healthcare, sanitation, and literacy. The foundation currently works in 156 villages in Andra Pradesh [17]. The Ambuja Cement Foundation (ACF), which works in over 700 villages in eight different states in India, trains youth groups in skills such as bamboo crafts, furniture making, or driving. The non-profit sector is also becoming heavily involved in vocational training for the urban poor in India in

445

two keys ways 1) training basic skill and leaving the trainees to seek wage employment or start their own enterprises or 2) “holistic” training which would include other services such as assistance in availing credit facilities [17]. The Goodwill International Association, for instance, based in Bangalore, trained unemployed youth and school dropouts in various disciplines such as fitting, welding, electric and plumbing. Likewise, Gram Vikas provides training to landless unskilled laborers in masonry or stone welding. Most Gram Vikas clients have managed to double their wages within two years of returning to the workforce. The Self Employed Women’s Association (SEWA) is another prime example that organizes women workers for full self-employment by imparting a variety of technical livelihood, management, and behavioral skills. 960,000 poor women from five states have benefited from SEWA’s services [18]. While such initiatives are noteworthy, it should also be noted that the demand for such services is much larger than such organizations are able to service, and few, if any, street children are easily able to enter into these programs, many of which require formal interactions with government offices. Mediated Video and Education Extension Other initiatives, such as Digital StudyHall and Digital Green have experience extending educational initiatives to larger audiences using tutored video instruction. In particular, such initiatives recognize that while video alone might add little to no value in information dissemination, mediated or tutoredvideo instruction, has shown to add incredible value to the value of information consumption by the poor in India. Tutored video instruction was pioneered by Jim Gibbons in the 1970’s [19]. Gibbons showed that students that came to classes where videos were mediated by a teaching assistant did better than those who just watched the videos passively, and better, even, than those who attended the live lecture. Two initiatives in India, Digital StudyHall and Digital Green, have shown promise in the use of mediated video to extend pedagogy and agricultural instruction, respectively, to the rural poor in India. Digital StudyHall digitally records live classes taught by good teachers in local languages and according to local curricula, collects them in a distributed database, and distributes them on DVDs to poor rural and slum schools which use TVs and DVD players to playback and mediate the content [20]. Digital Green (DG) disseminates targeted agricultural information using video and mediated instruction to buttress a system called agriculture extension, which seeks to spread expert agricultural advice to small and marginal farmers [21]. DG records live training sessions between agriculture extension workers and farmers, collects them in a database, and distributes them on DVDs to farming communities. In villages that receive these DVDs, farmers attend informal training groups where a village facilitator mediates DG content on a regular basis. Digital StudyHall and Digital Green both inspired the initial steps of the Social Enterprises project. SE was in a large part inspired by efforts made in these two initiatives to capture local training and educational programs in local languages and

using local audiences, and spread such knowledge to a larger client audience. Other projects involved in ICT for education extension that also inspired aspects of SE include the Deccan Development Society’s project based in Hyderabad, India which trains the illiterate to film their own videos, and Video Volunteers which partners with Indian non-profits to “empower communities” to address development issues using video [22,23]. ICT for Street Children In countries like India, it is becoming increasingly more popular for young and old, rich and poor alike, to attain IT related job skills. However many of these opportunities are left out of the hands of street children. They typically would not have the financial ability, nor the stability to enter a traditionally rigorous IT related training course. There is however, one example of a project that does IT training specifically for street children – in Ecuador, called iConnect Online. iConnect provides job-related computer and internet skills to street children who otherwise would not have access to such educational opportunities [24]. VII. CONCLUSION Based on initial ethnographic studies and experimentation, we have presented the Social Enterprises system, which combines video-content production, mediated instruction, and a mentoring framework to assist poor street youth move towards long-term financial sustainability by founding and running their own local enterprises. In a two month preliminary study, involving three local train station youth, we found that the SE system, with an initial capital investment of less than $200 dollars can provide youth with the tools necessary to find and begin start-up ventures with average profitability up to 100% higher than previous forms of employment. Local NGO field staff play a significant role in the overall implementation, monitoring and evaluation of the progress of such businesses, and are a key ingredient to the success of the system. However, these results are preliminary, and further study is necessary to better understand long-term emotional and financial sustainability of street youth in the SE system. To investigate further, additional content demonstrating villagelevel business opportunities are needed, along with a more localized and sustainable monitoring and evaluation system. We look forward to continuing collaboration with Praajak and finding increasingly cheap and scalable means of providing vocational tools and skills for the livelihood of millions of street children throughout the world. ACKNOWLEDGMENT The authors would like to thank Loreto Day School Sealdah and Groupe Development South Asia for initial introductions and feedback. Special thanks to Praajak and its executive

446

team for inviting us to work with them and for ongoing support and collaboration. REFERENCES [1] United Nations Childrens Fund (UNICEF) State of the Worlds Children Report, 2006 (New York: Oxford University Press, 2006) [2] United Nations Development Programme (UNDP), Human Development Report, 1993 (New York: Oxford University Press, 1993) [3] United Nations Development Programme(UNDP) , Human Development Report, 1996 (New York: Oxford University Press, 1996). [4] Directorate of Census Operations, West Bengal (2003). "Table-4: Population, Decadal Growth Rate, Density and General Sex Ratio by Residence and Sex, West Bengal/ District/ Sub District, 2001, Census of India [5] United Nations Millennium Project, A Home in the City: Task force report on improving the lives of slum dwellers, Executive Summary, Earthscan, London/Sterling, VA, 2005, p. 10. [6] World Bank, India: Achievements and Challenges in Reducing Poverty, 1997 [7] Human Rights Watch, Police Abuse and Killings of Street Children in India (New York: Human Rights Watch, 1996) [8] "State of the World's Street Children-Violence" www.streetchildren.org.uk. Retrieved February 16, 2009. [9] WHO, World Report on Violence and Health. Geneva: World Health Organization, 2002. [10] UNICEF, (2006) State of the World’s Children 2007: Women and Children. New York: UNICEF. [11] Natai, Kundu. Kolkata, India: Understanding Slums, Case Studies for the Global Report on Human Settlements, 2003. [12] Kidd, S., (2007) “Youth Homelessness and Social Stigma.” Journal of Youth and Adolescence, 36: 291–299. [13] Jones, G., Herrera, E. and Thomas de Benítez, S., (2007) “Tears, Trauma and Suicide: Everyday Violence Among Street Youth in Puebla, Mexico.” Bulletin of Latin American Research, 26(4): 462–479. [14] Droz, Y., (2006) “Street Children And The Work Ethic: New Policy for an Old Moral, Nairobi (Kenya).” Childhood: A Global Journal of Child Research, 13(3): 349-363. [15] Rede Rio Criança, (2007) Criança, Rua e ONG’s: Quem Faz e o que faz? Mapeamento de Ações das ONG’s Junto às Crianças e Adolescentes em Situacāo de Rua no Município do RJ. Rio de Janeiro: Criacāo Gráfica. [16] Evans, R., (2006) “Negotiating Social Identities: The Influence of Gender, Age and Ethnicity on Young People’s ‘Street Careers’ in Tanzania.” Children’s Geographies, 4(1): 109. [17] National Commission for Enterprises in the Unorganized Sector, India, Skill Formation and Employment Assurance in the Unorganized Sector, 2008. [18] Self Employed Women’s Association website, September 2008. http://www.sewa.org/

[19] J. F. Gibbons, W. R. Kincheloe, and K. S. Down, “Tutored videotape instruction: A new use of electronics media in education,” Science, 195(3):1139-1146, 1977. [20] Urvashi Sahni, Rahul Gupta, Glynda Hull, Paul Javid, Tanuja Setia,Kentaro Toyama, and Randy Wang. Using Digital Video in Rural Indian Schools: A Study of Teacher Development and Student Achievement, AERA 2008, New York, March 25-27, 2008 [21] Rikin Gandhi, Rajesh Veeraraghavan, Kentaro Toyama, Vanaja Ramprasad. Digital Green: Participatory Video for Agricultural Extension, ICTD 2007, Bangalore, India. December 15-16, 2007 [22] Deccan development society website, September 2008 http://www.ddsindia.com/www/default.asp [23] Video Volunteers website, September 2008 http://www.videovolunteers.org/ [24] iConnect website, September 2008. http://www.iconnect-online.org/Stories/Story.import4315

447

Speech vs. Touch-tone: Telephony Interfaces for Information Access by Low Literate Users

Jahanzeb Sherwani1, Sooraj Palijo2, Sarwat Mirza2, Tanveer Ahmed2, Nosheen Ali3, & Roni Rosenfeld1 1 Carnegie Mellon University 2 Health and Nutrition Development Society 3 Cornell University

Abstract—Information access by low literate users is a difficult task. Critical information, such as in the field of healthcare, can often mean the difference between life and death. We have developed spoken language interface prototypes aimed at low literate users, and tested them with community health workers in Pakistan. We present results showing that 1) in contrast to previous reports in the literature, well-designed speech interfaces significantly outperform touch-tone equivalents for both lowliterate and literate users, and that 2) literacy significantly impacts task success for both modalities. Index Terms—Speech technology, spoken language interfaces, low literate, illiteracy, information access, community health workers.

L

I. INTRODUCTION

literate users face great difficulty in accessing information that is often easily available to literate users. This is especially problematic in the developing world, where there are many more non literate users, and where the importance of information is often greater. Telephony-based spoken language interfaces offer a unique solution to information access problems for such users, where cell phones are quickly become ubiquitous, and low literacy renders textbased modalities nonviable. In our research on spoken language interfaces for low-literate users in the developing world, we have chosen to focus on a specific domain where information access is especially important: healthcare. Healthcare is a fundamental, yet often under-serviced need of citizens in developing countries. These regions have the highest maternal mortality and neonatal mortality rates in the world, and, not surprisingly, also have the world’s largest unmet need for health service providers. Given the high cost of training doctors and nurses, and the low number of medical schools in these parts of the world, many governments have begun community health worker (CHW) programs, where people (usually women) are chosen from their own communities, trained in basic health service provision for a few months, and sent back to provide health services in their communities. In some countries, especially in Latin America, their effectiveness is quite high, reducing infant mortality to below that of the US [1]. These CHWs vary greatly in literacy levels and receive little refresher training [2]. It is not OW

Manuscript received August 22, 2008.

surprising that the need for better information access by CHWs is widely agreed upon: “Providing access to reliable health information for health workers in developing countries is potentially the single most cost effective and achievable strategy for sustainable improvement in health care” [3]. Over the past three years, we have been researching spoken language interfaces for information access by low literate community health workers in Pakistan. In the first phase of our work we have established the importance of telephonybased information access systems, and pilot-tested one interface with one group of community health workers [4]. Since then, we have conducted a number of user studies testing speech interface prototypes in Urdu and Sindhi, in different urban and rural sites, with community health workers of varying literacy. In this paper, we present multiple lessons and findings from these studies: • While reinforcement of existing CHW skills is often cited by public health officials as their most important educational goal, we show that usability testing should not focus on skills reinforcement (Section III). • We describe novel improvements to the Poor Man’s Speech Recognizer, for rapidly achieving accurate speech recognition in a new language using a standard US English speech recognizer (Section V). • We describe a novel mobile setup (hardware and software) for carrying out field research in spoken interfaces, and argue that having such a modifiable system available at the field site is essential for rapid iterative development with participatory design (Section VI). • We present lessons learnt from three pilot experiments in various urban and rural field sites (Section VII). • We describe a novel method for quickly and effectively teaching novice, technology-shy participants how to use spoken interfaces, and show that an effective tutorial is crucial when conducting user studies on such interfaces. (Section VIII). • We present both qualitative and quantitative results from a comparative user study showing that a well designed speech interface significantly

448

outperforms a touch-tone equivalent for both lowliterate and literate users; and that literacy significantly impacts task success for both interfaces (Section IX). • Finally, we discuss the implications of these results and situate them in the larger context of research in ICT4D and SLT4D (Spoken Language Technologies for Development) (Section X). II. RELATED WORK There have been a number of approaches to GUI design for low-literate users. [5] presents design recommendations for non-literate users of a proposed PDA-like device, with many recommendations involving speech. However, these recommendations are not based on empirical evidence from evaluations with actual semi- or non-literate users – they are derived from a literature review of research on Western users. [6] focuses on extending access to digital libraries by nonliterate users, and also gives a short list of recommendations for such interfaces. However, usability tests reveal that users were not able to navigate information effectively, and result in recommendations for keyword search, audio-based help, and limiting the information set to lessen the cognitive load on users during navigation. [7] describes interface design guidelines, and a text-free interface that performed well in a usability test. [8] describes a PDA-based GUI designed for rural community health workers in India. While this may appear to have similarities to our work, their focus is on information entry, while ours is on information access. [9] describes a system for data entry as well as access to decision support by community health workers in India. This is in the same domain as our project, and has many similarities to our work. However, our focus is on speech interfaces in this domain, while their approach is GUI-based. [10] describes the iterative & collaborative design process for and evaluation of a GUI targeted to low-literate users for managing communitybased financial institutions in rural India. While the principles of GUI design do not carry across well to speech interface design, the collaborative design process described has lessons highly relevant to all interface design in such contexts. Speech interface research has resulted in a number of systems in various domains. While the most well known speech application is probably desktop dictation, this is just one point on a large multi-dimensional space of potential applications that can be made using speech. These dimensions include: choice of device (e.g., desktop, telephony, smartphone), task (e.g., information access, information entry), length of user training (often zero for commercial applications), vertical domain (e.g., stock prices, news, weather), acceptable user input (constrained, open-ended), interaction style (system initiative, user initiative, mixed initiative) and many others. For instance, Carnegie Mellon University’s Communicator travel information system [11] and MIT’s Jupiter weather information system [12] are two often-cited examples of speech-based information access systems usable over the telephone – these are mixed initiative systems that require zero user training, and accept a large range of user inputs, although as in all speech interfaces, acceptable user input is limited at each step. Most commercial

systems tend to be more constrained, since these are easier to build, although exceptions do exist, such as Amtrak’s “Julie” system which is unusually flexible. Contrasted to the above are call routing applications, which are used to direct a caller to a specific operator, given a few utterances [13]. The major push for speech interfaces in the developed world has come from the call center market, and that is what most research has focused on. However, since the needs of the populations that such systems serve are very different, there are entire domains that are still unexplored (e.g., access to books through speech). Thus, there is a need for research in domains relevant to emerging regions, targeted towards the specific needs and abilities of users in these regions [14, 15, 16]. The Tamil Market project undertaken by Berkeley’s TIER group was the first to design, develop and test a spoken language system with low-literate users in a domain (crop information access) relevant to them [17]. Results from a usability study of their speech interface suggest a difference in task success rates as well as in task completion times between groups of literate and non-literate users, though differences were not statistically significant. Further, [18] gives a strong indication that there are differences in skills and abilities between these two user groups, describes the linguistic differences in some detail, and suggests that further research is required to understand the nature of this difference and to derive principles of dialog design targeted towards such users. More recently, researchers at the Meraka Institute [19] have been working on speech and touch-tone interfaces for health information services in South Africa. Preliminary results suggest that touch-tone interfaces may be preferable to speech interfaces. A study by IBM Research India comparing speech and touch-tone interfaces reached a similar conclusion [20]. Taken together, these studies appear to suggest that speech interfaces may not be very useful for low-literate users in the developing world. Based on the results reported in this paper, we strongly disagree. [21] describes VoicePedia, a purely telephone-based speech interface for searching, navigating and accessing the entire Wikipedia web-site. An evaluation comparing VoicePedia with a GUI-based smartphone equivalent shows comparable task success across interface conditions, although the (highly literate) users in the evaluation invariably preferred the GUI alternative. [22] gives an excellent review of the potential contributions of CHWs in the developing world. Finally, [23] describes the difficulties low literate respondents face when asked questions that require abstract thought. III. HEALTH INFORMATION CONTENT Based on our prior ethnographic research, we had initially identified specific health topics on which to provide information through any automated interface [4]. However, our prior work was focused on urban community health workers with a minimum of 8 years of education. Since that time, we have shifted our focus to low literate, rural community health workers. In collaboration with our partner NGO in Pakistan, the Health and Nutrition Development Organization (HANDS), we initially opted to work with

449

reinforcing the material that the health workers were trained on (maternal and reproductive health), which is what a deployed system would eventually need to provide. Additionally, this seemed the prudent choice, as it is preferable to reinforce existing systems and practices than to create new ones. The following issues forced us to rethink this approach: 1. For the participant: In a user study, even though we clearly stated that “this is not a test of your knowledge”, especially when participants are tested on information they are supposed to already know, they believe that it is a test of their knowledge. In our experience, when participants were unable to give answers that they felt they should have known from before, they felt embarrassed and uncomfortable. 2. For the researcher: It is impossible to tell whether a response to a question-answer task is being given based on what the participant found through the system, or from prior knowledge. One way to cope with this issue is to conduct a pre-test of their knowledge, but this would further conflict with the previous issue. 3. For both: Reproductive health issues are extremely taboo in Pakistani society, and are rarely discussed in the presence of males. As the primary author (a male) needed to be present during the user studies, this presented a source of discomfort for user study participants (e.g., they sometimes leaned in to give a response privately to the female facilitator). Based on the above considerations in our pilot tests, we have now shifted to working with content that the community health workers have not been trained on before, without any taboo elements in it. IV. TELEPHONY INTERFACES FOR INFORMATION ACCESS To provide the information identified above, we have built two primary telephony interfaces that we have tested extensively. The first is a purely non-interactive system, which plays back a specific audio clip from beginning to end. This was primarily created as a baseline, to assess the cognitive load on the participants created by the length of the speech segment. The second interface is menu-based. It asks the user to select a given topic (e.g., malaria, diarrhea, or hepatitis), after which they are asked to choose from a specific sub-topic (e.g. general information, signs, preventative measures, treatment), after which they are given detailed content broken down into chunks of three bullet points at a time. The interface was created in two ‘flavors’: one using touch tone input for choosing between the options, and the other using speech input. Here is a sample call for both flavors, translated from Sindhi:

Speech Touch-tone Hello, I’m Dr Marvi, and I’m here to give you health information. What would you like to hear For information on Malaria, about? Malaria, Diarrhea, or press 2, for information on Hepatitis? Diarrhea, press 3, and for information on Hepatitis, press 4. User says Diarrhea User presses 3 Diarrhea. If this isn’t the Diarrhea. If this isn’t the topic you want, say ‘other topic you want, press 0. topic’. [Pause] [Pause] Let me tell you about Diarrhea. As a Marvi worker, you need to know that Diarrhea is a dangerous disease that can potentially be life threatening. You should know about its causes, its signs, its treatment, and how to prevent it. What would you like to learn To learn about the causes of about: causes, signs, diarrhea, press 2. To learn treatment, or prevention? about the signs of diarrhea, [Pause] To learn about a press 3. To learn how to treat different topic, say ‘other diarrhea, press 4. And to topic’. learn how to prevent diarrhea, press 5. [Pause] To learn about a different topic, press 0. User says ‘causes’ User presses 2 The causes of Diarrhea. If The causes of Diarrhea. If this is not the topic you want, this is not the topic you want, say ‘other topic’. [Pause] press 0. [Pause] Let me tell you about the causes of Diarrhea… [gives 3 bullet points on the topic]. To hear this again, say To hear this again, press 1. ‘repeat’. To hear more, say To hear more, press 2. ‘more information’. User says ‘more information’ User presses 2 [The system gives 3 more bullets on the topic, and this cycle continues until there are no more bullets, at which point the following instructions are given.] To hear this again, say To hear this again, press 1. ‘repeat’. For a different topic, For a different topic, press 0. say ‘other topic’.

V. IMPROVED “POOR MAN’S SPEECH RECOGNIZER” For speech recognition, we previously described a “poor man’s speech recognizer” [4], using a robust speech recognizer trained on US English speech. The basic principle of the approach is to map between phonemes in the desired language (Sindhi in our case) and the trained language (US English in our case). Thus a word such as ‘wadheek maaloomaat’ (transliterated Sindhi for “more information”) would be given the following US English phonetic pronunciation: W AH D I K M AA L U M AA DH. In our initially described approach, the choice of phonemes was left solely to the discretion of a language expert. We tested this approach with Microsoft Speech Server (MSS), although the

450

principle would work with any modern speech recognition system. This approach led to reasonable recognition rates, although it was not very robust, and prone to error when tested in the field. We have improved upon our approach significantly by incorporating a novel data-driven method, which we call the “Poor Man’s Speech Recognizer++”. The basic idea is to enable the developer and/or language expert to quickly generate new pronunciation definitions for words by varying any subset of phonemes in a given word’s pronunciation definition, and then testing these variants with limited amounts of data to empirically find the pronunciation definitions that would lead to the highest recognition accuracy. For instance, if the developer is unsure of the optimal choice for the last consonant in the word “maaloomaat”, she could specify a wildcard definition of “M AA L U M AA C?”, where the “C?” denotes an “any consonant” wildcard. Similarly, if the developer wants to test the optimal phoneme choice for the final consonant-and-vowel combination in the word “bachaao”, she may specify “B AX C? V?”, where “V?” denotes an “any vowel” wildcard. These pronunciation entries are automatically expanded to a speech recognition grammar consisting of all possible pronunciations based on the wildcards. Thus, if there are a total of 20 consonants in the phonetic dictionary for the source language (in this example, US English), “M AA L U M AA C?” would be transformed into a list of 20 pronunciations, each with a unique final consonant. This speech recognition grammar is then used to run a re-recognition pass over any sample utterance(s) of the given word, and the best matched pronunciations are then manually chosen by the user to be used as the optimized pronunciations in the final system. If there are multiple wildcards in the same entry, the combinatorial explosion would make it difficult for the speech recognizer to work with such a large grammar. For instance, if the developer was to try the entry “M V? L V? M V? C?”, if there are 20 total vowels and 20 total consonants, this would result in a 20*20*20*20 = 160,000 word grammar, which might be computationally intractable to run recognition on. In our experiments with MSS, such large grammars did not return recognition results even after 10 minutes on one word. A heuristic to solve this problem is to allow the developer to create arbitrary word boundaries, which would reduce the number of combinations in the final grammar. For instance, “M V? L V? / M V? C?” (the forward slash denotes an arbitrary word boundary) would result in a 20*20 + 20*20 = 800 word grammar, which is much quicker to compute. While the final result may lose some accuracy with the introduction of an arbitrary word boundary, it is a useful heuristic that works significantly faster (less than a few seconds for a recognition result with MSS). Using this heuristic, a narrower set of pronunciations can be derived, which can then be tested without the arbitrary word boundary. Preliminary results using this improved approach are described in Section VIII. VI. MOBILE USER STUDIES In our initial work, our prototype interface was running on a

server physically located in Karachi, accessible over the telephone line connected to a separate telephony server. Physically, this consisted of: • Windows server running Microsoft Speech Server, containing all the logic for the information access interfaces, also running a Voice-over-IP gateway • Linux server running Asterisk/Trixbox for Voiceover-IP support • Uninterrupted Power Supply (UPS) unit as backup in case of power failure • Monitors, keyboards, mice, routers, and network/power cables While this worked to some extent, it had the following problems: • Any power outage lasting longer than the maximum UPS backup time could potentially bring the system down. Running a Windows server for the speech components, and a Linux server for the telephony interface meant a high electrical load. • Any modifications to the system could not be made at the field site (often a health center) – they would have to be made in the city, away from the actual users. This did not facilitate iterative design with short feedback loops, nor did it enable participatory design. • Any software/hardware failure would require trained and available personnel at the server site. This was not always possible. • For extended field research, the above problems were compounded, and it became very unlikely for there not to be a problem • The phone line was also prone to temporary blackouts, sometimes for days on end • It was difficult to physically move the entire infrastructure to a remote field site, and such a move would not solve the power problems, nor the phone problem – in fact, a new phone line would have had to be provisioned, which could have taken months Based on the above observations, experiences and constraints, we realized the need for a mobile user study setup, where the actual system would be physically accessible in the field, without the power and telephony issues. This led to the following setup: • Laptop running Windows with Microsoft Speech Server, along with the Voice-over-IP gateway • Linksys SPA3102 device (around the size of a 4port network hub) connected to the laptop through one network cable, and connected to a telephone set through a standard phone cable • Power for the two devices Given the low power requirements for these two devices, we were able to get much longer backup times using the same UPS. Further, the portability of the setup meant it was simple to take it to any field site. Finally, interoperating with an

451

actual telephone set meant that we maintained the same physical interface as before, but removed all the intermediary components that were prone to failure. We tested this system in our final user study, and it worked without a problem. VII. PILOT STUDIES A. Description We conducted a number of pilot user studies over the past year, as described below: Month Jan Mar Jun

Place Memon Goth (town) Umarkot (rural) Dadu (rural)

Avg. Education 5-10 years

Language Urdu

Sample Size 10

<5 years

Urdu

10

<5 years

Sindhi

10

In these studies, we tested the relative effectiveness of printed text against the baseline speech system as described in Section III. The system would only play back audio comprising of an Urdu or Sindhi speaker reading out the text material verbatim. Users were given an information access task (e.g. name any one danger sign during pregnancy), and were then either given the relevant page (e.g. containing a list of danger signs during pregnancy) or played back the relevant audio clip on the telephone, to answer the question. These experiments were meant primarily to validate the content we had chosen (including the choice of language), as well as to provide a baseline against which further work could be measured. B. Findings Information presented orally needs to be short. Both low literate and literate users found it hard to hear long passages of text with the purpose of extracting small nuggets of information. When the length of passages were varied (a few sentences, to a page, to a pamphlet), the task became progressively more difficult. Low literate users were less likely to have ever used a phone. Also, low literate users were more hesitant when picking up the phone (more likely to ask for permission), and were more likely to hold the phone with the mouthpiece too high or too low. The national language is not always optimal. Initially, our partners had told us that Urdu (the national language of Pakistan) was a language that “most” of the target users would be familiar with and that it would be an acceptable choice for the system. The pilot studies showed that Urdu was not understood at all by 50% of the participants in Umarkot, and 66% of those in Dadu. Of the remaining participants, many still had difficulty since they were not completely familiar with Urdu. The regional language is also not always optimal. Based on our prior experience, we tested Sindhi content (text and

speech) in a rural health center in Dadu district (part of the Sindh province). However, our participants all belonged to migrant communities from Balochistan, and were native speakers of a minority dialect of Balochi without any written form. Thus, only those participants who had received at least some schooling had any knowledge of Sindhi (7 of the 10 participants). The remaining 3 participants did not understand Sindhi at all. Subjective feedback needs triangulation. When the nonSindhi speaking participants were asked if they would prefer a system in Balochi, none of them replied that they would – instead saying that the Sindhi system was fine the way it was. This was surprising, as they had not succeeded in any of the given tasks. Further probing and questioning showed that each had a different reason (however valid) for saying this – one said it due to peer pressure, thinking that the others would “blame” her as the reason why the system was not made in Sindhi. Another participant said that she assumed we were talking about official Balochi (unintelligible to speakers of their minority dialect), and said she would prefer a system if it were in her Balochi. This reinforces the need to triangulate all subjective feedback in ICTD research, as the sociocultural complexities inherent in such work are impossible to predict and account for in advance. Speech may be preferable to text, even for a baseline system. 60% of the participants in the Dadu study said they preferred the speech system, while 40% said that both speech and text were equal. No participant expressed a preference for text. Based on the previous point regarding triangulation, we must take this with a grain of salt – however, it is expected that users with limited literacy would prefer a system which doesn’t require reading. Also, there was no statistically significant difference in task success for these conditions in any of the studies – but it is important to note that the speech system was purposefully poorly designed as it was a baseline system without any interactivity. Training and working with local facilitators is essential. Over the course of these studies, we worked with user study conductors from the city as well as from the locality in which the research was conducted. While the local facilitators took more of an effort to train (requiring personalized attention, instead of assigned readings), they were much more effective in the user study process. Primarily, they were able to communicate very effectively with participants throughout the study, and were able to understand and translate their issues and feedback clearly to the research team. Additionally, they had deep knowledge of the community, the local context, and of the specific participants as well – so were able to think of complications before they happened, and were also able to provide extra information on past events when needed. Finally, the linguistic diversity (Sindhi and Balochi) that was required for the Dadu study meant that anyone other than a local community resident would not have been able to communicate effectively with all participants. Thus, we strongly recommend training and working with local facilitators for user studies.

452

VIII. FORMAL USER STUDY DESIGN In September 2008, we conducted a within-subjects user study testing the speech and touch-tone flavors of the menubased system described in Section III. The user study was conducted in Umarkot, Sindh, at a training center for community health workers. Participants were recruited through HANDS, and came from Umarkot and a nearby town, Samarro. A day before the actual study began we conducted a pre-study pilot with 3 participants A. Pre-study Pilot Our initial design was as follows. Participants would be introduced to the broad goals of the study, and the steps involved. Their verbal consent would be requested. Personal information would first be collected, including telephone use, educational history, and a short literacy test where the participant would read out a standard passage and be subjectively rated by the facilitator. They would then be verbally introduced to either flavor of the system (touch-tone or speech), and given a tutorial. After the tutorial, they would be given three tasks, with increasing complexity, on one disease. After this they would be introduced and taught the other flavor of the system, and would then be given three similar tasks on another disease. At the end of the tasks, they would be given a series of Likert scale1 questions to subjectively rate the systems on their own and in comparison with one another. Finally, the researcher and facilitator would conduct a short unstructured interview based on the participants’ experience in the user study. The tutorial for both flavors of the system consisted of three steps. In the first step, the participant would listen in (using earphones connected to an audio-tap2) on the facilitator using the system to complete a task. The facilitator would purposefully make a mistake (choosing the wrong disease) and would then correct it, and successfully complete the task. In the second step, the participant would be given a task to complete, while the facilitator would listen in, giving advice if the participant had any trouble. In the third and final step, the participant would be given 5 minutes to use the system as she pleased. The three tasks were roughly equivalent for both systems. The first task was general: “name any of the signs of disease X”. The second task was specific: “how many people are affected by disease X every year?” The third task was very complex, e.g., “is coughing a sign of Hepatitis?” – note that the answer for the third task was always no, meaning that the user would have to listen through all the signs for the disease, and would then need to deduce that since they did not hear it, it is not a sign. Our findings from this pre-study pilot, covering three participants, were as follows: 1 A standard tool used to elicit subjective feedback from participants. Participants are asked how strongly they agree or disagree with a given statement, by choosing a number, say 1 through 5, to represent their level of agreement. In our work, we adapted this tool for verbal presentation, and used a 3-point scale. 2 Also known as a Telephone Handset Audio Tap, or THAT.

• An effective tutorial is essential. Our tutorial did not teach participants how to use either system well. They were not able to complete the second task (on their own) effectively, and the 5 minute free-form practice was not helpful either. Thus, their performance on the actual tasks was abysmal, as they were not able to even navigate through the system effectively on the given tasks, much less answer the questions correctly. It was evident that we needed a better tutorial. • The tasks were possibly too difficult. Although it is uncertain whether this was due to the problematic tutorial, participants in the pilot were not able to succeed in any of the given tasks, being especially unprepared for the second and third tasks (the moderately difficult and difficult tasks). • The tasks were possibly too abstract. It is well known that low literate users have difficulty with abstract thinking [23]. Even the task of asking a question without any context (e.g. naming any symptom of a disease) is an abstract task. B. Changes to the Study Design Based on the above observations, we made some modifications to the user study design. The tutorial process was increased to three practice tasks instead of two. The “free-style” 5 minutes were removed. Further, each of the tasks was carried out by the participant, while the facilitator listened in on each dialog, and provided successively less assistance. Specifically, the facilitator gave explicit instructions on every step for the first task, less help on the second task, and almost no help (unless the participant was stuck) on the third task. The tasks themselves were shortened (to make up for the lengthened tutorial step) to two instead of three. These two were also made easier – with both tasks asking a “name any X of disease Y” form question, where X was one of: sign, prevention method, treatment method, cause, and Y was either Malaria or Hepatitis. Finally, we thought it may be pertinent to concretize the tasks by using the Bollywood Method [24]. In the Bollywood Method, user study tasks are given a dramatic and exaggerated back-story to excite the user into believing the urgency of the problem. We decided to apply this method to only the first of each pair of tasks. Thus, the tasks were given a back-story along the lines of: “Disease X has become prevalent in Khatoon’s neighborhood. She is worried about catching the disease and wants to know of any one method to prevent the disease. Using the system, find out any one method for prevention of disease X”. After making the above design changes, we conducted the formal study. We requested Sindhi-speaking participants, and worked with 9 participants over 3 days, and after two weeks, followed these with 11 more participants over 3 more days. The order of presentation of the two flavors of the system was counterbalanced.

453

IX. RESULTS Of the 20 participants, two were not able to speak Sindhi at all, and were unable to complete any of the tasks successfully – their data were removed from the final analysis. A. Personal Information Language: Of the remaining 18 participants, it is difficult to classify what language they spoke natively: not only is the local language (Thari) very similar to Sindhi, but there is also significant inconsistency in language and dialog naming. Many participants said they were native speakers of Sindhi, yet their Sindhi was very different from the Sindhi dialect used in the system. The fluidity of local dialects means that it is very difficult to tell with a high degree of certainty what dialect a particular person speaks by simply asking them. Age: The average age was 23 years (SD = 5.3), with a maximum of 32 and a minimum of 17. Years in School & Reading Ability: The average number of years in school was 6.3 (SD = 3.3), with a minimum of 0 and a maximum of 12. 3 participants were completely unable to read Sindhi, 5 were able to read with great difficulty, 7 were able to read with some difficulty, and 3 were able to read fluently. For the purpose of the analysis, the first two categories will collectively be referred to as ‘low literate’ participants, while the last two comprise the ‘literate’ participants. Thus, there were 8 low literate participants, and 10 literate ones. Telephone use: 15 participants had used telephones prior to the study, with 10 participants reporting using a phone at least once every two days. B. Quantitative and Qualitative Results Task success in the speech interface was significantly higher than in the touch-tone interface. There was a significant main effect for the interface type, F(1,68) = 6.79, p < 0.05, with 31 of 36 tasks (86%) successfully completed in the speech condition, and 22 of 36 (61%) in the touch-tone condition. These results are shown in Figure 1.

Fig. 1: Main Effects Plot for Task Success. There were main effects for both interface and literacy on task success. Task success for literate participants was significantly higher than for low-literate participants. There was a

significant main effect for literacy, F(1,68) = 10.61, p < 0.01, with 18 of 32 tasks (56%) successfully completed by low literate participants, and 35 of 40 tasks (86%) successfully completed by literate participants. These results are also shown in Figure 1. Literate participants had a perfect task success rate when using the speech interface. There were no interaction effects of literacy and interface. There was a difference of 25% in task success for both literacy groups between the touch-tone interface and the speech interface. Similarly, there was a difference of 32% in task success between low literate and literate participants. It is striking to note, however, that literate participants using the speech interface had a 100% task success rate (20 of 20 tasks), as shown in Figure 2.

Fig. 2: Interaction Plot for Task Success. Literate participants using the speech interface had 100% task success. There was no strong consensus on which interface was subjectively preferred. 10 users preferred the speech interface, while 8 preferred the touch-tone system. A sentiment echoed by a number of participants was “I don’t use the phone that often, and I am not used to using numbers – I prefer speaking instead”. However, other participants said “I am afraid I will say the wrong thing”, and that “it is hard to speak to it, because I say too much”. These participants understood what they were expected to say, but had a hard time saying it. Some participants said that speech might be problematic if they’re in a crowded area, since the system might end up hearing the sounds around them and get confused. The improved tutorial method worked well. All users were able to complete all of the tutorial steps, even though some took up to 3 tries on one task to get the correct answer. The problems they faced in initial practice tasks were successively corrected over the course of the three practice tasks, such that by the time they began the actual tasks, they were much better prepared to use the interfaces than in the pilot. Low-literate users expressed difficulty understanding the spoken language output from both interfaces. This was expressed only in the semi-structured interview at the end, when asked what main difficulties they faced. P9, for instance, said she understood the facilitator perfectly well, but didn’t

454

understand the majority of what the system said. During her tasks, it was evident that she wasn’t able to understand the instructions given to her by either system – as she was waiting without giving any input on certain prompts for up to 20 seconds at a time before hanging up. On further inquiry, it turned out that while P9 was a native speaker of Sindhi, her dialect of Sindhi (and in fact, the Umarkot dialect of Sindhi) is different from the “official” Sindhi that the system’s voice was recorded in. This includes both the accent as well as the word content – some words are significantly different in the local dialect. Additionally, the content included some Urdu words, which completely threw off the low literate participants. However, it was difficult to get the participants to explain what they found problematic, as they tended to blame themselves for the problems they faced, rather than blaming the system, or the designers of the system, for creating a system that didn’t match her language skills. Finally, it is important to note that when asked if her preference would change if the system was made in her language, P9 said that she would prefer the speech interface if both interfaces had been in her language. This sentiment was shared by other low literate participants for whom the system’s language was difficult to understand. Literate users said that the speech system required them to remember less. When asked why they preferred the speech system, the literate users responded that with the button system, they had to remember both the topic they were looking for, as well as the number they would need to press to get it. In some tasks they weren’t sure what the correct label was (e.g., when hearing the list of options in the task for naming a preventative method for Hepatitis, there was an initial topic titled “methods of transmission”, with the title “methods of prevention” coming later – the first topic was a potentially correct choice), and so they would have to remember two discrete bits of information for any option in the touch-tone case. Speech recognition accuracy was very high. While earlier experiments with the “poor man’s speech recognizer approach” had mediocre accuracy (around 50%), with the improvements described in Section III, the recognizer’s accuracy was 91% for the portion of the data that was transcribed. Specifically, this portion consisted of 150 total utterances, of which 133 were in-grammar (i.e., the user said something that the recognizer should have been able to recognize), and 17 were out-of-grammar. For the in-grammar utterances, 121 were correctly recognized, giving an accuracy of 121/133 = 91%. Further, of the 12 errors, only 2 were misrecognitions, while 10 were non-recognitions. Nonrecognitions are significantly easier to deal with, as the system can respond with “I didn’t understand what you said, please repeat that…” followed by a list of valid options. Misrecognitions are harder to recover from, as they result in the system confidently (yet incorrectly) assuming that the user said something else, and moves the dialog in the wrong direction (e.g., the user says “Diarrhea”, but the system hears “Malaria”, and takes the user to information on Malaria). Finally, of the 10 non-recognition errors, 4 were due to

acoustic issues caused by the telephony interface, which may be solved by tuning the parameters of the telephony interface device. X. DISCUSSION A. Task Success vs. Preference for Interface and Literacy While earlier studies have suggested an effect of literacy on interface use [17, 18], this study clearly demonstrates that literacy is a statistically significant determinant of task success. Moreover, the speech interface enjoyed a significantly higher task success rate than the touch-tone interface both for low-literate participants and for literate participants. Literate participants were able to solve every single task successfully using the speech interface, suggesting that lack of literacy constitutes a serious barrier to performance of these tasks, irrespective of the interface used. One of the surprising, and seemingly contradictory findings in the above results is that literate participants reported that they had to remember less with the speech interface and preferred it, yet low literate participants said that the speech interface was harder, and preferred the touch-tone one, even though they performed better with the speech interface on average. This is a known effect with evaluations of speech interfaces [16] although with continued use, it is expected that user preferences conform to match task success [25]. B. Orality and Literacy One of the frequently occurring themes in our research is that low literacy involves more than just the inability to read and write. Low literacy is the result of less schooling, and the experience of schooling imparts various skills beyond the mechanics of parsing written text, such as learning how to learn, learning the process of abstract thinking, learning to trust forms of knowledge other than experience-based knowledge, learning how to answer tests and exams (similar to a user study), and even learning to make sense of other languages, dialects and accents. We have found Ong’s framework of Orality and Literacy [26] to be a useful lens through which to analyze these issues. Ong proposes orality as a way of describing how people think, communicate, and learn in a culture where writing has not become internalized. Orality theory argues that writing has so fundamentally transformed consciousness in literate cultures that we (literate researchers) are unable to grasp how oral cultures and people operate. [27] summarizes Ong’s work on orality and discusses its implications for both interface design and user study design in developing regions. One of the key recommendations is the need to fundamentally redesign any content that was originally created for consumption by literate readers, since oral consumers require content with very different organization, presentation and context. Thus, orality provides a rich framework to understand why literacy makes a significant impact on task success, suggests ways to improve performance by low literate users, and also highlights the importance of localization.

455

C. Localization Localization refers to adaptation of content to local culture, dialect, terminology and language usage patterns. Although crucial, localization is quite tricky. We have found that even communicating about languages and dialects is non-trivial: a rural participant may self-identify as a “Sindhi” speaker, yet may be unable to understand a “Sindhi” pamphlet recorded by a “Sindhi” speaker from the city. The pamphlet may contain words from other languages (e.g. Urdu), the accent of the city speaker may be unfamiliar, and the dialects of the languages may be substantially different. Low literate participants are less likely to be exposed to alternative dialects, or to other languages, and find the urbanSindhi-accented system’s output more challenging than the literate participants. Even one unintelligible word can throw off a low-literate listener completely [27]. When participants found the system’s Sindhi difficult to understand, they were hesitant to speak at all after many prompts with the speech interface, though when given the touch-tone interface, they did attempt to press buttons – this may be because speech interfaces require the user to expose their confusion more publicly by verbalizing something potentially incorrect, versus pressing a button, which is less open to scrutiny (and social ridicule) than speech. The lesson, then, is that when designing a system for low literate users, it is crucial to choose both the language content and the system speaker (whose voice will speak that content) based on the local spoken dialect of the target user population. If there are multiple languages and dialects within the group of intended users, the system may need to be designed with multiple language or dialect support if low literate users are part of the user group. Further, any testing of the system must ensure that low literate users are adequately represented, as their experience of any system is qualitatively and quantitatively different from that of literate users, as shown by our research. This is substantially different than in the developed world, where one can often expect uniformity in language in a given region, given the conforming effect of schools and of universal access to mass media. Finally, the choice between speech and touch-tone may be a false dichotomy, as it may be optimal to provide both options, and let the user choose which option to use based on their current situation (e.g., when in a noisy environment, users may prefer to use touch-tone, but may switch to speech in a quiet place). This is common practice in the developed world. D. Literacy and User Study Design It is important to note that user study methodologies have been developed primarily with Western, literate participants in mind. Likert scales require the respondent to read and respond to the questions. User study instructions are recommended to be given uniformly, by reading aloud from a script – which is very foreboding and artificial sounding for a low literate user. Finally, the act of asking an abstract question (e.g., name any one sign of Diarrhea) and expecting an answer is also abstract, and would be harder for a low literate participant than a literate, schooled participant. While some work has been done

in this space (Likert mood knobs, Bollywood Method for task specification [24]), these methods have yet to be rigorously evaluated through multi-site experiments. The need to develop and improve methods for such research is urgent, and much work is needed in this direction. E. Significant Design Decisions The system described in this study is the outcome of various design decisions made over the course of more than year of testing various interface prototypes. Most notably, we attempted to optimize each flavor of the interface (touch-tone and speech) as much as possible. Thus, while touch-tone interfaces require a mapping between the specific choice (e.g. diarrhea) and the button to choose that option (e.g. “2”), speech interfaces do not have this requirement. This can be seen when comparing the equivalent prompts from both flavors: Speech: “What would you like to hear about? Malaria, Diarrhea, or Hepatitis?” Touch-tone: “For information on Malaria, press 2, for information on Diarrhea, press 3, and for information on Hepatitis, press 4.” These optimizations meant that both interfaces were optimized in their own right – it would not make sense to “cripple” the speech interface by forcing it to match the less natural phrasing of the touch-tone interface, since this is one of the very advantages of speech interfaces that we wished to test. Similarly, the system was chosen to have a persona (e.g., “Dr Marvi”), since it is much more natural for a low literate person to interact with a (virtual) person, than with an abstract system (e.g. “Health-Line”). Also, the information presented in both flavors was carefully adapted to make the content more conducive for hearing, e.g. simplifying sentence structures, and replacing difficult words with easier phrases. Additionally, the content was designed to give a high-level summary in the beginning, and to successively give both greater detail and larger amounts of content, as the user progressed through the hierarchy. This required a substantial redesign of the content by a content expert (a medical doctor who supervises health worker training), as the existing healthcare material we started with did not follow this paradigm. Finally, we chose a modified form of implicit confirmation strategy for dialog error-handling instead of explicit confirmation. In our approach, the system repeats what was recognized, and asks the user to take action only if recognition was incorrect (e.g. “Diarrhea. If this isn’t the topic you want, say ‘other topic’”). Explicit confirmations, on the other hand, force the user to state whether the recognition was correct or not (e.g. “Did you say Diarrhea? Please say ‘yes’ or ‘no’.”). Explicit confirmation is preferred when recognition accuracy is low, but is too tedious and distracting when accuracy is high, as is the case here. [28]. While not rigorously quantified, we believe that each of these design decisions was significant in making the user interface more usable by our end-users.

456

F. Speech Recognition Quality While speech recognition accuracy has been a persistent problem in our previous work, based on the improvements described in this paper, the system’s recognition accuracy (91%) was comparable to commercial systems deployed in the West that use robust recognition models trained on the language they are used for. We believe that robust speech recognition is a necessary (though far from sufficient) condition for the success of a speech system, and great care needs to be taken to improve speech recognition accuracy when conducting such research. G. The Importance of Effective Tutorials Through the pre-study pilot, we saw that the initial tutorial strategy we made was not at all effective. By improving the strategy, we saw large improvements in users’ ability to access information successfully. With an ineffective tutorial strategy, both interfaces may have been harder to comprehend for all participants, and this might have shifted their reported preference towards touch-tone, based on our earlier hypothesis. In this paper, we have proposed human-guided instruction in which users learn to use the system with a human mentor, and have shown that it worked successfully. Compared with our prior work using video tutorials, the interactivity and individually-tailored nature of the cooperative human-guided tutorial make it a better fit for both low literate and literate users. Further work is needed to rigorously prove it as a formal method for speech interface usability research. H. Rapid Iterative Development In our most recent study, we used our mobile user study infrastructure, which enabled rapid development and modification of the speech system while in the field. This meant that the feedback of local facilitators was used to make both major and minor modifications to the dialog flow of the system. Additionally, it meant that speech recognition tuning could be done locally and quickly. Finally, it was also possible to make minor changes after the pilot, as there were some issues that became obvious only when new users started to use the system. All of this underscores the need for having a system development setup that enables field-based modification of the system. We aim to use this method in all our future work. I. Comparison with Similar Research Our results contradict similar work in the field, most notably the study by Patel et al. [20] testing speech and touchtone interfaces for listening to pre-recorded radio shows and recording audio content for a radio talk show. In comparing our work with theirs, a number of factors need to be considered. First, in our system, the speech-input flavor was more conversational (e.g. “What would you like to hear more information about, diarrhea, malaria, or hepatitis?”) as compared to theirs (e.g., “To ask a question, say ‘question’; to listen to announcements, say ‘announcements’; to listen to the radio program, say ‘radio’”). It is this mapping of keyword to

semantics that touch-tone interfaces are forced to use (e.g. “For information about diarrhea, press 1”), though spoken interaction can avoid this requirement, making the interface more natural. We believe this difference in the interface is very significant for low literate and other technologically inexperienced users. Next, in their study with 45 participants, the only task that showed a significant benefit of touch-tone over speech was the one that required users to record their voice as the goal of the interaction. Speech interfaces that combine restrictive keyword-based grammars with open-ended “say anything” recording segments are very difficult for users [29], since it is not obvious when (or even why) it is not possible to speak in sentences in one part of the interaction, but it is required to do so in another part. Finally, based on our goals (a system for community health workers that can be trained), we were able to spend a considerable amount of time training participants in the user study on both the touch-tone and speech interfaces. Their system was designed and tested for users without any training, which is why their user study did not involve any training beyond a brief introduction. This difference is noteworthy, as even a limited amount of training can make a significant difference to the usability of an interface, as we saw during our pilot study. Thus, when comparing one study with another, it is important to keep the specifics of the design of the interface, study and tasks in mind, as well as of the larger goals of the system involved. Their study is an important and significant contribution insofar as it warns against the design of speech interfaces for tasks involving recording a spoken message in the context of untrained users. However, this should not be extrapolated to mean that touch-tone interfaces are inferior to speech interfaces in the developing world in general. Our study shows that speech interfaces can be significantly better than touch-tone interfaces for a different design of the interface, the task, and the user study. Finally, the study on the OpenPhone interface for HIV caregivers [19] suggests that users express preference for touch-tone interfaces when privacy is an issue. Privacy was never expressed as an important factor by participants in our study, and it is clear that such issues largely depend on the cultural context involved, as well as the specifics of the system’s domain (e.g., HIV vs. neonatal health). Thus, more work is needed to identify exactly where speech interfaces work well and where they do not. XI. CONCLUSION We draw two main conclusions from this work. The first is that the ability to perform information access tasks is strongly hampered by low literacy, and not just because of the inability to read. We derived empirical confirmation that literacy is a significant determinant of success in information access tasks. This by itself is not surprising, but our results further suggest that the problems associated with low literacy go far beyond the inability to read, since they also affect task performance using the speech interface, where no reading is necessary.

457

Our second conclusion is that, at least for some target populations, a well-designed speech interface can significantly outperform a touch-tone interface for both low-literate and higher literate users. Given the potential utility of information access and the pervasiveness of low literacy throughout many parts of the world, we hope to see many spoken dialog systems developed, evaluated and deployed in the near future. ACKNOWLEDGMENT This project was supported by Microsoft Research through its Digital Inclusion Initiative and by the U.S. Agency for International Development (USAID) under the Pakistan-U.S. Science and Technology Cooperation Program. We are grateful to both organizations for making this work possible. We would also like to thank all the members of the HANDS staff that facilitated this research, especially Dr Abdul Aziz, Bansi Malhi, Laal Jaan, and Dr Anjum Fatma. We are very grateful to Safieh Shah, Samia, Arbab, and Lata for facilitating the various user studies. Finally, we would like to thank the community health workers at Wahi Pandi, Memon Goth, and Umarkot who gave their time and feedback. REFERENCES [1]

H. M. Kahssay, M. E. Taylor, P. A. Berman. Community Health Workers: The Way Forward. World Health Organization, 1998.

[2]

S. Hunt. Evaluation of the Prime Minister's Lady Health Worker Program. Oxford Policy Management Institute. http://www.opml.co.uk/go.rm?id=380. Accessed Feb 1st, 2009.

[3]

N. Pakenham-Walsh, C. Priestley, and R. Smith. Meeting the Information Needs of Health Workers in Developing Countries. British Medical Journal, 314:90, January 1997.

[4]

J. Sherwani, N. Ali, S. Mirza, A. Fatma, Y. Memon, M. Karim, R. Tongia, R. Rosenfeld. HealthLine: Speech-based Access to Health Information by Low-literate Users. In Proc. IEEE/ACM Int'l Conference on Information and Communication Technologies and Development, Bangalore, India, December 2007.

[5]

M. Huenerfauth. 2002. Developing Design Recommendations for Computer Interfaces Accessible to Illiterate Users. Thesis. Master of Science (MSc). Department of Computer Science. National University of Ireland: University College Dublin. S. Deo, D. Nichols, S. Cunningham, I. Witten, 2004. Digital Library Access For Illiterate Users. Proc. 2004 International Research Conference on Innovations in Information Technology I. Medhi, A. Sagar, K. Toyama. Text-Free User Interfaces for Illiterate and Semi-Literate Users. Proc. International Conference on Information and Communications Technologies and Development, 2006. S. Grisedale, M. Graves, A. Grunsteidl, 1997. Designing a Graphical User Interface for Healthcare Workers in Rural India, ACM CHI 1997 V. Anantaraman, et al. Handheld computers for rural healthcare, experiences in a large scale implementation. In Proceedings of Development By Design, 2002.

[6] [7] [8] [9]

[10] T. Parikh, G. Kaushik, and A. Chavan, Design studies for a financial management system for micro-credit groups in rural India. Proc. of the ACM Conference on Universal Usability, ACM Press (2003).

[11] A. Rudnicky, E. Thayer, P. Constantinides, C. Tchou, R. Stern, K. Lenzo, W. Xu, A. Oh. Creating natural dialogs in the Carnegie Mellon Communicator System, in Proceedings of Eurospeech, 1999 [12] V. Zue, S. Seneff, J. Glass, J. Polifroni, C. Pao, T.J. Hazen, L. Hetherington, 2000 – JUPITER: A Telephone-Based Conversational Interface for Weather Information, in IEEE Transactions on Speech and Audio Processing, vol. 8, no. 1, January 2000. [13] A. L. Gorin, B. A. Parker, R. M. Sachs, J. G. Wilson. How May I Help You. Speech Communications, Vol. 23, pp. 113-127, 1997. [14] J. Sherwani, R. Rosenfeld. The Case for Speech and Language Technologies for Developing Regions. In Proc. Human-Computer Interaction for Community and International Development workshop, ACM CHI, Florence, Italy, April 2008. [15] F. Weber, K. Bali, R. Rosenfeld, K. Toyama. Unexplored Directions in Spoken Language Technology for Development. In Proc. Spoken Language Technology for Development workshop, SLT, Goa, India, 2008. [16] E. Barnard, M. Plauche, M. Davel. The Utility of Spoken Dialog Systems. Proc. Spoken Language Technology for Development workshop, SLT, Goa, India, 2008. [17] M. Plauche, U. Nallasamy, J. Pal, C. Wooters, and D. Ramachandran. Speech Recognition for Illiterate Access to Information and Technology. Proc. International Conference on Information and Communications Technologies and Development, 2006. [18] E. Brewer, M. Demmer, M. Ho, R.J. Honicky, J. Pal, M. Plauché, and S. Surana. The Challenges of Technology Research for Developing Regions. IEEE Pervasive Computing. Volume 5, Number 2, pp. 15-23, April-June 2006. [19] C Kuun, OpenPhone project piloted in Botswana http://www.csir.co.za/enews/2008_july/ic_05.html, accessed Feb 1st, 2009. [20] N. Patel, S. Agarwal, N. Rajput, A. Nanavati, P. Dave, T. Parikh, A Comparative Study of Speech and Dialed Input Voice Interfaces in Rural India. ACM CHI 2009. [21] J Sherwani, Dong Yu, Tim, Paek, Mary Czerwinski, Yun-Cheng Ju, Alex Acero, VoicePedia: Towards Speech-based Access to Unstructured Information, Interspeech 2007, Antwerp, Belgium. [22] A. Haines, D. Sanders, U. Lehmann, AK Rowe, JE Lawn, S. Jan, DG Walker and Z Bhutta. Achieving child survival goals: potential contribution of community health workers. The Lancet 369(9579): 21212131. 2007 [23] A.R. Luria. Cognitive Development: Its Cultural and Social Foundations. Harvard University Press, Cambridge, MA. 1976. [24] A. Chavan. 2007. Around the World with 14 Methods. http:// humanfactors.com/downloads/whitepapers.asp#CIwhitepaper. Accessed on August 22, 2008. [25] A. I. Rudnicky. Mode Preference in a Simple Data-Retrieval Task. Proceedings of the ARPA Workshop on Human Language Technology. San Mateo: Morgan Kaufmann, 1993, 364-369. [26] W. Ong. Orality and Literacy: The Technologizing of the Word. London: Routledge, 2002. [27] J. Sherwani, N. Ali, R. Rosenfeld. Orality-grounded HCID: Understanding the Oral User. Submitted to the Information Technology and International Development journal. [28] D. Jurafsky, J. H. Martin. Speech and Language Processing. Prentice Hall, 2008. [29] J. Sherwani, S. Tomko, R. Rosenfeld. Sublime: A Speech- and Language-based Information Management Environment. In Proc. IEEE Int.l Conference on Acoustics, Speech and Signal Processing, Toulouse, France, May 2006.

458

The Case for SmartTrack Michael Paik∗ , Ashlesh Sharma∗ , Arthur Meacham∗ , Giulio Quarta† , Philip Smith† , John Trahanas† , Brian Levine‡ , Mary Ann Hopkins† , Barbara Rapchak§ , Lakshminarayanan Subramanian∗ ∗ Courant

Institute New York University NY 10012 † School of Medicine New York University NY ‡ Department of Obstetrics and Gynecology New York Presbyterian Hospital Columbia University Medical Center New York NY 10032 § Leap of Faith Inc. 23 Brink St, Crystal Lake IL 60014

Abstract—Nearly 40 million people in Africa suffer from HIV/AIDS. African governments and international aid agencies have been working to combat this epidemic by vigorously promoting Highly Active Anti-Retroviral Therapy (HAART) programs. Despite the enormous subsidies offered by governments along with free Anti-RetroViral (ARV) drugs supplied by agencies, the introduction and implementation of HAART programs on a large scale has been limited by two fundamental problems: (a) lack of adherence to the ARV therapy regimen; (b) lack of accountability in drug distribution due to theft, corruption and counterfeit medication. In this paper, we motivate the case for SmartTrack, a telehealth project which aims to address these two problems facing HAART programs. The goal of SmartTrack is to create a highly reliable, secure and ultra low-cost cellphone-based distributed drug information system that can be used for tracking the flow and consumption of ARV drugs in HAART programs. In this paper, we assess the potential benefit of SmartTrack using a detailed needs-assessment study performed in Ghana, using interviews with 516 HIV-positive rural patients in a number of locations across the country. We find that a system like SmartTrack would immensely benefit both patients and healthcare providers, and can ultimately lead to improved patient outcomes and better accountability. Index Terms—SmartTrack, telehealth, drug tracking, drug monitoring, patient adherence

I. I NTRODUCTION According to the World Health Organization, effective HIV/AIDS care requires antiretroviral therapy (ART) [1] as a treatment method. Without access to antiretroviral therapy, people living with HIV/AIDS cannot attain the fullest possible physical and mental health and cannot play their fullest role as actors in the fight against the epidemic, because their life expectancy will be too short. While ART is commonplace in developed countries, these life saving medications have reached only a small percentage of the more than 40 million Africans infected with HIV or suffering from AIDS [2]. In order to be successful in combating this deadly disease, strict adherence to a highly active antiretroviral therapy (HAART) [3] program must be followed.

Despite the subsidies offered by pharmaceutical companies for expensive AIDS drugs, and the enormous effort of African governments to promote HAART programs, these initiatives have not been adopted on a wide-scale. The fundamental barriers to large-scale adoption are two-fold. The first problem is one of lack of accountability in the system, due to the theft of expensive ARV drugs, counterfeiting of drugs, and corruption. The Global Fund crisis that gripped much of Africa, with losses amounting to hundreds of millions of dollars was directly related to the lack of accountability in the system. In fact, many pharmaceutical companies that provide subsidized AIDS medications are demanding better accountability practices as a prerequisite to their continued participation. The second serious problem relates to the lack of patient adherence to the medication regimen. A typical HAART program requires every patient to consume two to five medications per day; in addition, required medication and dosages may change rapidly based on the side effects observed by the patient. Lack of medical oversight, or inappropriate use of ARVs not only harms the health of patients but may also encourage drug-resistant strains of HIV, posing a substantial public health risk. Hence, to improve patient outcomes, it is essential for doctors to continuously observe the health status and the medication consumption regimen of patients. In fact, it is relatively common for doctors and health-workers to physically track patients who have not reported for regular medical visits. In this paper, we motivate the case for SmartTrack, a project which aims to address these two fundamental problems with HAART programs. The vision of the SmartTrack project is to create a highly reliable, widely available and ultra low-cost cellphone-based distributed drug information system that can be used for tracking the flow and consumption of ARV drugs in HAART programs, ultimately leading to improved patient outcomes. The motivation to use cellphones in SmartTrack to track the flow of AIDS drugs is in large part driven by the explosive

459

growth in the adoption of low-cost cellphones in the rural developing world. For example, around 80% of Rwanda has cellphone coverage [4], [5] and the cost of cellphones is so low (less than $50) that the devices have become affordable by the poor [6]. Cellphones have enormous potential for enhancing rural healthcare, given their ability to act as a low-cost computing platform for distributed applications. The open-source movement in cellphone software [7], [8], [9] has also opened up the potential for a wide range of new applications targeting these low-cost devices. The vision of SmartTrack is to enable patients, healthworkers and doctors to use cellphones to record, track and transmit information about pharmacotherapy utilization. The idea is to tag every ARV drug bottle with a “smart” tag: an RFID or a barcode, that will enable patients to identify drugs easily and also enable them to report their drug consumption remotely using a cellphone. The drug information will be compiled on the device, sent to a central server, and stored at different levels (regional level, district level) to be used by the healthcare teams to tailor patient-specific therapies. In addition, the device will also be able to remind patient endusers about when to take their specific medications. Thus, SmartTrack will provide the patient with information about his or her specific regimen, and provide the care provider with information about patient compliance to the regimen. SmartTrack is based upon eMedonline1 , a technology previously developed and patented by our partner, Leap of Faith Technologies, Inc. [10]. This technology leverages cellphones and RFID technology to optimize medication compliance, track medication usage and extend patient care to remote settings. The previous project has been successfully used in the US for tracking adverse events and improving compliance with oral antineoplastic (anti-cancer) therapy; efficacy is currently being demonstrated among cardiovascular disease populations. A key requirement for the success of the SmartTrack is to first assess needs and user acceptability. In this paper, we describe our experiences in realizing this first important step in the SmartTrack project vision. We recently performed a detailed needs-assessment study, interacting with 516 AIDS patients in rural parts of Ghana to determine the potential role and effectiveness of cellphones in improving AIDS care delivery. To this end, we developed simple cellphone-based healthcare applications which we used as part of our study to understand user acceptability issues. Our study was conducted by medical students who interacted with each patient for over 30 minutes and used a translator (when language was an issue) to obtain answers to a detailed survey in combination with user interface testing. Based on our detailed patient study evaluations, we determined that a system like SmartTrack would immensely benefit both patients, healthworkers and doctors and can ultimately lead to improved patient outcomes and better accountability. 1 developed by Leap of Faith Inc., with support from the National Cancer Institute (Contract No.HHSN261200644005C) and the National Institute on Aging (Grant No.2 R44 AG022271-02A1)

II. M OTIVATION AND R ELATED EFFORTS The SmartTrack project vision is centered on three primary premises: (a) the need for better accountability in ARV drug distribution; (b) the need for improved patient adherence to medication regimen in HAART programs; (c) the utility of cellphones as a healthcare platform for monitoring the flow and consumption of drugs. In this section, we discuss these three aspects in more detail, along with the corresponding related efforts. A. Maintaining Accountability in the Supply Chain The first premise of SmartTrack builds on the position of the World Health Organization (WHO), which has argued for the need for a drug tracking system that can monitor the flow of medications from the supplier to the patient. The high cost of ARV drugs makes them an attractive target for theft. This is a particular threat in developing countries where corruption can be widespread and persistent. International HAART programs are not immune to theft, as was illustrated by the Global Fund scandal in Uganda in 2006; in that incident, an investigation revealed that ”tens of millions” of dollars in grants intended for AIDS treatment were misspent by officials, often to finance lavish lifestyles [11]. Similar problems have also been reported in other recipient nations [11], and the widespread theft of medicine for resale has been exposed in a number of instances [12], [13]. Perhaps more disturbing than outright theft is the growing practice of replacing legitimate medication with counterfeit drugs. While it is difficult, if not impossible, to determine the actual scope of this problem, WHO has estimated that in parts of Africa, Asia and Latin America, over 30% of all drugs sold are counterfeit [14]. Consuming counterfeit medication can lead to drug resistance and death, and many counterfeits have been found to contain highly toxic substances. In addition to the serious health consequences that can result from fake or stolen HAART drugs, the phenomenon could also lead to a loss of confidence in the health system as a whole. This threatens to undermine support for aid programs, as well as discourage patients from seeking out and adhering to treatment. A supply-chain management system that could track medicines from their acquisition point to the end user would be able to detect drug theft quickly, and would make it easier to discover sources of counterfeit drug substitution within the supply-chain. This added accountability could greatly improve the safety and effectiveness of HAART programs in developing countries. B. Tracking Patient Adherence and Symptoms Proper adherence to a HAART regimen is critical to prevent drug resistance and ensure survival [15]. However, HAART treatment requires multiple drugs, taken multiple times per day, and often results in unpleasant side effects. These difficulties, combined with the high cost of the drugs and the potential stigma of being identified as HIV+ can act as a deterrent to proper adherence. A number of studies have attempted to measure adherence rates in Sub-Saharan Africa,

460

and they have produced promising numbers, with an average of 77% of participants reporting proper and consistent use of the medication [16]. However, these studies rely on unverified self-reporting, which has been demonstrated to exaggerate the level of compliance in studies conducted in developed countries. Tracking adherence is especially difficult in developing regions, where lack of medical, transportation, and communication infrastructure can limit contact between patients and health-care workers [17]. For example, our study of HAART patients in Ghana found that more than 37% of participants went at least three months between clinic visits, that over 94% had never placed a phone call to their clinic, and that the same number had never received a phone call from their clinic. This lack of contact can make it very difficult for health workers to keep tabs on patient adherence and progress. Likewise, it can prevent patients from notifying health workers of critical symptom information that might require a trip to the clinic or a change in treatment. A low-cost method of tracking medication, adherence, and symptoms in developing countries could provide great benefit to its users. By preventing fraud, it could save precious financial resources for providers, and ensure that medicine gets to those who most desperately need it. By tracking adherence and symptoms, it could improve the responsiveness of healthcare workers to the needs of patients, and ultimately improve and prolong the lives of those receiving HAART treatment. C. Cellphones as a Healthcare Platform In the past few years, several research and developmental efforts around the world have explored [18], [19], [20], [21] the use of cellphones as a potential tool for improving healthcare in both the developed and the developing world. For a detailed overview of globalization and health related issues, please refer to [22]. In developed nations, the problem of tracking medication compliance has been addressed successfully with smartphone-based telehealth tools [23]. Recent growth in cellphone penetration in developing regions has made mobile telehealth solutions a real possibility for users in less-developed nations. In Sub-Saharan Africa in 2006, the overall cellphone penetration level was estimated at 15%, and it has more than doubled in the last two years [4], [5]. Furthermore, certain countries, such as South Africa, have attained adoption rates of over 70% [24]. In our own study, conducted among HAART recipients in Ghana, we found that cellphone usage was quite high in the cities and relatively large even in rural areas: Nationwide, 54% of participants reported using cellphones, with more urbanized areas such as the Greater Accra (77%), Northern (67%) and Volta (57%) regions ranking higher in usage than rural areas, such as the Central region (29%), and Upper West (26%). Several groups have examined the use of PDAs and cell phone-based tools to augment patient monitoring relative to ARV therapy, and to extend medical follow-up capabilities [17], [25], [26], [27], [18], [19], [20], [21]. One important project was led by Satellife [25], which tested the use of PDAs

in healthcare environments in Ghana, Uganda, and Kenya. The project put PDAs into the hands of physicians, medical students and community volunteers in different settings in order to demonstrate their viability and usefulness, especially for the collection of health data and dissemination of medical information. Bridges.org, a technology NGO, performed an independent evaluation of the Satellife PDA trial [28], in which it validated the use of handheld computers in healthcare environments in Africa. Specifically, it found handheld computers to be an appropriate technology for use in the African context, and concluded that they provide an inexpensive alternative to PCs in terms of computer power per dollar. The technology of Satellife was found to be simple to use and easily integrated into the daily routines of the healthcare professionals. Cell-Life [29], [30], a research team at the University of Cape Town, has demonstrated the effectiveness of an information gathering system using cellphones for improving the health outcomes in ART. Their system leverages health workers to collect real-time information on HIV/AIDS along with the spatial infrastructural requirements (person, their environment, and access to basic amenities). Cell-Life has been deployed in South Africa and Zambia. III. S MART T RACK The SmartTrack project was in part motivated by prior successful research and development by our partner, Leap of Faith Inc., of a cellphone-based telemonitoring system for improving medication adherence. This system, called eMedonline2 uses smartphones equipped with RFID scanners to read smart-tags attached medicine bottles. The phone provides reminders to the patient, and the patient scans each bottle with the phone when taking that particular medication, updating a remote database and providing doctors with detailed information on regimen adherence. The system validated the integration of cellphones and radiofrequency identification (RFID) as a therapeutic solution to medication compliance and supply chain management. Feasibility and functional tests of the system in a sample of oncology patients demonstrated that drug and compliancerelated data can be reliably collected, analyzed, and exported for use in other clinical monitoring systems. Patient acceptance and value of the system was very high. Building on this prior success, the SmartTrack project vision aims to develop and deploy a cellphone-based telemedicine and supply-chain sytem that exploits this trend, with the twin goals of tracking the flow of drugs to the field, and of monitoring the status and adherence of HAART patients. In addition to tracking adherence, we also envision a system that acts as a reminder system, that allows users to record symptom information which can be tracked and analyzed by their doctor, and that allows health workers to contact patients. There are, however a number of limitations on the ground that prevent 2 developed by Leap of Faith Inc., with support from the National Cancer Institute (Contract No.HHSN261200644005C), and with support from the National Institute on Aging (Grant No.2 R44 AG022271-02A1)

461 Trusted authority Pharmaceutical / Subsidized drug

Health organization

SmartTrack based tracking

us from simply deploying a modified version of our existing system or equivalent in rural Africa. • The prohibitive cost of smartphones: Smartphones are quite simply out of reach for most users in developing environments, so we must design a system for standard, low-end cellphones. • Limited data communication infrastructure: We must rely on simple Short Message Service (SMS) protocols to communicate between the local client and the central database. • Cost of communication: Every 160-byte message costs the patient money, so we must minimize the number of messages and be economical with data. • Language: In our Ghana study, we found speakers of 11 different languages, many of whom required interpreters to conduct the interviews. Our system would have to accomodate all of them. • Illiteracy: A great proportion of potential users are unable to read or write in any language, requiring us to develop a simple, non-textual interface. Later, we describe our detailed needs-assessment user study based on interactions with HIV positive patients in Ghana that seeks answers to these specific questions. Based on our user study assessments, we feel that it is possible to design an appropriate and usable SmartTrack system that can be used by (potentially illiterate) users, health workers and doctors to improve accountability and patient adherence.

local shops

Regional Hospitals

Rural clinics

Healthcare workers

rural patients

A. SmartTrack Architecture Figure 1 illustrates the supply chain flow of drugs in SmartTrack. Drugs in the supply chain system flow through a hierarchy of suppliers, distributors and customers. In SmartTrack, we require a mechanism to track the flow of drugs at every level in the drug supply chain system. The basic architecture of SmartTrack can be described as follows. Every drug bottle is tagged with a “smart” tag (RFID or barcode) which uniquely identifies a bottle. All the identities in the bottles should be generated by the supplier of the drugs. Each such identity should be generated from a secret key that makes it hard to generate new valid identities to primary deal with counterfeiting of bottles. The trusted central authority provides the medicines or drugs along with the tag (RFID or barcode) embedded in the bottle for tracking the flow of each drug. The Tag information of every item or supply is stored in the central server. Each intermediary point in the supply chain is managed by an Agent who is equipped with a cellphone and a tag reader (barcode or RFID reader) . Upon receipt of any goods, the Tag information is captured by the corresponding agent in the supply chain, who stores the Tag information in the cell phone. The cellphone acts as the local-store at each supply chain point. The central server or Trusted authority uses cellphones to remotely track the flow of goods. This type of remote authentication is possible by transferring the Tag information attached to the goods, from a cell phone to or from the central

National Hospitals

A generic design of tag based supply chain management system. The Trusted authority provides the Tag along with drug bottles. The users at each point in the supply chain use the cell phone based tag reader to identify and authenticate drugs. Trusted authority can remotely track items at every node in the supply chain and check for fraud using the cell phone based system. Fig. 1.

server using SMS/MMS or GPRS connectivity. To verify the authenticity of a tag, an agent has three options: online, offline, batch authentication. In online authentication, the agent signals the tag information to the central server using the cellular network and verifies the authenticity of the tag. In offline authentication, the agent prefetches the list of “admissible” tags from the server into its local-store and verifies each tag with the set of admissible tags. In batch authentication, the agent can assume that the goods are genuine and merely store the tag information in the cellphone locally. In this scenario, the agent can collect several tags and perform bulk verification in a lazy manner. An important benefit in this delayed batch verification is that it can significantly reduce the communication cost using cellphones and does not require immediate connectivity. In summary, this architecture represents a distributed data-store of cellphones, where each node maintains a local store of authorized signatures and nodes communicate with each other using cellular connectivity. While what we have described is a basic overview of the

462

SmartTrack architecture, we need to address several technical research challenges to make SmartTrack into a low-cost, secure, highly reliable and widely available system. We briefly outline the basic technical ideas that we intend to use in SmartTrack to achieve these properties. 1) Information aggregation: Using an inbuilt network cost model, SmartTrack will intelligently aggregate multiple updates from a cell-phone and semantically compress (not standard compression) to the smallest number of messages possible to reduce transmission costs. 2) Strong identity and secure updates: To provide strong security properties, the system will leverage public-key and symmetric-key cryptographic operations to develop an unforgable identity for every cellphone and also provide the ability to verify the validity of individual updates using aggregate signatures. 3) Privacy: To provide privacy, the system will develop an anonymized indexing system that cannot be traced back to individual patients without appropriate information. 4) Reliability and Availability: the system increases availability without incurring additional operational cost by maintaining a detailed local data-store within every cellphone that contains most of the appropriate information required by the end-user (health-worker, doctor or patient). This local-store is constantly synchronized with the main information store either using aggregate updates or physical synchronization. To improve reliability, SmartTrack will intelligently replicate and partition critical information in the data store between the local server and the smart-phones. While these technical issues are certainly important, a detailed discussion of them is outside the scope of this paper. The primary focus of this paper is to make the case for why rural users need a system like SmartTrack, and, if provided, will they actually use such a system. The rest of the paper will focus on answering these user-centric questions. IV. N EEDS -A SSESSMENT U SER S TUDY When working on any “technology for development” project, two fundamental questions always arise: the need for the technology and user acceptability of the technology. Many projects in the development space have been successful in the beginning phase but have often not been adopted on a wide scale. It is this barrier that we wish to cross in SmartTrack. Hence, we began with the endeavor of performing a detailed needs-assessment study to answer the two questions about need and user acceptance: (a) Do rural users need a system like SmartTrack? (b) If provided with a system like SmartTrack, will they use it? Rather, how would one build SmartTrack to maximize user acceptance. To address these questions, we began with the modest goal of designing a detailed survey covering a wide-range of questions. We also designed two simple user-interfaces as part of this study: Smartphone based system: We tailored our smartphone based telehealth system for cancer patients into a system for

AIDS patients. This system used a voice and text interface to prompt users to respond to various questions. In addition, the system used RFID tag identification which we did not use in our study. We used a HTC P6300 phone for this implementation. Cellphone based system: As a comparison point, we developed an alternative implementation using JavaME [31] on Nokia 3110 phones that used a text-free pictogram based user interface of specific healthcare symbols to primarily understand whether the end-user population would be conversant with a pictogram-based interface. A. Methodology The user study for SmartTrack was performed at several sites across Ghana. To perform the study, we developed a survey tool that was administered to cohorts of HIV+ patients in each location. The cohorts were patients affiliated with either the West Africa AIDS Foundation (WAAF) [32] or Korle Bu Teaching Hospital. Participants were referred to us by these organizations, and we did not attempt to implement a randomized study. The ages reported by patients ranged from 17 to 101, and 73.3% of the patients were female, 26.7% male. 77.3% of patients were currently receiving ARV therapy. We interviewed 516 HIV+ patients with a survey consisting of 72 questions divided into eight categories: 1) Laboratory values: CD4 count, viral load and WHO HIV stage (taken from blood tests/patient records). 2) Demographics: Age, housing, occupation, education, language skills, access to water and electricity (11 questions). 3) Health Care Access: Health practices, frequency of medical care, frequency of contact with health worker, distance to the hospital (9 questions). 4) Symptoms: HIV-related symptoms, typical response to certain symptoms, and histories of other conditions like tuberculosis or malaria (11 questions). 5) Medications: Medications being taken, adherence to regimen, acquisition of drugs (10 questions). 6) Narrative Text: Difficulties of the disease, difficulties of treatment, perceived effectiveness of treatment (9 questions). 7) Mobile Phone Usage: Access to mobile phone, usage habits, amounts paid for service and hardware (20 questions). 8) Patient Education: What is the patient’s primary source of health information (1 question)? The patients gender and use of a translator were not asked directly, but were recorded. When necessary the aid of a translator was enlisted. 278 patients were interviewed via healthcare associations for patients living with HIV, coordinated through WAAF. Researchers traveled to five cities in four different regions of Ghana to interview a total of six cohorts of patients: Volta Region (Ho), Central Region (Cape Coast), Northern Region (Tamale), Upper West Region (Wa and Lawra). Blood draws to be used for CD4 counts were provided free-of-charge by

463

80

percentage

60

40

20

0

ur

R al

an

Translator required

Fig. 2. Percentage of patients requiring translators, with urban compared to rural

100 Education percent of patients

80

percentage

In this section we present the results of our user study and observations made while interviewing 516 HIV+ patients in Ghana.

Yes No

rb

V. U SER S TUDY

100

U

WAAF in these locations. After having their blood drawn, the patients were asked to participate in the survey. Before beginning the survey, patients gave verbal consent. A further 238 outpatients at Korle Bu Hospital Fevers Unit were interviewed. The patients had come to the hospital to refill medication or to draw blood for CD4 T lymphocyte count. At that time, patients who were willing to participate were required to give written consent. The written consent forms were collected by Korle Bu Hospital staff to be included in the patients chart, and will not be published or disclosed in this or any other report to maintain patient confidentiality. The HIV+ patient population at Korle Bu Hospital was separated into groups of those taking ARV therapy, and those that had not yet started ARV therapy. Only patients who were currently taking ARV therapy were interviewed as their experiences are more relevant to the study. Adjustments were made to the survey and methods during the course of the study. The survey itself was edited after the first two administration sessions in order to more effectively elicit the target information. Several questions were altered, discarded, or added. The changes have been recorded and will be taken into account. The overwhelming majority of patients were interviewed individually. In certain situations however, there were time constraints and patients were interviewed in pairs. In these situations each patient answered individually.

60

40

20

A. Literacy & Education 0

ve ni U

S

SS

l

ity rs

l

o ho

ve

sc

-le O

o N

S JS

l

ve

rm

Fo

le

A-

h 6t

Illiteracy is one of the most serious challenges facing the developing world, and is particularly severe in rural areas. We gave participants a number of questions designed to gauge the level of literacy and education within our population. Questions we asked included: i) what languages does the participant speak? ii) in which languages is the participant able to read or write? iii) what is the participant’s highest level of education? We found that 55.6% of interviewees in urban areas, and 60% in rural areas needed a translator for our interviews. Interviews were conducted in English, which is also the official language of Ghana. Additionally, our survey showed that urban areas tend to have higher literacy rates than rural regions. Figure 3 shows the distribution of educational attainment among patients, and Figure 4 shows the percentages of patients able to read or write in each of the locally-used languages. Notably, at least 35% of patients had never attended school of any kind, and only 46% reported being able to read English. This high rate of illiteracy, particularly when combined with the lack of a common language among patients, presents a fundamental challenge to the deployment and acceptance of any telemedicine application. In order for SmartTrack to be effective, it must be operable by a patient with no reading skills whatsoever. Either a voice-based or a pictogram-based

Education level

Fig. 3.

Maximum education level

solution might solve the literacy problem, though the voicebased UI would not solve the multiple language problem, and would have the additional drawbacks of lack-of-privacy, in addition to feasibility and cost problems when using low-cost phones with expensive network access. To this end, we are exploring pictogram interfaces for this application. We developed a demonstration interface for our Ghanaian study in order to test our ability to convey healthcare-related concepts through graphical symbols. Although extremely preliminary, this demonstration was wellreceived by users and showed promise for expressing health concepts. Using simple pictograms, combined with shape or color symbols to represent different medications, it should be possible to make an effective interface that is usable despite barriers of literacy and language.

464

100

100 Languages known (read or write) percent of patients

percentage of ages of patients

80

60 60 percentage

percentage

80

40

40 20

20

0

i

Tw

e

on

N

ch

en

e i

nt

Tw

h,

is

i nt

a

G

Fr

Fa

Ew

gl

En i

ic

ab

Fa

Ar

h,

is

h,

is

gl

En

h

is

gl

gl

En

ic ab

En

Ar

0

3

-7

61

0

-6

51

0

It is impossible to divorce the problem of medical treatment in the developing world from the economic environment in which the patients live. The cost of care is only the most obvious element. The patient also must contend with the expense of travel to distant clinics, and the cost of missed work during time spend traveling. The frequent lack of electricity, clean water, and safe, sanitary conditions makes staying healthy even more difficult.

-5

C. Economic and social issues

41

A patient’s age is an important factor in determining proper medical care, and age statistics can offer some insight into the impact of AIDS on the society at large. Figure 5 shows the age distribution of the patients, and illustrates the disproportionate number of people seeking treatment who are in the 30 to 50 year-old range, with over 69% of patients falling within that group. For so many people in the prime of their work and family lives to be stricken can only have devastating effects on a society. The lost productivity of working-aged people slows the economic development of an already-depressed region, and the death or debilitation of so many parents will have social ramifications we have barely begun to understand [33]. Improving adherence can prolong lives and improve the quality of those lives. In addition to the direct and obvious benefit to the patient, this can also have the wider benefit of allowing the patient to be more productive economically, and to play a greater role in the lives of family and the community. Interestingly, 73% of patients were female. While it is estimated that almost 61% of HIV+ people in Sub-Saharan Africa are female[34], it is not clear if our higher number reflects the actual proportion of women with HIV in Ghana, or if it is a result of our non-random selection process. It is even conceivable to us that it is a result of the self-selection process of people seeking HIV treatment or getting diagnosis. [34]

0 Age

Percentage of patients able to read or write, by language

B. Age and Sex

-4

Fig. 4.

30

9

-2

20

Languages

Fig. 5.

Age distribution of participants

To better understand the economic situation of our population, we asked a number of questions pertaining to their basic employment, income, and living conditions. The following are some selected findings: •

• • • •

Type of house: 58.3% lived in a compound house, 24.4% in a self-contained house and 16.3% in a rented house. The rest were either homeless or lived in mud huts. Water source: 75.6% used piped water and 15% used well-water. Electricity: 26% did not have reliable access to electricity. Refrigeration: 56% lacked any sort of refrigeration. Occupation: The most common job categories reported by patients were petty trading(34.7%), service work(23.1%) and agriculture(12.6%).

To live in such a setting can be challenging enough for a healthy person, but for an HIV+ person, the difficulty is multiplied. People with AIDS can experience weakness and any number of other symptoms that interfere with their ability to work and earn money. Of our study participants, 64% responded that AIDS has a considerable effect on their ability to work, and 60% reported missing work as a result of HIV symptoms. Social stigma for those with HIV has been reported to be a serious issue in many cultures, particularly in parts of Africa, but interestingly, 61.5% of our participants said that they did not have any social problems due to AIDS. If true, this is wonderful news, but we suspect the response does not tell the whole story. One possibility is that patients were hesitant to admit that they had experienced such discrimination. Another is that they had not experienced discrimination because they had successfully hidden the fact that they had HIV. All of these details bring home a key point: Any designer of a telehealth system for the developing world must be keenly aware of the tenuous existence its users may lead. Costs must be kept to a bare minimum, as the pennies charged even to send simple SMS messages may accumulate to become a

465

serious burden. On the other hand, such systems also provide the opportunity to ease burdens by reducing the amount of travel required to see doctors, or by increasing the ease and affordability of communication with clinics. Furthermore, by helping to manage difficult medication regimens, a system like SmartTrack can improve the patient’s health and boost his or her ability to cope in a challenging world. 1) Regimen Compliance: Following a strict medication regimen for HAART treatment is essential for survival. To evaluate the current state of adherence among our population, we asked about frequency of missed doses, reminder methods for medications, and patient action upon missing a dose. 90% of patients admitted to frequently missing dosages, a disturbing figure given the risks associated with drug resistance against ARVs. 75% of the patients of did not use any kind of external reminders, relying on habit alone to take their pills at the scheduled time. In the case of a missed dose, 70% reported taking their pills when they remember. Findings like these boldly underline the need for better adherence management. Cellphone-based tracking tools have the potential both to serve as reminder systems, and to alert doctors to problems with compliance before they result in tragic consequences. 2) Response to drugs: In addition to adherence questions, patients were asked about their perceived response to and tolerance of ARV treatment, and their plans for future treatment. • 81.8% said that they felt much better after starting ARV medication. • 98% said that the medication was controlling their HIV symptoms. • 96% said that they had plans of continuing medication. Clearly, patients are happy with the results of treatment, and participants also expressed interest in a cellphone-based reminder system for taking medications, if one were made available to them. Perhaps with the right tools, patients would be better able to manage their ARV treatments and achieve better outcomes. D. Mobile phone usage We asked a number of mobile-telephone-related questions to better understand cellphone usage among the study population and gauge the potential for telehealth applications using mobile phones. • Usage of cellphones: 54% of patients use cellphones and the division of cellphone usage across various regions are as follows: Greater Accra (77%), Northern Accra (67%) and Volta(57%), which are urban areas rank higher than Central region(29%) and Upper West(26%), which are rural regions. In other uses of cellphones, only a small percentage of people used cellphones for playing games and listening to music. 82% of users used cellphones as alarms. • Would a cellphone reminder be useful? 91% said that it might be helpful in maintaining their HAART regimens. • Sharing of phones: About 80% of patients who use cellphones have their own phone and 16% share their phone

Disease/symptom Tuberculosis Malaria rashes/lesions weight loss pain while swallowing weakness stiffness,tickling,numbness Diarrhea shortness of breath coughing up blood

% of patients 84 98.8 52.7 60 26 60 41 34 28.5 9.4

% of doctor visits 85 85 81.2 79.4 85 74.1 84

TABLE I VARIOUS DISEASE & SYMPTOMS ; PERCENTAGE OF PATIENTS WHO HAVE THESE DISEASES OR SYMPTOMS AND PERCENTAGE OF PATIENTS WHO VIST A DOCTOR

with family and friends. 70% felt that a shared cellphone would still be useful for a healthcare application. • Frequency of calls & text messages: 75% make more than one call a day. 22% receive text messages daily and 18% send text messages daily. • Cost: 45% of users spend less than $50 on their cellphone handsets, and 69% spend less than $20 per month on cellphone services. The adoption of mobile phone technology among HIV+ patients in Ghana is quite widespread. This presents an outstanding opportunity to bridge some of the gaps that have prevented adequate healthcare from reaching those who most need it. Potential users have shown an interest and a willingness to try a system like SmartTrack. Patients are already quite reliant and comfortable with their phones, as shown by their frequent voice calls. In addition to the primary voice function, significant minorities of patients use their phones for music and games, and a large proportion of respondents use them for clocks and alarms. Few people use their phones for SMS texting, and any cost-effective telehealth system would be based on SMS for the forseeable future. However, the low rate of text-messaging is presumably related to the lack of literacy among users and language-barrier issues. This would not be a problem for an application that used SMS for data-transport, however, as the user would not be exposed to the underlying messages. With a current adoption rate of 54% among our participants in Ghana, and a worldwide growth rate in mobile phone adoption hitting 25% per year [24], cellphone-based telehealth systems promise to become more and more feasible as time goes on. E. Diseases & symptoms We also collected information on various symptoms, health practices, secondary diseases, and conditions not related to HIV. This was for the purpose of understanding symptom information that may be communicated by SmartTrack, assessing the applicability of SmartTrack to conditions beyond HIV, and getting a sense of the participant’s general health and medical habits. Symptom and disease information that we collected is summarized in Table I. These are symptoms and secondary

466

diseases which may be common to people suffering from AIDS, and the percentages of patients who reported having these conditions at any point. The second column contains the proportion of patients who traveled to see a doctor as a result of the corresponding condition. These trips can be quite difficult for patients, and might be avoided in certain cases if the doctor can consult with the patient remotely and recommend a course of action. Again, a cellphone-based health system could enable doctors to track patient symptoms remotely and act appropriately, contacting the patient if necessary. VI. U SER R EACTIONS AND L ESSONS L EARNED When we experimented with two different cellphone platforms with each user, the reactions were fairly varied across users. Many users were completely new to the concept of smartphones and had difficulties navigating the screen. This is not to rule out smartphones completely from such an environment, but it is well known that rural populations are not accustomed to sudden changes. The cellphone has been considered a user friendly device due to its similarity with the telephone. While many users found the voice option fairly attractive, our voice interface was based on English which was not understandable to over 50% of the users. The users we interviewed spoke several languages/dialects that made it hard for translators we used in our study to converse with local Ghanaians in rural areas. Another concern that patients expressed with voice was that a “talking” phone could disclose the HIV status of a patient. In contrast, users found pictograms very easy to understand. While we used only a limited set of universal healthcare symbols as pictograms in our study, it remains to be seen as to how many different types of questions can be phrased purely using pictograms. For example, to represent different drugs, we found that simple color and symbol coding would work sufficiently well for several illiterate users. Overall, we felt that a pictogram based user interface holds much promise and can also deal with the stigma issues that voice based interfaces sometimes might present. The overall response to a system like SmartTrack was very positive. The vast majority of patients (91%) seemed extremely positive and interested in a reminder system that could be placed on their cellphones. Typically, patients have to travel great distances to reach their treatment facilities; hence, being able to remotely communicate with the doctor is a huge positive factor for most patients. Social workers, nurses, community health workers, and doctors seemed very positive and amenable to the system. Many health-workers also wanted a similar system for other common diseases like malaria, TB and Cholera. Other important lessons we learned include: 1) From a cost perspective, smartphones are much more expensive than cellphones. In Ghana, users can purchase unlocked cellphones at extremely low cost, thus we felt the system should be built around people’s cellphones. 2) From a communication perspective, many rural areas we surveyed did not have GPRS coverage. Even if available,

it was much more expensive than SMS. Given that the information per patient is extremely limited, we felt that SMS is a much cheaper alternative than voice or data services. 3) The system can potentially be extended to other chronic illnesses like malaria, TB and Cholera. 4) It is critical to perform a needs-assessment study before performing the actual design. While this may seem obvious in retrospect, doing a detailed needs assessment study is a tedious and time consuming process that is often ignored. VII. C ONCLUSIONS As Bill Easterly wrote in The White Man’s Burden, “2.3 trillion dollars of aid over 50 years and we have nothing much to show for results.” [35] This so poignantly summarizes the current state of affairs of many philanthropic organizations funding AIDS relief efforts in Africa. While it is indeed true that some organizations have made significant progress in specific areas, this is more of an exception than the norm. Our effort is largely centered around technological issues facing the adoption of AIDS relief schemes. We believe that, a combination of technological and sociological approaches are needed to tackle the healthcare(HIV/AIDS) problem in developing regions. From a technical perspective, to make AIDS relief efforts more effective, it is essential to improve accountability in the health system, and to enhance patient adherence to medication regimen through constant monitoring. The vision of SmartTrack is to address these two problems, using a distributed cellphone-based platform to achieve these goals. Though this project is still in a relative state of infancy, it has undertaken a detailed needs-assessment study using interviews with over 500 AIDS patients in Ghana. Our next step in future work is the actual deployment of SmartTrack with the aid of the West Africa AIDS Foundation and Korle Bu Hospital. We plan to use the lessons learned from the needs-assessment study, with particular attention to the user interface issues, to deploy a widely acceptable version within Ghana to nonEnglish-speaking and non-literate populations. If successful, this effort can be replicated in other rural developing parts of the world. ACKNOWLEDGMENT We would like to thank Korlebu Hospital and the West Africa AIDS Foundation for enabling us to do the detailed user study and Microsoft Corporation for supporting this project. R EFERENCES [1] “World Health Organization,” http://www.who.int/3by5/en/.

[2] “HIV and AIDS in Africa,” http://www.avert.org/aafrica.htm. [3] “Treatment of HIV infection,” http://www.niaid.nih.gov/ factsheets/treat-hiv.htm. [4] “Upwardly mobile in Africa-Business Week,” http://www. businessweek.com/globalbiz/content/sep2007/gb20070913 705733.htm. [5] “Africa Telecom News,” http://whiteafrican.com/2008/08/01/ 2007-african-mobile-phone-statistics.

467

[6] “Low cost cellphones,” http://www.cameraphonesplaza.com/ new-low-cost-nokia-cell-phones/. [7] “Open Handset Alliance,” http://www.openhandsetalliance.com. [8] “Symbian OS: the open mobile operating system,” http://www. symbian.com/. [9] “Android - An Open Handset Alliance Project,” http://code. google.com/android/. [10] “Leap of faith technologies, inc.” [Online]. Available: http: //www.leapoffaith.com/ [11] R. Scheier, “African graft stings donors,” The Christian Science Monitor, Jun. 1 2006. [12] C. Mangwiro, “Mozambique links health officials to drug thefts,” Reuters, Aug. 28 2007. [13] F. Phiri, “Drug theft poses threat to battle against TB,” Inter Press Service News Agency, Mar. 24 2004. [14] “Counterfeit medicines fact sheet,” World Health Organization, Nov. 14 2006. [Online]. Available: http://www.who.int/ [15] P. Sax, “Patient adherence and response to therapy,” in AIDS Clinical Care August 1, 1999. [16] E. J. Mills, J. B. Nachega, I. Buchan, J. Orbinski, A. Attaran, S. Singh, B. Rachlis, P. Wu, C. Cooper, L. Thabane, K. Wilson, G. H. Guyatt, and D. R. Bangsberg, “Adherence to Antiretroviral Therapy in Sub-Saharan Africa and North America: A Meta-analysis,” JAMA, vol. 296, no. 6, pp. 679–690, 2006. [Online]. Available: http://jama.ama-assn.org/ cgi/content/abstract/296/6/679 [17] D. Werner, C. Thurman, and J. Maxwel, Where there is no doctor? [18] “OpenMRS,” http://www.openmrs.org. [19] “Partner in Health,” http://www.pih.org. [20] “Dimagi,” http://www.dimagi.com. [21] “D-Tree,” http://www.d-tree.org. [22] W. Kaplan, “Can the ubiquitous power of mobile phones be used to improve health outcomes in developing countries?” in Globalization and Health 2006.

[23] B. Rapchak, “Cell phone technology to enhance medication adherence and therapeutic outcomes,” in TEPR 2009, Medical Records Annual Meeting. [24] K. Kalba, “The adoption of mobile phones in emerging markets: Global diffusion and the rural challenge,” International Journal of Communication, vol. 2, Jun. 23 2008. [Online]. Available: http://ijoc.org/ojs/index.php/ijoc/article/view/216 [25] “Satellife: The Global Health Information Network,” http:// www.healthnet.org. [26] B. DeRenzi, N. Lesh, T. Parikh, C. Sims, W. Maokla, M. Chemba, Y. Hamisi, D. S. hellenberg, M. Mitchell, and G. Borriello, “E-imci: improving pediatric health care in lowincome countries,” in CHI 2008. ACM, pp. 753–762. [27] I. Medhi, A. Sagar, and K. Toyama, “Text-free user interfaces for illiterate and semiliterate users,” in ITID ’07: Information Technologies and International Development, 2007, pp. 37–50. [28] “Evaluation of the Satellife PDA Project, 2002: Testing the use of handheld computers for healthcare in Ghana, Uganda, and Kenya,” http://bridges.org. [29] S. Anand and U. Rivett, “Ict in the management of hiv treatment: Cell-life: a south african solution,” vol. 6, pp. 56–59, 2004. [30] “Cell-Life,” http://www.cell-life.org/. [31] “Java Micro Edition,” http://java.sun.com/javame/index.jsp. [32] “West Africa Aids Foundation,” http://www.waafweb.org/. [33] C. Bell, S. Devarajan, and H. Gersbach, “The long-run economic costs of AIDS : Theory and an application to south africa,” The World Bank, Policy Research Working Paper Series 3152, Oct. 2003. [Online]. Available: http: //ideas.repec.org/p/wbk/wbrwps/3152.html [34] “2007 AIDS epidemic update,” UNAIDS, Tech. Rep., Dec. 2007. [Online]. Available: http://data.unaids.org/pub/EPISlides/ 2007/2007 epiupdate en.pdf [35] W. R. Easterly, The white mans burden : why the Wests efforts to aid the rest have done so much ill and so little good. New York: Penguin Press, 2006.

468

Uses of Mobile Phones in Post-Conflict Liberia Michael L. Best‡∗ , Edem Wornyo† Thomas N. Smyth∗ and John Etherton‡ ∗ School

Interactive Computing of Electrical and Computer Engineering ‡ Sam Nunn School of International Affairs Georgia Institute of Technology Atlanta, Georgia 30332–0250 {mikeb, tsmyth3, jetherton}@cc.gatech.edu [email protected]

† School

A. Mobile phones in Africa Mobile phone penetration growth rates are today highest in Africa compared to all other continents [6]. The ITU

2004


2007
 2007


2001


2004


1998


1995


1989


1992


1986


1983


1980


Fixed
Lines
per
100
 Inhabitants


0.5
 0.4
 0.3
 0.2
 0.1
 0


Year


(a)

Mobile Teledensity 16
 12
 8
 4
 2001


1998


1995


1992


1989


1986


1983


0
 1980


I. I NTRODUCTION Considerable attention has been given to the role of information and communication technologies as tools for development within Africa, and increasing levels of excitement have concentrated on the use of mobile phones. With some fanfare the Economist [1] announced that the “real digital divide” was in terms of the differential access to mobile telephones while computers and the Internet were of less use. Many writers have disagreed with their pessimistic assessment of computers and the Internet (e.g. [2]). Nonetheless, it is clear that mobile telephones are playing a substantial and important role in development within the global south. Indeed, compelling evidence of the macro and microeconomic effect of mobile phones in low-income countries has been mounting [3]. For instance Waverman et al. [4] find that mobile phones offer a significant macroeconomic growth dividend and one that is “twice as large in developing countries compared to developed countries”. Microeconomic benefit is also evident. For instance Jensen [5] shows that mobile phone use by farmers in Southern India increases productivity, enhances revenues, reduces waste, and lowers consumer prices.

Fixed Line Teledensity

Subscribers
per
100
 Inhabitants


Abstract—Liberia is a country emerging from years of protracted and devastating civil conflict. Left without any fixed line telephone infrastructure, it relies solely on the mobile phone for telephony. This study investigates the usage of mobile phones in this immediate post-conflict setting. In particular, we adopt the uses and gratifications approach to media research, giving focus to both instrumental and intrinsic motivations for use. Mobile phone users in both the capital city of Monrovia and in various rural areas were surveyed using the Q methodology, which identified distinct perspectives within these urban and rural groups. Participants were then sorted into groups where each group contained users with similar perspectives on their mobile phones. These identified groups included sets of users who saw their phones as productivity enhancers, means of connectivity to family and friends, essential business tools, technological curiosities, and sources of personal security. The idea of a phone as a stylish object was markedly rejected, especially in rural areas. We contrast these Q-sort results from Liberia with previous work from Kigali, Rwanda, finding differences especially as related to security.

Year


(b) Fig. 1. Fixed-line and mobile teledensity in Liberia from 1980–2007. Steep drops in fixed lines are evident in 1991 and 2003. Introduction of competing mobile carriers in the mid 2000’s produced a soaring number of mobile subscribers. Note the difference in vertical scale between the two charts.

reports that average year-on-year growth rate for mobile phone subscribers in Sub Saharan Africa from 1999-2004 was double what it was in Europe. Indeed Sub Saharan Africa is a continent driven by mobile telephony and in 2001 the total number of mobile subscribers exceeded the number of fixed line subscribers [6]. In 2004, the mobile teledensity across all of Africa was 9.1, with the vast majority, 87%, making use of prepaid cards. Considering only Sub Saharan Africa the mobile teledensity is best approximated at 6.2% [7]. And while this number describes subscriber penetration it does not give an adequate sense of overall access and usage due to widespread sharing of phone subscriptions. Clearly, mobile

469

telephony is the central communication technology for much of Sub Saharan Africa. B. Mobile phones in Liberia Liberia, established as a state in 1847 by freed African slaves from the U.S.A., is situated on the Atlantic coast of West Africa with Sierra Leone, Guinea, and Cˆote d’Ivoire as bordering countries. A relatively small country with approximately 3.3 million inhabitants, it is attempting to right itself after decades of civil conflict. Unrest has been a staple within Liberia for more than 15 years with two major civil wars (1989-1996 and 19992003) in this time period. These years of conflict have seen nearly one-third of the population displaced and taken the lives of approximately 250,000 people. A peace was brokered and transitional government was established in 2003. A UN peacekeeping mission was positioned to keep this peace, and democratic elections were held in the fall of 2005. This resulted in the selection of Africa’s first elected female head of state, President Ellen Johnson-Sirleaf. An outcome of these years of civil conflict was the complete destruction of the fixed-line telephone infrastructure. The copper network was wholly destroyed or looted and all but one switch was destroyed [8]. It is clear from Figure 1(a) the steady decline in mainline penetration, starting from a very low level to begin with. The precipitous drops evident in 1991 and 2003 are the outcome of the two major civil wars such that by the time of the establishment of peace all mainlines where gone. On the other hand, mobile telephone adoption in Liberia has recently been growing at a staggering rate, as shown in Figure 1(b). It has been shown that teledensity phone penetration rates are likely to over count the number of actual subscribers (due to purchased but inactive accounts) and, as already mentioned, significantly undercount the number of actual users (due to sharing) [7]. In Liberia we estimate the subscribers to users ratio to be as high as one to five. Competition within the Liberian mobile phone sector is also robust with four active operators. Indeed, usage costs are reportedly the lowest in West Africa [34]. All county capitals and most other population centers currently receive signal from at least one of the mobile providers’ services, and two providers currently offer GPRS mobile internet services. Operators are actively extending both their networks and services. The striking success of Liberia’s mobile sector, which continues to develop at a feverish pace despite the resourcestrapped country, is a cause for optimism. C. Mobile phone usage in post-conflict settings Regrettably, civil conflicts such as those experienced in Liberia are not unusual in contemporary times. Indeed, while inter-state wars are increasingly less common, the incidence of civil conflict is on the rise [10]. Thus the study of ICT’s within countries emerging from civil conflict is an area of considerable importance though we note a paucity of work in this area [9].

Furthermore, retrospective empirical scholarship has demonstrated the critical nature of communication amongst the people of a nation if there is to be a lasting peace instead of an all-too-frequent return to civil conflict [10]. Modern information and communication technologies can therefore, on their face, serve as tools in this process of national reconciliation if they are ably applied to these communication activities. However, to understand what it would mean to “ably” apply modern ICT’s, including mobile telephony, to the process of post-conflict development requires at a minimum an adequate understanding of the current uses and meanings of mobile telephony in that environment. Such was the motivation for this study. II. U SES AND G RATIFICATIONS In seeking to uncover the everyday, micro-level motivations for mobile phone use among Liberians, this study draws inspiration from the uses and gratifications (U&G) research tradition. U&G as an approach originated in communications research in the mid-1970’s [11], advancing the view that consumers of mass media make active choices and selectively consume media in order to satisfy specific needs. More tersely, U&G can be said to focus on what people do with media, as opposed to what media does to people [12]. More recently, the U&G approach has been applied to study adoption and use of new media technologies, including telephones [13], the Internet [14], and mobile phones [15]. Also notable about the U&G tradition is attention to a broad range of motivations, including those which go beyond the purely instrumental or utilitarian (such as increased productivity or personal safety) to the intrinsic, social, or to quote McClatchey [16], ‘hedonic’ motivations for use. A typical U&G based study proceeds in one of two ways: either by starting with a hypothetical set of possible uses and seeking to confirm or deny each one, or in a more exploratory fashion, starting off with no such initial set. As will be seen, our study walks a line between these two alternatives. However, it must be noted that U&G in itself is not a method. Indeed, previous studies have employed a variety of different methods to investigate uses, including surveys [17], semistructured interviews [13], and focus groups [18]. A recent investigation of mobile phone usage among microentrepreneurs in Kigali, Rwanda [15] also drew upon the U&G approach. In using the Q-sort methodology (also used in this study and described in the next section), Donner identified four archetypal ‘factors’ which speak to predominant uses of mobile phones in Kigali. They were: convenient, intrinsic, indispensable, and productive. Donner remarked on the diverse nature of those factors, saying that they ‘’‘suggest numerous paths for future research.” Our research is intended as an extension of this body of research on uses of mobile phones to an immediate postconflict context. To our knowledge, ours is the first study of mobile phone uses and gratifications in such an environment. We believe that this context may give rise to unique motivations for use, especially given the vibrancy of Liberia’s

470

mobile sector as described above, and the obvious importance of communication to the task of rebuilding a nation. III. M ETHODS A. The Q-Sort Methodology The Q-sort method was employed to gain insight into the nature of mobile phone use in post-conflict Liberia. In this section, the concept of the Q-sort methodology is briefly explained. However, this paper does not present an in-depth treatment, as many relevant and well written expositions and examples of the Q-sort methodology are available elsewhere. The Q-sort methodology, which evolved from factor theory, was originally developed by the British physicist William Stephenson for psychological studies [19]. Despite earlier criticism of the technique in the academic community, the technique has gained increasing attention and acceptance as a tool for research in many areas from psychology [20], to medicine [25], communication [29], social sciences [30], and education [32]. In a Q-sort study, a subject is asked to arrange a set of statements (such as those shown in Table 1), pictures, or sounds, according to some perceptual metric. In most Q-sorts, the individual is requested to place a statement into one of the slots in a grid akin to Figure 2. This grid is designed to describe a quasi-normal distribution. Each column along the grid is given a relative position along some semantic differential, for instance from strongly disagree to strongly agree. The ordering of the statements in each column is irrelevant—only the lateral ordering carries meaningful information. Some studies use a rectangular grid pattern as opposed to a quasi-normal one. We chose the latter as we believe it forces the participant to think deeply in choosing the strongest points of agreement and disagreement. Data analysis in the Q-methodology establishes groups of individuals who sort particular traits in common places within the distribution. For example, consider a group of teenage mobile phone users in Tokyo who place great weight on their connections with friends and the stylish elements of their phone but who sort customer and work connections as low in importance; this group of people might be detected as a “factor” within the Q-sort methodology. Thus, Q-sort is said to be a person-oriented approach as opposed to a trait-oriented tool [33]. Previous work by Jonathan Donner [15] applied the Qmethodology to the study of mobile phone use among microentrepreneurs in Kigali, Rwanda. The present study is intended as an extension of that work, studying the case of mobile phone use in Liberia. In order to enable comparisons between the two studies we have used the same set of statements as was used in Rwanda (with only minor modifications). Both Rwanda and Liberia have emerged from recent civil conflict with Rwanda embarking on a path to peace starting in 1994 while Liberia saw conflict through till 2003. This study, therefore, examines how people perceive their mobile phones after only a few years of peace. And when we compare those perceptions to Donner’s study from Rwanda it is helpful

Fig. 2. An example Q-Sort, demonstrating the quasi-normal pattern into which statement cards are sorted. In practice, the full statement is printed on the front of the card, while a reference number is printed on the back. When the sort is complete, the cards are flipped and the pattern is recorded.

to recall that they have enjoyed an additional nine years of relative peace. Do those additional years explain some of the inter-state variation we have observed? B. Protocol The study required that participants arrange a set of statements as listed in Table I according to how these statements best describe their use of mobile phones. The statements were printed on flash cards for easy handling. The participants were advised to arrange the statements initially into three piles: “Describes me best”, “Neutral”, and “Describes me least”, and then to sort the piles into the appropriate categories in a quasinormal format similar as shown in Figure 2. In addition to the statements, demographic information was requested from the respondents. The time taken for the exercise ranged from 30 to 60 minutes per participant. The statements were in English. C. Participant Selection Unlike traditional quantitative survey techniques, Q-sorts can be carried out with a relatively small number of participants from a population space. We sought participants from both Monrovia, Liberia’s capital, and from various rural areas throughout the country. This is in contrast to Donner’s study which focused entirely on the capital of Kigali. In total, 63 participants were interviewed in Monrovia. Participants were chosen at random from passers-by on a street corner in downtown Monrovia. Fourteen responses were discarded due to incompleteness, leaving a total of 49. Outside Monrovia, a total of 36 participants were selected, also at random from busy areas. In total, we visited 13 towns and villages in several Liberian counties. The age of the respondents ranged from 19 to 62 years. Participants were given a US$5.00 mobile phone scratch card for their efforts, whether or not they completed the sort. Print literacy was a requirement for participation. In questionable cases, prospective participants were asked to read one of the statements from a flash card and describe its meaning before they were admitted to the study.

471 TABLE II FACTOR C HARACTERISTICS

TABLE I Q-S ORT S TATEMENTS Concept

Statement

Connectivity

I use my mobile phone to stay in touch with my customers. I use my mobile phone to stay in touch with my suppliers. My phone gives me access to new customers. I use my mobile phone to stay in touch with my family. My mobile phone helps me come and go without worrying about missing calls. I use my mobile phone to stay in touch with my friends.

Information

My mobile phone helps me find work. My mobile phone helps me keep informed about prices in my business.

Intrinsic

Having a mobile phone makes me feel more important. Having a mobile phone makes me feel more connected to the world. I like customizing my mobile phone with accessories like special sounds and carrying cases I enjoy talking to my friends and family on my mobile.* Having a mobile phone makes me happy. My mobile phone is stylish.

Productivity

My My My My My My

Security

I use my mobile phone for emergency calls. My mobile phone makes me feel more secure.

Other

Getting a mobile phone changed the way I do business. I am interested in learning about new features or mobile phone models. I can’t do business without my mobile phone. I was among the first of my friends and business associates. I give my mobile phone number to many people. I share my mobile phone with my family or friends. I keep my mobile phone with me at all times. My mobile phone gives me more control over who I talk to, and how/when I talk to them. I use my phone more for business more than for social calls.

business is easier now that I have a mobile phone. family is better off because I have a mobile phone. mobile phone saves me time. mobile phone lets me get more done during the day. mobile phone helps my business save money. mobile phone helps me make more money in a day.

* Due to a miscommunication, in the urban study, this statement was replaced with “I bought my mobile phone for business.” This change was incorporated into the analysis that follows.

D. Analysis Methods Q-analysis is usually performed using PQMETHOD, a software package developed specifically for the task. A typical Q-analysis involves several steps. Initially, a large correlation matrix is created, describing the similarities between the Qsorts of all pairs of participants. We then look for ways to reduce the information in this matrix into an interpretable form, a process which is both iterative and partially subjective. There are several routes to this goal, a review of which goes beyond the scope of this work. In our analysis we chose a procedure similar to Donner [15]; we performed a principal

F1

F2

F3

F4

Urban

# of Defining Participants % of Variance Explained

9 10%

13 13%

12 11%

4 7%

Rural

# of Defining Participants % of Variance Explained

13 20%

8 12%

5 11%

6 12%

components analysis to identify initial factors within the data, followed by a varimax rotation to arrive at the final set of factors. Each such factor can be thought of as an archetypal perspective; a sorting of the statements that defines one group of subjects against the others. Once these factors have been identified, a loading score is computed for each participant/factor combination, which measures the similarity of that participant’s perspective to the archetypal perspective of that factor. A participant is said to load on or define a factor if their loading score for that factor crosses a certain threshold. As a result of this process, each factor is associated with a set of participants defining it. In a sense the set of participants have now been clustered into a small number of factors (we find four factors in our study) with each participant assigned to that factor that best represents them. In the final step, the statement rankings for each participant are weighted according to that participant’s loading score for the factor they are assigned to (therefore if they define the factor more closely their ranking will have more weight). Then all of these weighted ranks are combined among the participants assigned to each factor such that each factor is then described by a list of single Z-scores, one per statement, along with a p value assessing the value’s statistical significance. Finally, these Z-scores are re-projected back into the original space of values from -4 to +4 (from “describes me least” to “describes me best”) as shown in Figure 2. And as an aid to comparison between the groups each set of Z-scores are projected into the space described by the other factors as well. PQMETHOD also determines for each a factor a set of “distinguishing statements” which differentiate the factor from the others. These statements are of special importance as they are most representative of the differences between the factors. It is in examining these representative statements that insight into the meaning of each factor can finally be gained. Tables III and IV, which we will go on to study below, show these sets of factors and their most distinguishing statements, along with those statements’ Z-scores and the -4 to +4 values associated with them. IV. R ESULTS The Q-sort data we obtained from urban and rural participants in Liberia have been analyzed separately. This has allowed us to examine differences in mobile phone use and perception between these populations. Following the procedures described above, PQMETHOD was used to perform our analysis. After the principal com-

472

Commonalities

Factor 3: Business F1 F2 F3 F4 2 1 2 4

Statement

I use my mobile phone for emergency calls.

Factor 1: Productivity Z F1 F2 F3 F4

Statement

Describes Me Best

∗∗I can’t do business without my mobile phone. ∗My phone gives me access to new customers. ∗∗I bought my mobile phone for business.

1.75 -3 3 1.49 1 -2 1.19 -1 -2

4 -2 3 -1 3 -1

Other Distinguishing (Relatively High) Statements

Describes Me Best

∗∗My mobile phone helps me make more money in a day.

Z F1 F2 F3 F4

Statement

1.57

3 -4

0 -4

Other Distinguishing (Relatively High) Statements

∗∗I use my mobile phone to stay in touch with my 0.8 friends. ∗∗My mobile phone lets me get more done during 0.49 the day. ∗My mobile phone helps me keep informed about 0.19 prices in my business. ∗∗My family is better off because I have a mobile -0.08 phone.

2

4 -3 -4

1 -4 -2 1 -2

3

2 -1

2 -1 1 -1 0 3 0 -2 0 -4

Other Distinguishing (Relatively Low) Statements

0

1 -2 -2

0

2

2

3

0

1 -2

2

∗∗I give my mobile phone number to many people. -1.27 -4 2 0 ∗∗My mobile phone is stylish. -1.49 -4 -2 -1

3 1

Other Distinguishing (Relatively Low) Statements

∗∗Having a mobile phone makes me feel more 0.17 connected to the world. ∗My mobile phone saves me time. -0.36

∗∗My mobile phone helps me keep informed about 0.83 1 -2 prices in my business. ∗My mobile phone makes me feel more secure. 0.21 -1 2 ∗I give my mobile phone number to many people. 0.14 -4 2 ∗I am interested in learning about new features or -0.15 -3 1 mobile models. ∗∗My mobile phone helps me make more money in -0.4 3 -4 a day.

Describes Me Least

∗∗My mobile phone lets me get more done during -0.78 the day. ∗My mobile phone saves me time. -0.82

1 -4 -2

3

1 -2

2

∗∗I use my mobile phone to stay in touch with my -0.91 4 4 -3 family. ∗∗Having a mobile phone makes me feel more -1.98 -1 -2 -4 important.

2

0

Describes Me Least

0

Distinguishing statements: ∗p < .05, ∗∗p < .01

Distinguishing statements: ∗p < .05, ∗∗p < .01

Factor 4: Security Factor 2: Connectivity

Describes Me Best

Z F1 F2 F3 F4

Statement

Z F1 F2 F3 F4

Statement

Describes Me Best

∗∗I use my mobile phone to stay in touch with my 2.18 friends. ∗I keep my mobile phone with me at all times. 1.29

2

4 -3 -4

0

3

0

4

Other Distinguishing (Relatively High) Statements

∗∗My mobile phone makes me feel more secure. 0.8 -1 ∗∗Having a mobile phone makes me happy. 0.78 0 ∗I give my mobile phone number to many people. 0.61 -4 ∗∗My family is better off because I have a mobile 0.5 0 phone. ∗I like customizing my mobile phone with acces- -0.29 -1 sories like special sounds and carrying cases.

2 1 -1 2 -1 -3 2 0 3 1 -2 -2 0 -1 -2

Other Distinguishing (Relatively Low) Statements

∗I use my mobile phone for emergency calls. 0.31 2 1 2 4 ∗∗My mobile phone helps me come and go without -0.3 -2 -1 -2 2 worrying about missing calls. ∗My mobile phone helps me keep informed about -1.15 1 -2 2 -1 prices in my business. Describes Me Least

∗∗I use my phone more for business more than for -1.33 social calls. ∗∗My mobile phone helps my business save money. -1.43 ∗∗My mobile phone lets me get more done during -1.6 the day.

3 -3

3

2

1 -3 1 1 -4 -2

1 3

∗∗I use my mobile phone for emergency calls. 2.31 2 1 2 ∗I keep my mobile phone with me at all times. 1.93 0 3 0 ∗∗My mobile phone lets me get more done during 1.57 1 -4 -2 the day. ∗I give my mobile phone number to many people. 1.16 -4 2 0

4 4 3 3

Other Distinguishing (Relatively High) Statements

∗∗I use my mobile phone to stay in touch with my family. ∗∗My mobile phone helps me come and go without worrying about missing calls. ∗I use my phone more for business more than for social calls. ∗∗My mobile phone is stylish.

4 -3

2

0.85 -2 -1 -2

2

0.63

3

2

0.16 -4 -2 -1

1

1.09

4

3 -2

Other Distinguishing (Relatively Low) Statements

∗∗I use my mobile phone to stay in touch with my -0.01 customers. ∗My mobile phone helps me keep informed about -0.51 prices in my business. ∗∗My phone gives me access to new customers. -0.95 Distinguishing statements: ∗p < .05, ∗∗p < .01

Distinguishing statements: ∗p < .05, ∗∗p < .01

TABLE III U RBAN FACTORS

2

3

1

0

1 -2

2 -1

1

3 -2

3

473

Commonalities

Factor 3: Intrinsic F1 F2 F3 F4 3 3 2 3

Statement

Having a mobile phone makes me feel more connected to the world. I use my mobile phone for emergency calls. I keep my phone with me at all times. My mobile phone is stylish

Describes Me Best

∗∗Having a mobile phone makes me happy. 4 3 2 1 2 2 2 4 -4 -4 -4 -3

Factor 1: Business Describes Me Best

∗∗My business is easier now that I have a mobile phone.

1.66

4

0 -2

2

∗I am interested in learning about new features or mobile models. ∗∗I like customizing my mobile phone with accessories like special sounds and carrying cases.

0.77

2 -1

0

1

2

0

0.3 -3 -3

0 -2

Other Distinguishing (Relatively Low) Statements

4

0 -2

2

0

0 -2

3

0

0

4

4

2

1

0

3

∗∗I give my mobile phone number to many people. -1.97 -1 -2 -4 -1 Distinguishing statements: ∗p < .05, ∗∗p < .01

∗∗I use my mobile phone to stay in touch with my 0.15 0 family. ∗∗I use my mobile phone to stay in touch with my -0.57 -2 friends.

Factor 4: Mixed

0 -3 -4 0

0 -2 0

Factor 2: Mixed Z F1 F2 F3 F4 Describes Me Best

4 -1 3 3

2

1 -2 1 -3

Other Distinguishing (Relatively High) Statements

∗I share my mobile phone with my family or friends. 0.34 -4 ∗∗I was among the first of my friends and business -0.07 -3 associates to get a phone.

∗∗I keep my mobile phone with me at all times. 1.68 2 ∗∗My mobile phone helps me find work. 1.42 0 ∗∗I use my mobile phone to stay in touch with my 1.23 -2 friends.

2 2 0 -2 1 0

4 3 3

∗I use my mobile phone to stay in touch with my 1.09 0 4 4 family. ∗∗I enjoy talking to my friends and family on my 1.03 -1 4 -1 mobile. ∗∗My business is easier now that I have a mobile 0.92 4 0 -2 phone. ∗∗I use my mobile phone to stay in touch with my 0.67 4 -2 4 customers. ∗Having a mobile phone makes me happy. -0.06 -2 -2 3

2

Other Distinguishing (Relatively High) Statements

1 -2

Distinguishing statements: ∗p < .05, ∗∗p < .01

∗∗I enjoy talking to my friends and family on my 2 -1 mobile. ∗∗My mobile phone makes me feel more secure. 1.57 1 ∗My mobile phone saves me time. 1.21 0

Z F1 F2 F3 F4

Statement Describes Me Best

Describes Me Least

∗I was among the first of my friends and business -1.01 -3 associates to get a phone. ∗∗My mobile phone gives me more control over who -1.53 -3 I talk to, and when I talk to them. ∗∗I share my mobile phone with my family or -1.83 -4 friends.

1 -2 0 0 -3 -4

2 2 1 0

Other Distinguishing (Relatively Low) Statements

∗∗I use my phone more for business more than for -0.23 3 2 2 0 social calls. ∗My mobile phone helps me come and go without -0.3 -2 -3 -3 -1 worrying about missing calls. ∗∗My mobile phone makes me feel more secure. -0.85 1 3 1 -2 Describes Me Least

Other Distinguishing (Relatively Low) Statements

∗∗My business is easier now that I have a mobile -0.06 phone. ∗My phone gives me access to new customers. -0.11 ∗∗I use my mobile phone to stay in touch with my -0.68 customers.

3

Describes Me Least

Other Distinguishing (Relatively Low) Statements

Statement

0.74 -1

∗∗My business is easier now that I have a mobile -0.79 phone. ∗∗My mobile phone helps me find work. -1.03

Other Distinguishing (Relatively High) Statements

∗∗My mobile phone helps my business save money.

1.4 -2 -2

Other Distinguishing (Relatively High) Statements

Z F1 F2 F3 F4

Statement

Z F1 F2 F3 F4

Statement

4

0 -2

1 0 4 -2

1 4

2 2 1

∗∗My mobile phone saves me time. -1.32 0 3 1 -3 ∗My mobile phone lets me get more done during the -1.65 -1 -1 -2 -4 day. Distinguishing statements: ∗p < .05, ∗∗p < .01

Distinguishing statements: ∗p < .05, ∗∗p < .01

TABLE IV RURAL FACTORS

ponent analysis, several factor rotations were computed and examined. For both data sets, a set of four factors was found to provide the best balance between explanatory power and succinctness. Table II displays the number of participants found to load on each factor, as well as the percentage of initial variance that each explains. Each factor can be taken to represent an archetypal perspective regarding phone use among Liberians. The explained variance proportions we obtained are similar to those obtained in previous studies. Tables III and IV show the commonalities across factors for both datasets, and the statements distinguishing each factor, ordered by Z-score. The computed ranks for each factor are also shown for each statement (F1, F2, F3, and F4). Statements

with ranks for the present factor of interest of +3 or +4, or -3 or -4, are placed under the headings ‘Describes Me Best’ and ‘Describes Me Least’, respectively. Other distinguishing statements with a high (or low) rank for the present factor relative to the other factors are placed under the heading ‘Relatively High’ (or ‘Relatively Low’). Below, we review the distinguishing statements for each factor in an effort to interpret the nature of the archetypal perspectives they represent. We also review commonalities across factors for the urban and rural groups. These commonalities are statements which are consistently ranked positively or negatively for each of the four factors, and thus indicate agreement across most participants.

474

A. Urban Commonalities across factors: The urban group exhibited few points of commonality across factors. The only statement which was ranked positively for all four factors was “I use my mobile phone for emergency calls,” which was rated +2, +1, +2, and +4. No statements received consistently negative rankings for all four factors. This lack of consensus could be due to the varied sample of participants and their divergent viewpoints. Factor 1: Productivity: The group of nine participants defining this factor was comprised mainly of business owners with less than 12 employees, except for a computer technology facility owner who had 30 employees. This factor describes responses where people perceive the statements “I use my mobile phone to stay in touch with my friends” and “My mobile phone helps me make more money in a day” as rather important where the normalized Z-scores are relatively high (+3 and +2). In contrast the prestige or intrinsic statement “My mobile is stylish” was rated quite low. It is clear that this particular group does not view the mobile phone as an object of style (-4) nor do they give their mobile phone numbers to many other people (-4). Instead, affiliates to Factor 1 recognize most significantly the business aspect of mobile telephony, followed by the security utility that the phone provides. This group, therefore, emphasized the importance of the mobile phone as a business tool rather than a fashion item. They also emphasized how the phone allows them to stay in touch with their friends (+2), perhaps as they go about conducting their business. Overall, this group seems to see the phone as a productivity tool. Factor 2: Connectivity: Thirteen participants affiliate strongly with this factor. This group of respondents has a demographic makeup of sole proprietors and other business owners with less than three employees, as well as several miscellaneous others. For this group, communication with friends is essential (+4), as is constant availability, as they admit to carrying their phone at all times (+3), and giving their number to many people (+2). The phone also makes them feel happy (+2), secure (+2), and makes their family better off (+1). On the other hand, this group rated statements related to business and productivity rather lowly, saying that they do not use the phone to find out about prices (-2), they do not use their phone more for business calls (-3), and they do not see the phone as enabling them to accomplish more in a day (-4). In sum, it seems that members of this group are primarily concerned with being available and in touch with their family and friends, and that they derive feelings of pleasure and security from that high level of connectivity. Factor 3: Business: The 12 participants defining this factor bought a phone for their business (+3), use the phone to gain access to new customers (+3) and stay informed about prices (+2), and in general, feel that they can’t do business without their phone (+4). On the other hand, perhaps due to the centrality of the phone in their business routine, they do not feel that the phone lets them get any more done in

a day (-2). Furthermore, they do not see the phone as a tool for connectivity with their family (-3), or as a stature booster (-4). Taken together, this set of statements indicates a business-minded group of individuals, for whom the phone is an essential tool. And indeed from our demographic notes, this group comprises business people in areas such as the oil industry, computer technology, money exchange (forex bureau), and others. Factor 4: Security: This group is defined by four participants, who assign the highest rank of +4 to the statements “I use my phone for emergency calls” and “I keep my phone with me at all times”. This suggests that the safety of themselves or their loved ones is of primary concern to members of this group. The group also views the mobile as a productivity enhancer given the relatively high rank assigned to the statements “My mobile phone helps me come and go without worrying about missed calls” (+2) and “My mobile phone lets me get more done during the day” (+3). Further, the group ranked the statements: “I use my mobile phone to stay in touch with my family” (+2), and “My mobile phone is stylish” (+1) as relatively important, while the statement “My phone gives me access to new customers” (-2) was seen as relatively unimportant. Thus this group has elements of connectivity oriented users as well as gratification since they use phones to stay in touch with family and find their phone stylish. Furthermore, they perceive the phone as offering some productivity enhancements such as getting more things done and staying in touch with customers. However, what is uniquely distinguishing about this factor is its emphasis on security with the top two statements associated with this factor concerned with this issue. B. Rural Commonalities across factors: One of the most striking things about the rural study was that during the initial sort of the cards into three piles, most participants sorted most cards into the ‘describes me best’ pile. The second stage of the sort then became a difficult exercise in prioritization. Many participants were visibly torn on which statements to promote to the highest levels and which to leave behind. Nonetheless, in contrast to the urban data which had few pervasive commonalities, several statements emerged as items of consensus for most participants. All four factors agreed that their mobile phones make them feel more connected to the world. That statement was ranked +3, +3, +2, and +3, respectively. Several participants spoke of communicating with family members in other countries in Africa and in the West as justification for their high rating of this statement. There was also widespread agreement on the phone not being an object of fashion. The statement “My phone is stylish” received rankings of -4, -4, -4, and -3. During the study, many participants openly scoffed at this statement upon reading it. Finally, while no distinct security or safety factor is identified in the rural data, there was nonetheless widespread reliance upon the phone for emergency use. The statement “I

475

use my phone for emergency calls” was ranked +4, +3, +2, and +1, while “I keep my phone with me at all times” was ranked +2, +2, +2, and +4. It was clear that many participants felt strongly about this function of the phone. Several participants offered compelling stories of using their phone to call for help during a robbery, to call for medical care for a loved one, or as a deterrent against sexual violence. Factor 1: Business: Thirteen participants defined this factor. Of the statements distinguishing it from the others, few were positive. Most prevalent among them was the assertion that business is easier to conduct thanks to the phone, which was rated +4. Participants in this group also claimed that their phones helped their business save money. On the other hand, members of the group do not share their phone with friends or family (-4), or especially rely on it for communication with family (0) or friends (-2). Overall, this group views the phone as a serious tool that has improved their ability to conduct business, much as in the urban factor of the same name. Factor 2: Mixed: This factor describes a variety of personal uses of the phone. Above all, the eight participants defining this factor enjoyed talking on the phone with their family and friends (+4). However, they rated the statement “My phone makes me feel more secure” quite high as well (+3), and they view the phone generally as a time saver (+3). No businessrelated statements were rated highly—the phone is not seen particularly as making business easier (0), and the phone is not used to stay in touch with customers (-2). We find this factor to be a mix of personal, productivity and security uses with a clear bias against business. Unlike the previous factors in this study, this group does not admit to a clear, single description. Factor 3: Intrinsic: The five participants defining this factor were clearly enthusiastic about intrinsic uses of their phones as technological artifacts. Their most highly rated distinguishing statement was “Having a mobile phone makes me happy” (+3), and they were also uniquely keen to learn about new features or models (+2), and, less strongly, to accessorize their phone (0). On the other hand, participants in this group were not very interested in business functions of the phone (-2), or in using it to find work (-2). They also reported not giving their number out to many people (-4), the reason for which is not clear. Factor 4: Mixed: Our analysis identified a large number of statements with broad meaning as distinguishing this factor. The six participants defining this factor keep their phone with them at all times (+4), use it to find work (+3), and stay in touch with friends (+3). They also find that it makes their business easier (+2), and they use it to stay in touch with their customers (+1). Thus whereas Factor 2 reveals a variety of personal uses with a bias against business uses, this factor suggests an even more general blend of valued uses. V. D ISCUSSION The results of our Q-sort analysis have highlighted a set of uses and gratifications for mobile phones in Liberia, some of which are particularly interesting in light of previous research.

I use my mobile phone for emergency calls Urban Liberia Rural Liberia Urban Rwanda

F1

F2

F3

F4

2 1 2

1 2 1

2 3 -4

4 4 -3

My mobile phone is stylish Urban Liberia Rural Liberia Urban Rwanda

F1

F2

F3

F4

-4 -4 -3

-2 -4 1

-1 -4 -3

1 -3 -3

Having a mobile phone makes me feel more connected to the world Urban Liberia Rural Liberia Urban Rwanda

F1

F2

F3

F4

0 3 2

2 3 1

2 2 3

3 3 3

TABLE V P OINTS OF COMPARISON BETWEEN RURAL AND URBAN L IBERIAN DATA , AND URBAN RWANDAN DATA FROM D ONNER [15].

We discuss them below. Table V presents comparisons of several noteworthy statements. Perhaps the most striking and unique result of this study is the prevalence of security and emergency use reported by Liberian mobile phone users. Data from Monrovia suggested a distinct factor emphasizing security, while rural data revealed security as an item of consensus. In both cases, the emphasis on security was considerably stronger than that reported by Donner [15] for Rwandan users, as shown in Table V. This greater emphasis could be due to Liberia’s much more recent civil conflict. From informal discussions with participants, it was clear that the safety and security of self, of loved ones, and of personal property is still a major concern in Liberia. This is by no means surprising. Despite the 15,000-strong UN peacekeeping mission, the country’s police force is still under development, and many ex-combatants have turned to crime as a source of financial support. In such a situation, it is understandable that a mobile phone is seen as providing security, as it allows the user to call a family member or an authority in the event of a crime or transgression. In several cases, even police officers themselves spoke of their mobile phones as a source of security. Of all the findings of this study, this emphasis on security carries the most implications for possible future technology designs. Current phones, while providing access to centralized network security services such as 9-1-1, were not designed for an environment with weak state institutions and a lack of centralized security apparatuses. Instead one could imagine a phone design incorporating a “panic” button feature, which emits a loud noise and flashing light, and automatically contacts other phones, either in the immediate geographical area, or on a predefined emergency contact list. This feature

476

could be activated in an attempt to stop a crime in progress, or to act as a deterrent against potential offenders. Another result of interest highlights a difference between urban and rural users within Liberia. While analysis of urban data revealed four fairly well defined factors, two of the four rural factors admitted to a more blurry description. We feel that this may be due to the multifaceted livelihoods characteristic of rural citizens and/or the fact that the mobile phone is often the only available ICT service in rural areas. We encountered many rural participants who reported a number of different occupations. One claimed to work for the Ministry of Immigration in addition to being a farmer. Another taught secondary school in addition to selling rubber. Several students reported also working in various family businesses. Such users are not likely to fit neatly into any one category, such as ‘business’ or ‘connectivity’—their use of the phone is more varied. On the other hand, rural users displayed an overwhelming rejection of the phone as an object of style. As reported, participants often openly scoffed at the “My phone is stylish” statement upon reading it from the card. It seemed that the idea of a phone being stylish was absurd, and to consider it so would be shameful. This is in contrast to both the urban Liberian and Rwandan data, both of which revealed a factor in which the same statement received a positive rating, as shown in Table V. Rural Liberians universally seem to view their phones as serious tools, not fashionable accessories. While in central Monrovia there exists a group of relatively successful business people that exhibit consumerist behaviors, consumerist populations are mostly absent throughout rural Liberia. This is due to poverty and the pervasive lack of a formal economy in many of these areas. We suspect that in such a context the idea of flaunting or fetishising a phone as stylish seems disassociated with local realities. Also due to limited resources, most participants owned the least expensive, most basic phone models, and seemed to be aware of the humbleness of their devices in comparison to the phones available in the city. This awareness is sure to preempt any pretensions of fashionability. With this in mind, the emergence of an intrinsic factor focusing on technological enthusiasms in the rural data becomes noteworthy. Members of that group expressed interest in learning about newer more advanced phones, despite the fact that they were likely to be unaffordable. Participants often spoke about this interest in technology as if it were a civic duty - that any good citizen should be up to speed with the latest technology. We suspect that this group is ultimately aspirational ascribing to a vision where technology serves as an engine of their personal, and the nation’s development. One finding from Donner’s study which was mostly replicated in our data was the feeling that the phone supports connectedness to the world, as also shown in Table V. This finding was especially prevalent among rural users, which is not surprising given the fewer options available to rural users for communicating internationally. Some parts of Liberia, such as Sinoe County, are without radio stations, newspapers, or

internet cafs, leaving the mobile phone as the only link to the outside world. Also common to the two studies was the finding of strong business-related factors. Our study identified clear business factors for both the urban and rural populations. In addition, several other factors rated productivity-related statements highly, such as ‘My mobile phone helps me find work.’ It is clear that the mobile phone plays a crucial role in the largely informal Liberian economy. VI. C ONCLUSION This study has employed the uses and gratifications approach and the Q-sort methodology to investigate mobile phone usage among urban and rural Liberians. As in previous work, several distinct user groups were identified. Urban users saw their phones variously as productivity enhancers, means of connectivity to family and friends, essential business tools, and security providers. A group of business users was also identified among rural users, as was a group of technoenthusiasts, and two groups which eluded definite description. The multifaceted nature of these groups replicates Donner’s principal finding [15] in highlighting the diverse uses and gratifications characteristic of mobile phone users in a lowincome region. However, the chief difference between these two studies— our identification of the prevalence of security use among Liberians—is potentially important in its own right. As stated, we believe that this finding may be related to the nature of Liberia’s immediate post-conflict environment. Not only does such a finding have implications for future technology designs (which we are interested to explore), it also suggests further research into the role of ICTs in the process of stabilizing and rebuilding a nation following a civil conflict. The unfortunate fact of widespread civil conflict in today’s world makes understanding such phenomena even more important. A more general theme that we encountered is the sheer indispensability of the phone for most users. In many cases, the mobile phone is their only option for communications other than physical travel, which is costly and time consuming. In other places where methods of communication are various and many, the idea of a single modality being so essential is harder to fathom. But many participants we spoke to related stories of the phone saving them many miles of travel. Businesspeople celebrated the time saved in ordering goods from their suppliers over the phone, instead of traveling by costly public transportation, sometimes only to find the supplier out of stock. One participant described a hypothetical situation in which her daughter had fallen ill and she was without a phone. How should she know where to take her, when the only doctor in the area could be in any of several different towns, each a considerable distance away? Add to this the security role they apparently perform, and it is clear that the phone is a truly indispensable item. A weakness of this work arises from the requirement that participants be able to read. This was the unfortunate reality, since performing a Q-sort requires rapid and repeated visual

477

scanning of the statement cards. Unlike in a traditional survey, we felt that simply reading the statement to the participant once would not be sufficient to support the sorting process. Especially in rural areas, this unfortunately excluded a significant number of potential participants. In future, we are interested in experimenting with study designs using iconography and/or sounds in order to allow non-literate participants to share their view. ACKNOWLEDGEMENTS We thank Jonathan Donner for kindly providing his data for the purpose of comparison. We also thank the various Liberian property owners who provided facilities for this study to take place, and the guards at the Carter Center for their help in pilot testing our study apparatus. Finally we thank the three anonymous reviewers for their constructive comments on this work. R EFERENCES [1] “The real digital divide,” The Economist, 2005, March 12. [2] E. J. Wilson, M. L. Best, and D. Kleine, “Moving beyond ‘The real digital divide’,” Information Technologies and International Development, vol. 2, no. 3, pp. iii–v, 2005. [3] M. L. Best and C. Kenny, “ICT’s, Enterprise and Development”. In ICT4D by T. Unwin. Cambridge, UK: Cambridge University Press, 2009. [4] L. Waverman, M. Meschi, and M. Fuss, The Impact of Telecoms on Economic Growth in Developing Countries. Vodafone, 2005. [5] R. Jensen, “The digital provide: IT, market performance and welfare in the South Indian fisheries sector,” Quarterly Journal of Economics, 2001. [6] V. Gray, The un-wired continent: Africa’s mobile success story. ITU, 2006. [7] J. James and M. Versteeg, “Mobile phones in Africa: How much do we really know?” Social Indicators Research, no. 84, pp. 117–126, 2007. [8] M. L. Best, K. Jones, I. Kondo, D. Thakur, E. Wornyo, and C. Yu, “Post-conflict communications: The case of Liberia,” Communications of the ACM, vol. 50, no. 10, pp. 33–39, 2007. [9] D. Thakur and M. L. Best, “The telecommunications policy process in post-conflict developing countries: The case of Liberia,” in The 36th Research Conference on Communication, Information, and Internet Policy (TPRC), 2008. [10] W. Long and P. Brecke, War and reconciliation: Reason and emotion in conflict resolution. Cambridge, Mass.: MIT Press, 2003. [11] J. G. Blumler and E. Katz, The uses of mass communications: Current perspectives on gratifications research. Beverly Hills: Sage Publications, 1974. [12] W. Chigona, G. Kankwenda, and S. Manjoo, “The uses and gratifications of mobile internet among the South African students,” in PICMET 2008: Portland International Conference on Management of Engineering & Technology, 2008, pp. 2197–2207. [13] J. W. Dimmick, J. Sikand, and S. J. Patterson, “The gratifications of the household telephone: Sociability, instrumentality, and reassurance,” Communication Research, vol. 21, no. 5, pp. 643–663, 1994.

[14] J. Eighmey and L. McCord, “Adding value in the information age: Uses and gratifications of sites on the World Wide Web,” Journal of Business Research, vol. 41, no. 3, pp. 187–194, 1998. [15] J. Donner, “Microentrepreneurs and mobiles: An exploration of the uses of mobile phones by small business owners in Rwanda,” Information Technology and International Development, vol. 2, no. 1, pp. 1–22, 2004. [16] S. McClatchey, “The consumption of mobile services by Australian university students,” International Journal of Mobile Marketing, vol. 1, no. 1, pp. 1–9, 2006. [17] K. Aoki and E. J. Downes, “An analysis of young people’s use of and attitudes toward cell phones,” Telematics and Informatics, vol. 20, no. 4, pp. 349–364, 2003. [18] F. Williams, H. Dordick, and H. Jesuale, Focus group and questionnaire development for exploring attitudes towards telephone service. Herbert Dordick and Associates, 1985. [19] W. Stephenson, The study of behaviour: Q technique and its methodology. Chicago: University of Chicago Press, 1953. [20] E. B. Brownlie, “Young adults’ constructions of gender conformity and nonconformity: A Q methodological study,” Feminism & Psychology, vol. 16, no. 3, pp. 289–306, 2006. [21] D. Shemmings, “‘Quantifying qualitative data: An illustrative example of the use of Q methodology in psychosocial research,” Qualitative Research in Psychology, vol. 3, no. 2, p. 147, 2006. [22] M. Lister and D. Gardner, “Engaging hard to engage clients: A Q methodological study involving clinical psychologists,” Psychology and Psychotherapy-Theory Research and Practice, no. 79, pp. 419–443, 2006. [23] P. H. D. Stenner, G. Bianchi, M. Popper, M. Supekova, I. Luksik, and J. Pujol, “Constructions of sexual relationships—a study of the views of young people in Catalunia, England and Slovakia and their health implications,” Journal of Health Psychology, vol. 11, no. 5, pp. 669– 684, 2006. [24] S. J. Snelling, “Q methodology and the political opportunity of feminist psychology,” Feminism & Psychology, vol. 14, no. 4, p. 519, 2004. [25] A. Risdon, C. Eccleston, G. Crombez, and L. McCracken, “How can we learn to live with pain? A Q methodological analysis of the diverse understandings of acceptance of chronic pain,” Social Science & Medicine, vol. 56, no. 2, p. 375, 2003. [26] L. D. Bryant, J. M. Green, and J. Hewison, “Understandings of Down’s syndrome: A Q methodological investigation,” Social Science & Medicine, vol. 63, no. 5, p. 1188, 2006. [27] R. M. Baker, “Economic rationality and health and lifestyle choices for people with diabetes,” Social Science & Medicine, vol. 63, no. 9, p. 2341, 2006. [28] I. Goldman, “Q methodology as process and context in interpretivism, communication, and psychoanalytic psychotherapy research,” The Psychological Record, vol. 49, no. 4, p. 589, 1999. [29] M. W. Kramer and P. M. Pier, “Students’ perceptions of effective and ineffective communication by college teachers,” The Southern Communication Journal, vol. 65, no. 1, p. 16, 1999. [30] N. J. A. van Exel, G. de Graaf, and W. B. F. Brouwer, “‘Everyone dies, so you might as well have fun!’ Attitudes of Dutch youths about their health lifestyle,” Social Science & Medicine, vol. 63, no. 10, p. 2628, 2006. [31] A. C. Borthwick, T. Stirling, A. D. Nauman, and D. L. Cook, “Achieving successful school-university collaboration,” Urban Education, vol. 38, no. 3, p. 330, 2003. [32] J. Szente, J. Hoot, and J. Ernest, “Parent/teacher views of developmentally appropriate practices: A hungarian perspective,” International Journal of Early Childhood, vol. 34, no. 1, p. 24, 2002. [33] A. C. McKennel, “Surveying attitude structures—discussion of principles and procedures,” Quality & Quantity, vol. 7, no. 2, pp. 203–294, 1974. [34] R. Southwood. “Liberia: Four mobile companies bring lowest prices in West Africa” Balancing Act. http://www.balancingactafrica.com/news/back/balancing-act 352.html#head (accessed February 20, 2009).

478

An Automated Braille Writing Tutor with Multilingual Exercises and Educational Games M. Bernardine Dias, M. Freddie Dias, Sarah Belousov, Mohammed Kaleemur Rahman, Saurabh Sanghvi, Imran Fanaswala, Wael Ghazzawi, Ameer Abdulsalam, Noura El-Moughny1 and S. Raghu Menon2

Abstract—This work advances the state-of-the-art in assistive technology for the visually impaired by enhancing an automated tutor designed to teach beginner-level braille writing skills. The hardware component of the tutor is a low-cost device that provides an intelligent tactile interface that connects to a laptop or desktop computer. We demonstrate the many uses of this tutor including its multilingual capabilities, interactive exercises, and educational games. Preliminary field tests of the tutor in India, Zambia, and Qatar have demonstrated the usefulness of this automated tutor, and its potential to enhance braille literacy in developing regions.

which monitors the student’s writing and transmits data in real time to a computer to provide immediate audio feedback to the user. The latest version of the BT consists of two rows of 16 braille cells and six buttons placed over the top of the two rows to work as an input area. The stylus is a standard braille stylus that connects to the BT by its metal tip. Moreover, students can press on one of two control buttons placed on the sides of the BT to perform a variety of mode changes. Further details of the BT are discussed in prior publications by Kalra et al. ([1] and [2]). A new design for the packaging the BT was also explored (Fig. 1.).

Index Terms— Assistive Technology, Braille, Visually Impaired.

B

I.

INTRODUCTION

RAILLE is a widely-used language that is the only means of literacy for blind people. Each braille character is formed using a subset of six embossed dots placed in a cell of two columns and three rows. The positions of the dots are universally numbered from one to six. Despite its significance and the accessibility it brings, learning to write braille still has a number of barriers. More than 90% of the world’s visually-impaired people live in developing communities and their literacy rate is estimated to be below 3% [1]. Poorer areas tend to have both a disproportionately high number of blind people and fewer resources for educating them [1]. Therefore, the need to improve literacy for the blind in affordable ways is paramount. The traditional method of writing braille itself creates formidable challenges to literacy [1]. In developed countries, braille is usually embossed with a six-key typewriter known as a Brailler. These devices are prohibitively expensive in developing countries where braille is almost always written with a slate and stylus and hence children must learn mirror images of letters, get delayed feedback, and visually impaired teachers have difficulty monitoring student progress and diagnosing some mistakes. Thus, educational braille writing technology relevant and accessible to developing communities could have significant impact.

II. AUTOMATED BRAILLE WRITING TUTOR In response to the observed need for enhancing literacy for the blind in developing regions, a group at Carnegie Mellon University developed an automated Braille Writing Tutor (BT) that uses audio feedback to provide guided practice for beginners learning to write braille [3]. The BT (Fig. 1.) consists of an electronic slate and stylus

Fig. 1. Prototypes and preliminary field testing of the Braille Writing Tutor

III. MULTILINGUAL EXERCISES AND EDUCATIONAL GAMES We collected ideas from teachers at three partner schools to understand the process for teaching beginners to write braille, and the use of educational games in the classroom. Based on this research we implemented a curriculum that includes six staged interactive exercises for students to practice basic skills needed for learning to write braille. These include practicing dot numbers, dot patterns, and the association of dot patterns with letters. This curriculum is now accessible in several languages, including English, Arabic, Chinese, and French. We also implemented some of the educational games used in braille-writing classes to improve the fun-factor of the tutor. These include an animal game that requires students identify an animal based on its sound, and then spell its name on the tutor, and the popular “hangman” game which helps children practice spelling while trying to guess a word chosen by the computer. We also created two new games that can be played using the BT. The dominos game, inspired by a popular game with children in Qatar, requires the student to match a letter chosen by the computer by picking a different letter which has the identical dot pattern in the left or right column as the specified letter. Since Chinese braille is based on a phonetic representation of the sounds of the language there are 3 categories of braille characters: Initials (the consonants that are only found at the beginning of a syllable), Finals (vowels), and Tones. The Chinese braille game challenges students to match a “final” to an “initial” chosen by the computer.

REFERENCES Manuscript received on January 30, 2009. 1 All authors except the last author were affiliated with Carnegie Mellon University during their involvement with the reported work. Author contact information: {mbdias, mfdias, sarahtbw, kaleem, scsanghv, imranf, wghazzaw, ameer}@cmu.edu and [email protected]. This publication was made possible in part by the support of the NPRP grant 30-6-7-91 from the Qatar National Research Fund. The statements made herein are solely the responsibility of the authors. 2 S. Raghu Menon is with Microsoft Research India, Bangalore ([email protected])

[1]

[2]

[3]

N. Kalra, T. Lauwers, and M. B. Dias, “A Braille Writing Tutor to Combat Illiteracy in Developing Communities,” in Artificial Intelligence in Information Communication Technology for Development workshop at IJCAI 2007, Hyderabad, Jan 2007. N. Kalra, T. Lauwers, D. Dewey, T. Stepleton, and M. B. Dias, Iterative Design of A Braille Writing Tutor to Combat Illiteracy, Proc. IEEE/ACM International Conference on Information and Communication Technologies and Development (ICTD), Bangalore, Dec 2007, pp 9-10. TechBridgeWorld Automated Braille Writing Tutor project website: http://www.techbridgeworld.org/brailletutor

479

Boosting European Market Access to Malian Mango Growers Saskia Harmsen

Abstract—FRUILEMA, a Malian fruit and vegetable economic interest group (EIG), is acquiring the tools and resources needed to meet the requirements (GLOBALGAP certification) for exporting Malian farm products to international markets. Starting with the highly marketable mango fruit, an ICT solution has been developed to enable the identification of producers, to trace farm plots and produce, and to obtain quantitative and qualitative information on produce to support analysis and decision-making. The information is accessible via a web-enabled platform and can be updated using Internet and PDAs. The information is disseminated to producers, trackers, collectors, importers, and other actors in the mango value chain. Index Terms—Agriculture, Geographic information systems, Information systems, Software

I. CASE (CONTEXT, NOVELTY AND SIGNIFICANCE)

A

GRICULTURAL exporters from Mali face many requirements and constraints if they wish to sell to the European market. They need to have exact knowledge about the quantity and quality of farm products, places and periods of production, production inputs used, etc. Meanwhile, international consumers increasingly want to be able to trace the products they buy back to the farmer and plot where it came from. For importers, it is important to be able to assess the quality and reliability of their exporting counterparts. For producers, it is important to know the exact surface of their production areas and the requirements of the market. As analysis during project formulation has shown, agricultural producers in Mali rarely have a correct estimate of their production capacity, often over- or underestimating their real production capacity. This has led to challenges in developing realistic production plans, procuring the right amounts of farm inputs like fertilizers, and being able to deliver the quantities of fruits promised to exporters. To minimize these challenges and optimize the meager resources available, IICD’s Senegalese partner Manobi has further developed and adapted its existing platform to suit FRUILEMA’s needs. As an association representing 790 small producers and five exporter groups throughout the country, FRUILEMA wants to use all the data gathered to S. Harmsen works for the International Institute for Communication and Development (IICD), P.O. Box 11586, 2502 AN The Hague, the Netherlands (phone: 31-70-311 73 11; fax: 31-70-311 73 22; e-mail: sharmsen@ iicd.org).

further inform partners about the supply process and improve north-south linkages that promote not only trade, but also mutual understanding and collaboration. This platform allows for the better identification of producers, to know their collective production forecasts, and to trace the entire chain of activities, from the purchasing of agricultural products back to the producer. During the first year of the system being in operation, FRUILEMA was able to use the platform to convince European fruit importers to work with them, resulting in two contracts for the 2008 mango harvesting season. The platform was developed during 2007-2008, and involved the identification and translation of all processes in the production and "conditioning" of mango products, software development as well as capacity building of the five exporting organizations involved. The platform was developed in collaboration with the five exporting organisations under FRUILEMA. Since the launch, the platform is being further developed to allow for the use of PDAs and other mobile devices to enable direct data collection and data uploading while in the field at the farmers’ site, thus expediting the data collection, management and dissemination processes. II. DEMONSTRATION The demo will showcase the possibilities offered by the platform, by showing the producer, plot and produce information; tracking plot locations and size using Geographical Positioning Systems (GPS); aggregating information for the use of export planning; and other functionality offered by the platform. If possible, the newly developed access modalities (mobile & PDA) will be showcased live as well. The different user levels and their respective tasks and information provision within the system will be demonstrated. Although the existence of a website strengthened the credibility of the exporting partners and even led to concrete exporting contracts with companies in the Netherlands and Belgium, developing a database with all the necessary information does not guarantee a successful market information system. Many other factors need to be taken into account such as access to credit, physical infrastructure, capacities and marketing. The lessons thus far gained will be discussed during the demonstration as well.

480

Creating a Mobile-Phone Based Geographic Surveillance System for Avian Influenza Yibo Lin and Claire Heffernan Abstract—Highly pathogenic avian influenza (HPAI) is not only a global thread to human and animal health but also disproportionately impacts poor livestock keepers in southern countries. While billions have been spent on the disease, response to the epidemic remains fragmented and information channels slow. As such, this demonstration details a geographic mapping system at the global and local levels to aid information transfer among policy makers, practitioners and the poor themselves regarding the control of this disease. The demonstration will include hands on case studies in which conference attendees will be asked to both make decisions and use the tool in a simulated outbreak situation. Index Terms—Avian influenza, Geographic information systems, Surveillance system, User-Generated Content

I. INTRODUCTION

ACCORDING to WHO [6] ‘Avian influenza is an infectious

disease of birds caused by type A strains of the influenza virus’. In recent years, the emergence of highly pathogenic avian influenza (HPAI) strain (H5N1) has changed global perceptions regarding poultry, poultry production systems and the risk of animal diseases being transmitted to man. Indeed, H5N1 is able to be transmitted to humans (presently at low levels) but has a high fatality rate [5]. On a practical level, it has been argued that the ability to respond to this epidemic remains weak [4]. And that preparedness for the future pandemic, if the disease does jump the species barrier, is poor [3]. Part of the problem is that poor livestock keepers, who globally are the most impacted by the disease, are often marginalized from services and information delivery [2]. Therefore, present surveillance information are ineffective. Indeed, researchers have already declared that a new system, which directly connects veterinary researchers/practitioners to decision-makers is required [1]. Nevertheless, direct information links between the poor and their service providers is also crucial to the success of any new system. Manuscript received February 2, 2009. Yibo Lin is a research associate at the Livestock Development Group, School of Agriculture, Policy and Development, University of Reading, UK (e-mail: yibo.lin@ reading.ac.uk). Claire Heffernan is the Director of the Livestock Development Group, School of Agriculture, Policy and Development, University of Reading, UK (Tel.: +44 1189 318213, e-mail: [email protected]).

As such, this demo illustrates a geographic mapping system developed at both the local, regional and global level to aid surveillance and remove communication barriers between stakeholders. Unique to the system is its ability to RSS global outbreak information from the WHO database to rapidly inform stakeholders of confirmed outbreaks. II. DEMONSTRATION During the demonstration, both the regional and global systems will be presented. The regional system illustrates how primary data can be gathered by local veterinarians and such geographic information may be shared by mobile phone. Conversely, the global system illustrates a five ‘layered’ map representing the five critical factors of an HPAI outbreak: geographic spread, human mortality, number of poultry culled and wildlife migration pathways. Use of the system will be linked to practitioners in the field. Further, conference goers can explore the use of the tool from the perspective of the veterinarian in an outbreak situation. Thus, attendees will be offered a choice of three case studies. In each case, conference attendees will be asked to use the tool to make decisions in an simulated outbreak situation – and the consequences of their decisions will be illustrated by an on-line mapping system which demonstrates the spread of the disease. REFERENCES [1]

[2]

[3] [4]

[5] [6]

S. Chang, J. Zhang, X. Liao, X. Zhu, D. Wang, J. Zhu, T. Feng, B. Zhu, G. F. Gao, J. Wang, H. Yang, J. Yu and J. Wang, "Influenza virus database (ivdb): An integrated information resource and analysis platform for influenza virus research", Nucleic Acids Research, vol.35, pp. D376-D380, October 25 2006. C. Heffernan and F. Misturelli, "The delivery of veterinary services to the poor: Preliminary findings from kenya", in Report for DFID's (Department for International Development) Animal Health Programme (AHP), Livestock Development Group, The University of Reading, Reading, UK, 2000. M. T. Osterholm, "Preparing for the next pandemic", The New England Journal of Medicine, vol.352, no.18, pp. 1839-1842, May 5, 2005 2005. I. Stephenson, K. G. Nicholson, J. M. Wood, M. C. Zambon and J. M. Katz, "Confronting the avian influenza threat: Vaccine development for a potential pandemic", The Lancet Infectious Diseases, vol.4, no.8, pp. 499-509 2004. WHO, "Highly pathogenic h5n1 avian influenza outbreaks in poultry and in humans: Food safety implications", INFOSAN, WHO, Geneva, Switzerland, 2005. WHO, "Questions and answers on avian influenza", in A Selection of Frequently Asked Questions on Animals, Food and Water, WHO, Geneva, Switzerland, 2006.

481

Design of a Blood Flow System A. Osuntogun, S. Thomas, J. Pitman, S. Basavaraju, B. Mulenga and S. Vempala

Abstract—Blood is a vital but often scarce resource in developing countries. It is crucial that safe blood is made available for transfusions in hospitals and clinics to prevent the spread of Transfusion-Transmitted Infections (TTI) such as HIV, Hepatitis and Syphilis. We present a system designed to promote hemovigilance in developing countries by monitoring the collection and usage patterns of blood, predicting collection and usage for upcoming time periods and finding an allocation assignment for blood distribution that is fair and efficient. Index Terms—Health, Developing Nations, Blood Safety I.

INTRODUCTION

B

is a precious resource needed to save lives during surgeries and other major operations at a hospital. In many developed countries, there are systems and protocols in place to ensure that the blood is collected safely and only safe and healthy blood is transfused to a patient. In developing countries, past research has shown that transfusion of unsafe blood units has led to the spread of Transfusion-Transmitted Infections (TTI) such as HIV, HBV, HCV and Syphilis. The WHO estimates that the risk of HIV infection through transfusion of unsafe blood is exceptionally high in Africa [1]. This knowledge has led to the development of protocols in developing countries to ensure that only safe blood is being transfused. In addition to this challenge, many developing countries do not have enough safe blood to meet the needs of the country hence the need for appropriate methods for collection, testing and allocation of a limited blood supply. LOOD

II.

PROBLEM DESCRIPTION

To understand and improve the blood supply chain in developing countries, we conducted a series of interviews at the Zambia National Blood Transfusion Service. Due to the scarcity of blood, units are often moved multiple times between collection and transfusion to regions where blood is needed. Blood allocation is currently being done in an ad hoc manner in which a blood safety officer uses past and current internal knowledge about the amount of blood available and

A. Osuntogun., S. Thomas and S. Vempala are with the Georgia Institute of Technology. Atlanta, GA 30328 (e-mail:[email protected], [email protected], [email protected]). J. Pitman and S. Basavaraju are with the Centers for Disease Control and Prevention, Atlanta GA 30333 (e-mail: [email protected],[email protected]). B. Mulenga is with the Zambian National Blood Transfusion Service, Lusaka, Zambia (e-mail: [email protected]).

possible needs to determine how much blood should be distributed to a hospital. To ensure fair and efficient distribution of blood, it is necessary to monitor and understand the patterns of blood collection and requests and to have up-todate data on available blood. The system aims to maximize blood usage while minimizing or eliminating wastage. III. CONSTRAINTS To create a system that is sustainable in a developing country, it is necessary to understand the constraints. An assessment of the user skills indicate that the system should be user friendly since the main users will be blood bank and hospital staff such as doctors, nurses and biomedical scientists who should not be expected to have complex technical skills. It is also important to consider network availability and power supply for a system that needs current data to function maximally. Other concerns for developing countries include funding for the maintenance and upkeep of the system. These issues were taken into consideration during the development of the system. IV. SOLUTION

Our work focuses on the design of a blood flow system that tracks a blood unit from collection to utilization. Aggregate information on a large pool of blood units can then be used to determine trends and build probabilistic models for blood flow. The major components of this system include monitoring the collection and usage pattern of blood units, predicting the collection and usage for upcoming time periods and finding a flow assignment for blood distribution that is fair and efficient. The system serves as a method for visualization and analysis of blood data as well as a decision-informing tool. A live and interactive demonstration of the software system showing tracking, prediction and allocation of blood units will be given in order to facilitate understanding of functionalities of the system. REFERENCES [1]

N. Dhingra, “Making Safe Blood Available in Africa” .World Health Organization, Blood Transfusion Safety. 2006. http://www.who.int/bloodsafety/makingsafebloodavailableinafricastate ment.pdf

482

DISHA: DISease and Health Awareness for Children on Multiple Input Devices Mohit Jain, Aakar Gupta, Navkar Samdaria, Praveen Shekhar and Joyojeet Pal

Abstract— Much recent work in multiple input use scenarios for children’s learning software has focused either on math or on English language learning. The persistence of under-information among children in the developing world on issues of hygiene and disease prevention remains a massive challenge within the scholarly community in public health, especially in the developing regions that multiple input learning technologies are designed for. DISHA is a collaborative platform for public health information for children in low-income regions using multiple mice. The system is designed towards collaborative use of screen resources. Index Terms— Computer Aided Learning, Multiple Mice, Developing Regions, Health Education.

G

I. INTRODUCTION

LOBAL health statistics show a staggering amount of deaths and disabilities due to preventable diseases including respiratory infections, diarrhea, malaria and tuberculosis that almost exclusively impact the developing world [1]. The issue of public health education is a relatively unexplored area within the domain of Computer Aided Learning (CAL), despite past evidence of effectiveness of audio-visual material for hygiene education, as well as the currently prevalent use of animation in static learning materials for children. With DISHA, we aim to supplement CAL programs, increasingly prevalent in the developing world including many areas with high risk of some of the aforementioned diseases. DISHA encompasses every aspect including the symptoms, prevention and cure of these diseases. Using a narrative-interactive loop format of story-telling followed by multiple choice Q&A, DISHA uses MultiPoint technology [2, 3] for healthcare learning material. The Q&A enforces collaboration, competition and turn taking as machine-induced interactive modules, following a game based approach. II. IMPLEMENTATION Giving each child a mouse of his or her own, DISHA aims Manuscript received January 30, 2009. Mohit Jain, Aakar Gupta, Navkar Samdaria and Praveen Shekhar are with Dhirubhai Ambani Institute of Information and Communication Technology, Gandhinagar, Gujarat, India (phone: +91 9879031798, +91 9904866639; e-mail: {miohit_jain, aakar_gupta, navkar_samdaria, praveen_shekhar} @daiict.ac.in). Joyojeet Pal is with University of Washington, Seattle, USA (phone: 510501-8679; e-mail: [email protected]).

to both spread the use of a single computer wider across all users as well as increase engagement between the children. We adopt a hybrid approach of racing and unity models [4], thus two teams of children play against each other by connecting to a server, thus ensuring collaboration among the team mates and competition among the teams. An Artificial Intelligence backbone tracks the performance of individual children and distributes on-screen prompts based on the performance of children and the areas they appear to need help. The idea of profiling performance over time used in DISHA also has an important machine language element in tabulating a child’s competencies that can be applied across various learning subjects for young children, which in turn can provide useful cross-domain data on learning skills. III. INTERFACE FEATURES We employ in our design features that have been seen as effective in past trial applications with education in the developing world. This includes some standard design principles of gaming including goal-based progression, competition, positive reinforcement, personalized scoring and other principles specifically relevant to ICTD such as Samelanguage Subtitling (SLS) and color and shape-coded mouse pointers. The interface is designed intuitively for a child’s use, assuming minimal availability of teachers. IV. DEMONSTRATION We demonstrate (a) on a single laptop with four mice the use of DISHA in a real world, (b) two teams playing against each other on two laptops connected to a server. REFERENCES [1] [2] [3]

[4]

Global Disease Statistics http://www.fightingdiseases.org/main/globalstats.php, Accessed January 30, 2009. U. S. Pawar, J. Pal and K. Toyama, "Multiple Mice for Computers in Education in Developing Countries," in Proc. of IEEE/ACM ICTD'06, 2006, pp. 64-71. The Microsoft Windows MultiPoint Software Development Kit (SDK). http://www.microsoft.com/downloads/details.aspx?FamilyID=A137998 B-E8D6-4FFF-B805-2798D2C6E41D&displaylang=en, Accessed January 30, 2009. S. Thota, J. Pal, U. S. Pawar, A. Joshi, M. Jain, S. Teja and S. Anikar, "From Pilot to Practice: Creating Multiple-Input Multimedia Content for Real-World Deployment," in Intelligent User Interfaces for Developing Regions, 2008, Canary Islands, Spain.

483

Freedom Fone: Dial-up Information Service Bev Clark and Brenda Burrell

Abstract— This demonstration highlights the latest advances of Freedom Fone: Dial-up Information Service. Freedom Fone provides organisations with a simple to use interface to manage Interactive Voice Response (IVR) menus and upload prerecorded content into an “audio magazine.” Members of the public can then phone in to listen to the latest information, or to leave a message. Freedom Fone prototype software has been piloted in Zimbabwe and South Africa, with sexual health, women’s rights and cholera awareness information. Index Terms—Africa, Information, Mobile communications

F

reedom Fone is a free open source software tool that can be used to build and update a dial-up information service in any language. Its easy to use interface lowers the barriers to using Interactive Voice Response (IVR) for outreach to communities small and large, near and far. Freedom Fone empowers non-technical organizations to build automated information services that are available to the public 24 hours a day, 7 days a week – power and connectivity allowing. Based on the concept of Interactive Voice Response (IVR), pre-recorded audio files are stored by Freedom Fone in a Content Management System (CMS). This is updated through a simple to use browser interface. Content is made available to telephone callers when they dial phone numbers associated with the service. Users are presented with an audio menu of options and using their telephone keypads select channels carrying information of interest to them. They are able to listen to a selection multiple times if they so choose. Deployment in any language is possible as key global files for menu prompts can be uploaded through the browser interface to the CMS. Individuals can contribute questions, content and feedback by leaving voice messages via the IVR interface. Freedom Fone can be operated as a collective, with different groups managing different channels (IVR menu options) of information from the same installation. Currently available as a prototype, Freedom Fone can be connected to telephony services via the internet using VoIP – this is the most scalable option; or via telephony cards to either analog or digital phone lines. Services can be offered at Manuscript received February 19, 2009. B. Clark is Creative Director of the Kubatana Trust of Zimbabwe (Phone: +263-4-776-008; fax: +263-4-746-418; e-mail: [email protected]). B. Burrell is Technical Director of the Kubatana Trust of Zimbabwe (Phone: +263-4-776-008; fax: +263-4-746-418; e-mail: [email protected]).

cost or free of charge to callers via toll free lines and callbacks. Members of the public can also request callbacks via SMS. The Freedom Fone prototype has been used to pilot the dissemination of sexual / reproductive health information with adolescents in Zimbabwe, and as an information service at the Association for Women’s Rights in Development conference in Cape Town in November 2008. We are currently trying to establish a cholera information service in Zimbabwe, in association with local and international health agencies. Elements used by Freedom Fone aren’t new, but their intuitive convergence makes it innovative and globally relevant. Global Voices' Ethan Zuckerman said recently "I'm surprised there hasn’t been more work done making interactive voice response systems usable for development purposes [1]." Freedom Fone provides exactly that. Other projects using Interactive Voice Response to make critical development information more available local communities are VoiKiosks, which “present an alternative platform to create and host such information kiosks in the telephony network,”[2] and MobilED, whose “Audio Wiki” programme uses a range of software products including Asterisk, Kannel and Flite to make audio information available to learners [3]. REFERENCES [1]

[2] [3]

Ethan Zucherman. “Delivering Ethiopian teff via Czech taxi?” My Heart’s in Accra Blog. October 13, 2007. http://www.ethanzuckerman.com/blog/2007/10/13/delivering-ethiopianteff-via-czech-taxi/ Sheetal Agarwal, Arun Kumar, Amit Anil Nanavati, Nitendra Rajput. “VoiKiosk: Increasing Reachability of Kiosks in Developing Regions”, WWW 2008 / Poster paper. Server – MobilED website http://mobiled.uiah.fi/?page_id=98

484

FrontlineSMS and Ushahidi – A Demo Ken Banks and Erik Hersman

Abstract—FrontlineSMS and Ushahidi are both free and open source tools which allow for the collection, dissemination and visualization of data collected to and from the field. This paper outlines details of live demonstrations of the tools at the ICTD 2009 conference in Qatar. Index Terms—FrontlineSMS, Ushahidi, data, messaging, open source tools, mobile, SMS

I.

INTRODUCTION

FrontlineSMS and Ushahidi are two fully free and open source tools which leverage information and communication technologies – of which mobile is a key component – to allow individuals or groups to collect and/or disseminate information to and from the field. The tools work independently, but can also work seamlessly together, as they are currently doing in a deployment in DRC. Ushahidi, which means “testimony” in Swahili, is a free-toaccess website that was developed to map reports of violence in Kenya after the post-election fallout at the beginning of 2008. Ushahidi's roots are in the collaboration of Kenyan citizen journalists during a time of crisis. The new Ushahidi Engine has been created using the lessons learned from Kenya to create a platform that allows anyone around the world to set up their own way to gather reports by mobile phone, email and the web - and map them. The core engine is built on the premise that gathering crisis information from the general public provides new insights into events happening in near real-time. It is being developed by a group of volunteer developers and designers, hailing primarily from Africa. So far there are representatives from Kenya, South Africa, Uganda, Malawi, Ghana, Netherlands and the US. A post-mortem study by the Harvard Humanitarian Initiative provided qualitative analysis on the added information gathering opportunities that this type of platform creates. A report titled, “Digitally Networked Technology in Kenya’s 2007–2008 Post-Election Crisis” from the Berkman Center for Internet and Society also used Ushahidi as an example of how mobile phone messaging can be used for good in times of crisis. FrontlineSMS is free software that turns a laptop and a mobile phone or modem into a central communications hub, allowing users to send and receive text messages with large groups of people through mobile phones. The software – originally

developed in 2005 and updated in 2007 – is being used around the world for a wide range of non-profit activities including the sending of market prices and other agricultural data to smallholder rural farmers in Aceh, Cambodia and El Salvador, the dissemination of news in Iraq, the sending of security alerts to fieldworkers in Afghanistan, for human rights work in places such as Zimbabwe, Pakistan and the Philippines, and the running of a rural healthcare network for 250,000 people in Malawi. Because the software can be used on a single laptop computer without the need for the internet, it has been widely adopted among the grassroots non-profit community and nominated for several awards. FrontlineSMS is currently the focus of a number of academic studies, and the recent UN Foundation/Vodafone report on mHealth described it as “one of the largest and most ambitious mHealth programs in the world”.

II. PRESENTATION / DEMO

Representatives from FrontlineSMS and Ushahidi will show how these two tools can be used independently, and in tandem, to change the way organizations large and small can communicate. This isn’t just technology for the developing world, but a way to understand and use technology on a local level that fits the everyday needs of ordinary people. Examples of users from various parts of the world will be shown, statistics from past deployments shared, and the technology will be demonstrated live. In the demonstration, a local tech hub (using a PC running FrontlineSMS) will be set up and the attendees will see a live display of real time submissions, response and plotting of this data on a global map immediately available to the world. It will show how powerful real-time data collection from disbursed and open communities can be visualized, and then acted upon by others, and how two-way group text messaging can be used to help run and co-ordinate field activities, and deliver information services to rural communities. Ken Banks is the Founder of kiwanja.net and developer of FrontlineSMS, and Erik Hersman the Technical Director at Ushahidi. Both have extensive experience working in the ICT4D field, and both actively maintain websites and Blogs on the application of technology in the developing world.

485

Global Youth Connectivity (GYC) – Using ICT for Peaceful Recovery and Long-term Change Annie Bertrand Abstract—ICT enables peaceful change in communities affected by conflicts, poverty and injustice. In this demo, we illustrate how young people use multimedia, cell phones and the internet to become global citizens responsible for building secure, productive and just communities at home and around the world. Global Youth Connectivity allows disaffected youth to express thoughts and feelings, while also scaling global citizenship education and civic engagement across borders. Thousands of Middle-Eastern and American youth have exchanged content and discussed online since 2005. Innovations and partnerships should enable Mercy Corps to empower millions more living in extreme vulnerability.

war, Ziad from Gaza City commented: "The situation is really

Index Terms— Civic engagement, Global Citizenship, Global Youth Connectivity

Many young people in the Middle-East suffer from repression, isolation and limited opportunities for social or civic engagement. Mercy Corps and IJMA35 are working on a grassroots news network with information organized by youth for youth. After a newspaper-based pilot in Iraqi villages, the media outlet will be scaled-up to reach millions of young Arabs across gender and social groups using internet (enewsletter, RSS feed, widgets) and GSM (SMS text and VOICE news). This digital peer-to-peer education process also enables anonymous feedback and asking of questions to be addressed. As the content evolves and feeds online dialogues, young reporters will also be trained on taking collaborative peaceful actions at the local, regional and international levels.

T

I.

INTRODUCTION

he Global Youth Connectivity1 (GYC) program called “WhyNot!” was established by Mercy Corps as part of a relief and development strategy to address psychosocial needs and reduce the widening knowledge gap in the Middle-East. Once computer centers are built2, we train young first-time users on digital photos, flip videos and PowerPoint to illustrate their culture, voice their concerns, and connect online with others living abroad. Global engagement helps those affected heal their wounds and change from “powerless beneficiaries” into active contributors to peace, productivity and justice at home. The curricula allow diverse youth to explore cultural differences and find common grounds around peaceful change. Such sharing and moderated dialogues also teach American youth about global issues from the perspective of those directly affected by them. II. EXPRESSION + GLOBAL CONNECTION = EMPOWERMENT

During the bombardments in Gaza earlier this year, we used cell phones to give internally displaced youth a chance to express concerns and be heard. Innovative software developed by Souktel3 allowed youth to ‘text’ insights to a central database, which fed into our blog. Reactions from peers abroad with inspiring words of sympathy had positive psychosocial effects on their healing process. In the midst of

bad in Gaza (…) I am happy about the posts on the blog. It is really nice to know that there are people who care about us and support us…this is encouraging.” Rebuilding hope is sometimes harder than infrastructure but a necessary condition for peaceful recovery. According to a UNDP report on Arab youth, global communication systems also provide a gateway for youth to delve into creativity, innovation, learning, professionalism, and democratic governance: “Their self-confidence is boosted when they expand their practical knowledge and communicate with their peers around the world” 4.

III.

IV. DEMONSTRATION

During the demo session, we will show how the Global Citizen Corps portal aggregates web, mobile, and other digital media content into a single, youth-friendly and accessible online resource. Inspiring extracts of youth content such as video-documentaries, photo-essays, and online dialogues will allow conference participants to witness the role that ICT plays in transforming young people into global citizens, capable of and responsible for building a secure, productive and just world now and across generations. REFERENCES [1] [2] [3]

Annie Bertrand is Director of Global Youth Development at Mercy Corps supporting youth transformation across 35 countries. She graduated with a MBA from Harvard Business School in 2007. (phone: +1 518 282 4783; email: [email protected])

INFORMATION + COMMUNICATION = PROGRESS

[4] [5]

Global Youth Connectivity www.globalcitizencorps.org/whynot “Centers for Excellence” in Iraq and Lebanon were established to provide access to over 500 computers connected to internet. Souktel uses SMS text messaging to link young people with jobs and aid agencies in Palestine and beyond www.souktel.org Arab Youth Strategising for the MDGs (2006), UNDP, p.37 IJMA3 is a partner of Mercy Corps in implementing ICT projects in the Middle-East www.ijma3.org

486

Implementing E-Government Accessible to Illiterate Citizens D. Kettani and A. El Mahdi known as: Electronic Fundamental Etat-Civil System (eFES). Abstract— Information and Communication Technologies (ICT) are becoming more and more popular as a tool in the effective administrations of local and national governments worldwide. Nevertheless, developing countries still lag behind. One main challenge is automating the back office associated with public administrations. Another challenge is designing and implementing ICT systems accessible to illiterate users. This Demo presents the Fez e-government platform deployed in Fez, Morocco. The intention is to first, explain the workings of the front office kiosk and how illiterate users request and procure municipal certificates and second, to outline how the back office is integrated into the kiosk and other regional administration. Index Terms— ICT-based municipal system, illiterate users, touch-screen kiosk, self-service technology

M

orocco, a North Africa developing country, has expressed interest in Information and Communication Technologies (ICT) since the beginning of the 1990s. Yet, the development of Morocco’s e-government national strategy took twelve years (1993-2005). One outcome for such a delay was the absence of concrete ICT actions to benefit the daily life of ordinary citizens. In response to Morocco’s low ICT penetration, action was taken on by the Fez municipality to collaborate with the ICT4D research laboratory at Al Akhawayn University in 2004. Thanks to research funding granted by Canada’s International Development Research Center (IDRC), collaboration focused on building an e-Government system for the city of Fez to initiate the introduction of ICT to Morocco’s local administration. In 2004, the research team opted for focusing on developing a platform automating the delivery of highly and widely requested municipal certificates, Manuscript received January 30, 2009. This work was supported by the International Development Research Center (IDRC) under Grant 104053. The project team hereby expresses its appreciation and gratitude to the IDRC for funding eFez project; without which this research work would have not been possible. Driss Kettani is an Associate Professor at the School of Science and Engineering and Head of the ICT4D Research Laboratory at Al Akhawayn University in Ifrane, Morocco. P.O.Box: 2122, Ifrane 53000, Morocco. (email: [email protected] ) Asmae El Mahdi is currently with Enhanced Technologies, Al Akhawayn University Technopark; a research team member at the ICT4D Research Laboratory at Al Akhawayn University in Ifrane, Morocco. Avenue Hassan II, B.P 2072, Ifrane 53000, Morocco. (corresponding author phone: 00212 35 86 2114; e-mail: [email protected] ).

The novelty of eFES system, completed in 2006 as a result of Public-Academia-Partnership, rests on the following: - Completing an ICT-based municipal system that enables the unprecedented shift from cumbersome manual service delivery to rapid, more efficient automated service delivery in Morocco’s developing country context - Succeeding the first-time introduction of “self-service technology” to local public administration. eFES project initiated the introduction of touch-screen kiosk to Fez municipal offices so that community members could have the new possibility of requesting the needed certificates without full dependency on employee intermediation (commonly known for abusive and citizen-unfriendly treatment). More importantly, the innovative qualities of eFES platform lie on initiating the introduction of “self-service technology” while taking into consideration Morocco’s high illiteracy rate. Specifically, eFES conducted extensive research to adapt the e-government touch-screen kiosk to an illiterate user profile. This involved exploring the different architectures, scenarios and technologies for designing ICT-based system adapted to illiterate users. It required also creating design principles and conducting usability testing in real life settings with real users (i.e. community members). Developing ICT solutions that address accessibility issues related to basic and digital illiteracy is a key in ICT growing use, adoption, and appropriation amongst the masses in developing countries. eFES significance comes from developing a methodology and architecture for e-government application implementation fulfilling the needs in the Arab World. This Demo intends to present Fez e-government system (commonly known as eFES); it aims to share the achievements of introducing ICT to local public administration in Arab developing country context. The eFES Demo would like to show the touch screen kiosk adapted to illiterate profile. It intends also to show video footage of kiosk uses in Fez real life settings.

487

Improving Data Quality With Dynamic Forms Kuang Chen, Harr Chen, Neil Conway, Heather Dolan, Joseph M. Hellerstein, and Tapan S. Parikh

Abstract—Organizations in developing regions want to efficiently collect digital data, but standard data gathering practices from the developed world are often inappropriate. Traditional techniques for form design and data quality are expensive and labour-intensive. We propose a new data-driven approach to form design, execution (filling) and quality assurance. We demonstrate U SHER, an end-to-end system that automatically generates data entry forms that enforce and maintain data quality constraints during execution. The system features a probabilistic engine that drives form-user interactions to encourage correct answers.

I. I NTRODUCTION Governments, companies, and individuals routinely make important decisions based on inaccurate data stored in supposedly authoritative databases. In healthcare, a simple error may have fatal consequences. While data quality can be addressed at every stage of the data life-cycle, from creation to archival, we believe that entry-time is the first and best opportunity to improve the quality of manually-entered data. There is much prior work on improving the quality of data that already resides in a database [1]. However, relatively little attention has been paid to improved techniques for data entry. Survey design [2] has long informed the design of data entry forms, applying principles for data encodings, constraints, and validation rules. For electronic forms, quality assurance during entry has centered on the ubiquitous and costly practice of double-entry [3]. Current standards have failed to take advantage of new technology: pervasive cellular networking and low-cost mobile devices allows even remote users to interact with data entry systems that could potentially provide rich feedback. For organizations with limited resources, existing standards are neither practical nor attainable. In such settings, designing data collection instruments is too often an ad hoc practice, consisting of mapping desired information elements to a set of entry widgets (text fields, combo boxes, etc.), guided only by the designer’s intuition. According to recent work on data collection in resource-poor settings, lack of expertise and difficulty of remote data collection are the chief obstacles to high data quality [4]. In our previous fieldwork with a well-funded HIV/AIDS treatment program in East Africa, we found that little thought was given to form design, and a haphazard double-entry program bottlenecked the data entry process to a substantial degree; in fact, the program’s health K. Chen, N. Conway and J. M. Hellerstein are with the Computer Science Division, University of California at Berkeley (email: {kuangc, nrc, hellerstein}@cs.berkeley.edu). H. Chen is with the Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology (email: [email protected]) H. Dolan and T. S. Parikh are with the School of Information, University of California at Berkeley (email: {dolan, parikh}@ischool.berkeley.edu) Manuscript accepted Feb 10, 2009.

clinic operations were limited to using paper forms to ensure timely information access. Only after a labor-intensive delay did the medical researchers enjoy the benefits of digital data for research and analysis. We have built a system called U SHER that maximizes data quality at entry-time using statistical data modeling, dynamic interfaces, and collaborative insight. Guided by prior data, U SHER learns probabilistic relationships in the data to train a model, which is then applied to automatically generate forms with the appropriate constraints. U SHER then provides realtime feedback during the data entry process to dynamically guide (or usher) the user toward better data quality. Based on a list of form questions and a sufficient set of answers, U SHER optimizes the form’s question-ordering and layout, mimicking survey design principles. During form entry, U SHER provides dynamic data-quality feedback to the user. When the user enters a value, U SHER automatically decorates the interface with hints and warnings if the answer is deemed “risky.” Decoration choices are probabilistically guided, and include auto-complete, correctness-thermometers, warning/error flags, and other scented widgets [5]. U SHER also invites the user to write and view comments about form questions or data instances for and by other users. Finally, U SHER mimics double-entry by choosing to re-ask questions with responses likely to be erroneous, based on the probabilistic model. II. D EMONSTRATION Our demonstration will show U SHER’s ability to approximate expert form design and double-entry based only on prior data, both on a PC and a mobile device. Using a real dataset from a rural health organization, users will be able to 1) automatically extract training data from a Microsoft Access database; 2) refine the automatically designed form; and 3) execute the forms with and without smart decorations and quality assurance. R EFERENCES [1] J. M. Hellerstein, “Quantitative data cleaning for large databases,” United Nations Economic Commission for Europe, 2008. [2] R. M. Graves, F. J. Fowler, M. P. Couper, J. M. Lepkowski, E. Singer, and R. Tourangeau, Survey Methodology. Wiley-Interscience, 2004. [3] S. Day, P. Fayers, and D. Harvey, “Double data entry: what value, what price?” Controlled Clinical Trials, 1998. [4] J. V. D. Broeck, M. Mackay, N. Mpontshane, A. K. K. Luabeya, M. Chhagan, and M. L. Bennish, “Maintaining data integrity in a rural clinical trial,” Controlled Clinical Trials, 2007. [5] W. Willett, “Scented widgets: Improving navigation cues with embedded visualizations,” IEEE TVCG, 2007.

488

IWB4D – Interactive Whiteboards for Development John Traxler and Lee Griffiths

Abstract — The work presented here describes the concept of ubiquitous interactive classrooms by demonstrating emerging, low-cost presentation technologies including the Nintendo Wii and palm-sized projectors. The authors also discuss how they can be used to promote better learning through ad-hoc digital interaction in traditional classrooms and the field. Index Terms — active white boards, mobile learning, whole class teaching

T

I. WHITEBOARDS

here is considerable literature, for example [1], [2], [3], [4], [5] & [6], showing the extent to which whole class teaching in core subjects can be enhanced by interactive whiteboards. These technologies are however extremely expensive and thus, untried in a development context. The Nintendo Wii games console has a unique form of interactivity. Since its release programmers have developed example systems which implement futuristic interfaces with the Wii console’s hand controller or Wiimote which can track moving objects with relative ease. One example of interest to educationalists working in ICTD is using the device to implement an interactive whiteboard. Our demonstration will illustrate teacher mobility and student interaction on a very low budget by demonstrating the ad hoc use of the Wii device to implement interactivity. Participants will see a live demonstration of a personal interactive whiteboard using the Nintendo Wii and discuss teacher mobility and student interaction in a development context using emerging technologies and how to set the system up in their own working environment. II. THE DEMONSTRATION The learning goals of the demonstration are that the participants will gain the technical skills and competence to procure and set up an inexpensive interactive whiteboard and will engage with ideas for improved and innovative teaching John Traxler is Visiting Scientist, Meraka Institute, Pretoria; Director, Learning Lab, Shropshire, European Chair ICTD Qatar & Reader in Mobile Technology for e-Learning, University of Wolverhampton; [email protected], +44 (0)1902 518591 Lee Griffiths is Faculty Learning Technologies Fellow and Lecturer in Computer Science, University of Salford; [email protected], +44 (0)161 2953658

including classroom practice and distant, remote or on-site mobile teaching. The topics covered will include a general introduction to interactive whiteboards, an introduction to the capability and functions of Nintendo's Wii, the development of a personal interactive whiteboard, the use multiple input devices and true mobility, a critique and discussion of whiteboard disadvantages, the set-up and download of the system, and how to enhance teacher mobility in diverse settings. REFERENCES [1]

[2]

[3]

[4] [5]

[6]

BECTA (2003). What the research says about interactive whiteboards. Coventry, England: British Educational Communication and Technology Agency. Retrieved 10 January2009 from http://www.becta.org.uk/page_documents/research/wtrs_whiteboards.pd f Glover, D., & Miller, D. (2001). Running with technology: the pedagogic impact of the large-scale introduction of interactive whiteboards in one secondary school. Journal of Information for Teacher Education 10 (3), 257–276. Glover, D., & Miller, D. (2002, November 11). The introduction of interactive whiteboards into schools in the United Kingdom: Leaders, led, and the management of pedagogic and technological change. International Electronic Journal for Leadership in Learning 6 (24). Retrieved January 5, 2009 from http://www.ucalgary.ca/~iejll/volume6/glover.html Glover, D., Miller, D., Averis, D., & Door V. (2005) The interactive whiteboard: a literature survey. Technology Pedagogy and Education 14 (2), 155–170. Hall, I., & Higgins, S. (2005). Primary school students’ perceptions of interactive whiteboards. Journal of Computer Assisted Learning 21 (2), 102-117. Retrieved February 2, 2007, from EBSCOhost Professional Development Collection. Hall, I., & Higgins, S. (2005). Primary school students’ perceptions of interactive whiteboards. Journal of Computer Assisted Learning 21 (2), 102-117. Retrieved February 2, 2007, from EBSCOhost Professional Development Collection. Tanner H., Jones S., Kennewell S., & Beauchamp G. (2005) Interactive whole class teaching and interactive white boards. In Building Connections: Theory, Research and Practice: Proceedings of the Annual Conference held at RMIT, Melbourne, 7th-9th July 2005, 720– 727.Pymble, N.S.W.: The Mathematics Education Research Group of Australasia. Retrieved January 12, 2009 from http://www.merga.net.au/documents/RP832005.pdf.

489

Livestock, Learning and Diagnostics: New Directions in Veterinary Tele-medicine Jun Yu and Claire Heffernan Abstract - It has been estimated that 2/3rds of the individuals living on less than $2/day own or keep livestock. However, research has illustrated the difficulties that this population face in obtaining accurate and relevant information regarding animal health. Therefore, the demonstration offers two tools to support the animal health needs of the poor in India. First, the creation of a mobile phone based knowledge transfer tool is described and second, a mobile phone-based diagnostic tool to enhance the capacity of the poor to diagnose diseases is also detailed. The demonstration of the tools will support the ‘hands-on’ exposure of conference goers with problems faced by the poor and the role of ICTs in their solution. Index Terms - Information Delivery System, Livestock Diseases, Mobile Phone, Poverty

I. INTRODUCTION

ICTs are critical to addressing the animal healthcare needs of

poor livestock keepers in the South [4]. First, such tools may be applied as a means of knowledge transfer and secondly, ICTs can help redress historical gaps in animal health service delivery to the poor. Indeed, research has identified that women and children, while primarily livestock care-takers are often the most excluded from knowledge transfer [5]. Equally, disease diagnosis is particularly problematic for the poor [5]. Therefore, in recent years there has been an increasing interest in tele-veterinary medicine to address animal healthcare gaps in southern countries. Mobile diagnostic technologies are a rapidly emerging field in tele-medicine (both human and animal) and are defined by two key characteristics: mobility and remote diagnosis [2]. In the following demonstration, the authors display three linked and inter-related tools to address these issues: first, a mobile phone version of the Livestock Guru, a multi-media learning program for poor farmers in the South; second, multi-media learning games for children and third, a Mobile Diagnostic System (MDS), which supports the diagnosis of Mastitis among poor dairy producers in India. The MDS enables farmers to take a picture of their sick animal by a mobile phone and then send this picture to a kiosk, which subsequently makes the diagnosis. The reason for the initial Manuscript received February 2, 2009. Jun Yu is a PhD student at the Livestock Development Group, School of Agriculture, Policy and Development, University of Reading, UK (e-mail: aar04jy@ reading.ac.uk). Claire Heffernan is the director of the Livestock Development Group, School of Agriculture, Policy and Development, University of Reading, UK (Tel.: +44 1189 318213, e-mail: [email protected]).

focus on mastitis was twofold. First, mastitis, historically, is difficult for farmers both in the North and in the South to diagnose. Second, delayed diagnosis has a detrimental impact on milk production and consequently income. As such, the disease has a large impact on the livelihoods of poor dairy producers, particularly in India [1]. The reason for choosing a mobile phone application is twofold. First, mobile phones are better from both a cost and battery life perspective [3]. Second, by connecting via Bluetooth to the Livestock Guru kiosk, the MDS can link to learning information provided by the Guru in relation to the livestock disease in question. Nevertheless, a version of the MDS entirely contained within a mobile phone was created and will be demonstrated as many poor livestock keepers may be too remote from a server (The Livestock Guru kiosk), in this manner, a farmer may take a picture of his or her animal and the mobile MDS will provide a diagnosis. The software was created based upon a combination of prototype matching and statistical pattern recognition. II. DEMONSTRATION

In the demonstration, a mobile phone based on Symbian S40/60 OS will be utilized to display the Mobile Livestock Guru. Its application among poor farmers in India (including women and children) will be detailed as part of the display. Additionally, use of the MDS will also be shown. As such, a display of symptom pictures will be offered and conference goers may utilize a mobile phone to take and send symptom pictures to a kiosk computer to make a diagnosis of a sick animal. Alternatively, a mobile phone stand-alone version MDS will be also available for conference goers to test. Further, a video of the tool in use in India will also be provided. REFERENCES [1]

[2] [3] [4] [5]

C. Heffernan, D. Pilling, J. Yu and Y. Lin, “Listening to the Voices of the Poor: Enhancing evidence-based decision-making in livestock development”, Nottingham University Press, Nottingham, United Kingdom, 2008. M. Papageorges, P. Herbert, “Telemedicine using standard internet technologies (telemedicine for less than $1000)”, Clinical Techniques in Small Animal Practice, vol. 16, no. 2, pp. 102-107, May 2001. L. Naismith, “Literature Review in Mobile Technologies and Learning: A Report for NESTA Futurelab”. Bristol: NESTA Futurelab, 2004. S. Warren, “A distributed infrastructure for veterinary telemedicine, in proceeding of the 25th annual International Conference of the IEEE EMBS, September 17th-21st, Cancun, Mexico, 2003. Yu, J. “Improving Development Information flows: The creation of ICTs for poverty alleviation in the livestock sector”, Thesis (submitted). The University of Reading, Reading, UK.

490

Metamouse: Multiple Mice for Legacy Applications Kurtis Heimerl (kheimerl)∗ , Divya Ramachandran (divya)∗ , Joyojeet Pal† , Eric Brewer (brewer)∗ and Tapan Parikh‡ ∗ Computer

Science Division University of California, Berkeley Berkeley, CA 94720-1776 (name)@cs.berkeley.edu

† School

of Information University of Washington Seattle, WA 98195-2840 [email protected]

Abstract—Traditional Single Display Groupware (SDG) solutions have been used to create software for developing regions. SDG allows for greater utilization of the limited infrastructure available in such areas. This technology is particularly appropriate in classrooms where groups of children often use educational software on a single computer. However, SDG has faced several challenges working with legacy applications. Our technology, called metamouse, takes a step toward an integrated multi-user application by allowing users to collaborate within unmodified legacy educational software.

I. P ROBLEM S TATEMENT Several million children, especially in the developing world, never use a computer without a partner sharing the mouse and keyboard [2]. Single Display Groupware techniques [4] are particularly relevant to this scenario, allowing for users to more effectively share computing resources by using multiple mice. Existing research has shown that these techniques are intuitive, usable, and beneficial in the developing world [3]. Despite these encouraging early findings, two important barriers remain in deploying shared input technologies in real world settings. First, most legacy software would have to be significantly modified to effectively use multiple mice, but among other issues, many business challenges exist in doing this. The second barrier, of interest from the HCI perspective, is that of efficiently encouraging coordinated onscreen decision-making using multiple mice. Most trials of multimouse, while emphasizing the importance of collaboration, have been impeded by working within a “fastest-fingerfirst” race-clicking, or repetitive click modes which require all users to click on the same link for the function to continue. Our challenge was to work with legacy software, and yet provide an alternate way for allowing coordinated clicking to capture the proven learning gains of collaboration. II. S OLUTION To solve these problems, we developed the metamouse. The metamouse does two things. First, it maps multiple user mice and cursors down to one metacursor that interacts with the application. This allows for the use of unmodified legacy applications. Secondly, we have to make sure that this mapping is intuitive and encourages collaboration among users. We began by trying to create an effective mapping for pointand-click applications. Tackling other tasks is left as future

‡ School

of Information University of California, Berkeley Berkeley, CA 94720-1776 [email protected]

work. To do this mapping effectively, we have to map mouse movements as well as mouse clicks. A. Movement To map the many cursors down to one, we use a well known SDG technique. We average the mouse locations, and display a metacursor at that position. With this, each user’s actions have a small, but noticeable, impact on the metacursor’s location. Likewise, if all of the users agree on the position of the cursor, the metacursor will be at that same location. This encourages collaboration, as the users must discuss and agree to place the metacursor where they want it. B. Clicking We can’t allow the users to click at any time, as the metacursor may be at an inappropriate location. To solve this, we only allow clicking when the mice are in close proximity to each other. When distant, all of the clicks are ignored. The intuition is that if the mice are close to each other, then the users have agreed on the correct location to make progress, and clicking should proceed. This encourages collaboration; all users must agree on a location before making progress. III. D EMONSTRATION We are going to demonstrate the point-and-click metamouse system running on Microsoft Windows XP. We will run a set of educational games from the following vendors using metamouse and four individual mice. • The Azim Premji Foundation [1] • Disney online • Leapfrog R EFERENCES [1] Azim Premji Foundation. http://www.azimpremjifoundation.org. [2] J. Pal, U. S. Pawar, E. A. Brewer, and K. Toyama. The case for multi-user design for computer aided learning in developing regions. In WWW ’06: Proceedings of the 15th international conference on World Wide Web, pages 781–789, New York, NY, USA, 2006. ACM. [3] U. S. Pawar, J. Pal, and K. Toyama. Multiple mice for computers in education in developing countries. In IEEE/ACM Intl Conf. on Information and Communication Technologies for Development, 2006. [4] J. Stewart, B. B. Bederson, and A. Druin. Single display groupware: a model for co-present collaboration. In CHI ’99: Proceedings of the SIGCHI conference on Human factors in computing systems, pages 286– 293, New York, NY, USA, 1999. ACM.

491

Mobile Phone Job Services: Linking Developing-Country Youth with Employers, via SMS Amber Houssian, Mohammad Kilany, and Jacob Korenblum Abstract—With its low cost and high penetration rates, mobile technology is uniquely poised to address developing-country labor market inefficiencies, by boosting young job-seekers’ access to employment information. Simple services like the SMS-based “job matching” application developed by Palestinian-Canadian NGO Souktel have transformed local labor markets--by giving young job-seekers real-time information about local jobs, and connecting them directly with employers.

people than conventional labor market methods. In its demonstration session, team members will use mobile phones to show the end-user software experience from jobseeker/employer vantage points, illustrating how the application matches youth with employers in five easy steps 1) Through short SMS surveys, youth users create basic “mini-CVs”. These include data on age, location, skills, education level, and career interests. 2) At the same time, employers create simple SMS “job ads” through a similar SMS Q & A process. 3) Both sets of data are sent directly from users’ phones to a central database. 4) Database matches youth “mini-CVs” with relevant “job ads”—pairing identical information from youth and employers. 5) Users then receive “job-match” messages directly to their mobile phones—with details about job opportunities or potential staff, along with phone numbers for follow-up interviews. Users can also browse for jobs or staff from their phone handsets.

I. CONTEXT: JOB MARKET GROWTH, BUT NO GOOD WAY FOR JOB-SEEKERS TO FIND INFORMATION ABOUT WORK

I

N the past decade, developing-country economies have grown impressively each year—a promising prospect for local youth, who often make up the bulk of new labour market entrants. However, the reality is much less rosy: While GDP is growing, young workers are still failing to find available jobs in key fields. In emerging markets, one of the main challenges is basic connectivity: How to give young job-seekers good information about available work, and how to link them efficiently with employers needing staff. Despite the acute need for labour market information/’matching’ mechanisms, few solutions have emerged. Most job-seekers lack web access, government initiatives are minimal, schools have no resources, and private-sector efforts are limited.

II. NOVELTY: JOB-SEEKERS MOBILE PHONE.

AND

EMPLOYERS MATCHED

BY

Mobile technology, however, is uniquely poised to ‘leapfrog’ many of these obstacles and help job-seekers in developing countries find key labor market information. One such solution is a new software application developed by engineers at Palestinian NGO Souktel. The application bridges information gaps by linking youth with job data through a simple text message-driven process. Piloted in 2006, the technology has now helped over 2,000 youth find work and internships through a process that saves users time and resources, while enabling them to connect with far more Manuscript received January 28, 2009. Amber Houssian, M.B.A., is the Regional Manager- Middle East for Souktel, and a past Strategic Planner at leading Canadian marketing firm Cossette (e-mail: [email protected]). Mohammad Kilany, B.Sc., is a co-founder of Souktel and a 2009 Synergos Arab World Social Innovation Fellow (e-mail: [email protected]). Jacob Korenblum, M.Ed., is a co-founder of Souktel and a past Harvard University Reynolds Fellow in Social Enterprise (e-mail: [email protected]). .

III. SIGNIFICANCE: BRIDGING THE DIGITAL REDUCING UNEMPLOYMENT; PROMOTING GROWTH.

DIVIDE;

Fast, easy to use, and widely accessible, phone-based job information services like Souktel offer four significant advantages over traditional job-find/hiring approaches: •

•

•

•

Faster, better access to information: Customized matching of youth with employers and “anytime, anywhere” access from mobile phone reduces job/staff search time from months to hours. Gender Sensitive: Virtual (non face-to-face) communication creates new, safe ways for women in traditional families to connect with female and male peers in the labour market. Wider Reach – Improved access to remote/rural communities: Mobile phone access option facilitates service use in areas with no Internet or poor communication/transport infrastructure. Ease of use in Conflict/Post-Conflict Zones: No new hardware required; no need for regular electricity or Internet connection. System can be scaled up or down quickly and cheaply, in response to usage patterns and user demands.

492

MultiMath: Numeric Keypads for Math Learning on Shared Personal Computers Sunil Garg, Charlotte Robinson, Clint Tseng, Heather Underwood, Richard Anderson, Joyojeet Pal University of Washington

Abstract—Past research has shown the benefits of increased engagement and collaboration for learning using multiple inputs for children sharing computers. Taking the work forth beyond clicks, we demonstrate the use of inexpensive numeric keypads as multiple input devices in shared computing scenarios, as many have argued are well-suited to the developing world, and show its utility in mathematical learning for young children.

W

I. INTRODUCTION

HILE a significant amount of work exists in the use of gaming in education, there is little literature on the value of multi-player gaming environments in developing world contexts. The work with multiple mice (Pawar, Pal et al. 2006) has succeeded in showing some gains with multiplechoice environments for young children constrained to sharing computers, and subsequent work has shown the impacts of split screens (Moed, Otto et al. 2009) in increasing the collaboration and engagement for learning. MultiMath is a multi-player educational game for basic arithmetic skills. Utilizing input via multiple numeric keypads and a splitscreen display, the application allows for multiple individuals to participate at their own speed and skill level while simultaneously introducing competition between players. The split screen concept enables the application to provide each player with personalized curriculum and also allows for the use of adaptive questioning.

II. TECHNICAL IMPLEMENTATION The MultiMath game is implemented as a desktop application using the Microsoft .NET framework. While Windows does support multiple mice and keyboards, it does not make any distinction between devices by default, as all mice control the same cursor and all keyboards provide a combined stream of input. However, Windows offers a RawInput API which can be used to simultaneously interface with multiple human interface devices by identifying each event, such as a key Sunil Garg, Charlotte Robinson, Clint Tseng, Heather Underwood, and Richard Anderson are with the Department of Computer Science and Engineering at the University of Washington, Seattle, WA 98195 USA. (email: {skgarg, solstinc, cxlt, hmu2, anderson}@cs.washington.edu) Joyojeet Pal is with the Center for Information and Society at the University of Washington, Seattle, WA 98195 USA. (email: [email protected])

press, as corresponding to a specific device identifier. Using this information, we can correlate each key input with a specific user and react accordingly. We employ USB numeric keypads because each unit costs approximately four dollars and can be used with virtually any modern computer, thus allowing for low-cost deployment in resource-constrained environments. Moreover, they offer a high degree of versatility, as they can be used for both numeric and directional input, and their keys can be remapped and labeled for specialized applications. Previous applications have used soft keyboards on the display which depend on mouse input (Moraveji, Kim et al. 2008) – we demonstrate that hardware input devices offer several advantages, such as by allowing faster input and by saving valuable screen real estate for other purposes. For display purposes, we split the screen area into portions reserved for each participant, allowing for individualized attention. Not only does this enable each player to complete problems presented to them at their own pace, but this also allows us to introduce adaptive questioning in order to best address the skill level of each individual.

III. DEMONSTRATION In this demonstration we will present an initial version of the MultiMath application, including multiple keypad input, split screen display, and a simple problem set. Using the implementation described above, the application will comprise of a game in which four players can individually complete randomly generated arithmetic questions and receive visual feedback of their progress and ranking within the game, through display elements such as a progress bar. At the demonstration, we also plan to display results from preliminary field tests with children during our iterative design process.

REFERENCES [1] [2] [3]

Moed, A., O. Otto, et al., "Reducing Dominance Behavior in MultipleMouse Learning Activities," 8th Int. Conf. on Computer Supported Collaborative Learning (CSCL 2009), Submitted, Rhodes, Greece. Moraveji, N., T. Kim, et al., "Mischief: Supporting Remote Teaching in Developing Regions," Proc. 26th Annu. SIGCHI Conf. on Human Factors in Computing Systems (CHI 2008), Florence, Italy. Pawar, U., J. Pal, et al., "Multiple Mice for Computers in Education in Developing Countries," 2006.

493

A New Generation of Open Source Data Collection Tools Yaw Anokwa, Carl Hartung, Adam Lerer, Brian DeRenzi, Gaetano Borriello

Abstract—Organizations in low income regions need tools for collecting and reporting on data. Existing paper-based approaches are often slow and incomplete when compared to data collection tools on mobile devices. In this demonstration, we present the authors’ Android-based system in addition to broader work by the OpenRosa Consortium – a community dedicated to building open source, standards-based tools for mobile data collection, aggregation, analysis, and reporting. Index Terms—data collection, mobile phones, ICT

I. I NTRODUCTION variety of tools are urgently needed to address the lack of data in low-income countries. This data is required for decision making and research in many sectors. Due to recent technological advancements, there is great excitement for using mobile devices to address current gaps in information. Phones and PDAs have proven to be dramatically faster and more complete than traditional methods of pen and paper, and can be more accurate and less expensive as well[5], [6]. The OpenRosa Consortium[1] is a group working together to foster open source, standards-based tools for mobile data collection, aggregation, analysis, and reporting. Participants include AED-Satellife, Cell-Life, CIDRZ, D-Tree International, DataDyne, Dimagi, Google, MRC-SA, University of Bergen, Makerere University, and the University of Washington. The group has active developers in Bangladesh, Kenya, India, Norway, Pakistan, South Africa, Tanzania, Uganda, and the United States primarily working on JavaRosa – an open source framework for data collection on Java-enabled phones. By developing open source solutions and conforming to standards based on the W3C XForms specification, OpenRosa projects can interoperate their components. So while many of the member organizations are currently using JavaRosa for form collection and management systems, others are building complex solutions like clinical trial software[4], data reporting on servers[3], and even community health worker management[2].

A

JavaRosa and AndroidRosa. We will also demonstrate ServerRosa, software which processes and hosts data submitted by mobile phones. To demonstrate JavaRosa, we will use a few examples of phones which support Java 2 Platform Micro Edition (J2ME). The Nokia 3110c and 6085 both sell for around $100 and are widely available in low income regions. The Nokia N95 is available for around $400 and is one of the most advanced smartphones available. We will demonstrate form filling and submission functionality designed for community health care workers on these phones. AndroidRosa will be demonstrated on the T-Mobile G1 phone running Android. Android is an open source operating system developed by Google that is more feature-rich and easier to develop for than J2ME. AndroidRosa uses the JavaRosa core code and leverages Android’s unique abilities. We will demonstrate form filling which builds on the G1’s input devices (touchscreen and keyboard) and sensors (SMS, location and images). To collect the completed form data sent from the phones, we will demonstrate ServerRosa. This server component is written in Java and can run on local servers as well as in the cloud. Phones can submit completed form data to the server, which can export submitted data and generate reports if necessary. The server can also summarize and aggregate submission data and send it to a separate Android application, which supervisors can use to monitor data collection. III. C ONCLUSION The OpenRosa Consortium is close to consolidating the necessary functionality for the first public release of JavaRosa, AndroidRosa and ServerRosa. Together, we are working towards solutions to allow organizations with minimal technical capacity to collect data using a variety of mobile phones. ACKNOWLEDGMENT The authors thank all members of OpenRosa, especially Neal Lesh, Drew Roos and Clayton Sims.

II. D EMONSTRATION In this demonstration, we will present some of the data collection and management tools under development by the authors for the OpenRosa Consortium. We will show form filling, submission and monitoring on mobile phones using Yaw, Carl, Brian, and Gaetano are affiliated with the University of Washington and Adam is affiliated with the Massachusetts Institute of Technology. This work was done while Yaw, Carl, Adam and Gaetano were interns at Google. All authors can be reached at [email protected].

R EFERENCES [1] [2] [3] [4] [5]

OpenRosa, http://openrosa.org CommCare, http://commcare.cs.washington.edu Gather, http://blog.gatherdata.org OMEVAC, http://epihandy.org/index.php/OMEVAC K. Shimira et al., The use of personal digital assistants for data entry at the point of collection in a large household survey in southern Tanzania, Emerging Themes in Epidemiology, 2007. [6] T. Parikh, et al., Mobile Phones and Paper Documents: Evaluating a New Approach for Capturing Microfinance Data in Rural India, CHI, 2006.

RuralScope: An Information System for Tracking Rural Disbursements 1

Sai Gopal Thota, Rabin Patra, Murali Medisetty, Sivananda Reddy, Vivek Mungala, Joyojeet Pal

Abstract—Tracking and ensuring disbursements make it to their intended recipients is among the problems development professionals are most concerned about. To examine the potential monitoring role playable by technology, we present RuralScope, an information repository designed to handle the information needs of a newly started welfare scheme offering daily wage payments to over 31 million families in rural India. Since the problem of monitoring development extends beyond project specific needs to a general sense of awareness, RuralScope works on two levels. First, for project auditors, RuralScope acts as an easily searchable, disconnected database of disbursements that helps speed up the audit process. Second, for a more general audience, an organized and annotatable front end for casual viewers, policy-makers, and researchers alike to keep track of documents and data related to the project RuralScope.

M

I. INTRODUCTION

UCH recent interest in development has focused on the

need for robust systems that can enable efficient governance of public funding in various parts of the world. Increasing transparency within the functioning of governments and public finances is seen as an important goal for development (Qureshi 2004 [1], Moody-Stuart 1997 [2]) Gaps in adequate information, legal loopholes in the implementation of such schemes have been cited as reasons for weakened monitoring of corruption several developing nations (Levin and Satarov 2000[3]). The National Rural Employment Guarantee Act (NREGA) was passed by the Government of India in 2005 to address the problems of seasonal employment in Rural India. The scheme broadly offers 100 days of employment at minimum wage to every rural household through projects that create local infrastructure, often roads or schools. The scheme officially already reaches 31 million households, making it the largest welfare administration scheme ever in the world. But within the first few years of its progress, many questions have arisen on disbursement of the funds and the need for better transparency. Two kinds of problems have been highlighted: first the misrepresentation of welfare payments and siphoning of funds. A second, and more fundamental issue faced by rural development issues, has been the limited coverage of the scheme and its impacts by media. media. The only effective way of tracking disbursements in NREG has been through social audits, perilous processes in which activists have been known to get murdered during the time they spend in actual audits as they make a lot of enemies during the process of recovering the misused funds through these audits.

1

494

II. IMPLEMENTATION RuralScope is an information system that helps make the audit process more efficient and less risky for the auditor by arranging the data and making it available on disconnected devices, replacing cumbersome muster rolls and dramatically reducing the amount of time an auditor needs to spend on the ground. On the front end, RuralScope offers a single gateway into all the information on outlays and disbursements for NREG allowing researchers the ability to identify trends using graphs and maps and annotate visualizations. Finally, by allowing researchers to plug in relevant demographic, agricultural produce and related data from other sources, RuralScope makes it easier for researchers and casual users alike to contextually understand the disbursements. The descriptions for all visualizations and annotations created by users as stored as structured XML, and we use here data from the state of Andhra Pradesh for 2007 disbursements representing 28000 villages and over 6 million workers. III. DEMONSTRATION We will demonstrate use cases for casual online users with limited understanding of the NREG system or database through queries to the RuralScope database that can create visualizations. For sophisticated users familiar with the NREG or activist/monitors, we demonstrate a working model for audit queries to trace by individual job card holder or family ID exactly how much a family has been paid over a certain time period. To emphasize the value addition in the audit process that RuralScope brings, we plan to contrast the new interface with the currently prevalent muster roll data outputs. This feature is specifically targeted at making social audits quicker. Finally, we would like to demonstrate the disconnected version of RuralScope designed for a mobile computing device, on which data can be collected or viewed from local storage as needed, and thereafter synchronized in areas of network connectivity. This last feature has much relevance to several prevalent ICTD project domains including delay tolerant networking and OpenMRS REFERENCES [1] [2] [3]

Qureshi, Z. (2004). Millenium Development Goals and Monterrey Consensus: from vision to action. ABCDE Conference of the World Bank, Monterrey, Mexico. Moody-Stuart, G. (1997). Grand corruption: how business bribes damage developing countries. Levin, M. and G. Satarov (2000). "Corruption and institutions in Russia." 16(1): 113-132.

Sai Gopal Thota([email protected]), Rabin Patra([email protected]) , Murali Medisetty([email protected]), Sivananda Reddy([email protected]), Vivek Munagala([email protected]), Joyojeet Pal([email protected])

495

T-Cube Web Interface in Support of Real-Time Bio-surveillance Program Artur Dubrawski, Maheshkumar Sabhnani, Michael Knight, Michael Baysek, Daniel Neill, Saswati Ray, Anna Michalska and Nuwan Waidyanatha

Abstract—T-Cube Web Interface is a generic tool to visualize and manipulate large scale multivariate time series datasets. The interface allows the user to execute complex queries quickly and to run various types of statistical tests on the loaded data. We show its utility in an important application scenario: real-time bio-surveillance system designed to support rapid detection and mitigation of bio-medical threats in developing countries. Index Terms—bio-surveillance, event detection, interactive analytics, data cubes.

I

nfectious diseases such as tuberculosis or avian influenza pose a persistent threat to human population all over the world. Their mitigation and containment is especially difficult in the least developed countries, where existing surveillance systems cannot gather and process the necessary information in time for public health authorities to effectively respond to rapidly developing bio-events. For example, the system currently in use in Sri Lanka is paper-based and relies on the postal service to communicate community health reports to the central office for analysis, which may take even ten days. Communication latency is exacerbated by limitations of analytic resources (in terms of their quantity as well as training) available to process the available information. We believe that that affordable, existing information and communication technologies can play the key role in collecting, communicating and analyzing the relevant data, enabling reliable and timely identification of emerging threats and, therefore, effective containment of their consequences. The International Development Research Centre of Canada supports the Real-Time Bio-surveillance Program (RTBP), a Manuscript received January 30, 2009. This work was supported in part by the International Development Research Centre of Canada (p;roject number 105130), the U.S. Department of Agriculture (award 1040770), Centers of Disease Control and Prevention (award R01-PH000028), National Science Foundation (under grant IIS-0325581) and by the U.S. Department of Defense (AFOSR award FA8650-05-C-7264). A. Dubrawski, M. Sabhnani, M. Knight, M. Baysek, D. Neill, and S. Ray are with the Auton Lab, Carnegie Mellon University, Pittsburgh, PA 15213 USA (Corresponding author: A. Dubrawski, phone: 412-268-6233; fax: 412268-6233; e-mail: [email protected]) (Other authors email: [email protected], [email protected], [email protected], [email protected], [email protected]). A. Michalska is with the Institute of Computer Science, University of Warsaw, ul. Banacha 2, 02-097 Warszawa, Poland (e-mail: [email protected]). N. Waidyanatha is with LIRNEasia, 12 Balcombe Place, Colombo 08, Sri Lanka (e-mail: [email protected]).

pilot project currently under development in Sri Lanka and Tamil Nadu state of India. It will consist of information gathering system based on mobile handheld devices and wireless networking, and of a specialized IT system designed for automated surveillance of the incoming data and for interactive navigation through data and extracted information. In this demonstration, we present a prototype application of the T-Cube Web Interface in monitoring of public health data in Sri Lanka and India. T-Cube Web Interface [1] is a generic tool for interactive analysis, visualization and manipulation of large scale multidimensional temporal and spatio-temporal datasets commonly encountered in public health domain. It allows the users to execute complex queries quickly and to perform various types of statistical tests against the loaded data. The underlying data representation technology, the T-Cube [2], is an in-memory data structure designed to improve the response time to ad-hoc time series queries against large datasets. TCube has been tested on synthetic and real-world datasets containing millions of records and hundreds of dimensions. Results show that its response time can be orders of magnitude shorter than that of the state-of-the-art commercial database tools. The attainable speedups have been shown to qualitatively change the way the users interact with their data, leading to better situational awareness, comprehension, and interpretability of data and results of analyses. It has already been successfully used in equipment maintenance, logistics and food safety domains. We intend to leverage the benefits of T-Cube technology in RTPB. Demonstrated capabilities of the tool include geospatiotemporal visualization of syndromic data, navigation through different levels of data aggregation, and selected statistical analyses including spatial scan for rapid detection of emerging outbreaks of diseases. Presentation emphasizes user-perceived utilities due to computational efficiency of the analytic algorithms and the underlying data representation technique. REFERENCES [1]

[2]

S. Ray, A. Michalska, M. Sabhnani, A. Dubrawski, M. Baysek, L. Chen, J. Ostlund, T-Cube Web Interface: A Tool for Immediate Visualization, Interactive Manipulation and Analysis of Large Sets of Multivariate Time Series, 2008 AMIA Annual Symposium, Washington, DC, 2008. M. Sabhnani , A. Moore, and A. Dubrawski, Rapid processing of ad-hoc queries against large sets of time series. Advances in Disease Surveillance 2, 2007.

496

Web Search over Low Bandwidth Jay Chen, Lakshminarayanan Subramanian, and Jinyang Li

Abstract— Web search and browsing have been streamlined for a comfortable experience when the network connection is fast. Existing systems, however, are not optimized for scenarios where connectivity is poor, as is the case for many users in developing regions where fast connections are expensive, rare, or unavailable. We examined the challenges faced by users behind one of these connections in a previous study, and found that the existing web interface is incapable of providing a good experience when the connection is extremely slow. In this demonstration we present a prototype implementation of a web browsing system that incorporates what we learned. Our system helps the user’s search offline as much as possible, and in a single search query when information from the Internet is required. Index Terms—Low Bandwidth, Search, Internet, Intermittent Network

I. INTRODUCTION

S

low, expensive, or non-existent Internet access is a fact of life for many computer users in the developing world where the physical infrastructure has yet to catch up to the increasing demand. In these regions connectivity is often intermittent due to power cuts [1], and bandwidth is generally an expensive and rare commodity because none of the traditional wire-line connectivity solutions (fiber, broadband and dial-up) are economically viable for rural regions with low user densities [2]. Existing work on bringing the Internet to the developing world generally focuses on connectivity issues [3, 4], but even after Internet connectivity is established via any of these physical channels, the link is slow. Systems typically attempt to mask the poor connection from the application using some combination of traditional methods such as compression, caching, filtering, and prefetching [5, 6]. However, when the connection is extremely slow (hours or days for mechanical backhauls) the iterative process of web search [7] becomes impractical due to the long round trip time. The existing literature is unclear on how to further improve the web browsing interactivity when the process is nearly or completely asynchronous. Our work identifies this problem as pervasive and unsolved in developing countries, and as a result we

developed a system that explicitly exposes the slow network connection to the user while providing features to make progress despite it.

II. DEMONSTRATION In this demonstration we show a prototype of our web search system over a low bandwidth connection. Our system is comprised of two proxy servers: one local proxy that is behind the low bandwidth link, and one remote proxy that is well connected to the Internet. Requests are made at the local proxy and sent to the remote proxy for processing and eventual return of optimized results. This architecture in itself is not new, and similar to the TEK project [5]. What is novel about our system is that it has several unique features designed with asynchrony as a primary consideration. Due to lack of space we list only the search process along with a few of the most visible application level features to be demonstrated here. The web search interface provides local search and query construction assistance. These two features enable users to avoid using the network if the result may be satisfied locally; if query cannot be satisfied locally then the search query has already been iteratively refined to be as specific as possible. The query construction interface has fields for users to explicitly specify the type of response expected by the user. The remote proxy uses this additional information to compress, filter, and sometimes prefetch results based on the user’s preferences. Once a request is issued to the remote proxy, the local proxy also provides an estimated time until request completion. Users are free to queue up requests and return after the queries are satisfied. Finally, after results are returned, they are integrated into the local cache and available to the user for browsing. REFERENCES [1]

[2]

[3] [4]

Manuscript received January 28, 2009. J. Chen is with New York University, New York, NY 10003 USA (e-mail: [email protected]). L. Subramanian is with New York University, New York, NY 10003 USA (e-mail: [email protected]). J. Li is with New York University, New York, NY 10003 USA (e-mail: [email protected]).

[5] [6] [7]

S. Surana, R. Patra, S. Nedevschi, M. Ramos, L. Subramanian, and E. Brewer, “Beyond Pilots: Keeping Rural Wireless Networks Alive,” NSDI, 2008. S. Mubaraq, J. Hwang, D. Filippini, R. Moazzami, L. Subramanian, and T. Du, “Economic analysis of networking technologies for rural developing regions,” Workshop on Internet Economics, 2005. K. Fall, “A Delay Tolerant Network Architecture for Challenged Internets,” Computer Communication Review, 2003. R. Patra, S. Nedevschi, S. Surana, A. Sheth, L. Subramanian, and E. Brewer, “WiLDNet: Design and Implementation of High Performance WiFi Based Long Distance Networks,” NSDI, 2007. W. Thies et al., “Searching the World Wide Web in Low-Connectivity Communities,” WWW, 2002. Loband, http://www.loband.org. T. Joachims, L. Granka, Bing Pan, and G. Gay, “Accurately Interpreting Clickthrough Data as Implicit Feedback,” In ACM SIGIR, 2005.

497

Author Index Abdulsalam, Ameer 478 Agarwal, Sheetal K. 56 Ahmed, Tanveer 447 Ali, Nosheen 447 Anderson, Richard 492 Anderson, Ruth 233 Anokwa, Yaw 493 Aoki, Paul 35, 118 Asuntogun, A. 280 Aziz, Sherif M. 288 Balakrishnan, Ravin 337, 406 Banks, Ken 484 Barnard, Etienne 95 Basavaraju, S. 481 Basu, Anupam 201 Baysek, Michael 495 Belousov, Sarah M. 129, 478 Bertrand, Anne 485 Best, Michael L. 192, 468 Bhatnagar, Subhash C. 183 Biswas, Manna 437 Blagsvedt, Sean 150 Borriello, Gaetano 233, 493 Braa, Jørn 379 Brewer, Eric 490 Brunette, Waylon 233 Brunskill, Emma 74 Bubrawski, Artur 495 Burrell, Brenda 483 Bussell, Jennifer 173 Canny, John 139 Chabossou, A. 392 Chakraboty, Dipanjan 85 Chauhan, Himanshu 85 Chen, Harr 487 Chen, Jay 495 Chen, Kuang 487 Chen, R. 300 Chib, Arul 328 Chu, Gerry 337 Clark, Bev 483 Clayton, Peter 320 Coles, J. 300 Conway, Neil 487

Crow, Ben 212 Dasgupta, Tirthankar 201 Davis, James 212 DeRenzi, Brian 493 Dias, M. Frederick 129, 478 Dias, M. Bernadine 129, 478 Dolan, Heather 487 Donner, Jonathan 17 El Mahdi, A. 486 El-Moughny, Noura 478 Escobari, Marcela 17 Etherton, John 468 Fanaswala, Imran 478 Findlater, Leah 406 Gandhi, Rikin 337, 406 Garg, Sunil 492 Ghadially, Rehanan 310 Ghazzawi, Wael 478 Gigler. Björn-Sören 267 Golly-Kobrissa, Romain Tohouri 379 Gong, Haijun 129 Griffiths, Lee 488 Grover, Aditi Sharma 95 Gunawardane, Prabath 212 Gupta, Aakar 482 Harmen, Saskia 479 Hartung, Carl 493 Hefferman, Claire 480, 489 Heimerl, Kurtis 490 Hellerstein, Joseph M. 487 Helmer, Scott 407 Hersman, Erik 484 Ho, Melissa R. 35, 118 Hollow, David 27 Hopkins, Mary Ann 458 Houssain, Amber 491 Hussain, Faheem 252 Hutchinson, Kelly 163 Kim, Matthew 139 Jain, Mohit 482 Jain, Shirley 139 Javid, Paul 437 Kettani, D. 486 Khan, Farida 310

498

Kilany, Mohammad 491 Kleine, Dorothea 108 Knight, Michael 495 Kolko, Beth E. 46, 192, 233 Komathi, A.L.E. 328 Korenblum, Jacob 491 Kossi, Edem Kwame 379 Kumar, Arun 56, 85, 139 Kuun, Christiaan 95 Kuriyan, Renee 66 Lee, J. 300 Lerer, Adam 493 Levine, Brian 118, 458 Li, Jinyang 496 Lin, Yibo 480 Lodha, Suresh 212 Lustig, Caitlin 233 Luk, Rowena 118 Maitland, Carleen F. 427 Masperi, Paola 27 Mathur, Akhil 139 Meacham, Arthur 458 Medisetty, Murali 494 Meirer, Patrick 3 Menon, S. Raghu 337, 478 Michalska, Anna 495 Middleton, Erin 212 Mills-Tettey, G. Ayorki 129 Mirza, Sarwat 447 Molla, Alemayehu 163 Mostow, Jack 129 Mulenga, B. 280, 481 Mungala, Vivek 494 Nanavati, Amit Anil 56 Nedevschi, Sergui 357 Neil, Daniel 495 Nyella, Edwin 243 Okon, Uduak A. 367 Osuntogun, A. 481 Owusu, Emmanuel K. 35 Paik, Michael 222, 458 Pal, Joyojeet 357, 414, 482, 490, 492,494 Palijo, Sooraj 447 Parikh, Tapan S. 487, 490 Patnaik, Somani 74 Patra, Rabin 357

Pawar, Udai Singh 150 Pitman, J. 280, 481 Planché, Madelaine 95 Poon, Anthony 233 Putnam, Cynthia 46 Quarta, Giulio 458 Rahman, Mohammed Kaleemur 478 Rajput, Nitendra 56 Rao, F. R. 300 Ramachandran, Divya 490 Rapchak, Barbara 458 Ratan, Aishwarya Lakshmi 150 Ratra, Rabin 494 Ray, Isha 66 Ray, Saswati 495 Reddy, Sivananda 494 Robinson, Charlotte 492 Rosenfeld, Roni 447 Sabhnani, Maheshkumar 495 Sæbo, Johan 379 Sahyraoui, Sofiane M. 348 Samdaria, Navkar 482 Sanghvi, Saurabh 478 Satpathy, Sambit 150, 337 Sharma, Ashlesh 458 Shekhar, Praveen 482 Sherwani, Jahanzeb 447 Singh, Gusharan 406 Singh, Nupur 183 Smyth, Thomas N. 468 Sweet, Tracy Morrison 129 Stodden, Victoria 3 Stork, C 392 Stork, M. 392 Subramaniam, Lakshminarayanan 222, 458, 496 Subramaniam, Thanuja 150 Terzoli, Alfredo 320 Thakur, Dhanaraj 192 Thies, William 74 Thinyane, Mamello 320 Thomas, S. 280, 481 Thota, Sai Gopal 494 Titlestad, Ola 379 Tongia, Rahul 252 Toyama, Kentaro 150, 337, 406, 427 Trahanas, Philip 458

499

Traxler, John 488 Tseng, Clint 492 Underwood, Heather 492 van Gorp, Annemijin 427 Vempala, S. 280, 481 Waidyanatha, Nuwan 495 Wornyo, Edem 468 Yu, Jun 489 Zaharia, Matei 118 Zahonogo, Z. 392 Zia, Lilian 150