Thesis Ssd Defence

Planning of Rural Feeder Service to Bus Stop Thesis submitted in partial fulfillment Of the requirement for the award of the degree of

Doctor of Philosophy

in Civil Engineering

By Sudhanshu Sekhar Das

Department of Civil Engineering Indian Institute of Technology, Kharagpur Kharagpur – 721302, India January 2008

Indian Institute of Technology

Department of Civil Engineering INDIAN INSTITUTE OF TECHNOLOGY, KHARAGPUR

CERTIFICATE This is to certify that the thesis entitled Planning of Rural Feeder Service to Bus Stop that is being submitted by Shri Sudhanshu Sekhar Das to the Indian Institute of Technology, Kharagpur for the award of the degree of Doctor of Philosophy is a record of bonafide research work carried out by him under my supervision and guidance. Shri Sudhanshu Sekhar Das has worked on the problem for over four years and the thesis, in my opinion, is worthy of consideration for the award of the degree of Doctor of Philosophy in Engineering in accordance with the regulations of the Institute. The results embodied in this thesis have not been submitted to any other University or Institute for the award of any Degree or Diploma.

(Dr. Bhargab Maitra)

Associate Professor Department of Civil Engineering Indian Institute of Technology, Kharagpur

Dedicated to The Almighty & My Parents

ACKNOWLEDGEMENT I take this opportunity to express my deep sense of gratitude to my supervisor Dr. Bhargab Maitra for his invaluable guidance, advice and constant encouragement through out the period of present research work. I express my sincere thanks to Prof. B. B. Pandey, Prof. K. Sudhakar Reddy and Dr. Amarnatha Reddy for their suggestions and help throughout the present research work. I wish to record my indebtedness to Prof. S. P. Das Gupta, Head of the Civil Engineering Department for providing all facilities and extending help during my research work. I express my sincere thanks to my doctoral scrutiny committee members, Prof. R. N. Dutta and Dr. D. J. Sen for their valuable suggestions during different stages of my research. I owe my deep sense of gratitude to my friends and co-scholars Amar, Debabrata, Debasis, Kishore, Phani, Sridhar, Tusar, Umesh for their valuable and timely help in many aspects. Special thanks to Mr. Asim, for his help in carrying out the surveys.

At last but not least, I express my heartfelt gratitude and regards to my father (Dr. S. N. Das), mother (Mrs. K. Das) their love, patience, sacrifice and constant inspiration and hearty thanks to my wife (Lasyamayee) and son (Satyajeet) for their patience, tolerance and understanding my difficulties and other family members, who have extended their active cooperation by taking a lot of pain and anxiety during my Ph.D study without

which

this

work

can’t

be

fruitful

one.

I

acknowledge

the

appreciation, understanding, inspiration and support I received from them.

Date:

(Sudhanshu Sekhar Das)

ABSTRACT

In rural areas of developing countries, the private vehicle ownership is generally low and therefore, the travel need is largely served by public transportation system. In India, all the major roads are generally served by bus transportation system. But, in most of the rural areas, feeder services are not available for providing transportation linkage between village settlements and bus stop. In the present work, an investigation is carried out on planning of rural feeder service to bus stop. The work is demonstrated with reference to a case study in rural India. Two small carriers namely Trekker and Tempo are considered as feeder vehicles. Three different forms of operation of feeder vehicles are investigated: fixedschedule, dial-a-ride and dial-a-slot. Dial-a-ride and dial-a-slot are designed as demand-responsive forms of operation. Travel behaviour analysis constitutes a significant part of the work. As a part of travel behaviour analysis, a stated choice survey instrument is designed with three alternative forms of operation of feeder vehicles and two alternative vehicle types. Both quantitative and qualitative attributes are included in stated choice experiment. The stated choice data collected from rural trip makers are analyzed in two stages. In Stage-I, the stated choice data is analyzed using Multinomial Logit, Nested Logit, Covariance Heterogeneity Nested Logit and Random Parameter Logit (RPL) model specifications, and willingness-to-pay (WTP) values are estimated for all the attributes of hypothetical feeder service. In RPL models, all the random parameters are assumed to follow the constrained triangular distribution due to its advantages and attempts are made to take into account the effect of socio-demographic of trip makers’ on WTP values. In Stage-II analysis, WTP values obtained from Stage-I are used judiciously to reduce the number of variables describing the hypothetical feeder service. Separate utility equations are developed for bicycle and motor cycle users.

i

Using the utility equations and WTP values obtained from travel behaviour analysis, operationally viable feeder routes are designed assuming fixedschedule form of operation of vehicles. Generalized cost and passenger-km are the two alternative measures of effectiveness considered during the design of feeder service. Several alternative scenarios are formulated considering

external

subsidy

and

cross

subsidy

as

the

two

policy

instruments. A heuristic approach is followed for the selection of feeder routes and vehicle for all these scenarios. The uncertainties associated with the use of utility equations developed from stated preference data for demand estimation is duly considered, and the recommended routes are classified as stable and unstable routes. Finally, a comparison is made among three different forms of operation of feeder vehicles. For this purpose, passenger movements along the selected routes are simulated under different forms of operation of feeder vehicles. Altogether, a comprehensive approach is demonstrated for the planning of rural feeder service to bus stop with due consideration to travel demand, characteristics of feeder service including type of vehicle and form of operation, behaviour of trip makers, operational viability of feeder service, and role of policy measures.

ii

Content

Content Title

Page

Certificate Dedication Acknowledgement Abstract

i

Contents

iii

List of Table

vii

List of Figure

xi

1.1

Introduction MOTIVATION

1

1.2

OBJECTIVE AND SCOPE OF THE WORK

2

1.3

FEEDER VEHICLES FORMS OF OPERATION

4

1.4

STUDY AREA

6

1.5

ORGANIZATION OF REPORT

8

1.6

SUMMARY

8

Chapter 1.

Chapter 2.

Approach and Methodology

9

2.1

INTRODUCTION

9

2.2

TRAVEL BEHAVIOUR ANALYSIS

9

2.2.1

Design of Experiment

10

2.2.1.1

Type of Data and Preference Elicitation Technique

11

2.2.1.2

Attributes and Their Levels

15

2.2.1.3

Choice Sets

16

2.2.1.4

Questionnaire Design and Pilot Survey

19

2.2.2

Collection of Data and Development of Database

19

2.2.3

Analysis of Data

21

2.2.3.1

Organization of Data

21

2.2.3.2

Econometric Models

22

Multinomial logit

22

Nested logit

23

Covariance heterogeneity nested logit

24

Random parameter logit

24

iii

Content

Title

Page

Distributions of random parameters

26

Selection of points for RPL model

27

2.2.4

Valuing of Attributes

28

2.2.5

Comparison of Utility Equations

31

2.2.5.1

The Likelihood Ratio Test

32

2.2.5.2

Goodness of Fit

32

2.2.6

Selection of Models for Estimation of Ridership

33

2.2.7

Development of Generalized Cost Equations

34

2.3 2.3.1

DESIGN OF FEEDER SERVICE

35 35

Database

2.3.1.1

Travel Demand to Bus Stop

36

2.3.1.2

Temporal Variation of Demand

36

2.3.1.3

Road Network

37

Road network to bus stop

37 38

Cutoff Revenue

2.3.1.4 2.3.2

Measurement of Effectiveness, Alternative Scenarios

38

2.3.3

and Fare Level Selection of Feeder Route and Vehicle

39

2.3.3.1

External Subsidy

40

2.3.3.2

Cross Subsidy

44

2.3.4 2.4 Chapter 3.

Forms of Operation of Feeder Vehicles

47 48

SUMMARY Travel Behaviour Analysis

3.1

INTRODUCTION

49

3.2

DESIGN OF EXPERIMENT

49

3.2.1

Attributes and Their Levels

50

3.2.2

Choice Sets

51

3.2.3 3.3

Questionnaire and Pilot Survey

52

COLLECTION OF DATA AND DEVELOPMENT OF DATABASE

54

3.4

ANALYSIS OF DATA: STAGE-I

55

3.4.1

Organization of Data

56

3.4.2

Multinomial Logit Model

57

3.4.3

Nested Logit Model

58

iv

Content

Title

Page

3.4.4

Covariance Heterogeneity Nested Logit

61

3.4.5

Random Parameter Logit Model

62

3.4.6

Estimation of WTP Value

64

3.4.7

Comparison of Utility Equations

68

3.5

ANALYSIS OF DATA: STAGE-II

70

3.5.1

Model for Bicycle Users

71

3.5.2

Model for Motorcycle Users

72

3.5.3

Estimation of WTP Values

73

3.6

GENERALIZED COST OF TRAVEL

74

3.7

SUMMARY

75

Chapter 4.

Design of Feeder Service

4.1

INTRODUCTION

76

4.2

DATABASE

76

4.2.1 4.2.1.1

Travel Demand to Bus Stop Modeling of Trip Rates

76 77

Revenue generating trips

77

Educational trips

78

Household trips

79

Total trips

80

4.2.1.2

Validation of Modeled Trip Rates

80

4.2.1.3

Estimation of Trips

81

4.2.2

Temporal Variation of Demand

82

4.2.3 4.2.3.1

Road Network

83

4.2.3.2

Base Network

83

Road Network to Bus Stop

83 85

Cutoff Revenue

4.2.4 4.3

MEASURES OF EFFECTIVENESS, ALTERNATIVE

87

4.3.1

SCENARIOS AND FARE LEVELS Measure of Effectiveness

87

4.3.2

Alternative Scenarios

88

4.3.3 4.4

Fare Levels

90

FEEDER SERVICE WITH GENERALIZED COST AS MEASURE OF EFFECTIVENESS

v

90

Content

Title 4.4.1

Feeder Routes and Vehicle

Page 91

4.4.1.1

Scenario-I

91

4.4.1.2

Scenario-II

92

4.4.1.3

Scenario-III

93

4.4.1.4

Scenario-IV

93

4.4.1.5

Scenario-V

94

4.4.1.6

Scenario-VI

94

4.4.1.7

Comparison of Different Scenarios

95

Effect of ASC on Operational Viability

96

4.4.2 4.5

FEEDER SERVICE WITH PASSENGER-KM AS MEASURE OF

100

4.5.1

EFFECTIVENESS Feeder Routes and Vehicle

100

4.5.1.1

Scenario-I

100

4.5.1.2

Scenario-II

101

4.5.1.3

Scenario-III

101

4.5.1.4

Scenario-IV

102

4.5.1.5

Scenario-V

103

4.5.1.6

Scenario-VI

103

4.5.1.7

Comparison of Different Scenarios

104

Effect of ASC on Operational Viability

105

4.6

FORMS OF OPERATION OF FEEDER VEHICLES

108

4.7

SUMMARY

111

4.5.2

Chapter 5.

Conclusions 112

5.2

INTRODUCTION CONCLUSION

5.3

FUTURE SCOPE OF THE WORK

117

References

119

Annexure-A

137

Annexure-B

153

5.1

112

Resume

vi

Content

LIST OF TABLES Table No.

Title of Table

Page 7

Table 1.1

Descriptive Statistics of the Study Area

Table 3.1

Attributes and Their Levels

50

Table 3.2

Salient Features of Data Used for Model Development

55

Table 3.3

Estimation Results of MNL Models (Stage-I)

57

Table 3.4

Estimation Results of DGNL Model

58

Table 3.5

Estimation Results of NL Models

60

Table 3.6

Estimation Results of CHNL Models

62

Table 3.7

Estimation Results of RPL Models (Stage-I)

64

Table 3.8

WTP Estimates from RPL1 Model (Stage-I)

65

Table 3.9

Estimated WTP Values from Stage-I Models

66

Table 3.10 Log-likelihood Test of Logit Models (Stage-I)

69

Table 3.11 Estimation Results of Models for Bicycle Users

72

Table 3.12 Estimation Results of Models for Motorcycle Users

73

Table 3.13 Estimated WTP Values from Stage-II Models

74

Table 4.1

Average Income for Different Categories of Household

77

Table 4.2

Estimated Trip Rates: Revenue Generating Trips

77

Table 4.3

Estimated Trip Rates: Educational Trips

78

Table 4.4

Bus Stops and Their Influence Areas

84

Table 4.5

Fixed Cost of Feeder Vehicles

86

Table 4.6

Revenue Required for Covering Fixed Cost

86

Table 4.7

Revenue Required for Covering Running Cost

87

Table 4.8

Selected Fare Combinations

90

Table 4.9

Details of Feeder Service in Scenario-I: GC as MOE

91

Table 4.10 Summary of Feeder Service in Scenario-I: GC as MOE

92

Table 4.11 Summary of Feeder Service in Scenario-II: GC as MOE

93

Table 4.12 Summary of Feeder Service in Scenario-III: GC as MOE

93

Table 4.13 Summary of Feeder Service in Scenario-IV: GC as MOE

94

Table 4.14 Summary of Feeder Service in Scenario-V: GC as MOE

94

Table 4.15 Summary of Feeder Service in Scenario-VI: GC as MOE

95

Table 4.16 Attributes of Recommended Feeder Service: GC as MOE

96

Table 4.17 Effect of ASC on Recommended Feeder Service in Scenario-I: GC as MOE

97

Table 4.18 Effect of ASC on Revenue Surplus: GC as MOE vii

100

Content

Table No.

Title of Table

Page

Table 4.19 Summary of Feeder Service in Scenario-I: Passenger-km as MOE

101

Table 4.20 Summary of Feeder Service in Scenario-II: Passenger-km as MOE

101

Table 4.21 Summary of Feeder Service in Scenario-III: Passenger-km as MOE

102

Table 4.22 Summary of Feeder Service in Scenario-IV: Passenger-km as MOE

102

Table 4.23 Summary of Feeder Service in Scenario-V: Passenger-km as MOE

103

Table 4.24 Summary of Feeder Service in Scenario-VI: Passenger-km as MOE

104

Table 4.25 Attributes of Recommended Feeder Service: Passenger-km as MOE

105

Table 4.26 Effect of ASC on Revenue Surplus: Passenger-km as MOE

107

Table 4.27 Comparison of GC Saving for Bus Stop Bound Trips

110

Table 4.28 Comparison of GC Saving for Both Directions of Travel

111

viii

Content

LIST OF FIGURES Figure No.

Title of Figure

Page

Figure 1.1

Feeder Vehicles Considered in the Present Work

5

Figure 1.2

Outline of the Study Area

7

Figure 2.1

Schematic Diagram for Travel Behaviour Analysis

10

Figure 2.2

Schematic Diagram for Design of Feeder Service

36

Figure 2.3

Selection of Routes and Vehicle with External Subsidy

41

Figure 2.4

Selection of Routes and Vehicle with Cross Subsidy

45

Figure 3.1

Three Level Form of Operation Based Tree Structure

59

Figure 3.2

Three Level Vehicle Based Tree Structure

60

Figure 3.3

Two Level Form of Operation Based Tree Structure

60

Figure 3.4

Two Level Vehicle Based Tree Structure

60

Figure 4.1

81

Figure 4.2

Comparison of Modeled and Observed Revenue Generating Trips Comparison of Modeled and Observed Household Trips

Figure 4.3

Estimated Trips to Bus Stops

82

Figure 4.4

Temporal Variation of Travel Demand

83

Figure 4.5

Road Network to Bus Stops

85

Figure 4.6

Comparison of Different Scenarios: GC as MOE

95

Figure 4.7

Effect of ASC on GC Saving: GC as MOE

98

Figure 4.8

Effect of ASC on Passenger-km: GC as MOE

98

Figure 4.9

Effect of ASC on Passenger Served: GC as MOE

98

Figure 4.10 Stability of Routes with GC as MOE

81

99

Figure 4.11 Comparison of Different Scenarios: Passenger-km as

104

Figure 4.12 Effect of ASC on GC Saving: Passenger-km as MOE

105

Figure 4.13 Effect of ASC on Passenger-km: Passenger-km as MOE

106

Figure 4.14 Effect of ASC on Passenger Served: Passenger-km as

106

Figure 4.15 Stability of Routes with Passenger-km as MOE

107

ix

Chapter 1 Introduction

1.1 MOTIVATION A country’s ability to unleash economic potential is closely linked to the effectiveness of its transportation system. Over several decades, lack of transportation infrastructure has been a major bottleneck for accelerating the economic growth of developing countries. In the recent years, developing

countries

like

India

have

realized

the

need

for

road

infrastructure, and taken up several highway development projects (Maitra et al. 2002; 2003). As rural population constitutes around 70% of the country’s population, Government of India also initiated a rural road development programme, in the year 2000, called Pradhan Mantri Gram Sadak Yojana (PMGSY) to connect all villages with a population of more than 500 by all weather roads.

It is aimed to provide single all weather

road connectivity to 160,000 villages in India by construction of rural roads under PMGSY (Sikdar 2002). Private vehicle ownership is very low in rural India. Therefore, alongwith the improvement of road connectivity, it is also necessary to develop suitable passenger transportation systems in rural areas. Presently, most of the major roads like National Highways, State Highways and Major District Roads are served by bus transportation systems. However, transportation linkage between village settlements and bus stops is largely a missing component. With the development of rural roads, paratransit carriers have started operating in some areas for providing access to nearest bus stop or higher order settlements. However, a systematic approach for the development of feeder service, based on planning principles, is missing. Several studies

Chapter 1

have been reported in literature on routing, scheduling and design of feeder service in urban areas (Wirasinghe 1980; Geok and Perl 1988; Martins and Pato 1998; Shrivastava and Dhingra 2001;

Shrivastava and O’Mahony

2006). However, there is little information available regarding planning of rural feeder service in developing countries. While in rural areas the demands are scattered over large geographical areas, the socioeconomic characteristics of rural population are also distinctly different from urban population. Therefore, it is necessary to develop a systematic approach for planning of rural feeder service with due consideration to demand pattern and socioeconomic characteristics of trip makers. Users’ benefit is a driving force for improvement planning of transportation system. For estimation of benefit, it is required to understand users’ valuation of travel attributes or willingness-to-pay (WTP). Several studies have been reported in literature on value of travel time savings (Bradley and Gunn 1990; Carlsson 1999; Hensher 1994; Hensher 2001a; Hess et al. 2005).

In some cases, valuation of qualitative travel attributes has also

been attempted (Hensher and Sullivan 2003; Hunt 2001; Phanikumar and Maitra 2007). However, there is very little information available regarding valuation of travel attributes by rural trip makers in developing countries. Therefore, in the process of planning of rural feeder service it is also necessary to investigate trip makers’ valuation of travel attributes or WTP. 1.2 OBJECTIVE AND SCOPE OF WORK The objective of the present work is to develop a framework for planning of rural feeder service to bus stop. In order to satisfy the objective, it is necessary to address several related aspects like type of feeder vehicle, form of operation, estimation of demand, estimation of users’ benefit, selection of routes, fare level and operational viability. The carrying capacity, fixed cost and operating cost are different for different vehicle types. Accordingly, service characteristics like headway and fare, and operational viability are also likely to be influenced by the selection of feeder vehicle. In the present work, two small carriers are considered as

2

Introduction

feeder vehicles. The users’ benefit resulting from feeder service may be influenced by the form of operation of feeder vehicles. In the present work, along with the traditional fixed-schedule form of operation, two demandresponsive forms of operation of feeder vehicles are investigated. The demand-responsive forms are called as ‘dial-a-ride’ and ‘dial-a-slot’. Investigation on travel behaviour of trip makers is an integral part of planning of rural feeder service. It is necessary to carry out travel behavior analysis for rational estimation of users’ benefit and estimation of demands to be served by feeder service. It is a common practice to design survey instrument, collect behavioral data (stated preference and/or revealed preference) from commuters, and analyze the same using suitable model specifications. Although behavioral data can be analyzed by different model specifications, the scope of the present work is limited to the use of three logit model specifications namely Multinomial Logit (MNL), Nested Logit (NL) and Random Parameter Logit (RPL). Users’ benefit in the context of transport improvement may be perceived as a reduction in the disutility of travel. Disutility of travel is generally expressed using several quantitative and qualitative attributes. These attributes have different measuring units, and therefore, need to be transformed to have a common unit for comparison or aggregation purpose. When monetary attribute is involved, the transformation is simple and the transformed value associated with each attribute is termed as WTP. Aggregation of WTP values for all the attributes, describing an alternative, is termed as Generalized Cost (GC). A reduction in the GC may be considered as a measure of users’ benefit. The WTP values may be influenced by the socioeconomic and/or trip characteristics of trip makers. Therefore, it is necessary to investigate the effect of socioeconomic and/or trip characteristics on WTP values. As a part of the total demand is expected to use feeder service, it is necessary to model total passenger demand from each village settlement to bus stop.

3

Chapter 1

The selection of route, vehicle type and form of operation are also required to be judged in the light of operational viability and other policy measures. The scope of the present work is summarized as follows: Travel Behaviour Analysis •

Design of experiments

•

Collection of stated preference data

•

Development

of

utility

equations

using

different

Logit

model

different

Logit

model

specifications: o

Multinomial Logit

o

Nested Logit

o

Random Parameter Logit

•

Estimation of willingness-to-pay values

•

Comparison

of

models

obtained

from

specifications •

Selection of models for estimation of ridership

•

Development of generalized cost equations

Design of Feeder Service •

Development of Database

•

Selection of feeder routes and vehicles

•

Comparison of different forms of operation of feeder vehicles o

Fixed-schedule

o

Dial-a-ride

o

Dial-a-slot

The work is demonstrated with reference to a case study in rural India. 1.3 FEEDER VEHICLES AND FORMS OF OPERATION Rural settlements are located in a scattered manner covering large geographical area. As a result, demand levels are generally low on various roads connecting rural settlements to the nearest bus stop. In low dispersed demand scenario, only small carriers are considered suitable as feeder vehicles. Two alternative vehicles are considered (Figure-1.1): ‘Tempo’ with carrying capacity of 6 persons and ‘Trekker’ with a carrying capacity of 10

4

Introduction

persons. It may be mentioned that both capital cost and operating cost are low for ‘Tempo’ and ‘Trekker’. Therefore, these vehicles are widely used as paratransit modes in India.

Tempo

Trekker

Figure-1.1 Feeder Vehicles Considered in the Present Work Alongwith the traditional fixed-schedule form of operation, two demandresponsive forms of operation of feeder vehicles namely; ‘dial-a-ride’ and ‘dial-a-slot’ are investigated. In ‘fixed-schedule’, the arrival time of the next vehicle is known to commuters but the availability of seat is not assured due to limited seat capacity. As the next vehicle arrival time is known but the seat availability or travel opportunity in that vehicle is not assured, commuters’ waiting time is described as ‘anxious waiting at stop’.

In ‘dial-a-ride’, commuters are assumed to inform service provider about the origin and the destination for a ride along the route using toll free telephone available at stop provided by Government or service provider. In response, service provider informs commuters about the vehicle allotted for trip, but starts the vehicle only when the capacity utilization of the vehicle along the route is assured to a desired level. Therefore, both operator and commuters are benefited. The operator provides the service with desired utilization of seat capacity and commuters are benefited as the seat availability is assured in a specified vehicle. As the seat availability is assured in a specified vehicle, the waiting is described as ‘Relaxed Waiting at Stop’.

5

Chapter 1

In ‘dial-a-slot’, the span of operation is divided into suitable time slots. Commuters are assumed to inform service provider in advance about their preferred time slot for journey by dialing a toll free telephone number from home. The service provider considers all such requests, schedules a vehicle ensuring acceptable usage of vehicle capacity along the route, and informs commuters about the allocated time slot and vehicle. In the process, some commuters may be allocated time slots other than the requested ones. Deviation from requested time slot, if any, is considered as disutility to commuters. As the seat availability is assured in a specified vehicle and arrival time is also known, commuters can wait at home. The time deviation or waiting is considered as ‘Relaxed Waiting at Home’. In this form of operation, both operator and commuters are benefited. The operator provides the service with acceptable utilization of seat capacity along the route, and commuters are benefited as the seat availability is assured and waiting is a relaxed-waiting at home.

1.4 STUDY AREA A 194.3 square-km geographical area in the state of West Bengal, India is selected as the case study. The study area includes parts of three administrative blocks namely Narayangarah, Kesiary and Dantan-I in Kharagpur Subdivision of West Medinipur District. The study area is located outside the urban fringe and represents typical rural characteristics. The nearest urban centre (i.e. Kharagpur town) is about 37 km from the northeastern corner, and 26 km from the north-western corner of the study area. The study area is bounded by National Highway-60 (NH) in the Eastern side, Belda–Kesiary road (major district road, MDR) in Northern side, Kesiary–Bhasra road (major district road, MDR) in Western side and river Subarnarekha in the Southern side (Figure-1.2). The NH and MDRs are served by bus service. However, the roads within the study area are not served by public transport system and therefore, a distinct travel patterns is observed from village settlements to bus stops located on roads representing the study area boundary. Descriptive statistics of the study area are shown in Table-1. 6

Introduction

Figure-1.2 Outline of the Study Area

Table-1 Descriptive Statistics of the Study Area

Number of villages*

173 *

Number of households

22,434

Total population*

115755 *

Average household size

5.16

*

58%

Rate of literacy

Total population with age less than 6 years* *

Total working population

17142 45208

*

Total main work force

30517 *

Agriculture dependable main work force

68.96%

Casual work force*

14661 *

Agriculture dependable casual work force

89.3%

Number of villages at a distance beyond 2 km from bus stop

126

Population living beyond 2 km from bus stop

74401

Number of households beyond 2 km from bus stop

14274

7

Chapter 1

Road length (within the study area)

253.9km

National Highway (a part of the study area boundary)

29.1km

Major District Road (a part of the study area boundary)

21.9km

*Source: 2001 census, Govt. of India

1.5 ORGANIZATION OF REPORT The motivation, objective and scope of the present work along with a brief description of feeder vehicles and forms of operation are discussed in Chapter 1. A description of the study area is also included in Chapter 1. Chapter 2 deals with the approach and methodology followed for travel behaviour analysis including preference elicitation, selection of attributes and their levels, generation of alternatives, data collection and analysis. Besides, it also includes a discussion on the approach adopted for the selection of feeder routes and vehicles. The approach followed for the comparison of different forms of operation of feeder vehicles is also included in Chapter 2. Chapter 3 presents in details the econometric analysis of behavioral data using various model specifications, and estimation of trip makers’ willingness-to-pay with respect to attributes of hypothetical feeder service. Chapter 4 demonstrates estimation of demand, selection of feeder routes and vehicles under different policy scenarios, and comparison of different forms of operation of feeder vehicles. Chapter 5 presents conclusive remarks and the findings from the work, with a note on scope of future works. 1.6 SUMMARY This chapter justifies the need for developing a framework for planning of feeder service to bus stop. The objective of the study is outlined and the steps required to be carried out to fulfill the objective are enumerated under the scope of the work. The types of feeder vehicle and forms of operation considered for investigation are also discussed. The study area is described and the organization of report is also mentioned.

8

Chapter 2 Approach and Methodology

2.1 INTRODUCTION This chapter presents the approach and methodology followed to develop a framework for planning of rural feeder service to bus stop. There are two major components of work namely ‘Travel Behaviour Analysis’ and ‘Design of Feeder Service’. Section 2.2 describes methodology followed for travel behaviour analysis, which includes selection of attributes and their levels, preparation

of

choice

sets,

design

of

questionnaire,

collection

and

organization of data, analysis of data using different model specifications, estimation

of

willingness-to-pay

(WTP)

values

and

development

of

generalized cost (GC) equations. The methodology followed for design of feeder service is described in Section 2.3. This section includes development of database required for design of feeder service, measure of effectiveness for selection of feeder routes and vehicles, policy instruments, and comparison of different forms of operations of feeder vehicles. Finally, the work presented in this Chapter is summarized in Section 2.4. 2.2 TRAVEL BEHAVIOUR ANALYSIS The travel behaviour analysis includes selection of attributes and their levels, preparation of alternatives, formation of choice sets, collection of data and development of database, analysis of behavioural data using different model specifications, estimation of WTP values and development of generalized cost equations. Figure-2.1 shows a schematic diagram of the methodology followed for travel behaviour analysis. Various steps of the methodology are discussed in subsequent sections.

Chapter 2

Design of Experiment Type of Data and Preference Elicitation Technique, Attributes and their Levels, Choice Sets, Questionnaire Design and Pilot Survey Collection of Data and Development of Database Sampling Strategy, Data Collection and Database Development Analysis of Data Organization of Data, Econometric Models Valuing the Attributes/ willingness-to-pay (WTP) values Comparison of utility equations Selection of Models for Estimation of Ridership Development of Generalized Cost Equations

Figure-2.1 Schematic Diagram for Travel Behaviour Analysis 2.2.1 Design of Experiment Design of experiment aims to combine attribute levels into profiles of alternatives and choice sets. It provides a structure that allows estimation of choice parameters in models and a ‘highly structured method of data generation’ (Hanley et al. 1998), relying on carefully designed tasks or ‘experiments’ to reveal the factors that influence choice. In a choice experiment, individuals are asked to choose their preferred alternative among several alternatives presented in a choice set. Each alternative is described by a number of attributes or characteristics. A monetary value is included as one of the attributes, along with other attributes of importance, while describing the profile of the alternative presented. Few alternative profiles are then assembled in a choice set and presented to respondents, who are asked to state their preferred profile in each choice set (Hanley et al. 1998; Louviere et al. 2000; Bennett and Blamey 2001).

Thus, when

individuals make their choice, they implicitly make trade-offs among the levels of the attributes in the different alternatives presented in a choice set. The attribute in monetary form enables estimation of the value of the 10

Approach and Methodology

other attributes in terms of respondents’ WTP.

So, design of experiment

includes type of data, preference elicitation method, attributes and their levels, design of alternatives and choice sets. 2.2.1.1 Type of Data and Preference Elicitation Technique In behavioural analysis, it is necessary to collect preferences of people in the form of either Revealed Preference (RP) or Stated Preference (SP) data (Adamowicz et al. 1994; Bates 1982; Kroes and Sheldon 1988; Louviere 1988a; Hensher 1994; Holguin-Veras 2002). RP refers to the observation of preferences revealed against real articles. It is required to have a present demand for the article in question in order to apply RP. In RP, the attributes may be collinear, making it difficult or impossible to predict the effect of independent variation in an attribute. Also, RP data may be inappropriate as they cannot accommodate non-existing attributes or variability of attributes which in-turn does not permit to establish their influences. RP methods capture only ‘use value’. In addition, RP data requires large number of observations leading to expensive and time consuming data collection process. RP methods include approaches such as the hedonic pricing (Pommerehne 1988; Bateman et al. 2000; Howarth et al. 2001) and the travel cost method. In the hedonic pricing method, an article is assumed to be formed by a set of attributes and the value of article is considered as a function of each attribute. The value of an attribute is called an implicit price or a hedonic price of the attribute, as it cannot be observed directly in a real market. It is possible to estimate the price by analyzing the prices of an article that has different quantities of each attribute in the market. Basic problems with the hedonic pricing method include omitted variable bias, multicollinearity, functional form, market segmentation and restrictive market assumptions. In the travel cost method, a value of non-market article is estimated by using consumption behaviour in a related market, where travel costs are used as a measure of preferences for the article. Basic problems with the travel cost method include choice of dependent variable, multi-purpose trips, holiday-makers versus residents, calculation of distance costs and the value of time, and statistical problems.

11

Chapter 2

SP refers to observation of preferences stated against real and/or hypothetical article.

SP facilitates inclusion of hypothetical attributes and

variability of attributes and requires fewer observations than RP. In addition, it allows complete control over choices offered and their attributes and ensures sufficient variation in data. SP methods include Contingent Valuation Methods (CVM) and Conjoint Methods (Pommerehne 1988). In CVM, the entire article is valued by eliciting a person’s WTP directly. As a result, nothing is revealed about the value of the different attributes that comprise the article. CVM may be classified as open ended CVM and referendum CVM. In the open ended CVM, a person is asked to state his/her WTP without giving any amount. The WTP can be estimated by simply taking the mean of the WTP stated. In the referendum contingent valuation, an amount is given and a person is asked to state whether he is willing to pay or not (yes/no). The data is generally analyzed using binary logit model.

Conjoint Methods include Conjoint Rating, Ranking and Choice (Discrete Choice Experiment). These three methods differ in theoretical assumptions, methods of analysis and experimental procedures (Louviere et al. 2000; Blamey et al. 2002). The basic design of alternatives is same in the three techniques and respondents must decide which of mutually exclusive multiattribute alternative(s) they prefer. In Rating, respondents are asked to evaluate a series of alternatives, one at a time, using a numerical ratings scale. The degree of task complexity is higher as the respondents have to place a value (characterizing the strength or degree of preference) on each alternative

(Louviere

et

al.

2000).

Contingent

rating

provides

the

respondent with the opportunity to rate alternatives equally and thus to indicate indifference among alternatives. The data is analysed by regressing the rating scores against the attributes and using Ordinary Least Squares (OLS), to estimate regression parameters. Ratings data are ordinal (only the ordering matters: the difference in ratings does not measure the strength of preference for one alternative over another) and discrete (as opposed to continuous variables) and these characteristics of the data 12

Approach and Methodology

violate the assumptions underlying OLS. Furthermore, because respondents are not required to choose a particular alternative, but to simply rate each one on a preference scale, the model cannot be used to predict choice behaviour or level of demand for a particular alternative (Adamowicz et al. 1998). There exists a second variant of rating method called ‘paired comparison’, where respondents are presented two alternatives and asked to rate their preference for the alternatives on a five or ten-point scale (Likert scale e.g., 1= highly preferred, 5= highly not preferred). These numbers may not represent the actual or true choice behavior of individuals due to the lack of strong theoretical foundation consistent with economics (Adamowicz et al. 1998). In most applications it is a standard practice to

vary both levels and types of attribute over the series of questions, such that respondents are only required to consider two or three attributes at a time.

In Ranking, respondents are presented with three or more alternatives in one question and asked to rank the alternatives from most to least preferred,

therefore,

provide

a

complete

preference

order

(strongly

ordered). A series of these ranking exercises is administered to the respondent. Conjoint ranking is not much used because of theoretical difficulties in analyzing the data (Louviere and Timmermans 1990). The use of ranking and rating techniques suffers from potential theoretical and practical obstacles. These concerns include the difficulty individuals might experience in ranking/rating all the alternatives, and the fact that rating

tasks

in

particular

involve

difficulty

in

making

interpersonal

comparisons and departure from the choice contexts that are faced by consumers in the real world (Bennett and Blamey 2001). Several SP studies used traditional ranking or rating based preference techniques (Hunt 2001; Lai and Wong 2000; Praveen and Rao 2002). Discrete choice experiment (DCE) is the simplest of the choice techniques and thus its biggest advantage is the low cognitive complexity – the degree

13

Chapter 2

of task complexity and difficulty arising from the experiment. The DCE experiments provide a framework for estimating the relative marginal disutility of variations in attributes, and their potential correlations (Louviere et al. 2000). The DCE method involves consumers making mutually exclusive choices from a set of substitutable articles. Moreover, DCE has strong theoretical foundation based on economic theory, and is an established approach for understanding and predicting consumer tradeoffs and choices in marketing research. DCE methods have been used extensively to model the behavior of individuals. For example, in the field of transportation, for valuing travel time savings (Carlsson 1999; Hensher 2001a, 2001b; Hensher 2004; Hess et al. 2005; Greene et al. 2006), for mode

choice

modeling

(Bhat

1995;

Brownstone

and

Train

1999;

Brownstone et al. 2000; Alpizar and Carlsson 2001; Arne and Felix 2004; Train and Winston 2004; Koppleman and Sethi 2005), for route choice (Yai et al. 1997; Hensher and Sullivan 2003), and several other nontransportation fields such as households’ response to rebates on energyefficient appliances (Revelt and Train 1997), impact of fish stock, which is affected by water quality, on anglers’ choice of fishing site (Train 1998) recreational saltwater fishing site (Mellisa et al. 1998), tourism (Robin and Adamowicz 2003), valuing wetland attributes (Carlsson et al. 2003), demand for genetically modified food (Rigby and Burton 2003; Carlsson et al. 2004; Onyango et al. 2004), and consumers’ willingness to pay for water service improvements (Hensher et al. 2004; MacDonald. et al. 2003), to avoid power outages (Carlsson and Martinsson 2004), radical Islamic terrorism (Barros and Proença 2005), British general elections (Garrett 2001), voting behavior (Ding-Ming Wang 2001), etc. The attribute based DCE technique is therefore, adopted in the present study for collecting the preferences of users. The term ‘discrete choice’ arose from the distinction between continuous and discrete variables for denoting a set of alternatives. The word ‘discrete’ indicates that the choice is discrete in its nature, meaning that it is only possible to choose one alternative. A discrete choice situation is defined as

14

Approach and Methodology

one in which the respondent faces a choice among a set of alternatives meeting the following criteria (Train 1993): •

The number of alternatives in the set is finite

•

The alternatives are mutually exclusive

•

The set of alternatives is exhaustive (all possible alternatives are included)

The DCE is characterized as a method in which the article in question is described by a number of attributes. The attributes and their levels must be constructed so that they force the respondent to trade. It is important to note that each time a level is changed, a new scenario arises. By securing a certain variation in the scenarios, it becomes possible to examine the degree to which each attribute influences the choice of the decision-maker; that is, to estimate the marginal rates of substitutions of the attributes (Louviere et al. 2000). Attributes can possess either positive or negative utility, and to varying degrees. Choice experiments can thus be used to examine the response of an individual to changes in the scenario attributes. 2.2.1.2 Attributes and Their Levels For the development of choice sets, it is necessary to identify suitable attributes and define their levels. A well-designed behavioral experiment requires significant pre-testing for identifying attributes, their levels and important interactions (Louviere 1988b).

Bennett and Blamey (2001)

specified that the attributes should be relevant to the requirements of the policy makers, but need to be meaningful and important to the respondents. To satisfy these requirements, it is important to carry out review of literature, group discussions, and interviews with key persons such as policy makers, experts and users. Blamey et al. (2002) suggested that the preference should be given to those attributes which are demand-relevant, policy relevant and measurable. The cost attribute plays an important and distinct role in the DCE. The inclusion of a cost attribute provides the DCE with a special quality as it becomes an elicitation procedure for WTP. This implies that benefits are estimated in monetary terms and causes the DCE to be consistent with welfare economics. Results from different studies can

15

Chapter 2

then be compared. Inclusion of a cost attribute makes it possible to indirectly obtain the respondent’s WTP for either the article in its entirety (an alternative) or the respondent’s WTP for the attribute respectively, i.e. marginal WTP (Bennett and Blamey 2001; Carlsson et al 2003; Hanemann 1984) In this context, it is important to determine the way in which levels are to be presented: either qualitatively or quantitatively. Moreover, it is necessary to decide whether the quantitative attributes should be presented in absolute or relative terms. Green and Srinivasan (1990) mentioned that the levels have to be acceptable such that levels that will be dominated at any stage are avoided. Ryan (1999) described three key factors while choosing the levels for each attribute: •

The levels must be plausible to the respondents

•

The levels must be actionable to the respondents

•

The levels must be constructed so that the respondents are willing to make trade-offs between combinations of the attributes.

In the present study, alternatives of feeder services are designed on the basis of type of feeder vehicle and form of operation. The attributes of two vehicles and three forms of operation as discussed in Chapter 1 are taken for the design of alternatives. The significant attributes and their levels for generation of alternatives are selected based on discussions with trip makers and experts. 2.2.1.3 Choice Sets This stage includes the formation and pairing of alternatives. Various methods can be used to design and reduce (if required) the number of alternatives to be included in the questionnaire. One of the crucial objectives of the experimental design is that the number of alternatives is minimized while being able to infer utilities for all possible alternatives – which imply keeping the choice task simple to the respondents and at the same time being able to extract all the necessary information from the

16

Approach and Methodology

choices. Design of experiment is a way of manipulating attributes and their levels to permit rigorous testing of certain hypotheses of interest (Louviere et al. 2000). The most popular way of combining the levels of the attributes is the factorial design. The questionnaire is designed so that each level of each attribute is combined with every level of all other attributes. A factorial design is simply the factorial enumeration of all possible combinations of attribute levels. Factorial design may be ‘full factorial’ and ‘fractional factorial’. Full factorial design refers to a design in which all possible alternatives are represented. If there are ‘M’ attributes each with ‘L’ levels, then the full factorial technique would result into LM alternatives. Full factorial design has very attractive statistical properties as it guarantees that all attribute effects of interest are truly independent (i.e. attributes are independent by design). However, full factorial design is only a real possibility for small experiments that involve a limited number of attributes or levels. With more attributes and levels, it may be necessary to reduce the size of the design. This can be done by the use of fractional factorial design (Louviere et al. 2000). Fractional factorial design involves selection of a subset (a fraction) of the full factorial design, in which the properties of the full factorial design are maintained in the best possible manner, such that the effects of interest can be estimated as efficiently as possible. However, all fractional designs involve some loss of statistical information. This loss of information can sometimes be significant, as fractional factorial designs limit the ability to take higher order effects into account, i.e. interactions among two or more attributes (Louviere et al. 2000). Several studies used main effect fractional factorial design, in which it is assumed that interactions among attributes are insignificant in all two-way and higher order interactions. Louviere et al. (2000) stated that the exclusion of interaction effects does not necessarily lead to biased result, because: •

Main effects typically account for 70% - 90% of the explained variance

17

Chapter 2

•

Two-way interactions typically account for 5% - 15% of the explained variance

•

Higher-order interactions account for the remaining explained variance.

Even if interactions are significant, they rarely account for much of the explained variance and hence may not significantly affect design efficiency. When it comes to the efficiency of designs, Huber and Zwerina (1996) mentioned that level balance, minimal overlap, utility balance and orthogonality are the four properties that characterize efficient choice designs. Level balance requires that the levels of each attribute occur with equal frequency in the design. A design has minimal overlap when an attribute level does not repeat itself in a choice set. Minimal overlap relates to the statistical properties when pairing the alternatives. Utility balance requires that the utility of each alternative within a choice set is equal. Utility balance is difficult to incorporate in the design as it demands a priori knowledge of respondents’ preferences. Finally, orthogonality can be considered as the most important aspect of efficiency of experiments. An orthogonal design is one in which the levels of different attributes across profiles are uncorrelated. Such designs assure that an estimate of one attribute is unaffected by the estimate of other attributes (Huber 1987). The type of design or appearance of choice sets in the questionnaire is another issue to be given due consideration. Three types of designs are available for preparing choice sets: designs with fixed, randomized or individualized

choice

sets.

In

a

fixed

experiment

approach,

each

respondent faces exactly the same choice sets at exactly the same stage of the choice task. In a randomized experiment, each respondent also receives the same choice sets but the order differs over respondents. In an individualized experiment each respondent receives his/her own choice sets, generally, pivoting on his/her previous responses. Depending on the data collection method and analysis, either of the design types may be adopted.

18

Approach and Methodology

Another issue is whether to present the alternatives in the choice sets in a generic (alternatives A, B, C) or alternative specific form (Brand A, Brand B, Brand C). Blamey et al. (2000) discussed advantages of these two approaches. An advantage of using alternative specific labels is familiarity with the context and hence the cognitive burden is reduced. However, the risk is that the respondent may not consider trade-offs between attributes. This approach is preferred when the emphasis is on valuation of the labeled alternatives. An advantage of the generic model is that the respondent is less inclined to only consider the label and thereby focus more on the attributes. Therefore, this approach is preferred when the emphasis is on the marginal rates of substitution between attributes. It is necessary to check the questionnaire with the help of a pilot study and redesign the questionnaire, if necessary. In the present study alternatives are generated using fractional factorial orthogonal main effects only technique. The choice sets are prepared using fixed experiment approach in alternative specific form. 2.2.1.4 Questionnaire Design and Pilot Survey The questionnaire should include socioeconomic and trip characteristics of respondents as well as the SP choice sets. Before going for the main survey, the design questionnaires should be taken for pilot survey to see the understanding of targeted population.

Considering low literacy rate,

and exposure of the people of the area to such type of survey for the first time, an added emphasis is given in the present work on pilot surveys. Process of presenting choice set, number of choice set in the questionnaire, etc. are finalized based on pilot surveys. 2.2.2 Collection of Data and Development of Database Before carrying out data collection, it is necessary to resolve the issue of sampling. All surveys need to be based on the application of strict sampling theory (Kish 1965; Yates 1981). This permits quite small samples to be reprehensive of population from which the samples are drawn. Without

19

Chapter 2

representativeness, it would normally necessary to conduct a complete census, which is an expensive option, probably to the point that transportation data would never be collected. The choice of survey population obviously depends on the objective of the survey. Given the survey population, a sampling strategy should be determined. Sampling strategy may indicate either simple random sampling or choice based sampling. Random sampling is one in which all individuals from the sample have equal opportunity to be chosen as potential respondents. Choice based sampling/ stratified sampling is one in which individuals are chosen depending upon characteristics/ strata such as gender, income, occupation, residential location, etc. (Louviere et al. 2000). The motivation for randomness is that if the sample is random, the sampling distribution will be the normal distribution with the ‘true’ mean of the population. If the sample is not random, a bias is introduced which causes a statistical sampling or testing error by systematically favoring some observations over others. When conducting a DCE, it is important to consider the data collection procedure. The methods available for collecting data are (Bennett and Blamey

2001):

questionnaires,

Face-to-face email/internet,

interview, gathering

telephone in

interview,

‘central

mailed

facilities’

and

combination of the above. Face-to-face interviews are characterized by the interviewer and respondents sharing both time and space. Besides generating very high response rates, the advantage of this method is that the interviewer can lead the respondent through the hypothetical scenario and elaborate if the respondent does not understand the task. All other techniques involve low response rate, high costs and time, especially in developing countries. In the present study, stratified random sampling based on occupation of head of household is accepted for conducting survey. A face-to-face personal interview with head of the household is adopted to collect the data. The main reason for selecting this sampling method is due its capacity to

20

Approach and Methodology

include different strata of people along with maintaining randomness in the sample; thus contributing to minimum bias (Hensher 1994). In developing country like India, walking is the most common form of travel for a distance of up to 1 to 2 km (Iles 2005). Therefore, behavioral samples are targeted from villages which are located beyond 2 km from bus stops. A proper database needs to be prepared with the data collected from the surveys for an easy and efficient estimation of models and other analysis. The database to be developed should include both primary and secondary data. Primary data include socioeconomic and trip making characteristics of respondents along with stated choice responses. Trip making data from households for revenue generating and non-revenue trips are required to be collected from primary sources. Secondary database should include bus routes, road network and village level characteristic of the study area. 2.2.3 Analysis of Data Analysis of data includes organization of data, model specifications, model development and interpretation of results. 2.2.3.1 Organization of Data Data consisting of socio-demographic variables, preferences and other details

should

be

well

organized.

Generally,

each

choice

contains

information related to the level of each attribute in the alternative, and the chosen alternative. Several coding styles are available to decide how levels of attributes enter into the models. Numerical/Quantitative attributes (e.g., time, price) can enter in cardinal linear form (i.e. on a continuous scale) and take actual values. Qualitative attributes on the other hand can take either dummy coding (0, 1) or effects-type coding (1, 0, -1) specifications. In dummy coding, the presence of level in the design is coded as ‘1’ and absence as ‘0’ whereas in effects-type coding, presence of level in the design is coded as ‘1’ and absence as ‘-1’ when there are two levels; and presence of the first level as (1, 0) the second as (0, 1) and the third as (1, -1) in case of a three level attribute. This kind of coding has an

21

Chapter 2

advantage that all effects are stated in deviation from some average; and interaction terms, if present in the design, are uncorrelated with the main attributes. Also, the sum of the part worth across attributes is zero (Haaijer 1999). 2.2.3.2 Econometric Models Commonly used models to estimate discrete choice experiments are Logit and Probit models. Based on their evaluation technique they are further categorized as closed form, partial closed partial simulation and complete simulation (Train 2003), as given below: Complete closed form

Binary Logit /Multinomial logit (MNL) /Nested logit (NL)

Partial closed form/ partial simulation Mixed logit, (ML) / (RPL) Complete simulation

Binary

Probit/Multinomial

probit

Heteroscedastic extreme value (HEV) The scope of the present work is limited to analyzing behavioural data using different logit model specifications namely MNL, NL and RPL. Therefore, only these model specifications are introduced in this Chapter. The theoretical foundation related to the development of these econometric models is discussed in Annexure-A. Multinomial logit (MNL): In econometric models based on Random Utility Theory (Thurstone 1927; McFadden 1974), the utility of each element consists of an observed (deterministic) component and a random (disturbance) component. If the random error terms are assumed to follow extreme value type-I (Gumbel) distribution, and be independently and identically distributed (IID) across alternatives and cases (or observations), the multinomial (or conditional) logit (MNL) model (McFadden 1974) is obtained. This model can be estimated by maximum likelihood techniques, and is useful for modeling choice behavior due to its simple form for choice probabilities. However,

22

Approach and Methodology

several limitations apply to this model. The most severe of these is the Independence of Irrelevant Alternatives (IIA) property, based on the assumption that the error terms are independent across alternatives, choice sets, and respondents (Louviere and Woodworth 1983; Kamakura and Srivastava 1984), which states that a change in the attributes of one alternative changes the probabilities of the other alternatives in proportion. This substitution pattern may not be realistic in all settings. Secondly, the coefficients of all attributes are assumed to be the same for all respondents in a choice experiment, whereas in reality there may be substantial variability in how people respond to attributes. Nested logit (NL): The most widely known model which relaxes IIA of the MNL model is the nested logit (NL) model which allows interdependence between the pairs of alternatives in a common group (McFadden 1978; Ben-Akiva and Lerman 1985; BoÈ rsch-Supan 1990). In Nested Logit Model, the set of alternatives are divided (and sub-divided) into exclusive groups (nests), where some aspect only pertains to members of that particular group. Derivation of NL model is based upon the assumptions of MNL, except that correlation of error terms is assumed to exist among predefined groups of alternative. Such error correlations arise if an observed factor influences the utility of all members in the group. The NL model can be written as product of a series of MNL choice models defining each level of tree structure.

NL

relaxes IIA by organizing similar alternatives into groups and allowing different correlation patterns between groups than within group. By allowing correlation among subsets of utility functions, the IIA problem of MNL is alleviated partially. It retains restrictions that alternatives in a common nest have equal cross-elasticity and alternatives not in a common nest have cross-elasticity as for the MNL. The NL model arises as a random utility model in which the random component of utility has the generalized extreme value distribution.

23

Chapter 2

NL model has been used in several studies for analyzing preference such as new vehicle choice (Mccarthy and Tay 1998), debit, credit, or cash use for gasoline purchases (Carow and Staten 1999), alternative models of automobile purchases (Wojcik 2000), pre-work trip-making and home departure time choice (Yun et al. 2000),

airport and airline choice for

passengers departing from a large metropolitan area (Pels et al. 2000), choice of a graduate business school (Montgomery 2002), combined-mode choice, location choice, and route choice (Lo et al. 2004), air craft choice (Wei and Hansen 2005). Covariance heterogeneity nested logit (CHNL): The NL model imposes restriction of equal correlation in random utility components among nested alternatives across respondents. The degree of (increased) sensitivity (i.e. cross elasticity) between the alternatives present in one branch to alternatives present in another branch differs based on socioeconomic characteristic of respondents. In general, ignoring the variation of covariance among nested alternative across respondents making choice (i.e. ignoring covariance heterogeneity in a nested logit model) may produce biased and inconsistent estimates of variables. Allowing heterogeneity across individuals in the covariance of nested alternatives in the estimation of NL model leads to the development of CHNL model (Bhat 1997). Random parameter logit (RPL): Modifications to the MNL model to overcome the limitations lead to the development of RPL. RPL is a highly flexible model that can approximate any random utility model (McFadden and Train 2000). RPL model allows for a more heightened level of flexibility by specifying taste coefficients to be randomly distributed across individuals (Revelt and Train 1998 ; Louviere et al. 2000). The utility expression for RPL is the same as that for MNL model except that the analyst may nominate one or more taste parameters (including alternative-specific constants, i.e. ASCs) to be treated as random parameters with the variance and mean to be estimated. The RPL form has

24

Approach and Methodology

important behavioral implications. The attributes with random parameters induce a distribution around the mean that provides a mechanism for revealing

preference

heterogeneity

in

the

sampled

population.

This

heterogeneity takes the form of a random effects version of unobserved heterogeneity that may be refined by making it a function of observed variables such as income, sex, age, trip purpose, etc. This is a way of revealing the specific sources of variation in unobserved heterogeneity across a sampled population. RPL can also account for correlation among alternatives. There are several advantages of RPL model specification. •

The model does not exhibit the IIA property

•

The model can be derived from utility maximizing behavior

•

The model can account for uncontrolled heterogeneity in tastes across respondents

•

The model allows the unobserved factors to follow any distribution

•

The model (with normally distributed coefficients) can approximate multinomial probit models

RPL models have been used for analyzing preferences in numerous contexts, which include valuing of travel time savings (Algers et. al. 1998; Carlsson 1999; Hensher 2001a ; Hensher 2001b; Hess et al. 2005; Cherchi and Polak 2005), travel mode choice (Alpizar and Carlsson 2001; Arne and Felix 2004) households’ response to rebates on energy-efficient appliances (Revelt and Train 1997), the impact of fish stock, which is affected by water quality, on anglers’ choice of fishing site (Train 1998), recreational saltwater fishing site (Mellisa et al. 1998), tourism in Uganda (Robin and Adamowicz 2003), valuing wetland attributes (Carlsson et al. 2003), demand for genetically modified food (Rigby and Burton 2003; Carlsson et al. 2004; Onyango et al. 2004), and consumers’ willingness to pay for water service improvements (MacDonald et al. 2003; Hensher et al. 2004), to avoid power outages (Carlsson and Martinsson 2004),

consumers’

choice of vehicle (Brownstone and Train 1999, Brownstone et al. 2000, Train and Winston 2004), radical Islamic terrorism (Barros and Proença

25

Chapter 2

2005), British general elections (Garrett 2001), voting behavior in Taiwan (Ding-Ming Wang 2001), etc. Distribution of random parameters: In RPL model, it is necessary to make an assumption regarding the distribution of random parameters/coefficients. This assumption causes much concern in RPL model development process. A specific distribution is selected with a sense that the ‘empirical truth’ somewhere lie in their domain. The commonly used distributions are normal (Revelt and Train 1997; Algers et al. 1998; Carlsson 1999;

Alpizar and Carlsson 2001;

Hensher 2001a; Carlsson et. al. 2003; Cherchi and Polak 2005), log-normal (Revelt and Train 1997; Hensher 2001a; Alpizar and Carlsson 2001), uniform and triangular (Revelt and Train 2000; Hensher and Greene 2001; Train 2001; Garrett 2001). However, there are other distributions, like Johnson’s

SB

distribution

(Hess

et

al.

2005),

discrete

distribution

(Chintagunta et al. 1991), which are also attempted by researchers. A brief discussion on various distributions is also included in Annexure-A. It may be mentioned that every distribution has its strength and weakness. The weakness is usually associated with the spread or standard deviation of the distribution. At its extremes include behaviorally unacceptable sign changes for the symmetrical distributions. The lognormal has a long upper tail. The normal, uniform, and triangular may give the ‘wrong’ sign for some parameters depending on the standard deviation. One appealing solution is to make the spread or standard deviation of each random parameter a function of mean. This way a conversion to truncated or constrained distributions appears to be the most promising direction of research in the future (Hensher and Greene 2003). For example, the usual specification of a normal distribution is β i = β + sν i , where β is the mean as s is the spread (standard deviation) and ν i is the random variable. The constrained specification would be β i = β + βν i so that the standard distribution is made equal to the mean. The constraint specification concept can be applied to other distribution also. For example, a constrained triangular distribution is 26

Approach and Methodology

a generalization of the uniform distribution, allowing for a peak in the density function with two endpoints of the distribution are fixed as zero and 2*mean. It is bounded below zero, bounded above at a reasonable value that is estimated and symmetric such that the mean is easy to interpret. It is appealing for handling of attribute values/ WTP values (Hensher and Greene 2001). Also with β i = β + βν i , where ν i has support from -1 to +1, it does not matter if β

is negative or positive. A negative coefficient on

ν i simply reverses all the signs of the draws, but does not change the interpretation. The advantages of constrained triangular distribution (say where mean equals spread) may be enumerated as follows. •

A constrained triangular distribution assures that the sign of the mean is constant throughout the sample

•

Unlike normal or lognormal distributions this is bounded in nature which results in early convergence (less computational time)

•

Estimation of WTP value is simple as the impact of spread is negligible due to constraint. Ratio of mean coefficient of any attribute over mean coefficient of cost directly gives WTP unlike with normal or lognormal distributions where standard deviation has significant effect on WTP.

In the present work, all RPL models are developed assuming constrained triangular distribution of random parameters. Selection of points for RPL model:

The RPL model does not have a closed form expression (unlike the MNL model) and therefore, it is approximated numerically through simulation by the method of Simulated Maximum Likelihood (SML). Numerous procedures have been proposed for taking intelligent/smart draws from a distribution (Morokoff and Caflish 1995). Random draws are commonly adopted using pseudo-random sequences for the discrete points in the distribution. Bhat (2001) showed that the coverage of random utility space is more representative by a quasi-monte carlo approach that uses non-random and more uniformly distributed sequences within the domain of integration. This procedure, known as ‘Halton’ sequences, offers the potential to reduce the 27

Chapter 2

number of draws that are needed for estimation of RPL models, thereby reducing run times and/or reducing the simulation error that is associated with a given number of draws. The number of draws is also another important aspect and required to secure a stable set of parameter estimates. In general, it appears that as the model specification becomes more complex in terms of the number of random parameters and the treatment of preference heterogeneity around the mean, the number of required draws increases. There is no standard number but experience suggests that a choice model with three alternatives and one or two random parameters (no preference heterogeneity) can produce stability with as low as 25 intelligent draws (i.e. Halton sequences, Bhat 2001; Train 2003), although 100 appears to be a good number (Hensher and Greene 2003). The best test however is to estimate models over a range of draws (say, 25, 50, 100, 250, 500, 1000 and 2000) for confirmation of stability/precision. This kind of study is particularly important when deriving empirical distributions for WTP indicators. Train (1999) and Bhat (2001) showed that the simulation variance in the estimated parameters was lower using 100 Halton numbers than 1,000 random draws. With 125 Halton draws, they found the simulation error to be half as large as with 1,000 random draws. Hensher (2000) investigated Halton sequences involving draws of 10, 25, 50, 100, 150 and 200 (with three random generic parameters) and compared the findings in the context of value of travel time savings (VTTS) with random draws. In all the models investigated, Hensher concluded that a small number of draws (as low as 25) produces model fits and mean VTTS that are almost indistinguishable. This is a phenomenal development in the estimation of complex choice models. 2.2.4 Valuing of Attributes In discrete choice models, valuing of an attribute measure is relatively straightforward as it is given by the ratio of partial derivatives of the constant utility function with respect to that attribute and travel cost (i.e. marginal rate of substitution between the attribute and travel cost at constant utility).

28

Approach and Methodology

Suppose, the deterministic part ‘V’, of the utility function in the model contains travel-time TT and travel-cost attribute TC. Then value of traveltime (VOT) is simply computed as under. VOT =

∂V / ∂TT

(2.1)

∂V / ∂TC

With the commonly used linear-in-variables utility function, the above formula reduces to βTT / βTC , where βTT and βTC are the coefficients of travel time and travel cost respectively. It is important to appreciate that the justification for this approach rests on a substantial body of microeconomic theory that addresses the issue of how individuals allocate time and its variation amongst alternatives. Valuing of an attribute from MNL, NL and CHNL model estimates is done by the above said marginal rate of substitution at constant utility. On the other hand, valuing of an attribute from RPL model estimate is not straight forward but depends to a great extent on the assumption of the random distribution of an attribute. A judicial decision is required on the assumption of random parameters to get an acceptable (say positive distribution for time attribute) distribution and the value (i.e. estimating coefficients with plausible sign) an attribute. If the mean estimate of random parameter needs to be of specific sign (i.e. non-negative), then lognormal distribution is favored. But the disadvantage of this distribution lies in its long upper tail that is behaviorally implausible for valuation (Hensher 2000) especially in estimating standard deviation. A uniform distribution with a (0, 1) bound is sensible, when dummy variables are to be estimated. But if either of normal, uniform, triangular distribution is used, then the disadvantage lies with the wrong sign to some shares with the estimated value due to the spread/standard deviation of the distribution. This disadvantage can be avoided

by

imposing

a

constraint

on

the

distribution

by

making

spread/standard deviation a function of the mean (Hensher and Greene 2001). Investigation on a constrained distribution has become an ongoing line of research (Hensher and Greene 2003; Greene et al. 2006; Hensher 2006; Basu and Maitra 2007; Phanikumar and Maitra 2007) for estimating a

29

Chapter 2

plausible value of an attribute. In RPL, valuing of an attribute is constructed using either the unconditional parameter estimate or the ‘common-choicespecific’

conditional

parameter

estimate

(Hensher

et

al.

2005).

In

unconditional parameter estimate, population moments are used to obtain a distribution of the parameter estimate. Here each sampled individual is randomly assigned along the continuous distribution. On the other hand, if the individuals are assigned based on subjective priors, i.e. knowledge of the chosen alternative, then valuing of attribute becomes conditional. Such an approach (popular in Bayesian paradigms) enables the analyst to identify ‘common-choice-specific’ parameter estimate. In this case, there is a possibility to observe that given a sample size, the frequency distributions do not resemble the priori continuous distributions for the random parameters (Sillano and Ortuzar 2005). On the other hand, unconditional parameter estimates simply refer to the random distributions assumed for RPL model development. Therefore, the present work considers only unconditional parameter estimates as the scope of the investigation. Accordingly, the RPL coefficients are estimated over the population using classical estimation procedure. The value of travel time measure from an unconditional parameter estimate may be constructed as given from Equation 2.2 to 2.4. This way it is possible to calculate value considering point estimates of coefficients but this approach ignores sampling variance in these point estimates (as correlation between random parameters are not considered). Case 1: When travel cost is fixed parameter with a coefficient estimate β TC and travel-time is random parameter with a mean coefficient estimate βTT and standard deviation/spread STT VOT = (βTT + STT × Dr ) / βTC

(2.2)

Case 2: When travel cost is random parameter with a mean coefficient estimate β TC and

standard

deviation/spread STC ,

30

and

also

travel-time

is

random

Approach and Methodology

parameter

with

a

mean

coefficient

estimate

βTT

and

standard

deviation/spread STT VOT = (βTT + STT × Dr ) /(βTC + STC × Dr )

(2.3)

Case 3: When travel cost is fixed parameter with a coefficient estimate β TC and travel-time is random parameter with a mean coefficient estimate βTT , standard deviation/spread STT and also heterogeneity coefficient estimate

δ TD ( H ) around the mean estimate of travel time for a heterogeneity study against ‘z’ observed variable (z can be defined as a dummy variable interacting with travel time attribute).

VOT = (βTT + δTT (H ) × z + STT × Dr ) / βTC

(2.4)

Dr in all the above cases is a draw from an assumed distribution of random parameter travel time. For example in triangular distribution, the draw is obtained

from

a

standard

uniform

distribution

Dr = 2V − 1 if V < 0.5 else Dr = 1 − 2(1 − V ) .

In

case

V = U[0,1] of

by

triangular

distribution, VOT can be calculated either by spread (like Case 1) or by standard deviation formulae. For a triangular distribution, the standard deviation is measured as spread / 6 . So, VOT can then be measured

as (βTT + (STT

6) × Dr ) / βTC .

For

a

triangular

or

constrained

triangular

distribution (where say the mean equals the spread), then the distribution of VOT for case 1/case 3 will be another triangular and/or constrained triangular distribution.

2.2.5 Comparison of Utility Equations

Selection of the utility equations obtained from different models is a key issue. In the present work, selection of model is done considering goodness of fit (ρ2), t-statistics, log-likelihood and the rationality of modeling technique.

The utility equation selected in this manner from different

model specifications, is used for further analysis.

31

Chapter 2

2.2.5.1 The Likelihood Ratio Test

The most common test undertaken to compare any two models is the likelihood ratio (LR) test. The LR statistics is given as: LR = -2 (LL1 - LL 2 )

(2.5)

Where, LL1 and LL2 are the log likelihood at convergence for model 1 and model 2 using same data set. The statistic used is chi-squared distributed with (K2-K1) degrees of freedom, where K is the number of estimated parameters. If the value of the LL-test exceeds the critical chi-squared value then one can conclude that the two models are statistically different, rejecting the null hypothesis of no difference. It may be noted that it is only possible to compare log-likelihood estimates for models that share common distributional assumption, e.g. the MNL model versus the more general RPL model. 2.2.5.2 Goodness of Fit

One of the most well-known measures of goodness of fit is the pseudo R2. It is defined as follows:

ρ2 = 1 −

LL(1) LL(0)

(2.6)

The nominator is the value of the log likelihood function at the estimated parameters and the denominator is its value when all the parameters are set equal to zero. The larger the difference between the two log likelihood values, the more the extended model adds to the very restrictive model. Ben-Akiva and Lerman (1985), however, point out that interpreting rho square whenever some additional independent variables are added is problematic. They suggest the calculation of rho squared bar (adjusted rho square) Adjusted

ρ2 = 1 −

LL(1) − K LL(0)

Where, K is the number of parameters

32

Approach and Methodology

2.2.6 Selection of Models for Estimation of Ridership

Utility equations developed using only SP data are acceptable for estimation of WTP values. The attributes and variables contained within RP data set are likely to be ill conditioned (e.g. largely invariant), and therefore,

parameter

estimates

(other

than

the

alternative-specific

constant terms, i.e. ASCs) obtained from model using RP data are likely to be biased. On the other hand, the attributes of SP data sets are likely to be of good condition and hence, the associated parameter estimates obtained from model using such data are likely to be unbiased. Nevertheless, the ASCs estimated from SP data are likely to be behaviorally meaningless, while those obtained from RP data sources likely to be of substantive behavioral value (Hensher et al. 2005). The ASCs obtained from a discrete choice models represent not only the average unobserved effect for each alternative but also reflect the choice shares within the data set from which it was estimated. For SP data, the choice shares will be obtained over a number of hypothetical choice sets

derived from some underlying

experimental design, each of which is given to multiple individuals. Beyond representing the average unobserved effects, the ASCs obtained from SP data may be meaningless (particularly for studies involving demand forecasting). On the other hand ASCs, acquired from RP data with or without other attributes should reflect the true choice shares observed over the population. Louviere et al. (2000) also argued on the same line that analyst should exploit strengths of both data sources while discarding the weakness displayed by each. Therefore, it is desirable to develop a joint SP-RP model for the demand estimation purpose. Although the use of joint SP-RP model is desirable for demand estimation purpose, it may not be possible to develop such joint models in all the applications primarily due the non-availability of RP database. In the present case, no feeder service is operational in the study area and as such commuters do not have any choice of mode while accessing bus stops from village settlements. In the absence of RP data, the role of ASCs become extremely important as SP data may not reflect true market shares

33

Chapter 2

observed in real market (Hensher et. al 2005). If the SP data include new alternatives and one truly believes that these data reproduce correctly the market shares of the population in forecasting, then the ASC (both for the existing and for the new alternatives) should be adjusted to match the SP market shares (Cherchi and Ortu´zar 2006).

Conversely, if the market

shares to match are unknown then relying on estimation results is the only alternative, i.e., as long as theoretical analysis are satisfied, considerations on the ASC from the model that provides the best statistical fit left with analysts judgment (Cherchi and Ortu´zar 2006). In the present case study it is not possible to reliably re-estimate the ASCs of SP model and therefore, the utility equations developed from SP data are used for the estimation of demand. The estimated demand is subsequently used to identify operationally viable feeder routes. As the ASCs present in such utility equations may result biased estimation of demand, the operational viability of recommended feeder routes is also checked assuming 50% and 25% of the ASC values in utility equations. Accordingly, routes are classified as ‘stable’ or ‘unstable’. The routes which remain operationally viable with the use of 100%, 50% and 25% of the ASC values, are defined as stable routes. These routes are called ‘stable’ because even if the ASC values are reduced to 25% of the values used for recommending these routes, the routes will still remain operationally viable. By classifying recommended routes as ‘stable’ or ‘unstable’, the uncertainty associated with

demand

estimation

and

its

effect

on

operational

viability

is

acknowledged, investigated (at least to some extent) and linked with recommendations. 2.2.7 Development of Generalized Cost Equations

Disutility of travel is generally expressed using several quantitative (e.g. waiting time) and qualitative (e.g. travel comfort) attributes. These attributes generally have different measuring units, and therefore, need to be transformed to have a common unit for comparison or aggregation purpose. When monitory attribute is involved, the transformation is simple and the transformed value associated with each attribute is generally

34

Approach and Methodology

termed as WTP value. Aggregation of such WTP values, for the attributes describing an alternative or system, is termed as Generalized Cost (GC). In the present case, valuing of attributes are used for developing generalized cost equations. An example of GC equation is given below: GCij =a1 (ttij) + a2 (wtij) + Fij

(2.7)

Where, GCij= generalized cost of travel between ‘i’ to ‘j’ ttij = travel time for travel between ‘i' and ‘j’ wtij = waiting time for travel between ‘i' and ‘j’ Fij

= direct cost of travel from ‘i' to ‘j’

a1 and a2 are the WTPs with respect to travel time and waiting time respectively. GC is a comprehensive measure of disutility and therefore, a reduction in GC is considered as a measure of users’ benefit. 2.3 DESIGN OF FEEDER SERVICE

This section describes the methodology followed for the design of feeder service. Figure-2.2 shows a schematic diagram of the methodology followed for the design of feeder service. The activities are discussed in subsequent sections. These include database, measure of effectiveness, alternative scenarios for operation, selection of routes and vehicles, and comparison of different forms of operation of feeder vehicles. 2.3.1 Database

It is necessary to develop a database for the design of feeder service. The database should include travel demand to bus stop, road network, link flows, temporal distribution of demand, and cutoff revenue required for viable operation.

35

Chapter 2

Database Travel Demand to Bus Stop, Temporal Variation of Demand, Road Network, Cutoff Revenue Measure of Effectiveness Alternative Scenarios Fare Levels Selection of Feeder Routes and Vehicles

Comparison of Different Forms of Operation Figure-2.2 Schematic Diagram for Design of Feeder Service 2.3.1.1 Travel Demand to Bus Stop

For planning of rural feeder service, it is essential to estimate travel demands to bus stops from various village settlements. The travel demand from each village to the nearest bus stop is modeled using socio-economic and demographic data. This is essentially the potential base demand, and a part of this demand is expected to use feeder service. Trips generated for bus stop and beyond from each village should be estimated separately for different categories of trip namely revenue generating and non-revenue generating. Regression analysis and/or trip rate analysis may be used for modeling trip generation (Ortuzar and Willumsen 2002). 2.3.1.2 Temporal Variation of Demand

The travel demand in the context of feeder service is expected to vary during different hours of a day. The variation of demands for bus stop bound and village bound trips are also unlikely to be the same. It is necessary to understand the temporal variations of bus stop bound and village bound travel demands for identifying the span of operation of feeder service and estimating the number of trips/vehicles required during the peak period.

36

Approach and Methodology

2.3.1.3 Road network

Road network of the study area is an essential input for the design of rural feeder service. Road network of the study area and other spatial information may be included in the geographical information system (GIS) database. Road network to bus stop: It is aimed to provide single all weather road connectivity to all villages through the on-going rural road development programme called the PMGSY. ‘Single all weather road connectivity’ essentially means the development of an all weather spanning tree road network. In the present work, this spanning tree is developed with due consideration to the preference of rural trip makers. The spanning tree includes links which are the most preferred links for travel to bus stop. An iterative approach, as given below, is followed to obtain the spanning tree from the base road network. It is assumed that the links included in the selected spanning tree will be developed under the PMGSY and other road development programmes. Step-1: Consider all roads links included in the base network for providing connectivity between a bus stop and all villages located in the bus stop influence area. Say, this network is represented by ‘m’ road links connecting ‘n’ nodes, where m>(n-1) Step-2: Assign travel demand from all village settlements to the bus stop using all or nothing assignment Step-3: Calculate the link flows Step-4: Temporarily delete the link with the minimum link flow Step-5: Check the connectivity of the network. If connectivity is not affected then go to Step-6, otherwise go to Step-7

37

Chapter 2

Step-6: Permanently delete the link identified in Step-4 and update the network (i.e. m=m-1). If m= (n-1) then proceed to Step-8, else proceed to Step-2 with the updated network. Step-7: Put back the link deleted temporarily in Step-4, identify the link with the next higher flow, temporarily delete that link and proceed to Step5. Step-8: With the updated network having (n-1) link connecting n nodes, assign travel demand from all village settlements to the bus stop using all or nothing assignment and obtain the final link flows. 2.3.1.4 Cutoff Revenue

It is aimed to design feeder service which will be operationally viable. Therefore, it is essential to estimate the cutoff revenues. The cutoff revenue is considered as the minimum required earning to cover the fixed cost and running cost of vehicle alongwith a minimum profit for operator. The cutoff revenue is an essential input for selecting feeder routes and vehicle. Also, as two different vehicle types are considered, cutoff revenues are estimated separately for two types of vehicle. 2.3.2 Measure of Effectiveness, Alternative Scenarios and Fare Level

It is necessary to consider an appropriate measure of effectiveness (MOE) while selecting feeder routes and vehicles. An attempt is made to optimize the MOE in the process of identifying operationally viable feeder routes. The MOE primarily represents the objective of providing feeder service in a quantitative manner. The selection of MOE is therefore, a policy matter. Different objective functions/ quantitative measures which are used in transportation studies include generalized cost (Silman et al. 1974; DuffRiddell and Bester 2005), generalized time (Lampkin and Saalmaas 1967; Dubois et al. 1979; Mandl 1980; Ceder and Wilson 1986), consumer surplus (Hasselström 1981), number of trips (VanNes et al. 1988), service coverage

38

Approach and Methodology

(Spasovic et al. 1994; Ramirez and Seneviratne 1996), etc.

In the present

work, selection of feeder routes and vehicles are demonstrated considering two alternatives MOEs namely, generalized cost and passenger-km served. It is also necessary to appreciate the role of policy instruments in the design of rural feeder service. The MOE may be further optimized using policy instruments. In the present work, ‘External Subsidy’ and ‘Cross Subsidy’ are considered as two policy instruments. With external subsidy, the cutoff revenue is lowered up to an amount equivalent to the amount of external subsidy per vehicle. With cross subsidy, the revenues earned by some vehicles/routes in excess of the cutoff revenue are utilized to cover more areas under feeder service in the light of the MOE.

Several alternative

scenarios are formulated and analyzed in order to demonstrate the role of policy measures in the design of feeder service. The fare has a vital role to play in the selection of operationally viable routes and types of vehicle. Therefore, it is also necessary to identify a suitable fare range for each vehicle type and consider practical fare levels during the analysis. 2.3.3 Selection of Feeder Route and Vehicle

Researchers have used different techniques for solving transit network design (TND) problem. Lampkin and Saalmans (1967), Silman et al. (1974), Dubois et al. (1979), Hsu and Surti (1976), Dhingra (1980), Mandl (1980), Baaj and Mahamassani (1990; 1995), and Shrivastava and Dhingra (2001) have used heuristic approaches for solving TND problem. Use of genetic algorithm-based approaches have been attempted by Pattnaik et al. (1998), Chien et al. (2001), Tom and Mohan (2003), Fan and Machemehl (2004), Khasnabis and Tom (2006). Other soft computing tools have been used by Hasselström (1981), VanNes et al. (1988), Baaj and Mahmassani (1991), Shih et al. (1998), and Zhao and Zeng (2006) for solving TND problem. In the present work, a heuristic approach is followed while optimizing the MOE in the process of selecting feeder routes and vehicles.

39

Chapter 2

With external subsidy, it is required to judge the operational viability of individual routes. With cross subsidies, it is required to judge the overall operational

viability

of

all

vehicles/routes

rather

than

individual

vehicle/route. Therefore, different approaches are necessary for selecting routes and vehicles under these two cases. 2.3.3.1 External Subsidy

The methodology followed for route and vehicle selection under external subsidy is shown in Figure-2.3 and the activities are discussed in subsequent sections. Step-0: Database The database includes the following: •

Present travel demand

•

Temporal variation of demand

•

Village settlements and their distance from bus stop as per the spanning tree road network

•

Capacity, journey speed and cutoff revenue for Tempo and Trekker

•

Layover time of feeder vehicles

•

Fare levels for Tempo and Trekker

Step-1: Select a Fare Level Different fare levels are investigated. For each fare level, repeat Step-2 to step-12. Step-2: Select a feeder vehicle Two types of feeder vehicle are considered. For each type of feeder vehicle repeat Step-3 to Step-12. Step-3: Select a bus stop A number of bus stops are located defining the boundary of the study area. For each bus stop repeat Step-4 to Step-12

40

Approach and Methodology

Database Select a Fare Level Select a Feeder Vehicle Select a Bus Stop Define a Feeder Route Peak Period Demand

Number of Vehicle Trips

No

Average Time Deviation

Re-estimate the Demand Yes

No

Converge

Calculate Number of Vehicles Check the Operational Viability Compare Feeder Routes from a Bus Stop

All Routes

Yes

All Bus Stops

Yes

Yes

All Vehicles All Fare Levels

Yes

No

Recommend Fare Combination, Feeder Routes and Vehicles

Yes

No

No

Generate Fare Combinations

All Fare Combinations Select a Fare Combination

Select Route and Vehicle

No

Figure-2.3 Selection of Routes and Vehicle with External Subsidy

41

Chapter 2

Step-4: Define a Feeder Route Several villages are located in the influence area of a bus stop. A feeder route will have one of these villages at one end, and the bus stop at the other end. Describe all possible feeder routes taking each of these village settlements as one end. For each feeder route repeat step-5 to Step-11 Step-5: Estimate Peak Period Demand At this stage, the headway of operation is not known. Therefore, assume that all trips from villages to bus stop are to be made by feeder service. Accordingly, estimate the daily demand for the busiest link (as per Section 2.3.1.2). Then, identify the peak period on the basis of the observed temporal variation, and estimate the peak period demand. Step-6: Calculate Number of vehicle Trips Using the demand and vehicle capacity, calculate the number of vehicle trips required to serve the peak period demand. Step-7: Calculate average time deviation Using the number of vehicle trips and the duration of peak period, calculate the service headway of vehicles and assume half the headway as average waiting time or time deviation for commuters. Step-8: Re-estimate the Demand Re-estimate the demand from each village to be served by feeder service and the resulting peak period demand for the busiest link. For the reestimation of demand using logit model, the average time deviation may be taken from Step-7, walking distance from each village may be taken according to the description of feeder route in Step-4, and fare as per Step1.

42

Approach and Methodology

Step-9: Check for Convergence If the difference between the re-estimated demand in Step-8 and the demand used in Step-6 and Step-7, is within an acceptable limit then go to Step-10, otherwise with the re-estimated demand repeat Step-6 to Step-9. Step-10: Calculate Number of Vehicles Calculate the round trip time required for a vehicle on the basis of route length, speed of vehicle and minimum layover time. Based on round trip time, calculate the number of vehicles required to maintain the service headway (Step-7). Step-11: Check the Operational Viability Calculate total daily revenue earned by the route. For this purpose, consider both directions of travel, and both peak & off-peak period demand. Assuming that this revenue will be distributed equally among all the vehicles, calculate the daily revenue earned per vehicle. A route is considered as viable, if the revenue eared per vehicle is at least equal to the cutoff revenue. If a route is viable then calculate the passenger served, passenger-km served, generalized cost, number of vehicles, etc. Step-12: Compare feeder routes from a bus stop Among all feeder routes investigated (Step-4) from a bus stop, identify only the viable feeder route(s) (Step-11). If there is more than one viable route, then select the route which is optimizing the MOE. The selected route is the best viable route from a bus stop considering a vehicle type and a fare level. Step-13: Generate fare combinations Repetition of Step-4 to Step-12 as per Step-3 will generate the best viable routes, if any, from all the bus stops considering a particular vehicle type and a fare level. Step-2 and Step-1 will ensure generation of such best viable routes for each vehicle type and fare level. For each vehicle type, take all the fare levels which indicate at least one viable feeder route with

43

Chapter 2

saving in GC for commuters in the study area. Considering two vehicle types and all such fare levels, generate possible fare combinations for further investigation. Each fare combination should include one fare level for Tempo and another fare level for Trekker. Step-14: Select routes for a fare combination For a fare combination, there is a choice to select either Trekker or Tempo with respective fare levels. The selected route must be operationally viable and ensure GC savings to commuters of the study area. Now, for each bus stop influence area, select the alternative which will optimize the MOE. The same vehicle type may not be selected for all routes. Also, some bus stop influence areas may not be served by feeder service. Finally, calculate the MOE considering all the selected feeder routes in the study area. Step-15: Recommend Fare Combination, Feeder Routes and Vehicle Among all fare combinations, the one indicating optimum value of the MOE considering all the selected feeder routes in study area is recommended. 2.3.3.2 Cross Subsidy

The methodology followed for route and vehicle selection under cross subsidy is shown in Figure-2.4 and the activities are discussed in subsequent sections. Step-0: Database Same as the database indicated in Section-2.3.3.1 Step-1: Select a Fare Level Different fare levels are investigated in the present work. For each fare level, repeat Step-2 to step-11. Step-2: Select a feeder vehicle Two types of feeder vehicle are considered in the present work. For each type of feeder vehicle repeat Step-3 to Step-11.

44

Approach and Methodology

Database Select a Fare Level Select a Feeder Vehicle Select a Bus Stop Define a Feeder Route Number of Vehicle Trips

Peak Period Demand

No

Average Time Deviation Re-estimate the Demand Yes

No

Converge

Calculate Number of Vehicles

Calculate Revenue, Passenger-km served, GC, etc

All Routes

Recommend Fare Combination, Feeder Routes and Vehicles

Yes

All Bus Stops

Yes

No Yes

Yes

All Vehicles All Fare Levels

Yes

No All Fare Combinati ons

No Generate Fare Combinations

Select a Fare Combination

Select Routes and Vehicle

No

Figure-2.4 Selection of Routes and Vehicle with Cross Subsidy

45

Chapter 2

Step-3: Select a bus stop A number of bus stops are located defining the boundary of the study area. Taking every bus stop and village settlements located in its influence area, a spanning tree is developed (Scetion-2.3.1.2). For each bus stop repeat Step-4 to Step-11 Step-4 to Step-10: Same as what is given in Scetion-2.3.3.1 Step-11: Calculate Revenue, Passenger-km served, GC, etc. Calculate the passenger served, revenue generated, passenger-km served, generalized cost, number of vehicles, etc. for each route. Assuming that the daily revenue earned will be distributed equally among all the vehicles, calculate the daily revenue earned per vehicle. A route is considered as viable, if the revenue eared per vehicle is at least equal to the cutoff revenue. The viability is checked at this level only for identifying fare levels suitable for developing fare combinations. Step-12: Generate fare combinations Repetition of Step-4 to Step-11 as per Step-3 will generate the all possible routes from all the bus stops for a vehicle type and a fare level. Step-2 and Step-1 will ensure generation of such routes for each vehicle type and fare level. For each vehicle type, take all the fare levels which indicate at least one viable feeder route with saving in GC for commuters in the study area. Considering two vehicle types and all such fare levels, generate possible fare combinations for further investigation. Each fare combination should include one fare level for Tempo and another fare level for Trekker. Step-13: Select routes for a Fare combination For each fare combination, there is a choice to select either Trekker or Tempo with respective fare levels. For all bus stop influence areas, initially select the routes and vehicle types keeping in mind only the MOE (i.e. not considering viability of selected routes). Now, calculate the average revenue

46

Approach and Methodology

earned per vehicle considering all vehicles on all selected feeder routes. It is likely that the average revenue eared per vehicle will not satisfy the requirement of the cutoff revenue. Therefore, it may be necessary to accept alternative set of feeder routes with lesser MOE in order to satisfy the criteria for cutoff revenue. This is done in an iterative manner. For each bus stop influence area, the next best route in terms of the MOE is considered for investigation, and the impact of this on the overall revenue earned per vehicle is analyzed. The bus stop influence area indicating the maximum increase in average revenue earned per vehicle for unit change in MOE is finally selected.

The process in repeated several times in an iterative

manner till the minimum cutoff revenue criterion is satisfied. In the process, route in a bus stop influence area may be selected repeatedly and finally some bus stop influence areas may not even be served by feeder service. Calculate the overall (i.e. considering all the finally selected routes) GC, passenger-km served, passenger served, vehicles, etc. Step-14: Recommend Fare Combination, Feeder Routes and Vehicles Among different fare combinations, the one indicating optimum value of the MOE considering all the selected feeder routes in the study area is recommended. 2.3.4 Forms of Operation of Feeder Vehicles

In the methodology presented in Section-2.3.3, a fixed-schedule form of operation of feeder vehicles is assumed while selecting feeder routes and vehicle. Also, an average time deviation equals to half the headway is taken for the estimation of demand for feeder service. With the approach presented in Section-2.3.3, it is not possible to compare different forms of operation. Therefore, an approach of simulating passenger movements along feeder routes is followed for comparing different forms of operation. The passenger arrival at stops is described with the help of a suitable distribution, and passenger movements along feeder routes are simulated under different forms of operation.

47

Chapter 2

The time deviations are distinctly different for different forms of operation. In fixed-schedule, the time deviation is represented by the difference between arrival time of vehicle at stop and arrival time of passenger at stop. In dial-a-ride, arrival time of the passenger at a stop along the route ensuring the capacity utilization of a feeder vehicle is taken as the starting time for the vehicle. Time deviation for all passengers traveling in that vehicle is calculated by taking the difference of time between arrival time of the vehicle at stop and arrival of passenger at stop. In dial-a-slot, starting time of a vehicle is selected in such a manner so as to minimize the total time deviation for all passengers traveling in a vehicle. The time deviation for individual passengers of the vehicle may be taken as the time difference between the intended slot and the actual slot of journey. 2.4 SUMMARY

This chapter deals with the approach and methodology followed for planning of rural feeder service to bus stop. The methodology focuses on two broad aspects of the work namely behavioural analysis and design of feeder service. The methodology followed for travel behavioural analysis includes type of data, preference elicitation technique, design of discrete choice experiment, modeling technique, estimation of rural users’ WTP, and development of GC model. A brief introduction of Multinomial Logit, Nested Logit and Random Parameter Logit model specifications is also included. The methodology followed for design of feeder service includes database, measure of effectiveness, alternative scenarios for operation, selection of routes and vehicles, and comparison of different forms of operation of feeder vehicles. ‘External Subsidy’ and ‘Cross Subsidy’ are considered as two policy instruments, and the approaches followed for selection of routes and vehicles with these policy instruments are discussed. The methodology followed for comparing different forms of operation of feeder vehicles is also included.

The methodologies described in this chapter are employed for

analyzing behavioral data in Chapter-3 and design of feeder service in Chapter-4.

48

Chapter 3 Travel Behaviour Analysis

3.1 INTRODUCTION This chapter presents collection and analysis of behavioural data for valuing of attributes of rural feeder service and prediction of mode choice behaviour. A detail description of design of experiment with pilot survey is presented in the Section 3.2. This includes attributes and their levels, preparation of choice sets and questionnaire. Section 3.3 describes the collection of data and development of database. Section 3.4 and 3.5, deals with analysis of stated choice data and development of utility equations using

Multinomial

Logit

(MNL),

Nested

Logit

(NL),

Covariance

Heterogeneity Nested Logit (CHNL) and Random Parameter Logit (RPL) model specifications. While developing RPL models, attempts are made to take into account the heterogeneity associated with the mean estimate of the random parameter(s). Utility equations developed in Section 3.4 and Section 3.5, are then used for valuing of attributes or estimation of willingness-to-pay with respect to rural feeder service and prediction of mode choice behaviour of rural commuters. Generalized cost models are discussed in the Section 3.6 followed by a summary in Section 3.7. 3.2 DESIGN OF EXPERIMENT Survey instrument was designed for the collection of stated choice (SC) data from rural trip makers by describing hypothetical feeder services to bus stop with suitable attributes and their levels. The choice instrument was subjected to considerable scrutiny before finalizing the attributes and their levels. A detailed discussion on the design of stated choice experiment is given below.

Chapter 3

3.2.1 Attributes and Their Levels In rural areas, the travel speed for feeder service is expected to be low due to frequent stoppage of vehicles at intermediate stops. Also, during reconnaissance survey it was observed that rural users did not consider travel time or travel speed a major decision variable. This is because the difference in journey times for short distance trips by different motorized modes is unlikely to be different significantly. The attributes and their levels for the design of stated choice (SC) experiment were decided after discussion with experts and trip makers. The attributes included fare, access walking distance, seating discomfort (within a vehicle), time deviation (i.e. time difference between intended and actual start of journey), and waiting discomfort. Attributes and their levels considered for SC experimentation are given in Table- 3.1. Table-3.1 Attributes and Their Levels Attribute

Levels

Fare per Km

Rs.1.00, Rs.1.50, Rs.2.00, Rs.2.50

Seating Discomfort

Comfortable Seating, Congested Seating

Access Walking Distance

0-0.5km, 0.5km-1km., 1km-1.5km, 1.5km-2km.

Time Deviation

0-15min, 15-30min, 30-45min, 45-60min

Waiting Discomfort

Anxious Waiting at Stop, Relaxed Waiting at Stop, Relaxed Waiting at Home

It may be mentioned that only small vehicles are considered as feeder modes, and traveling as standee is not a viable option for such vehicles. However, often in rural India such vehicles are found to carry more passengers

than

the

seat

capacity

specified

by

manufacturer.

Accommodating more passengers causes discomfort to passengers and the

50

Travel Behaviour Analysis

travel condition is described as ‘congested seating’ in stated choice experiment. When vehicles carry passenger only upto the seat capacity as specified by the manufacturer, the travel condition is described as ‘comfortable seating’. Three possible forms of operation of feeder vehicles namely ‘fixedschedule’, ‘dial-a-ride’ and ‘dial-a-slot’ are investigated. In all the three forms of operation, there may be a difference between the intended time and the actual time of starting a journey. This is described as ‘time deviation’ in the choice experiment. It may be mentioned that in ‘fixedschedule’, the arrival of the next vehicle is known to commuters waiting at a stop. However, the travel opportunity in that vehicle is not assured due to the limited seat capacity. As travel opportunity is not assured, the waiting is described as ‘anxious waiting at stop’. In ‘dial-a-ride’, the seat availability is assured in a specified vehicle only after arrival of passengers at a stop. As seat availability is assured in a specified vehicle, the waiting is described as ‘Relaxed Waiting at Stop’. In ‘dial-a-slot’, the seat availability is assured in specified vehicle and the schedule is known to trip makers before arrival at a stop. Such waiting is described as ‘Relaxed Waiting at Home’. 3.2.2 Choice Sets A full factorial design (Louviere et al. 2000) considering one attribute with two levels, another attribute with three levels, and other three attributes with four levels each, would produce 384 alternatives. However, it was neither necessary nor practically possible to include all these combinations in the SC experiment. Therefore, some alternatives were eliminated using fractional factorial technique (Green et al. 2001). Fractional factorial orthogonal design using SPSS 7.5 (as described in Hensher et al. 2005) was used to produce the alternatives, using all attributes and their corresponding levels, with an assumption that all interaction effects are negligible (Hensher et al. 2005; Street et al. 2005). These alternatives were used to prepare 10 choice sets, each containing 6 SC alternatives in

51

Chapter 3

an alternative specific form to represent three forms of operation (i.e. fixed-schedule, dial-a-ride and dial-a-slot) each with two alternative vehicles (i.e. Trekker and Tempo). Respondents were requested to state their choice considering 6 SC alternatives presented in a questionnaire. Subsequently, respondents were also requested to indicate their choice considering 7 alternatives including the mode used for the most recent trip to bus stop as another alternative. 3.2.3 Questionnaire and Pilot Survey A questionnaire consisting of four parts (Part A to Part D) was designed. Part A to Part C of the questionnaire was to develop the database required for SC analysis. Part D was to collect data for developing trip rates for different kinds of trip to bus stop. The development of database and analysis of data pertaining to Part A to Part C, are discussed in this Chapter. The analysis of data collected in Part D is discussed in Chapter 4. Part A was to collect information related to respondents’ most recent trip characteristics including information related to the mode used to access bus

stop.

Part

B

was

to

record

the

respondents’

socio-economic

information. Part C was to observe respondents’ choices from the various stated choice scenarios presented to them. Initially, 3 choice sets containing 7 alternatives were included in this part. Part D was to collect the trip diary of the previous week, covering details of (i) revenue generating trips by each earning member of the household, (i) educational trips and (iii) other trips for household purpose. During Feb-Mar, 2005 a paper-pencil (face-to-face) based pilot survey was carried out in phases at different locations in the study area. The objective of the pilot survey was to identify demerits or problems, if any, associated with the questionnaire design and the data collection process. An emphasis was given on checking respondents’ understanding of the questionnaire, way of presenting the questionnaire, number of stated choice observations to be included in a questionnaire considering the respondents’ fatigue,

52

Travel Behaviour Analysis

strategic locations for data collection to get a likely distribution of the population, etc. Pilot surveys also ensured adequate training to survey team members (enumerators). Pilot survey revealed difficulties of rural respondents to choose one out of seven alternatives given in a choice set. Therefore, respondents were requested to make choices in the following sequence. Step-1: Choose one from four alternatives representing fixed-schedule and dial-a-ride each with two alternative modes

Step-2: Choose one from four alternatives representing fixed-schedule and dial-a-slot each with two alternative modes Step-3: Choose one from four alternatives representing dial-a-ride and diala-slot each with two alternative modes

Step-4: Choose one from seven alternatives representing fixed-schedule, dial-a-ride, dial-a-slot each with two alternative modes and the mode used for the most recent trip to bus stop

Step-5: If a respondent choose the mode used for the most recent trip to bus stop, then he/she was asked to choose one from remaining 6 SC alternatives.

Respondents, who were initially unable to choose one from seven choices, were comfortable to give their choices when sequential approach was followed. Choices indicated by a respondent in sequential approach also helpful to check the consistency of choices made by the same respondent. However, in order to avoid fatigue (Carson et al. 1994), only one choice set containing seven alternatives with its sequences was included in the questionnaire.

53

Chapter 3

3.3 COLLECTION OF DATA AND DEVELOPMENT OF DATABASE The data collection included both primary as well as secondary sources. For the primary data, survey was conducted during April-May 2005, with the help of trained enumerators. In total, 998 households responded out of 1144 households contacted for survey. Other than SC choices, respondents were requested to

provide information pertaining to socioeconomic

characteristics and trips made in the last seven days by households. Trip diaries for revenue-generating trips were collected only from earning members. Revenue-generating trip details for 1691 earning members engaged in different occupations were collected. Non-revenue generating trip details were collected from 5353 persons representing 998 households. Road network, village locations, socioeconomic and demographic data, and bus stop locations were collected from secondary sources like census, district authority and regional transport authority. The village locations in the study area were obtained from topographical sheets of the Survey of India. Village level socioeconomic and demographic data were collected from census and block offices. Village level data included number of households and classification of households as per occupation of the head of household. Population, number of persons employed in each employment sector, private vehicle ownership and population in different age group were also obtained from secondary sources. The database prepared for analysis included primary data collected through SC surveys with respondent’s socio economic characteristics such as age, occupation, personal income, household size, household income, trip characteristics such as trip purpose, number of trips, time of the trip and SP choices. The secondary data such as village level data, bus route and road network characteristics were included in the database. Though 998 head of the household were interviewed, responses of some respondents

were

removed

due

to

incomplete

information

and

inconsistency in stated choice making. This resulted in only 674 refined

54

Travel Behaviour Analysis

responses for the purpose of model development. Salient features of the SC data used for model development are given in Table-3.2.

Table-3.2 Salient Features of Data Used for Model Development Total sample

674

Cultivator household

335

Agriculture/daily Labour household

121

Business & Service household

218

Household monthly Income ≤ Rs.4000

454

>Rs.4000

220

Trip Characteristics of Sample Maximum trip length

13 km

Minimum trip length

2.0 km

Average trip length

6.3km

Geographic Distribution of Sample >2.0 km and ≤5.0 km

37.7%

>5.0 km and ≤ 7.0 km

33.8%

>7.0 km and ≤ 9.0 km

19.1%

>9.0 km and ≤ 13.0 km

9.3%

Sample Share of Choice Trekker: Fixed-schedule

45.8%

Trekker: Dial-a-Ride

13.4%

Trekker: Dial-a-Slot

27.6%

Tempo: Fixed-schedule

6.2%

Tempo: Dial-a-Ride

1.5%

Tempo: Dial-a-Slot

5.5%

3.4 ANALYSIS OF DATA: STAGE-I Initially, an attempt was made to develop utility models using all the seven alternatives (i.e. six SC alternatives and the presently used mode) by LIMDEP 8.0. But, a model convergence could not be obtained. Therefore, the analysis is carried out in two stages. In stage-I, database including only six SC alternatives related to hypothetical feeder service is considered 55

Chapter 3

for the analysis. Stated preference data is rich and acceptable for valuation purpose. Therefore, analysis of database consisting of only SC alternatives is used for valuing of attributes of rural feeder service. In stage-II, the database consisting of seven alternatives is analyzed. In this stage, the number of variables describing the SC alternatives related to hypothetical feeder service is reduced judiciously by using the knowledge of Stage I analysis. This is used for valuation as well as mode choice prediction. In Stage-I analysis, the stated choice data is analyzed using three logit model specifications namely MNL, NL and RPL. Heteroskedasticity in NL model is also investigated with respect to socioeconomic characteristics of trip makers by developing CHNL model. In the wake of RPL model development process, RPL model with independent choice set (say, RPL1) and

RPL

model(s)

with

preference

heterogeneity

(say,

RPL2,3)

are

investigated. Preference heterogeneity study is carried out against all observed but relevant socioeconomic variables defining dummy variable, which is allowed to interact with random attribute(s). RPL models are estimated using constrained triangular distribution (Hensher and Greene 2001; 2003). 3.4.1 Organization of Data In the development of logit models, walking distance and cost are entered in cardinal-linear (i.e. continuous scale) form. The attribute, seating discomfort is effects-coded (-1 if congested seating, 1 if comfortable seating). The time deviation under three waiting discomfort levels are represented by three dummy variables namely anxious waiting at stop, relaxed waiting at stop, and relaxed waiting at home, multiplied by the time deviation. The three forms of operation are effects-coded through two variables called ‘system1’ (1 if ‘dial-a-slot’ 0 if ‘dial-a-ride’, -1 if ‘fixed scheduled’) and ‘sytem2’ (0 if ‘dial-a-slot’, 1 if ‘dial-a-ride’, -1 if ‘fixed scheduled’). Two vehicle types are also effects coded by a variable called ‘type’ (1 if ‘Tempo’, -1 if ‘Trekker’).

56

Travel Behaviour Analysis

3.4.2 Multinomial Logit Model Initially a MNL model (called as MNL1) is developed (Table-3.3). It is evident from Table-3.3, that the t-statistics (more than 1.96) of all estimates except for ‘system1’ and ‘sytem2’ are statistically significant with a confidence level more than 95%. The model MNL1 is re-estimated excluding ‘system1’ and ‘sytem2’. The re-estimated (called as MNL2) model is also shown in Table-3.3. The overall goodness-of-fit of estimated models is carried out based on pseudo R2 (ρ2) and t-stat (non-zero test) of coefficient estimates. A ρ2 value in the range of 0.2 to 0.4 is said to be good model fit (Louviere et al. 2000). The signs of the attribute estimates are as expected and in agreement with the actual condition of the study area. Here, negative signs of the quantitative attributes indicate that the utility for trip makers decreases with an increase in the magnitude of respective attributes. For the qualitative attribute ‘seating discomfort’, the positive

sign

‘comfortable

indicates seating’

that

changing

increases

the

from

utility.

‘congested Similarly,

seating’

negative

to sign

associated with the estimate of ‘type’ indicates trip makers’ preferences towards ‘Trekker’.

Table-3.3 Estimation Results of MNL Models (Stage-I) Attributes

MNL1

System1

-0.124 (0.558)

Sytem2

0.157 (0.526)

Type Seating Discomfort

MNL2

-0.608 (8.919)

-0.601 (9.264)

1.08 (9.321)

1.083 (9.452)

Access Walking Distance

-0.00037 (2.67)

-0.00039 (2.868)

Anxious Waiting Time at Stop

-0.093 (9.636)

-0.093 (11.091)

Relaxed Waiting Time at Stop

-0.07 (5.642)

-0.065 (9.051)

Relaxed Waiting Time at Home

-0.051 (5.448)

-0.054 (8.491)

Cost

-0.049 (13.803)

-0.049 (14.015)

Log likelihood function

-747.638

-747.829

0.20479

0.20459

2

ρ

Note: t-statistics are shown in parenthesis.

57

Chapter 3

3.4.3 Nested Logit Model An attempt is made to analyze the SC data by developing nested logit (NL) models. Hensher (1998) suggested the use of Heteroscedastic Extreme Value (HEV) model as a search engine for identifying the most likely tree structure. An attempt was made to develop a HEV model (Bhat 1995; Hensher 1998), but the model convergence could not be obtained. As the convergence problem of HEV models has also been reported earlier and is found to be quite common, Henser et al. (2005) suggested the use of inclusive value parameters obtained from a fully degenerated tree structure NL model as guidance for identifying the likely tree structure. Accordingly, a degenerated NL (DGNL) model is developed as shown in Table-3.4.

Table-3.4 Estimation Results of DGNL Model Attributes Type

-2.069

Seating Discomfort Access Walking Distance Anxious Waiting Time at Stop Relaxed Waiting Time at Stop Relaxed Waiting Time at Home Cost

1.074 -0.0003 -0.091 -0.059 -0.049 -0.047

Inclusive Value Parameters Fixed schedule: Tempo

0.841

Fixed schedule: Trekker

1.096

Dial-a-ride: Tempo

0.836

Dial-a-ride: Trekker

1.120

Dial-a-slot: Tempo

0.856

Dial-a-slot: Trekker

1.099

The inclusive value parameters of ‘Tempo’ for all forms of operation of feeder vehicles are similar. Inclusive value parameters of ‘Trekker’ for all forms of operation of feeder vehicles are also similar. Based on these observations, a vehicle based tree structure (i.e. Figure-3.2 or Figure-3.4) is considered suitable for the development of NL model. However, further

58

Travel Behaviour Analysis

investigations are carried out on the selection of tree structure for NL model.

Greene (2000) argued that often there is a natural partition of alternatives which can guide the tree structure for NL model. In the present work, four possible tree structures guided by natural partition of alternatives (i.e. vehicle based and form of operation based) are obtained as shown in the Figure-3.1 to Figure-3.4. The NL models (called as NL1, NL2, NL3 and NL4) representing these tree structures are estimated and the estimation results are shown in Table-3.5. While estimating these models, all the attributes of MNL2 are included in the utility of the lower level. The signs of the attribute estimates may be interpreted in the similar manner as mentioned in Section 3.4.2. A comparison of model estimates shown in Table-3.5 indicates that for the two-level vehicle based tree structure (i.e. model NL4), the inclusive value parameters of branch ‘Tempo’ and ‘Trekker’ are distinctly different and less than unity. Therefore, model NL4 is accepted from random utility maximization point of view. It may be mentioned that analysis of DGNL model as per Henser et al. (2005) also indicated a vehicle based tree structure for NL model. Therefore, both the approaches i.e. Henser et al. (2005) and Greene (2000) indicated the same tree structure for NL model.

Travel Demand Responsive

Fixed Scheduled Tempo

Trekker

Dial-a-Ride

Tempo

Dial-a-Slot

Trekker Tempo

Trekker

Figure-3.1 Three Level Form of Operation Based Tree Structure

59

Chapter 3

Travel Tempo Fixed Scheduled

Trekker

Demand Responsive

Dial-a-Ride

Fixed Scheduled

Dial-a-Slot

Demand Responsive Dial-a-Slot

Dial-a-Ride

Figure-3.2 Three Level Vehicle Based Tree Structure

TRAVEL Fixed Schedule

Tempo

Trekker

Dial-a-Slot

Dial-a-Ride

Trekker

Tempo

Tempo

Trekker

Figure-3.3 Two Level Form of Operation Based Tree Structure

Travel Tempo

Trekker

Fixed Scheduled Dial-a-Ride Dial-a-Slot

Fixed Scheduled Dial-a-Ride Dial-a-Slot

Figure-3.4 Two Level Vehicle Based Tree Structure

Table-3.5 Estimation Results of NL Models Attributes Type Seating Discomfort Access Walking Distance Anxious Waiting Time at Stop Relaxed Waiting Time at Stop Relaxed Waiting Time at Home

NL1

NL2

NL3

-0.619 (9.85) 0.714 (6.081) -0.0001 (1.027) -0.0564 (4.938) -0.0557 (5.465) -0.0229 (3.478)

-0.087 (0.035) 1.209 (4.407) -0.0002 (0.781) -0.084 (3.074) -0.066 (4.244) -0.042 (3.576)

-0.69 (11.403) 0.66 (6.135) 0.00001 (0.147) -0.046 (4.565) -0.046 (5.089) -0.018 (2.959)

60

NL4

NL5

0.101 (0.063) 1.122 1.119 (9.008) (9.399) -0.00043 -0.00043 (2.908) (3.009) -0.097 -0.096 (9.905) (10.106) -0.068 -0.068 (8.471) (8.615) -0.056 -0.056 (7.452) (7.46)

Travel Behaviour Analysis

Cost Lib Level IV Parameters Fixed- scheduled*/Tempo** Demand Responsive */Trekker** Branch Level IV Parameters Fixed Ride Slot

-0.0295 (6.549)

-0.039 (3.953)

0.638 (0.000)* 0.515 (3.123)*

4.28 (1.022)** 0.257 (1.352)**

2.713 (0.000) 2.933 (3.442) 3.386 (3.64)

Fixed Tempo Fixed Trekker Demand Responsive Tempo*/Tempo** Demand Responsive Trekker*/Trekker** Log likelihood function ρ2

-733.69 0.22

-0.024 (5.176)

-0.050 (13.026)

-0.05 (13.325)

2.661 (4.19) 2.322 (5.049) 2.65 (4.506) 0.253 (1.166) 2.133 (1.757) 0.221 (0.966)* 2.112 (1.909)* -738.25 0.215

-737.5 0.216

0.867 (4.577)** 0.735 (2.598)** -747.19 0.205

0.865 (4.599) ** 0.748 (3.917) ** -747.42 0.205

Note: t-statistics are shown in parenthesis. * ,** name and value of corresponding IV parameters,

The vehicle specific variable ‘type’ is not significant in NL4. Therefore, the model is re-estimated excluding ‘type’, and the re-estimated model (called as NL5) is also shown in Table-3.5. The NL5 model is accepted considering significance and signs of parameter estimates. Estimates of NL5 are also in conformity with MNL2 and random utility maximization.

3.4.4 Covariance Heterogeneity Nested Logit Heteroskedasticity

in

NL

model

is

investigated

with

respect

to

socioeconomic and trip characteristics of trip makers. However, only household income (defined as 1 for household income ≤ Rs.4000 per month, and 0 for household income >Rs.4000 per month) is found to have statistically significant effect. The parameter estimates of CHNL model are shown in Table-3.6 and the signs of the attribute estimates may be interpreted in the similar manner as mentioned in Section 3.4.2. The estimates of CHNL model are significant with acceptable ρ2 value. It may be noted that the ρ2 value increased from MNL model to NL model, and again from NL model to CHNL model.

61

Chapter 3

Table-3.6 Estimation Results of CHNL Model Attributes

CHNL

Seating Discomfort

1.12 (8.971)

Access Walking Distance

-0.00043 (3.086)

Anxious Waiting Time at Stop

-0.096 (9.366)

Relaxed Waiting Time at Stop

-0.067 (8.45)

Relaxed Waiting Time at Home

-0.055 (6.937)

Cost

-0.051 (11.481)

IV parameters Tempo

0.705 (4.608)

Trekker

0.625 (4.175)

Heterogeneity Parameters Household Monthly Income

0.495 (3.231)

Log likelihood function

-741.805

2

ρ

0.211

Note: t-statistics are shown in parenthesis

3.4.5 Random Parameter Logit Model The SC data is also analyzed using RPL model specification. Several RPL models are developed assuming all the attributes other than the cost as random parameters. Ruud (1996) pointed out that RPL models have a tendency to be unstable when all attributes are allowed to vary. Fixing the cost parameter resolves this instability. If the cost parameter is allowed to vary, the distribution of other attribute value is the ratio of two distributions,

a

Cauchy

distribution,

which

has

no

finite

moments

(Brownstone 2000). With a fixed cost parameter, the distribution of other attribute value is same as the parameter of the attributes (Hensher 2001a). Therefore, in the present work all attributes other than travel cost are considered random. All the random parameters in RPL models are assumed to follow the constrained triangular distribution. RPL models are estimated with simulated maximum likelihood technique, using intelligent (Halton) draws with 500 replications. Several researchers (Amador et al. 2005; Phanikumar and

Maitra 2007; Hensher et al. 2007) have captured the

effect of socioeconomic/trip characteristics on coefficient estimates of RPL models and WTP values. Therefore, the decomposition effect of all the

62

Travel Behaviour Analysis

relevant socioeconomic variables on the mean estimates of random parameters is investigated. However, only the ‘household monthly income’ and the ‘present mode used to access bus stop’ are found to have decomposition effects on the mean estimates of random parameters. Accordingly, three RPL models are developed as follows: a) RPL1: without considering mean heterogeneity, b) RPL2: considering mean heterogeneity with respect to ‘household monthly income’ c) RPL3: considering mean heterogeneity with respect to ‘present mode used to access bus stop’ In the development of RPL3, the ‘present mode used to access bus stop’ is described as ‘motorcycle’ and ‘others’ (predominantly bicycle).

Table-3.7 shows the estimation results for RPL models. The signs of the attribute estimates may be interpreted in the similar manner as mentioned in Section 3.4.2. It may be observed that the attributes of all three models (i.e. RPL1, RPL2 and RPL3) are statistically significant and signs of the estimates are as expected. The ρ2 values indicate that these models are good in fit. In RPL2 model, the household monthly income (defined as 1 for household monthly income ≤ Rs.4000 and 0 otherwise) is found to have statistically significant decomposition effect on the mean estimates of ‘type’, ‘seating discomfort’, ‘access walking distance’ and ‘anxious waiting time at stop’. In RPL3 model, the present access mode (defined as 1 for user of motorcycle and 0 otherwise) is found to have statistically significant decomposition effect on the mean estimates of ‘seating discomfort’, ‘access walking distance’ and ‘anxious waiting time at stop’, ‘relaxed waiting time at stop’ and ‘relaxed waiting time at home’. An increase in ρ2 is observed from CHNL to RPL1 model, RPL1 to RPL2, and again RPL2 to RPL3.

63

Chapter 3

Table-3.7 Estimation Results of RPL Models (Stage-I) Attributes

RPL1

RPL2

RPL3

Random parameters (Constrained T-Dist.) Type

-0.711 (7.880) -0.5 (3.594)

Seating Discomfort

1.649 (5.708)

2.286 (5.423)

-0.784 (8.447) 1.256 (5.166)

Access Walking Distance

-0.001(3.241)

-0.002 (4.245) -0.00088 (2.949)

Anxious Waiting Time at Stop

-0.147 (5.95)

-0.201 (5.427) -0.125 (5.602)

Relaxed Waiting Time at Stop

-0.12 (5.129)

-0.145 (4.989) -0.105 (4.546)

Relaxed Waiting Time at Home

-0.097 (5.079)

-0.118 (4.897) -0.084 (4.456)

Non-random parameters Cost

-0.075 (6.871)

-0.092 (6.239) -0.076 (7.435)

Spread of random parameter Type

0.711 (7.88)

0.5 (3.594)

0.784 (8.447)

Seating Discomfort

1.649 (5.708)

2.286 (5.423)

1.256 (5.166)

Access Walking Distance

0.001(3.241)

0.002 (4.245)

0.00088 (2.949)

Anxious Waiting Time at Stop

0.147 (5.95)

0.201 (5.427)

0.125 (5.602)

Relaxed Waiting Time at Stop

0.12 (5.129)

0.145 (4.989)

0.105 (4.546)

Relaxed Waiting Time at Home

0.097 (5.079)

0.118 (4.897)

0.084 (4.456)

Heterogeneity around the Mean of Random Parameters HIN Type

-0.39 (2.443)

Seating Discomfort

-0.431 (2.079)

MC

1.483 (5.116)

Access Walking Distance

0.0013 (3.54) -0.00076 (2.585)

Anxious Waiting Time at Stop

0.031(2.01)

-0.081 (5.158)

Relaxed Waiting Time at Stop

-0.044 (2.585)

Relaxed Waiting Time at Home

-0.028 (1.769)

Log likelihood function ρ2

-745.149

-724.642

0.20744

0.22925

-698.52 0.25704

Note: t-statistics are shown in parenthesis. HIN = Household Monthly Income, MC = Motorcycle User

3.4.6 Estimation of WTP Value Valuing of attributes of feeder service from MNL, NL and CHNL model estimates is straight forward as it is given by the ratio of partial derivatives 64

Travel Behaviour Analysis

of the utility function with respect to attributes and travel cost. In the present work, attribute values for RPL models are also calculated based on unconditional parameter estimate (i.e. simulating a population moment) using the spread of triangular distribution. As in RPL models, all the attributes except the cost is considered as random and assumed to follow constrained T-distribution, the marginal values of the attributes presented in Table-3.8 are calculated as given below. Value of Seating Discomfort = (1.649 + 1.649 × Tc ) /− 0.075 Value of Access Walking Distance = (−0.001 + 0.001 × Tc ) /− 0.075 Value of Anxious Waiting Time at Stop = (−0.147 + 0.147 × Tc ) /− 0.075 Value of Relaxed Waiting Time at Stop = (−0.12 + 0.12 × Tc ) /− 0.075 Value of Relaxed Waiting Time at Home = (−0.097 + 0.097 × Tc ) /− 0.075 Table 3.8 WTP Estimates from RPL1 Model (Stage-I) Attribute

Willingness-To-Pay Estimates+

Mean !

Seating Discomfort

Access Walking Distance Anxious Waiting Time at Stop Relaxed Waiting Time at Stop Relaxed Waiting Time at Home

Std. Deviation !

43.76

++

82.98 12.2

Unit

++ ++

09.96

++

08.05

18.17!

Paise/Km

++

Paise/Km

++

Paise/min

++

Paise/min

34.16 05.02

04.10

++

03.31

Paise/min

+ Based on 20000 Random Draws ! WTP value is for change in level from ‘congested seating’ to ‘comfortable seating’ ++The WTP estimates are for average trip length

An attempt is also made to estimate the mean values of attributes or WTP from the same RPL model using simple division rule, and the estimated values are presented in Table-3.9. A comparison of values reported in Table-3.8 and Table-3.9 shows that the mean values of attributes calculated from RPL models’ estimate by simple division rule almost approximate the mean values calculated by unconditional parameter estimate basis. This happens due to the assumption of the constrained Tdistribution for all random attributes, and travel cost as non-random, which 65

Chapter 3

makes the distribution of value of the attributes also another constrained triangular

distribution.

Based

on

this

observation,

henceforth

the

unconditional mean value estimates of attributes in all RPL models are calculated by simple division rule.

WTP values obtained from MNL2, NL5, CHNL, RPL1, RPL2 and RPL3 models, are reported in Table-3.9. It may be observed that relaxed waiting at stop in

comparison

improvement.

with

anxious

Relaxed

waiting

waiting

at

at

home

stop is

is

considered

considered

as

as

an

further

improvement in comparison to relaxed waiting at bus stop. In RPL2 model, commuters from high income households are found to have higher WTP for (i) reduction in access walking distance, (ii) reduction in anxious waiting time at stop, and (iii) comfortable seating. In RPL3 model, commuters using motorcycle are found to have higher WTP for (i) reduction in access walking distance, (ii) reduction in waiting times, and (iii) comfortable seating. It may be mentioned that motorcycle users are predominantly from high income households and therefore, heterogeneity study carried out with either ‘household monthly income’ or ‘present access mode’ may be accepted. But, when ‘present access mode’ is considered, heterogeneity is observed on all the three types of waiting time, and the ρ2 is also improved indicating a better goodness of fit (Table-3.7). Table-3.9 Estimated WTP Values from Stage-I Models Attribute

Willingness-To-Pay Estimates MNL2

Seating Discomfort

!

Access Walking Distance Anxious Waiting Time at Stop Relaxed Waiting Time at Stop Relaxed Waiting Time at Home

44.3

NL5 44.5

49.2+

53.1+

11.8+

12.0+

8.2+

8.4+

6.8+

6.9+

CHNL

RPL1

RPL2

RPL3

49.4 / 71.8#/ 40.1** 32.9## 52.9+ 83.2+143.8*+/ 134.9#+/ 56.1**+ 72.3##+ + + 11.8 12.1 13.7*+/ 17.0#+/ 11.6**+ 10.3##+ + + 8.3 9.9 9.9+ 12.3#+/ 8.6##+ + + + 6.8 8.0 8.1 9.3#+ / 7.0##+ 44.4

43.7

*

Unit Paise/Km Paise/Km Paise/min Paise/min Paise/min

! WTP value is for change in level from ‘congested seating’ to ‘comfortable seating’ +The WTP estimates are for average trip length , *For high income household , ** For low income household, # For motorcycle user, ## For bicycle user (1$US = 45INR exchange rate on 2005 , 1 INR = 100 paise)

66

Travel Behaviour Analysis

It may be observed from Table-3.9 that WTP values obtained from NL model are more than those from MNL model. Similar observation was made by Bhat (1997) in intercity mode choice behavior for weekday business travel in the Toronto-Montreal corridor of Canada. However Bhat (1998a) and Chattopadhyay (2000) observed lower value in NL model. No significant difference is observed in WTP estimates obtained from CHNL models. WTP values for reduction in waiting times as obtained from the RPL1 model are higher than those obtained from MNL, NL and CHNL models. Higher WTP value is also obtained from RPL1 model for reduction in access walking distance. The WTP value for comfortable seating as obtained from RPL1 model is similar to the value obtained from MNL model. Higher WTP values from RPL models were reported by Bhat (1998b) in mode choice modeling, Train (1998) in the context of recreational demand, Carlsson (1999) while estimating value of travel time for business class, Revelt and Train (1998) in household appliance study, Haefen (2003) in the context of welfare estimate of clean up eutrophic sites, Hensher (2001a; 2001b; 2001c) and Amador et al. (2005) in estimating value of travel time saving and Phanikumar and Maitra (2007) in valuing attributes of rural bus service. There may be some unobserved effects (Hensher 2001b), which trip makers consider and correlate with travel attributes other than cost attribute. RPL models capture the unobserved effects and such effects are generally more correlated with attributes other than the cost attribute. Therefore, higher order values may be obtained from RPL model specification. Among all attributes, the change in WTP value across different model specification is found more for the access walking distance. The unobserved effects captured in RPL models may not be correlated uniformly with all the noncost attributes. In the present case study, such unobserved effects are more correlated with access walking distance and accordingly, the maximum change in WTP estimate is obtained with respect to access walking distance. It may be mentioned that higher order WTP estimate from RPL model is not a consistent observation obtained in all WTP studies. Underestimation of values of attributes has also been reported in the context of value of travel

67

Chapter 3

time (VOT) (Algers et al. 1998). Alpizar and Carlsson (2001) found that the VOT could be underestimated or overestimated on the chosen mode. The international experience in gain or loss in measurement of the value of an attribute using RPL is an interesting part to note (Amador et al. 2005), but there is not a general conclusion, because the value depends on the nature of the data and specifications used in each study. Altogether, variations are observed in WTP estimates obtained from different model specifications. Therefore, it is important to appreciate the role of model specification while calculating WTP values. It is difficult to judge which model is the best in terms of WTP values. Validation of WTP values is again a difficult task to achieve in reality. Estimation techniques, which impose less structure on the error term (as in RPL model) hold more appeal from an analytical point of view. In order to be more consistent with statistical

models

of

human

behavior,

the

RPL

model

is

created.

Heterogeneities of sample are also captured within the framework of RPL model. Moreover, directly modeling unobserved preference heterogeneity in RPL results in significant improvement in the precision of the parameter estimates

and

WTP

values

(Layton

2000).

Therefore,

RPL

model

specification may be preferred over MNL or NL model specifications.

3.4.7 Comparison of Utility Equations Sections 3.4.2 to 3.4.5 describe model development for valuing of attribute(s) of rural feeder service. A comparative analysis between logit models (say, parameter richer Model2 vs simpler/restricted Model1; Model2Model1) is done using log-likelihood ratio test to identify the model that statistically represents the data better. The null hypothesis (Ho) for this test

may

be

defined

as

‘The

restricted/simpler

Model1

statistically

represents the data better than the parameter richer Model2’. It may be mentioned that a direct test between RPL model with independent choice set (say, RPL1) and MNL model (i.e. RPL1-MNL) is not possible given the use of constrained T-distribution in the estimation of random parameter(s) in RPL1 model. This is because the spread of random parameter distribution 68

Travel Behaviour Analysis

is constrained to equal the mean of the distribution and so no additional parameters are estimated in the RPL1 model (Greene et al. 2006). As such any statistical test will have zero degrees of freedom. On the other hand, RPL models with heterogeneity study (say RPL2,3)

estimate more free

parameter(s) than MNL/RPL1/NL/CHNL model. Hence a log-likelihood test between RPL2,3, and other models (say MNL, RPL1, NL, CHNL) are possible. In the present work, the null hypothesis is tested (against critical χ 2 Chisquare value at 99% confidence level) for ‘N’ (extra parameters estimates for Model2) degrees of freedom. The additional interaction term introduced to uncover mean heterogeneity in RPL3, however, allow for tests of statistical significance between these model, the MNL, NL, CHNL, RPL1 and RPL2 models in this work. It is also observed (Tables-3.10) that the parameter richer models (i.e. at heterogeneity level study) represent the stated choice data statistically better than corresponding simpler/restricted models. So the null hypothesis is rejected, which implies the existence of preference heterogeneity in the perception of choice of the sampled population.

Table-3.10 clearly shows the parameter richer RPL3 model

statistically represents the data better than any other model.

Table-3.10 Log-likelihood Test of Logit Models (Stage-I) Test

2*(Lnl_Model2 Critical χ 2 Value# - Lnl_Model1)

Degrees of freedom

Null Hypothesis

RPL3 - MNL

98.6

15.086

5

Rejected

RPL3 - NL

97.8

13.277

4

Rejected

RPL3 - CHNL

86.6

11.341

3

Rejected

RPL3 – RPL1

93.2

15.086

5

Rejected

RPL3 – RPL2

52.2

6.635

1

Rejected

# Value at 99 percent confidence level

It may also be observed that RPL3 model produced the highest ρ2 among the MNL, NL, CHNL and RPL models developed in the present work. Hence, WTP values obtained from RPL3 model are finally accepted for further analysis.

69

Chapter 3

3.5 ANALYSIS OF DATA: STAGE-II The analysis of database with seven alternatives is discussed in this section. It may be mentioned that majority of the attributes (say, access walking distance, waiting discomfort and time deviation) describing six SC alternatives related to hypothetical feeder service, are not directly relevant to describing the present travel by bicycle or motorcycle. Therefore, in Stage-II analysis, the number of variables describing the SC alternatives is reduced judiciously using the result of Stage-I analysis. The following attributes, which are not relevant to the present vehicle used are represented by a single explanatory variable (called as individual specific accessibility) using the knowledge of WTP values obtained form Stage-I analysis. •

Access walking distance

•

Time deviation under different levels of waiting discomfort

It may be mentioned that representing several variables in terms of a single explanatory variable, as attempted in the present work, has been used in many studies related to four stage transportation planning and activity based planning. Carrier (2003) developed a GC model and used the GC for the next stage of work i.e. revenue calculation for airlines using passenger origin-destination simulator. Yao and Morikawa (2003) included the expected maximum utility derived from mode/route choice stage as an explanatory variable in the utility function of destination choice. Kato et al. (2005) also used GC of earlier analysis in the origin destination simulation of urban rail project.

Bicycles and motorcycle are the two modes presently used for accessing the bus stop which are located beyond 2.0 km from the village settlements. The Stage-I analysis (especially RPL3) indicated that these two are different user groups and their WTP are different. Also, the RPL3 model alongwith the estimated WTP values is finally accepted for subsequent analysis. Therefore, in Stage-II analysis, two separate RPL models are developed for representing choice behaviour of bicycle and motorcycle users. 70

Travel Behaviour Analysis

It is mentioned in Table-3.2 that responses obtained from 674 respondents are used for SC analysis. 500 out of those 674 respondents indicated the use of bicycle as the present access mode. Therefore, responses obtained from those 500 respondents are analyzed in Section 3.5.1. The remaining 174 respondents indicated the use of motorcycle as the present access mode, and the responses obtained from those are analyzed in Section 3.5.2.

3.5.1 Model for Bicycle Users The data is analyzed by developing MNL and RPL model specifications. In the development of models, six alternative specific constants are taken considering ‘bicycle’ as the base mode. The distance traveled to bus stop under different discomfort levels is described and entered in model with the help of three dummy variables called DL1c, DL2c, and DL3c. •

DL1c: Distance traveled under comfortable seating in a vehicle

•

DL2c: Distance traveled under congested seating in a vehicle

•

DL3c: Distance traveled using bicycle

Individual Specific Accessibility (ISA) and direct cost of travel are the other two variables entered in the models in cardinal linear form. The MNL (called as MNLc) model estimates obtained for bicycle users are reported in Table-3.11. Subsequently, a RPL (called as RPLc) model is also developed and reported in Table-3.11. The models show that all attributes considered are significant. Also, ρ2 values indicate that these models are good in fit. The result indicates the highest disutility for DL3c i.e. distance traveled by bicycle.

A decrease in disutility is indicated for ‘traveling in

bicycle’ to ‘traveling in congested seating in a vehicle’. A further decrease in disutility is indicated for ‘traveling in congested seating in a vehicle’ to ‘traveling in comfortable seating in a vehicle’. The negative sign associated with ‘ISA’ shows increase of disutility with increase in value of ‘ISA’. The RPL model is accepted for further analysis from an analytical and estimation technique point of view.

71

Chapter 3

Table-3.11 Estimation Results of Models for Bicycle Users Attributes

MNLc

RPLC Random parameters (Constrained T-Dist.)

DL1c DL2c

1.025 (4.426)

0.9 (6.635)

0.605 (2.928)

0.594 (4.539)

DL3c

-0.524 (3.141)

-0.551 (3.141) Non-random parameters

ISA

-0.0066 (7.073)

-0.0066 (6.281)

-0.009 (11.539)

-0.0094 (10.285)

2.03 (3.24)

2.411 (3.533)

3.351 (5.236)

4.206 (6.348)

2.421 (3.4)

2.948 (3.735)

3.495 (5.5)

4.052 (5.75)

1.734 (2.554)

2.549 (3.633)

3.76 (6.608)

4.272 (7.073) Spread of random parameter

Direct Cost of Travel ASC Fixed-Schedule Tempo ASC Fixed-Schedule Trekker ASC Dial-a-Ride Tempo ASC Dial-a-Ride Trekker ASC Dial-a-Slot Tempo ASC Dial-a-Slot Trekker DL1c

0.9 (6.635)

DL2c

0.594 (4.539)

DL3c Log likelihood function ρ2

0.551 (3.141) -567.376

-571.568

0.22597

0.22025

Note: t-statistics are shown in parenthesis.

3.5.2 Model for Motorcycle Users In the development of models for motorcycle users, six alternative specific constants are taken considering ‘motorcycle’ as the base mode. The distance traveled to bus stop under different discomfort levels is described and entered in model with the help of three dummy variables called DL1m, DL2m, and DL3m. The descriptions for DL1m and DL2m are similar to DL1c and DL2c respectively. DL3m represents the distance traveled using motorcycle. Individual Specific Accessibility (ISA) and direct cost of travel are the other two variables included in the models in cardinal linear form. The MNL (called as MNLm) model developed using the above attributes is reported in Table-3.12. Subsequently, a RPL (called as RPLm) model is also developed and reported in Table-3.12. The models show that all attributes

72

Travel Behaviour Analysis

considered are significant in both the models. The ρ2 values also indicate acceptable goodness of fit. The estimated result shows ‘traveling in motorcycle’ is considered utility in comparison to ‘traveling in congested seating in a vehicle’. The utility value for ‘traveling in motorcycle’ is observed to be almost similar to that of ‘traveling in comfortable seating in a feeder vehicle’. The negative sign associated with ‘ISA’ indicates that disutility increases with an increase in the value of ‘ISA’. The RPL model is found acceptable and therefore, selected for further analysis.

Table-3.12 Estimation Results of Models for Motorcycle Users Attributes

MNLm

RPLm Random parameters (Constrained T-Dist.)

DL1m DL2m

-0.607 (1.729*)

-0.463 (2.217)

-0.85 (2.769)

-0.787 (4.129)

DL3m

-0.607 (2.115)

-0.468 (2.91) Non-random parameters

ISA

-0.0044 (4.826)

-0.005 (4.939)

-0.0036 (3.513)

-0.0048 (5.479)

2.0811 (2.181)

3.086 (2.962)

3.754 (3.733)

4.827 (4.375)

3.059 (3.4)

4.17 (4.108)

Direct Cost of Travel ASC Fixed-Schedule Tempo ASC Fixed-Schedule Trekker ASC Dial-a-Ride Tempo ASC Dial-a-Ride Trekker ASC Dial-a-Slot Tempo ASC Dial-a-Slot Trekker

3.428 (3.772)

4.454 (4.382)

*

2.749 (2.775)

1.589 (1.708 ) 3.315 (3.916)

4.412 (4.894) Spread of random parameter

DL1m

0.463 (2.217)

DL2m

0.787 (4.129)

DL3m

0.468 (2.91)

Log likelihood function ρ2

-239.726 0.20174

-237.92 0.20776

Note: t-statistics are shown in parenthesis. * Less than 90% confidence interval

3.5.3 Estimation of WTP Values The WTP values obtained in Stage-II analysis form RPLc and RPLm models are reported in the Table-3.13. It may be observed from Table-3.13 that 73

Chapter 3

WTP values for ‘congested seating’ to ‘comfortable seating’ are 32.7 and 67.7 paise per km for bicycle and motorcycle users respectively. A comparison of these WTP values with the WTP values obtained in Stage-I analysis (refer Table-3.9) indicates that WTP values estimated in Stage-I and Stage-II analysis are consistent and comparable. Table-3.13 Estimated WTP Values from Stage-II Models Attributes

Change of travel from bicycle/ motorcycle to comfortable seating in a vehicle Change of travel from congested seating to comfortable seating in a vehicle

Willingness-To-Pay (WTP) Estimates

Unit for WTP

MNLc 172.1

MNLm 0

RPLc 155.0

RPLm 1

Paise/Km

46.7

67.5

32.7

67.7

Paise/Km

3.6 GENERALIZED COST OF TRAVEL Based on the WTP values estimated in Stage-I and Stage-II analysis, generalized cost equations are developed for bicycle and motorcycle users. The generalized cost equations in paise are given below.

GCc= αc *D + 0.702 (72.3*AWD +10.3*AWS +8.6*RWS +7.0*RWH ) + Cost

(6)

GCm= αm*D + (134.9*AWD +17.0*AWS +12.3*RWS +9.3*RWH ) + Cost

(7)

D = Distance traveled, in Km αc = 0, if travel is under comfortable seating in a vehicle 32.7, if travel is under congested seating in a vehicle, and 155.0, if travel is by bicycle αm= 0 if travel is under comfortable seating in a vehicle 67.7 if travel is under congested seating in a vehicle, and 1 if travel is by motorcycle AWD = Access walking distance, in km AWS = Anxious waiting at stop, in min

74

Travel Behaviour Analysis

RWS = Relaxed waiting at stop, in min RWH = Relaxed waiting at home, in min Cost = Total cost of travel, in paise

3.7 SUMMARY This chapter

deals

with

the

design

of questionnaire,

collection

of

behavioural data, development of database, analysis of travel behavioral data and estimation of WTP values. The behavioural data collected from rural users are analyzed by developing utility models in two stages. In Stage-I, only the SC alternatives related to hypothetical feeder service are analyzed by developing MNL, NL, CHNL, and RPL model specifications. While developing RPL models, preference heterogeneity associated with the mean estimates of random parameters is investigated considering all relevant socioeconomic variables and trip characteristics. A comparison of WTP values obtained from different model specifications is also included in Stage-I analysis. WTP values calculated in Stage-I, are used judiciously to reduce the number of variables in Stage-II analysis. In Stage-II, seven alternatives (i.e. six alternatives representing hypothetical feeder service, and one representing the present access mode) are considered for analysis. Two separate models representing bicycle and motorcycle users are developed in Stage-II. Separate generalized cost models are also presented for bicycle and motorcycle users.

Model developed and WTP

value calculated in this chapter are used in Chapter-4 for the design of rural feeder service to bus stop.

75

Chapter 4 Design of Feeder Service

4.1 INTRODUCTION The selection of routes and type of vehicle for rural feeder service to bus stop is discussed in this chapter. Input database developed for this purpose primarily includes travel demand to bus stop, road network and cutoff revenue. Development of database is discussed in Section 4.2. In the process of designing feeder service, it is necessary to select an appropriate measure of effectiveness and consider relevant policy issues. These aspects are elaborated in Section 4.3. In the present work, planning of rural feeder service is demonstrated considering two measures of effectiveness. The selection of feeder routes and type of vehicle with these measures of effectiveness are discussed in Section 4.4 and Section 4.5. A comparison of different forms of operation of feeder vehicles is reported in Section 4.6, which is followed by the summary in Section 4.7. 4.2 DATABASE The total travel demand from different villages to the nearest bus stop is the potential base demand for feeder service. A part of the base demand is expected to use the feeder service. The road network of the study area, village and bus stop locations, routes followed to bus stop and link flows are the vital inputs for the design of feeder service. The temporal distribution of demand is a necessary input for deciding the span of operation and the frequency of service. The viable cutoff revenue is also an essential input. All these inputs are included in the database. 4.2.1 Travel Demand to Bus Stop This part describes the estimation of trips generated from villages located in the influence area of a bus stop. The demands are estimated using trip

Design of Feeder Service

production models. As the base demand estimated at this stage is used to assess the demand for feeder service, it is aimed to model only those trips which are generated from villages and terminated either at the bus stop or proceeded beyond the bus stop. 4.2.1.1 Modeling of Trip Rates The trip rates are estimated separately for revenue generating and nonrevenue generating trips. Non-revenue generating trips are further classified in two categories namely educational trips (i.e. trips made for educational purpose only), and household trips (i.e. trips made for social, recreational and household purposes). For the estimation of trip rates, working population is classified under three categories namely cultivator, daily labour and service/business. Households are also categorized in the same manner based on the nature of occupation of the head of household. Trips generated from a village are directly related to socioeconomic characteristics of the working population or households. Average household income of each category is given in Table-4.1. The estimation of different categories of trip generated from villages to bus stop is explained below. Table-4.1 Average Income for Different Categories of Household Household Category Cultivator Daily Labour Service/Business

Monthly Household Income in Rupees 3070 1715 5140

Revenue generating trips: Revenue generating trip data collected from 1691 persons are analyzed. The estimated revenue generating trip rates are shown in Table-4.2. These trip rates are used to estimate the total revenue generating trips from different villages to bus stop. Table-4.2 Estimated Trip Rates: Revenue Generating Trips Category (j) Service and Business Cultivator Daily Labour

No. of Samples 531 844 316

Total Work Trip per Week 1183 111 104

77

Av. Trip per Day per Person (RTj) 0.318 0.019 0.047

Chapter 4

It may be observed from Table-4.2 that the trip rate is the highest for service and business category followed by daily labours and cultivators. Service and business population is expected to travel more frequently than labours/ cultivators. It may be noted that even for service and business population, the observed trip rate does not indicate daily travel to bus stop. This may be justified by the fact that due to lack of transport connectivity and other reasons, a part of the total service population stay in town and visit home (in village) once in a week or so. The trip rates as reported in Table-4.2 also indicate that very few daily labours go out of the village for work. Cultivators also travel to bus stop only once in a while for purchasing of agricultural requirements such as pesticides, seeds, fertilizer etc. Educational trips: Educational trip rates for different categories of household are shown in Table-4.3. The estimated trip rates along with the number of households under each category in a village are used to estimate the total educational trips.

Table-4.3 Estimated Trip Rates: Educational Trips Category of Household (j) Service and Business Cultivator Daily Labour Total

Source: Primary surveys

No. of Samples 348 465 185 998

Trips per Week 559 655 123

Average Trips per Day per Household (ETj) 0.229 0.201 0.095

The trips reported in Table-4.3 are primarily for higher education purpose. The educational trip rates may be related to the income of household (Table-4.1) category. Higher income households pay more emphasis on higher education. Daily labour households do not usually send their children for higher education. Accordingly, a low trip rate is estimated for this category.

78

Design of Feeder Service

Household trips: Socioeconomic characteristics of household and village level attributes are used to model household trip rates. A regression model with household income, family size and bus stop distance as explanatory variables is developed (Equation 4.1).

HTij = 0.0001 + 0.00003 × HIij + 0.041 × FSij − 0.036 × log(BSDi )

(4.1)

Where,

HTij = Trip rate for household category ‘j’ in village ‘i’ HIij = Household monthly income of category ‘j’ in village ‘i’ FSij = Family size of category ‘j’ in village ‘i’ BSDi = Distance to bus stop from village ‘i’ The t-statistics of constant, coefficient of HIij, FSij and log(BSDi) are calculated as 1.29, 16.28, 18.24 and 2.12 respectively. F-value is much higher than the critical value of acceptance. R2 value of the model is 0.696, which is acceptable for travel demand modeling point of view. The tstatistics of all parameter estimates except the constant are statistically significant with a confidence level more than 95%. Therefore, the model is re-estimated by neglecting the constant. The t-statistics of coefficients of HIij, FSij and log(BSDi) for the re-estimated model are calculated as 18.431, 20.212 and 3.753 respectively. R2 value of the re-estimated model is 0.693. Therefore, the household trip production per day for category ‘j’ household in village ‘i’ can be written as Equation (4.2).

HTij = 0.00003 × HIij + 0.04 × FSij − 0.033 × log(BSDi )

(4.2)

The signs of the coefficient estimates of the model are logical. The positive sign of coefficient estimates of household income and family size indicate that number of trips increases with an increase in household income and family size. The negative sign of the coefficient of bus stop distance from village indicates a decrease in the number of trips with an increase in the distance from bus stop. The average household incomes for different

79

Chapter 4

categories of household are used to calculate the daily household trip production from a village. Total trips: Total trips generated from a village are the sum of revenue and nonrevenue generating trips. The total trips generated from a village ‘i’ to the nearest bus stop may be expressed as follows: Ti =

∑

3 j =1

ETj *HHij +

∑

3 j =1

HTij *HHij +

∑

3 j =1

RTj *Pij

(4.3)

Where, Ti = Total trips to nearest bus stop from a village ‘i’ HHij = Number of households in a village ‘i’ under household category ‘j’ ETj = Educational trip rate for household category ‘j’ HTij = Household trip rate in village ‘i’ for household category ‘j’ RTj = Revenue generating trip rate for household category ‘j’ Pij = Number of workers in a village ‘i’ under worker category ‘j’ 4.2.1.2 Validation of Modeled Trip Rates

Before using the trip rates for estimation purpose, it was necessary to validate them. Validation was necessary to ensure that the model meets the intended requirements in terms of the method employed and the results obtained. Validation of trip rates was done by counting the number of trip makers leading to bus stop and recording their trip purposes. Two bus stops namely Dantun and Kalabani were selected for this purpose. The validation survey was carried out during June 2006. The modeled and observed revenue generating trips to bus stops are shown in Figure-4.1. A comparison of modeled and observed household trips to bus stops is shown in Figure4.2. The educational trips could not be validated as during validation survey, the educational institutes were closed due to summer vacation. It may be observed from Figure-4.1 and Figure-4.2 that the modeled trips are generally in good agreement with the observed trips. Therefore, the modeled trip rates are accepted for estimating trip generation from villages to bus stop.

80

Design of Feeder Service

450

Modeled Trips

Number of Trips

400

Observed Trips

350 300 250 200 150 100 50 0

Dantun

Kalabani

Figure-4.1 Comparison of Modeled and Observed Revenue Generating Trips

Number of Trips

700

Modeled Trips

Observed Trips

600 500 400 300 200 100 0 Dantun

Kalabani

Figure-4.2 Comparison of Modeled and Observed Household Trips 4.2.1.3 Estimation of Trips

Different categories of trip generated from 126 villages in the study area are estimated and classified according to bus stop influence area. The trips generated from the influence area of each bus stop are summarized in Figure-4.2.

81

Chapter 4

Syamalpur

RevenueGenerating Trips

Sarisa

Household Trips

Salajpur

Educational Trip

Panchyar Nachipur Monaharpur

Name of Bus Stop

Kukai Khokra Kalabani Ektarpur Dantan Daihara Bhasra 0

100

200

300

400

500

600

Number of Trips per Day

Figure-4.3 Estimated Trips to Bus Stops 4.2.2 Temporal Variation of Demand

The temporal variations of inward (i.e. from bus stop to village) and outward (i.e. from village to bus stop) travel demand as obtained from the primary survey (trip-diary) are shown in Figure-4.4. It may be observed that for outward trips the peak is in the morning, while for inward trips the peak is in the evening. 83% of the outward trips were observed during the morning hours (i.e. between 5.30 am to 00.30 pm). On the other hand around 81% of inward trips were observed during the evening hours (i.e. 00.30 pm to 7.30 pm). The inward and outward trips between 7.30 pm and 5.30 am were negligible. Therefore, the span of operation of feeder service is considered between 5.30 am to 7.30 pm.

82

7.30 pm

After 7.30 pm

Time

6.30 pm

5.30 pm

4.30 pm

3.30 pm

2.30 pm

1.30 pm

12.30pm

11.30 4am

10.30 am

8.30am

7.30am

6.30am

5.30 am

1600 1400 1200 1000 800 600 400 200 0

9.30 am

Outward Trips from Village Inward Trips to Village

Before 5.30 am

Number of Trips

Design of Feeder Service

Figure-4.4 Temporal Variation of Travel Demand 4.2.3 Road Network

This section deals with the development of road network to bus stop. This is done in two stages. Initially, all the road links are coded to develop a base network. Subsequently, a spanning tree road network (called as road network to bus stop) is derived from the base network. 4.2.3.1 Base Network

A database is developed in GIS environment including information related to roads

links,

topographical

village

location

sheets.

For

this

and

bus

purpose,

stop

location

the scanned

obtained image of

from the

topographical sheets is digitized using ERDAS IMAGINE software. As the topographical sheets are of 1980s, road network database information is updated with limited field surveys. All available fair weather roads up to cart track are included in the database. The road network thus developed is called the base network. 4.2.3.2 Road Network to Bus Stop

As mentioned in Chapter-2, it is aimed to provide single all weather road connectivity to all villages through the on-going rural road development programme called the PMGSY. ‘Single all weather road connectivity’ 83

Chapter 4

essentially means the development of an all weather spanning tree road network. In the present work, this spanning tree is developed with due consideration to the preference of rural trip makers. An iterative procedure, as mentioned in Section 2.3.1.3, is followed to derive the road network to bus stop or spanning tree network from the base network. For carrying out all or nothing assignment, the shortest path is generated in the GIS environment (Arcview-3.2) by implementing the Dijkstra’s algorithm using avenue script. Spanning trees obtained for different bus stop influence areas are shown in Figure-4.5. The road length of spanning tree in the influence area of each bus stop along with number of villages and population served are shown in Table-4.4. Two distinct routes (without any common road link) terminate at Khokra bus stop. Therefore, two spanning trees (i.e. Khokra_1 and Khokra_2) are developed for this bus stop. The villages in the influence area of Ektarpur bus stop are located within 2.0 km from the Highway. Therefore, no feeder service is considered for these villages. Instead, it is assumed that more bus stops will be provided in the future along the National Highway to bring these villages within 2.0 km from bus stop. Table-4.4 Bus Stops and Their Influence Areas Bus Stop No. of Villages Population Road Length in km+ Bhasra 12 11162 12.3 Daihara 6 3151 7.2 Dantan 15 9541 14.2 Ektarpur 13 8880 13.6 Kalabani 11 5590 8.3 Khokra 14 8695 15.3 Kukai 8 4188 5.5 Monaharpur 7 5937 9.0 Nachipur 5 2406 5.2 Panchyar 7 3529 5.6 Salajpur 12 5773 12.8 Sarisa 7 2700 5.0 Syamalpur 9 2849 6.3 126 74401 120.3 Total + Road length of the spanning tree

84

Design of Feeder Service

Figure-4.5 Road Network to Bus Stops 4.2.4 Cutoff Revenue

The cutoff revenue is considered as the minimum required earning to cover the fixed cost and running cost of vehicle alongwith a minimum profit for operator. Fixed cost of vehicle includes capital repayment (equitable monthly installment), insurance, road tax, permit and crew cost. Cost associated with fuel, lubricant, tyre and maintenance are considered under running cost. The cutoff revenue is calculated with the following base assumptions. •

Feeder Vehicles are to be purchased new, taking loan from Bank

•

Rate of Interest: 10.5% per annum compounded annually

•

Period of repayment: 7 year for ‘Trekker’ 5 year for ‘Tempo’

•

Days of Vehicle Operation: 30 days per month

•

Maintenance: No maintenance cost for new vehicle in the first year

•

Diesel price: Rs.36/lt 85

Chapter 4

•

Lubricant Price: Rs.100/lt

•

Fuel Efficiency: 15 km/lt for Trekker’ 30 km/lt for ‘Tempo’

•

Crew cost: Rs.3000/month

•

Profit for Operator: Rs. 3000/month

The cutoff revenue for Tempo and Trekker are estimated on the basis of the above assumptions and data collected from Regional Transportation Office, Bhadrak, Orissa as well as operators of such types of vehicle. The fixed costs of vehicle are summarized in Table-4.5. Table-4.5: Fixed Cost of Feeder Vehicles Trekker

Tempo

Vehicle Cost

220000

120000

Insurance cost(per Year)

9500

3000

Permit (Per Year)

1750

1200

Road Tax(per Year)

540

480

Cost are expressed in rupees

Based on the fixed cost components, the required minimum daily revenues are estimated as shown in Table-4.6. The crew cost is also included while estimating the revenue requirement in Table-4.6. The revenue required to cover the running cost of vehicle is summarized in Table-4.7. Table-4.6 Revenue Required for Covering Fixed Cost Trekker

Tempo

3710

2580

Insurance Cost (Per Month)

792

250

Permit (Per Month)

146

100

45

40

Crew Cost (Per Month)

3000

3000

Revenue Per Month

7693

5970

Revenue Per Day

247

199

Equitable Monthly Installment

Road Tax (Per Month)

Cost are expressed in rupees

86

Design of Feeder Service

Table-4.7 Revenue Required for Covering Running Cost Trekker

Tempo

Fuel

2.40

1.20

Tyre

4000/40000 km = 0.10

2000/20000km = 0.10

Lubricant

200/1000 km = 0.20

2000/2000 km = 0.10

Miscellaneous

0.10

0.10

Total

2.80

1.50

Cost are expressed in rupees per km

The cutoff revenue (CR) required per day per vehicle is estimated to meet fixed cost, running cost and minimum profit for operator.

Cutoff revenue

per day in rupees for ‘Trekker’ and ‘Tempo’ are given in the Equation 4.4 and 4.5 respectively. CR(Trekker) = 247 + 2.8 * d + P

(4.4)

CR(Tempo) = 199 + 1.5 * d + P

(4.5)

Where, d is the distance traveled per day in km and P is the minimum profit for operator. P is assumed as Rs. 100 per day per vehicle (or Rs. 3000 per month per vehicle) 4.3 MEASURE OF EFFECTIVENESS, ALTERNATIVE SCENARIOS AND FARE LEVELS

In order to design rural feeder service, it is necessary to select an appropriate measure of effectiveness (MOE). Also, it is also necessary to consider related policy measures in order to investigate their role on MOE. The MOEs, policy scenarios with assumptions and fare levels considered for the operation are discussed below. 4.3.1 Measure of Effectiveness

The selection of feeder routes and vehicle type is done considering two alternative MOEs namely generalized cost (GC) and passenger-km. A reduction in GC is a rational measure of users’ benefit. Therefore, the selection of routes and vehicle is attempted to bring down the overall GC, or enhance the GC saving. Alternatively, it may be aimed to enhance the passenger-km served by feeder service. Accordingly, passenger-km served

87

Chapter 4

by feeder service is also considered as another MOE. With GC as MOE, the routes and vehicle types are selected in order to minimize the GC or maximize the GC saving. On the other hand, with passenger-km as MOE, the routes and vehicle types are selected in order to maximize the passenger-km served by feeder service. 4.3.2 Alternative Scenarios

From viability point of view, it may not be possible to cover all road links in the influence area of a bus stop or all the bus stop influence area by feeder service. Therefore, ‘Cross Subsidy’ and ‘External Subsidy’ are the two relevant policy measures considered in order to cover more road links in a bus stop influence area or more routes in a geographical area by feeder service and thereby enhance the MOE. With cross subsidies, the overall operational viability of all vehicles/routes rather than individual vehicle/route, is considered. In this mechanism, the surplus earned in excess of the cutoff revenue by some vehicles/routes is utilized to cover more areas under feeder service in the light of the MOE. It is ensured that the average revenue earned per vehicle (considering all the vehicles operating on all the feeder routes) satisfies the criteria of cutoff revenue. The cross subsidy principle may work satisfactorily, if all the vehicles/routes are operated by a single operator. External subsidy is another mechanism considered for enhancing the MOE. In this process, the cutoff revenue required per vehicle is lowered up to an amount equivalent to the amount of external subsidy per vehicle. Different scenarios are formulated for analysis based on external subsidy and/or cross-subsidy. These scenarios are mentioned below. •

Scenario-I: No external subsidy or cross subsidy

•

Scenario-II: Only external subsidy of up to Rs. 20.00 per vehicle per day

88

Design of Feeder Service

•

Scenario-III: Only external subsidy of up to Rs. 40.00 per vehicle per day

•

Scenario-IV: Only cross subsidy

•

Scenario-V: Cross subsidy and external subsidy of Rs. 20.00 per vehicle per day

•

Scenario-VI: Cross subsidy and external subsidy of Rs. 40.00 per vehicle per day

Scenario-I may cover some bus stop influence areas only partially by feeder service. Also, some bus stop influence areas may not be served at all. Scenario-II and III may help to further improve the MOE by extending feeder routes to cover more villages and/or bus stop influence areas. Scenario-IV in comparison to Scenario-I is expected to optimize the MOE by covering more villages and/or bus stop influence areas using the strength of cross subsidy. Scenario-V and VI explore the strength of both cross subsidy and external subsidy to further improve the MOE. For Scenarios-I, II and III, feeder routes and vehicle are selected to maximize the GC saving following the procedure described in Section 2.3.3.1. For Scenarios-IV, V and VI, feeder routes and vehicle are selected to maximize the GC saving following the procedure described in Section 2.3.3.2. The selection of feeder routes and vehicle under different scenarios is carried out with the following base assumptions. •

Span of operation: 5.30 am to 7.30 pm

•

Peak period duration: 5.30 am to 12.30 pm for outward trips 12.30 pm to 7.30 pm for inward trips.

•

Total demands during the peak and off-peak period are uniformly distributed over the peak and off-peak period respectively

•

Speed of feeder vehicle: 20 km per hour

•

Minimum layover time: 10 min at each end for fixed form of operation

•

Seating Discomfort: Comfortable seating

89

Chapter 4

4.3.3 Fare Levels

The fare directly influences patronage and revenue. Therefore, selection of viable feeder route and vehicle is also influenced by the selection of fare. In the present work, a fare range from Rs. 0.50 per km to Rs. 2.0 per km is considered. However, only practical fare levels represented in multiple of Rs. 0.25 are taken for investigation. 4.4 FEEDER SERVICE WITH GENERALIZED COST AS MEASURE OF EFFECTIVENESS

The six scenarios as described earlier are analyzed by taking the GC saving as MOE. The analysis presented in this section considers fixed form of operation of feeder vehicles. Although, all the above scenarios are investigated considering all the practical fare levels from Rs. 0.50 per km to Rs. 2.00 per km, fare levels of Rs. 0.50 per km and Rs. 0.75 per km did not produce any viable feeder route for any of the scenarios. On the other hand, fare levels beyond Rs. 1.25 per km did not produce any saving in GC. Therefore, Rs. 1.00 per km and Rs. 1.25 per km are the only two practical fare levels identified in the present study which produced viable feeder route(s) with GC saving. Accordingly, the following fare combinations (Table-4.8) are designed. In the following sections, the results are reported only for these fare combinations. The process of designing feeder service for all the six scenarios with the fare combinations mentioned in Table-4.8 is discussed in Section 4.4.1. Table-4.8 Selected Fare Combinations Fare Combinations FC-I FC-II FC-III FC-IV

Fare of Trekker 1.00 1.25 1.25 1.25

Fare of Tempo 1.00 1.00 1.00 1.25

Fares are in Rs. per km

In Section 4.4.1, the mode choice models developed from the analysis of only SP data in Stage-II analysis of Chapter 3 are employed. There are inherent uncertainties associated with the use of such SP based models for 90

Design of Feeder Service

prediction purpose. The alternative-specific constant(s) (ASC) of these models does/do not convey any practical implication in terms of prediction purpose (Hensher et al. 2005). Therefore, the effect of ASC on operational viability of feeder routes (selected in Section 4.4.1) is also investigated and reported in Section 4.4.2. 4.4.1 Feeder Routes and Vehicle

The selection of feeder routes and vehicle for all the scenarios are discussed below. 4.4.1.1 Scenario-I

The details of viable routes and vehicle for different bus stop influence areas are reported in Table-4.9. It may be noted that with fare combination FC-I, no viable route is obtained for any bus stop influence area. Under FC-II, three viable routes are identified. The number of viable routes is found to increase under FC-III and FC-IV. A summary of feeder service under different fare combinations is reported in Table-4.10. It may be observed from Table-4.10 that the GC saving is the highest under FC-IV. Therefore, FC-IV and its corresponding routes and vehicle are recommended for the Scenario-I. FC-IV is expected to offer viable feeder service covering part of the seven bus stop influence areas. It may be observed from Table-4.10 that the fare combination (i.e. FC-IV) producing the highest GC saving do not maximize the route length and passenger-km. This indicates that the selection of routes and vehicle is likely to be different if passenger-km is considered as the MOE. Table-4.9 Details of Feeder Service in Scenario-I: GC as MOE Fare

Bus Stop

Vehicle

Route

Passenger

Passenger

GC

Surplus

Type

Length

-km

Served

Saving

per Vehicle (Rs.)

(km) FC-II

FC-III

Bhasra

Tempo

8

4982

757

1231

6.0

Daihara

Tempo

5.9

1180

212

215

34.5

Dantun

Tempo

7.9

3107

483

753

7.0

Bhasra

Trekker

9.7

5430

771

1188

23.4

Daihara

Trekker

6.6

1277

218

165

68.1

91

*

Chapter 4

FC-IV

Dantun

Trekker

8.9

3273

493

714

21.8

Kalabani

Trekker

6.3

1802

372

337

-2.1

Manoharpur

Trekker

5.8

2293

513

425

16.1

Salajpur

Trekker

6.3

2290

485

482

7.5

Syamalpur

Trekker

3.2

642

208

9

11.6

Bhasra

Tempo

8

4982

757

1231

6.0

Daihara

Tempo

5.9

1180

212

215

34.5

Dantun

Tempo

7.9

3106

483

753

7.0

Kalabani

Trekker

6.3

1802

372

337

-2.1

Manoharpur

Trekker

5.8

2293

513

425

16.1

Salajpur

Trekker

6.3

2290

485

482

7.5

Syamalpur

Trekker

3.2

642

208

9

1.6

*Earning in excess of cutoff revenue

Table-4.10 Summary of Feeder Service in Scenario-I: GC as MOE Fare FC-II FC-III FC-IV

Route Passenger Trekker Length (km) -km 21.8 9269 0 46.8 17007 43 43.8 16296 19

Tempo 33 0 33

Total Passenger Vehicle Served 33 1453 43 3060 52 3031

GC Saving 2198 3320 3451

4.4.1.2 Scenario-II

The routes and vehicle in Scenario-II are selected considering a reduction in cutoff revenue by Rs.20.00 as compared to the cutoff revenue in Scenario-I. The feeder service selected for each fare combination is summarized in Table-4.11. Like Scenario-I, fare combination FC-I did not result any viable feeder route. FC-IV shows the highest GC saving among all the fare combinations. Therefore, FC-IV and its corresponding routes and vehicle are recommended in Scenario-II. A comparison of recommended feeder services under Scenario-I and Scenario-II clearly indicates that additional GC saving is possible if an external subsidy of up to Rs.20.00 per vehicle is provided. The route length and passenger-km also improved in Scenario-II. The route wise details of recommended feeder service in Scenario-II are given in Table-B.1 of Annexure-B.

92

Design of Feeder Service

Table-4.11 Summary of Feeder Service in Scenario-II: GC as MOE Fare FC-II FC-III

Route Length (km) 39.3 55.3

FC-IV

55.3

Passenger Trekker Tempo -km 15058 0 59 18549 48 0 18393

18

46

Total Vehicle 59 48

Passenger Served 2780 3587

GC Saving 3551 3404

64

3542

3674

4.4.1.3 Scenario-III

The routes and vehicles in Scenario-III are selected considering a reduction in cutoff revenue by Rs.40.00 as compared to the cutoff revenue in Scenario-I. The feeder service selected for each fare combination is summarized in Table-4.12. FC-II shows the highest GC saving among all the fare combinations. Therefore, FC-II and its corresponding routes and vehicle are recommended in Scenario-III. A comparison of GC saving of the recommended routes and vehicle under Scenario-II and III (Table-4.11 and Table-4.12) indicates that the GC saving is enhanced in Scenario-III. The route wise details of the recommended feeder service in Scenario-III are given in Table-B.2 of Annexure-B. Table-4.12 Summary of Feeder Service in Scenario-III: GC as MOE Fare FC-I FC-II FC-III FC-IV

Route Length (km) 21.9 51.7 55.0 61.7

Passenger Trekker Tempo -km 9668 18 4 18057 21 36 17323 42 4 19457 7 66

Total Vehicle 22 57 46 73

Passenger Served 1486 3508 4091 4001

GC Saving 4501 6179 3387 4055

4.4.1.4 Scenario-IV

The summary of routes and vehicle in Scenario-IV is reported in Table-4.13. With FC-I, no viable route is selected for any of the bus stop influence areas. FC-III shows the highest GC saving among all the fare combinations. Therefore,

FC-III

and

its

recommended in Scenario-IV.

corresponding

routes

and

vehicle

are

It is also observed that the GC saving for

the recommended service is enhanced in Scenario-IV as compared to the same in Scenario-I. Some of the bus stops, which were not covered by feeder service in Scenario-I, are now covered in Scenario-IV.

93

Chapter 4

Table-4.13 Summary of Feeder Service in Scenario-IV: GC as MOE Fare FC-II FC-III

Route Length (km) 34.9 72.3

FC-IV

72.9

Passenger Trekker Tempo -km 12264 0 45 22435 58 3 22390

58

4

Total Vehicle 45 61

Passenger Served 2030 4709

GC Saving 2801 4173

62

4702

4004

A comparison of recommended feeder services in Scenario-I and IV indicates that apart from enhancing GC saving, the cross subsidy also becomes instrumental to cover more roads under feeder service and enhance the passenger-km.

The route wise details of the recommended

feeder service in Scenario-IV are given in Table-B.3 of Annexure-B. 4.4.1.5 Scenario-V

In this scenario, the combined effect of external subsidy up to Rs.20.00 per vehicle per day and cross subsidy on feeder service is examined. The routes and vehicles in Scenario-V are selected considering a reduction in cutoff revenue by Rs.20.00 as compared to the cutoff revenue in Scenario-I or Scenario-IV. The summary of routes and vehicles selected in Scenario-V is reported in Table-4.14. Like Scenario-IV, fare combination FC-I did not result any viable feeder route. FC-III shows the highest GC saving among all the fare combinations. Therefore, FC-III and its corresponding routes and vehicle are recommended in Scenario-V. For the recommended feeder service, an improvement in GC saving is observed in Scenario-V as compared to Scenario-IV. The route wise details of the recommended feeder service in Scenario-V are given in Table-B.4 of Annexure-B. Table-4.14 Summary of Feeder Service in Scenario-V: GC as MOE Fare FC-II FC-III FC-IV

Route Length (km)

Passenger Trekker Tempo -km

54.7 71.4 72.8

16790 22304 22031

0 54 23

68 11 56

Total Vehicle

Passenger Served

GC Saving

68 65 79

3374 4704 4610

3625 4982 4561

4.4.1.6 Scenario-VI

The routes and vehicles in this scenario are selected considering a reduction in cutoff revenue by Rs.40.00 as compared to the cutoff revenue in 94

Design of Feeder Service

Scenario-I or Scenario-IV. The feeder service optimizing the GC saving for each fare combination is summarized in Table-4.15. FC-III shows the highest GC saving among all the fare combinations. Therefore, FC-III and its corresponding routes and vehicle type are recommended in Scenario-III. An increase in GC saving is observed in Scenario-VI as compared to Scenario–V. The route wise details of the recommended feeder service in Scenario-VI are given in Table-B.5 of Annexure-B. Table-4.15 Summary of Feeder Service in Scenario-VI: GC as MOE Fare

Route Length (km)

FC-I FC-II FC-III FC-IV

Passenger Trekker Tempo -km

Total Vehicle

Passenger Served

GC Saving

26 72.8

10392 22119

21 11

0 75

21 86

1697 4587

4635 6005

73.4 75.9

22379 22291

36 0

36 96

72 96

4698 4571

6448 4828

4.4.1.7 Comparison of Different Scenarios

A comparison of the recommended feeder services in six scenarios is presented in this section. GC saving in six scenarios is shown in Figure-4.6.

7000

6448

6179

GC Saving in Rs.

6000 4981

5000 4000

4173 3674

3417

3000 2000 1000 0 Scenario-I

Scenario-II

Scenario-III

Scenario-IV

Scenario-V

Scenario-VI

Figure-4.6 Comparison of Different Scenarios: GC as MOE

The lowest GC saving is observed in Scenario-I. With the introduction of external subsidy, the GC saving is increased in Scenario-II. An increase in the amount of external subsidy resulted higher GC saving in Scenario-III as compared to Scenario-II. The introduction of cross subsidy in Scenario-IV resulted

higher

GC

saving

in

comparison 95

to

Scenario-I.

With

the

Chapter 4

introduction of external subsidy along with the cross subsidy, a further increase in GC saving is observed in Scenario-V and VI. In Scenario-VI, the maximum GC saving is expected. Other attributes of the recommended feeder services in different scenarios are reported in Table-4.16. Although no systematic trend is observed in attribute values under different scenarios, it is evident that the route length covered by feeder service may also be enhanced using the strength of cross subsidy. Figure-4.6 and Table-4.16 clearly show that the cross subsidy is a potential policy instrument, which should be used to enhance the benefits to rural trip makers. Even when an external subsidy is provided, feeder service with cross subsidy is expected to bring higher users’ benefit than with no cross subsidy. “First cross subsidy and then external subsidy” appears to be a rational policy for providing feeder service to bus stop in rural areas. Table-4.16 Attributes of Recommended Feeder Service: GC as MOE Fare Scenario-I Scenario-II Scenario-III Scenario-IV Scenario-V Scenario-VI

FC-IV FC-IV FC-II FC-III FC-III FC-III

Route Length (km) 43.8 55.3 51.7 72.3 71.4 73.4

Passenger Trekker Tempo Total Passenger -km Vehicle Served 16296 18393 18057 22435 22304 22379

19 18 21 58 54 36

33 46 36 3 11 36

52 64 57 61 65 72

3031 3542 3508 4709 4704 4698

4.4.2 Effect of ASC on Operational Viability

A comparison of all the relevant attributes of recommended feeder routes in Scenario-I with 100%, 50% and 25% of the ASC values is given in Table4.17. It may be observed that for each bus stop influence area except Syamalpur, there is a surplus earning per vehicle even if the ASC values are taken as 50% and 25% of the ASC values used in Section 4.4.1. The routes which remain viable with 100%, 50% and 25% of the ASC values (as used in Section 4.4.1) are defined as stable routes. These routes are called ‘stable’ because even if the ASC values are reduced to 25% of the values used for recommending these route (in Section 4.4.1), the routes will still remain viable. A similar analysis is carried out for all the selected routes

96

Design of Feeder Service

under Scenarios-II to VI and reported in Table-B.1 to Table-B.5 of Annexure-B. Table-4.17 Effect of ASC on Recommended Feeder Service in Scenario-I: GC as MOE Bus stop

GC Surplus/ Route %ASC Passenger Vehicles Demand km Saving Vehicle Stability

Bhasra

100 50 25

4982 4738 4491

18 17 16

757 717 677

1231 1160 1091

6.0 8.6 11.3

Stable Stable Stable

Daihara

100 50 25

1180 1085 984

4 4 4

212 195 177

215 190 161

34.5 13.9 3.9

Stable Stable Stable

Dantun

100 50 25

3107 2947 2785

11 11 10

483 456 429

753 708 663

7.0 -3.6 6.4

Stable Stable Stable

Kalabani

100 50 25

1802 1718 1561

5 5 4

372 353 317

337 311 266

-2.1 -9.1 23.2

Unstable Unstable Unstable

Manoharpur

100 50 25

2293 2180 1976

6 6 5

513 485 436

425 393 340

16.1 8.4 30.8

Stable Stable Stable

Salajpur

100 50 25

2290 2196 2018

6 6 5

485 461 417

482 455 402

7.5 -0.3 28.9

Stable Stable Stable

Syamalpur

100 50 25

642 576 447

2 2 2

208 187 145

9 -12 -42

11.6 -11.8 -65.5

Unstable Unstable Unstable

The overall variation in GC saving due to the change in ASC is reported in Figure-4.7. It is evident that there is a decrease in the value of GC saving with the reduction in ASC values. The reason may be attributed to less demand served by feeder service. The variations of passenger-km and passenger served are reported in Figure-4.8 and 4.9 respectively. Both passenger-km and passenger served show a declining trend with a reduction in ASC values.

97

GC Saving in Rupees

Chapter 4

100%ASC

7000

50%ASC

25%ASC

6000 5000 4000 3000 2000 1000 0 Scenario-I

Scenario-II

Scenario-III

Scenario-IV

Scenario-V

Scenario-VI

Passenger km Served

Figure-4.7 Effect of ASC on GC Saving: GC as MOE 100%ASC

25000

50%ASC

25%ASC

20000 15000 10000 5000 0 Scenario-I

Scenario-II

Scenario-III Scenario-IV

Scenario-V

Scenario-VI

Figure-4.8 Effect of ASC on Passenger-km: GC as MOE

100%ASC

Demand Served

5000

50%ASC

25%ASC

4000 3000 2000 1000 0 Scenario-I

Scenario-II

Scenario-III

Scenario-IV

Scenario-V

Scenario-VI

Figure-4.9 Effect of ASC on Passenger Served: GC as MOE

The stability of all the recommended routes in Section 4.4.1 under all the study scenarios is summarized in Figure-4.10. It is evident that all the recommended routes in Section 4.4.1 are not stable. In other words, if the true ASC is significantly different from the estimated ASC of SP based models, then all the routes recommended in Section 4.4.1 may not be viable. Accepting the uncertainties associated with the ASC, the information 98

Design of Feeder Service

presented in Figure-10 may be utilized to select only the stable routes for initial operation of feeder service. Once the feeder service is operational, the ASC may be recalibrated using the then RP database and accordingly, the feeder service may be extended to cover more routes or bus stop influence areas.

Bus stop Scenario-I Scenario-II Scenario-III Scenario-IV Scenario-V Scenario-VI Bhasra Daihara Dantun Kalabani Khokra_1 Khokra_2 Kukai Manoharpur Nachipur Panchiyar Salajpur Sarisa Syamalpur Route Not Selected

Stable Route

Unstable Route

Figure-4.10 Stability of Routes with GC as MOE

The variation of average surplus earned per vehicle per day (considering all the vehicles on all feeder routes) with ASC is reported in Table-4.18. However, such a calculation is relevant only for Scenarios-IV, V and VI with cross subsidy. It is interesting to note that even though some individual routes are unstable (Figure-4.10), the average surplus earned per vehicle per day (Table-4.18) indicates viable operation. Therefore, although there are uncertainties associated with the ASC, all the routes recommended in Section 4.4.1 may be taken up for operation of feeder service for ScenariosIV, V and VI. However, without cross subsidy (i.e. Scenarios-I,II and II), it is desirable to select only the stable routes (Figure-4.10) for initial operation of feeder service and later on extend the feeder service to cover more routes or bus stop influence area based on the recalibration of ASC using the then RP data. Figure-4.10 and Table-4.18 again shows the strength of cross-subsidy in the context of rural feeder service to bus stop.

99

Chapter 4

Table-4.18 Effect of ASC on Revenue Surplus: GC as MOE %ASC

Average Revenue Surplus/ Vehicle in Rs. Scenario-IV

Scenario-V

Scenario-VI

2.2 -4.1 7.5

6.0 3.8 10.0

1.7 -1.0 1.3

100 50 25

4.5

FEEDER

SERVICE

WITH

PASSENGER-KM

AS

MEASURE

OF

EFFECTIVENESS

The six scenarios described earlier in Section 4.3.2 are analyzed with passenger-km as MOE in this section. The analyses presented in this section also consider fixed form of operation of feeder vehicles. Feeder routes and vehicle are selected on the principle of maximizing the passenger-km for each fare combination. The fare combinations investigated are same as given in Table-4.8. The process of designing feeder service for all the six scenarios are discussed in Section 4.5.1. As discussed in Section 4.4.2, the effect of ASC on operational viability of feeder routes (selected in Section 4.5.1) is investigated and reported in Section 4.5.2. 4.5.1 Feeder Routes and Vehicle

The selection of feeder routes and vehicle for all the scenarios are discussed below with passenger-km as MOE. 4.5.1.1 Scenario-I

A summary of all the viable routes under different fare combinations is reported in Table-4.19. It may be noted that under fare combination FC-I, no viable route is obtained for any of the bus stop influence areas. It may be observed from Table-4.19 that, both FC-IV and FC-III served maximum passenger-km. Other attributes of the selected routes in both the fare combinations

have

the

same

value.

This

is

because

in

both

fare

combinations only Trekker is selected as a feeder vehicle and the fare level for Trekker is same in FC-III and FC-IV. It may be observed from Table4.20 that the fare combinations producing the maximum passenger-km also maximize the route length and GC saving. The route wise details of

100

Design of Feeder Service

recommended feeder service in Scenario-I are given in Table-B.6 of Annexure-B. Table-4.19 Summary of Feeder Service in Scenario-I: Passenger-km as MOE Fare

Route Length (km)

FC-II FC-III FC-IV

Passenger Trekker Tempo -km

22.8 48.9 48.9

9566 17007 17007

0 43 43

34 0 0

Total Vehicle

Passenger Served

GC Saving

34 43 43

1453 3060 3060

2243 3320 3320

4.5.1.2 Scenario-II

The routes and vehicle in Scenario-II are selected considering a reduction in cutoff revenue by Rs.20.00 as compared to the cutoff revenue in Scenario-I. The feeder service selected for each fare combination is summarized in Table-4.20. Like Scenario-I, fare combination FC-I did not result any viable feeder route. Both FC-III and FC-IV show the highest passenger-km served among all the fare combinations. Other attributes of the selected routes in both combinations have the same value. Like Scenario-I, in both fare combinations only Trekker is selected as a feeder vehicle and the fare level for Trekker is same in FC-III and FC-IV. A comparison of passenger-km in Table-4.19 and Table-4.20 indicates that passenger-km is further enhanced in Scenario-II. The route length and GC saving is also improved in ScenarioII. The route wise details of the recommended feeder service in Scenario-II are given in Table-B.7 of Annexure-B. Table-4.20 Summary of Feeder Service in Scenario-II: Passengerkm as MOE Fare FC-II FC-III FC-IV

Route Passenger Trekker Tempo Total Passenger GC Length (km) -km Vehicle Served Saving 43.8 58.0 58.0

16026 19289 19289

0 50 50

64 0 0

64 50 50

2980 3868 3868

3680 3415 3415

4.5.1.3 Scenario-III

The routes and vehicles in this scenario are selected considering a reduction in cutoff revenue by Rs.40.00 as compared to the cutoff revenue in

101

Chapter 4

Scenario-I. The feeder service for each fare combination is summarized in Table-4.21. The maximum MOE is observed in both FC-III and FC-IV. Like previous two scenarios, all other attributes of the selected routes in both combinations have the same value. A comparison of passenger-km in Table4.20 and Table-4.21 indicates that passenger-km is increased further in Scenario-III.

The route wise details of recommended feeder service in

Scenario-III are given in Table-B.8 of Annexure-B. Table-4.21 Summary of Feeder Service in Scenario-III: Passengerkm as MOE Fare FC-I FC-II FC-III FC-IV

Route Passenger Trekker Tempo Total Passenger GC Length (km) -km Vehicle Served Saving 23.1 57.3 67.3

9797 19492 20635

21 10 54

0 62 0

21 72 54

1488 3905 4092

4485 5384 3569

67.3

20635

54

0

54

4092

3569

4.5.1.4 Scenario-IV

The summary of routes and vehicle selected in Scenario-IV is reported in Table-4.22 for different fare combinations. Both FC-III and FC-IV produced the maximum MOE. Like previous three scenarios, all other attributes of the selected routes in both combinations also have the same value. It is evident from Table-4.22 that passenger-km, route length and GC saving improved further as compared to the same in previous three scenarios. Some of the bus stops, which were not covered by feeder service in Scenario-I are now covered in Scenario-IV. The details of the routes selected in this scenario are reported in Table-B.9 of Annexure-B. Table-4.22 Summary of Feeder Service in Scenario-IV: Passengerkm as MOE Fare FC-II FC-III FC-IV

Route Passenger Trekker Tempo Total Passenger GC Length (km) -km Vehicle Served Saving 32.2

11935

0

44

44

2013

2414

74.1 74.1

22538 22538

61 61

0 0

61 61

4714 4714

3954 3954

102

Design of Feeder Service

4.5.1.5 Scenario-V

In this scenario, the combined effect of external subsidy upto Rs. 20.00 per vehicle per day and cross subsidy is examined. The routes and vehicle in Scenario-V are selected considering a reduction in cutoff revenue by Rs.20.00 as compared to the cutoff revenue in Scenario-I and Scenario-IV. The summary of routes and vehicle selected for each fare combination in Scenario-V is reported in Table-4.23. Table-4.23 shows FC-IV has the highest MOE. So, FC-IV and its corresponding routes and vehicle are recommended in Scenario-V. MOE, route length and GC saving also improved in this scenario as compared to the same in Scenario-IV. The route wise details of recommended feeder service in Scenario-V are given in Table-B.10 of Annexure-B. Table-4.23 Summary of Feeder Service in Scenario-V: Passenger-km as MOE Fare FC-II FC-III FC-IV

Route Passenger Trekker Tempo Total Passenger GC Length (km) -km Vehicle Served Saving 58.2 67.8 81.8

18328 22772 23121

0 56 64

74 9 0

74 65 64

3568 4341 4714

4059 4544 4056

4.5.1.6 Scenario-VI

The routes and vehicle in this scenario are selected considering a reduction in cutoff revenue by Rs.40.00 as compared to the cutoff revenue considered in Scenario-I and Scenario-IV. The routes and vehicle selected in each fare combination are summarized in Table-4.24. Table-4.24 shows FC-IV has the highest MOE among all the fare combinations. Therefore, FC-IV and its corresponding routes and vehicle are recommended in this scenario. In comparison to Scenario-V, a further improvement in MOE is observed in Scenario–VI. The route wise details of recommended feeder service in Scenario-VI are given in Table-B.11 of Annexure-B.

103

Chapter 4

Table-4.24 Summary of Feeder Service in Scenario-VI: Passengerkm as MOE Fare

Route Passenger Trekker Tempo Total Passenger GC Length (km) -km Vehicle Served Saving

FC-I FC-II FC-III FC-IV

25.1 75.4 81.6 88.3

10325 22284 23278 23421

23 0 52 67

0 96 19 0

23 96 71 67

1697 4570 4708 4714

4605 4821 5236 4102

4.5.1.7 Comparison of Different Scenarios

A comparison of the recommended feeder services under six scenarios is made in this section. MOEs (in passenger-km) obtained under six scenarios are shown in Figure-4.11. The lowest MOE is estimated in Scenario-I. With the introduction of external subsidy, MOE increased from Scenario-I to Scenario-II. An increase in amount of external subsidy resulted higher MOE in Scenario-III as compared to Scenario-II. Introduction of cross subsidy in Scenario-IV further enhanced the MOE as compared to previous three scenarios. With the introduction of external subsidy along with the cross subsidy, a further increase in MOE is observed in Scenario-V and VI. In Scenario-VI, the highest passenger-km is served.

Passenger km Served

25000 20000

22538 19289

23121

23421

Scenario-V

Scenario-VI

20635

17597

15000 10000 5000 0 Scenario-I

Scenario-II

Scenario-III Scenario-IV

Figure-4.11 Comparison of Different Scenarios: Passenger-km as MOE

Other attributes of the recommended feeder services in different scenarios are reported in Table-4.25. A systematic trend is observed in attribute values under different scenarios. It is evident that the route length covered

104

Design of Feeder Service

by feeder service may also be enhanced using the strength of cross subsidy. Figure-4.11 and Table-4.25 clearly show that the cross subsidy is a potential policy instrument, which should be used to enhance the benefit to rural trip makers. Table-4.25 Attributes of Recommended Feeder Service: Passengerkm as MOE Fare Scenario-I Scenario-II Scenario-III Scenario-IV Scenario-V Scenario-VI

FC-IV FC-IV FC-IV FC-IV FC-IV FC-IV

Route Trekker Tempo Total Passenger GC Length (km) Vehicle Served Saving 48.9 58.0 67.3 74.1 81.8

43 50 54 61 64

0 0 0 0 0

43 50 54 61 64

3060 3868 4092 4714 4714

3320 3415 3569 3954 4056

88.3

67

0

67

4714

4102

4.5.2 Effect of ASC on Operational Viability

A comparison of all the relevant attributes of recommended routes of feeder service in all the scenarios with 100%, 50% and 25% of the ASC values is reported in Table-B.6 to Table-B.11 of Annexure-B. The overall variation in GC saving with change in ASC value is reported in Figure-4.12. It is evident that there is a decrease in the value of GC saving with a reduction in the value of ASC. The reason may be attributed to less demand served by feeder service. The variations of passenger-km and passenger served are reported in Figure-4.13 and 4.14 respectively. Both passenger-km and

GC Saving in Rupees

passenger served show a declining trend with a reduction in ASC value.

4500 4000 3500

100%ASC

50%ASC

25%ASC

Scenario-IV

Scenario-V

3000 2500 2000 1500 1000 500 0 Scenario-I

Scenario-II

Scenario-III

Scenario-VI

Figure-4.12 Effect of ASC on GC Saving: Passenger-km as MOE 105

Passenger Km Served

Chapter 4

100%ASC

25000

50%ASC

25%ASC

20000 15000 10000 5000 0 Scenario-I

Scenario-II Scenario-III Scenario-IV

Scenario-V

Scenario-VI

Demand served

Figure-4.13 Effect of ASC on Passenger-km: Passenger-km as MOE

100%ASC

5000 4500 4000 3500 3000 2500 2000 1500 1000 500 0 Scenario-I

Scenario-II

Scenario-III Scenario-IV

50%ASC

Scenario-V

25%ASC

Scenario-VI

Figure-4.14 Effect of ASC on Passenger Served: Passenger-km as MOE

The stability of all the recommended routes in Section 4.5.1 under all the study scenarios is summarized in Figure-4.15. It is evident that all the recommended routes in Section 4.5.1 are not stable. In other words, if the true ASC is significantly different from the estimated ASC of SP based models, then all the routes recommended in Section 4.5.1 may not be viable. Accepting the uncertainties associated with the ASC, the information presented in Figure-15 may be utilized to select only the stable routes for initial operation of feeder service. Once the feeder service is operational, the ASC may be recalibrated using the then RP database and accordingly the feeder service may be extended to cover more routes or bus stop influence area.

106

Design of Feeder Service

Bus stop Scenario-I Bhasra Daihara Dantun Kalabani Khokra_1 Khokra_2 Kukai Manoharpur Nachipur Panchiyar Salajpur Sarisa Syamalpur Route Not Selected

Scenario-II Scenario-III Scenario-IV Scenario-V

Stable Route

Scenario-VI

Unstable Route

Figure-4.15 Stability of Routes with Passenger-km as MOE

The variation of average surplus earned per vehicle per day (considering all the vehicles on all feeder routes) with ASC is reported in Table-4.26. However, such calculation becomes relevant only for Scenarios-IV, V and VI with cross subsidy. It is interesting to note that even though some individual routes are unstable (Figure-4.15), the average surplus earned per vehicle per day (Table-4.26) indicates viable operation with the strength of cross subsidy. Therefore, although there are uncertainties associated with the ASC, all the routes recommended in Section 4.5.1 may be taken up for operation of feeder service for Scenarios-IV, V and VI. However, without cross subsidy (i.e. Scenarios-I,II and II), it is desirable to select only the stable routes (Figure-4.15) for initial operation of feeder service and later extend the feeder service to cover more routes or bus stop influence area based on the recalibration of ASC using the then RP data. Figure-4.15 and Table-4.26 again shows the strength of cross-subsidy in the context of rural feeder service to bus stop. Table-4.26 Effect of ASC on Revenue Surplus: Passenger-km as MOE %ASC 100 50 25

Average Revenue Surplus/ Vehicle in Rs. Scenario-IV 0.7 -2.3 4.9

Scenario-V 0.0 1.2 3.1

107

Scenario-VI 0.5 4.4 6.3

Chapter 4

4.6 FORMS OF OPERATION OF FEEDER VEHICLES

In Sections 4.4 and 4.5, feeder routes are selected assuming fixed-schedule form of operation of feeder vehicles. In this section, a comparison is made among three forms of operation of feeder vehicles namely fixed-schedule, dial-a-ride and dial-a-slot. In Sections 4.4 and 4.5, the average waiting time for commuters is assumed as half the headway of vehicles; and a heuristic approach is followed for estimating demands for feeder service. It is difficult to compare three forms of operation following a similar approach because the performance of dial-a-ride or dial-a-slot forms of operation directly depends on the time deviation of individual passengers. Therefore, passenger movements on the entire feeder routes are simulated for comparing three different forms of operation of feeder vehicles. Although feeder routes are designed for six scenarios considering two different measures of effectiveness, in the present section Scenario-IV with GC as MOE is only selected for comparison of different forms of operation of feeder vehicles. The demand for feeder service is a function of the time deviation (or waiting time), waiting discomfort, seating discomfort etc. Therefore, the demand for feeder service may be influenced by the form of operation of feeder vehicles. However, in the present analysis the demand for feeder service is assumed to remain unaltered for different forms of operation. The demand for feeder service is taken as per Section 4.4.1.4. The span of operation of feeder service is divided into two time periods: 5:30 am to 00:30 pm (morning hours), and 00:30 pm to 7:30pm (evening hours). During morning hours, the demand is largely for bus stop bound trips. Therefore, with “dial-a-ride” or “dial-a-slot”, vehicles are assumed to move to bus stop depending on the temporal demand generated at villages and return back to village after dropping off passengers at bus stop primarily to ensure the availability of vehicles for bus stop bound trips. In the evening hours, the demand is largely for village bound trips. There are inherent uncertainties associated with the time of return trips as villagers

108

Design of Feeder Service

normally go to town for several works and may not know in advance the exact time of return. Also, it may be practically difficult to introduce toll free telephone facility (required for demand responsive operations) beyond the study area. Therefore, in the evening hours, vehicles are assumed to move from bus stop once the seat capacity is exhausted, and return to bus stop after dropping off passengers in villages primarily to ensure availability of vehicles at bus stop end. In fixed-schedule form of operation, the movements in both directions are governed by the schedule only. In the process of simulating passenger movements for three forms of operation of feeder vehicles, the following additional assumptions are made. •

Layover time: 2 min at bus stop end during 5:30 am to 00:30 pm and at village end during 00:30 pm to 7:30pm for demand responsive form of operation.

•

The intended time of journey is assumed to follow a uniform random distribution. Intended journey times are generated separately for each direction of travel. The journey times are also generated separately for morning (5:30 am to 00:30 pm) and evening (i.e. 00:30 pm to 7:30 pm) periods.

A comparison of the estimated GC saving obtained from heuristic and simulation approaches with fixed form of operation of feeder vehicles is reported in Table-4.27. The estimated GC saving reported in Table-4.27 is only for the bus stop bound trips. The GC saving using heuristic approach is the same as what is reported in Section 4.4.1.4. It may be observed from Table-4.27 that lower GC saving is estimated for all the feeder routes when simulation approach is employed. In heuristic approach, the number of passengers arriving at a stop is uniformly distributed over the time period (say, morning peak or evening peak). In simulation, the intended time of journey is generated randomly following a uniform random distribution. As a result, all vehicles are not loaded uniformly. While the seat capacity is not exhausted for some vehicles, some other vehicles are overloaded with ‘congested seating’. GC is increased due to travel under congested seating.

109

Chapter 4

Moreover, due to the limited seat capacity some passengers are required to wait for the next vehicle. As a result, a lower order GC saving for all the routes is estimated when passenger movements are simulated. However, it may be mentioned that the results obtained from simulation are more realistic that the results obtained from heuristic approach. A comparison of different forms of operation of feeder vehicles in terms of GC saving for bus stop bound trips is also reported in Table- 4.27.

It is

observed from Table-4.27 that for all the routes, GC saving is enhanced when dial-a-ride form of operation is deployed over fixed-schedule. A further increase in GC saving is estimated for all the routes in a consistent manner if dial-a-slot is deployed instead. Table-4.27 Comparison of GC Saving for Bus Stop Bound Trips Bus Stop Bhasra Daihara Dantan Kalabani Khokra_I Khokra_II Kukai Manoharpur Nachipur Panchiyar Salajpur Sarisa Syamalpur All Bus Stops

Heuristic Approach Simulation Approach Fixed-Scheduled Fixed-Scheduled Dial-a-Ride Dial-a-Slot 1165 997 1210 1338 153 144 191 259 695 617 727 801 337 237 369 445 409 355 444 519 100 73 136 204 71 20 112 183 477 383 505 585 -24 -50 5 78 40 -19 75 147 512 414 547 623 222 143 235 277 17 -1 68 134 4174 3313 4624 5593 GC saving in Rupees

The GC saving for both directions of travel is also estimated and reported in Table-4.28. Table-4.28 also indicates that in terms of GC saving, “dial-aride” and “dial-a-slot” are superior to “fixed-schedule” form of operation. Between two demand responsive forms of operation, a higher GC saving is expected by operating vehicles under “dial-a slot”. The difference in GC saving among different forms of operation is primarily due to the difference in waiting discomfort. 110

Design of Feeder Service

Table-4.28 Comparison of GC Saving for Both Directions of Travel Bus Stop Bhasra Daihara Dantan Kalabani Khokra_I Khokra_II Kukai Manoharpur Nachipur Panchiyar Salajpur Sarisa Syamalpur All Bus Stops

4.7.

Fixed-Scheduled 1980 268 1239 492 657 146 23 825 -99 -16 869 306 -12 6678

GC saving in Rupees

Dial-a-Ride 2424 381 1455 742 888 272 222 1016 9 146 1098 368 120 9141

Dial-a-Slot 2601 477 1559 847 992 371 325 1123 114 249.8 1204 545 216 10624

SUMMARY

Design of rural feeder service is demonstrated in this chapter taking the necessary inputs from travel behaviour analysis in Chapter 3. Design of feeder service primary includes identification of viable feeder routes and vehicle. The required input database is developed including travel demand to bus stop, road network and cutoff revenue. Considering external subsidy and cross subsidy as two policy instruments, six hypothetical scenarios are formulated. The selection of feeder routes and vehicle under all these scenarios are demonstrated for two different measures of effectiveness namely generalized cost and passenger-km. For all the scenarios and measures of effectiveness, the stability of recommended feeder routes is also investigated. Finally, passenger movements along the feeder routes are simulated for comparing different forms of operation of feeder vehicles.

111

Chapter 5 Conclusion

5.1. INTRODUCTION The findings and conclusions drawn from the present study are reported in this chapter. A brief outline of the work is also included in the conclusion section. The conclusion section is followed by the scope of future works. 5.2. CONCLUSION Vehicle ownership is generally low in rural areas of developing countries and therefore,

travel

need

in

rural

areas

is

largely

served

by

public

transportation system. In India, all the major roads are generally served by bus transportation system. But, transportation linkage between village settlements and bus stops is a missing component. In the present work, an investigation is made on planning of rural feeder service to bus stop. Travel behaviour analysis constitutes a significant part of the work. As a part of travel behaviour analysis, a stated choice survey instrument is designed with three alternative forms of operation of feeder vehicles and two alternative vehicle types. The stated choice data collected from rural trip makers are analyzed using different logit model specifications. Willingnessto-pay (WTP) values with respect to attributes of rural feeder service are calculated from these logit models. Attempts are made to account for the effect of socio-demographic characteristics of trip makers on WTP values. Utility models and WTP values are the two major outcomes from travel behaviour analysis. Using utility models and WTP values, viable feeder routes are designed assuming fixed-schedule form of operation of vehicles. Generalized cost and passenger-km are the two alternative measures of effectiveness (MOEs) considered for the design of feeder service. Several alternative scenarios

Conclusion

are analyzed while selecting feeder routes and vehicle. Subsequently, a comparison is made among three different forms of operation of feeder vehicles. For this purpose, passenger movements along the selected routes are simulated under different forms of operation of feeder vehicles. The conclusions and observations drawn from the present work are summarized below. •

A comprehensive approach is demonstrated for the planning of rural feeder service to bus stop with due consideration to travel demand, characteristics of feeder service including type of vehicle and form of operation, behaviour of trip makers, operational viability of feeder service, and role of policy measures.

•

The need for carrying out travel behaviour analysis in the process of planning of feeder service is established. The utility models and WTP values obtained from travel behaviour analysis are used as essential inputs for the selection of feeder routes and vehicles. The estimated WTP values are also used for comparison of different forms of operation of feeder vehicles.

•

The travel behaviour analysis carried out in the present work has established/strengthened

(i)

the

importance

of

selecting

suitable

attribute(s) for stated choice experimentation, (ii) the role of model specification in analyzing stated choice data, and (iii) the need for investigating the effect of socioeconomic attributes on WTP estimates. The specific experiences and observations related to travel behaviour analysis are enumerated below.

113

Chapter 5

¾

The stated choice method is applied for eliciting preferences of people taking a case study in rural India. The literacy rate in the study area is only 58% (Census 2001, Govt. of India). The pilot survey initially revealed difficulties of respondents in choosing one out of seven alternatives given in a choice set. However, when a sequential approach is followed (refer Section 3.2.3), all respondents were comfortable in stating their choices. The responses thus obtained from rural commuters are also found consistent. The experience indicates that stated choice method may be applied for elicitation even in rural areas of developing countries. Sequential approach might become useful in such cases, if the number of alternatives present in a choice set is more.

¾

The qualitative aspects of travel are not given due consideration in planning and operation of transportation services in developing countries like India. Small vehicles like tempo or trekker carrying more passengers than the seat capacity, overcrowding of buses, and poor vehicle condition are the common features of transportation system in India. Under this circumstance, it is interesting to note rural trip makers’ willingness-to-pay (WTP) for a change in travel condition from ‘congested seating’ to ‘comfortable seating’. The present work justifies the need for (i) including qualitative attributes in SP experiment, (ii) estimating users’ WTP with respect to those attributes, and (iii) including them while estimating user benefits or carrying out improvement planning of transportation system.

¾

An analysis of stated choice data by Multinomial Logit (MNL), Nested Logit (NL), Covariance Heterogeneity Nested Logit (CHNL) and Random Parameter Logit (RPL) model specifications clearly indicate that WTP estimates may vary depending on model specification. In the present work, less restrictive RPL model specification is generally found to produce higher order values of attributes. RPL models capture some unobserved effects and such effects are generally more

114

Conclusion

correlated with attributes other than the cost attribute. Therefore, higher order values may be obtained from RPL model specification. Also, the unobserved effects may not be correlated uniformly with all the non-cost attributes. In the present work, such unobserved effects are more correlated with access walking distance and accordingly, the maximum change in WTP estimate is obtained with respect to access walking distance.

¾

The valuing of time deviation under different levels of waiting discomfort indicates that relaxed waiting at stop (representing dial-aride form of operation) in comparison to anxious waiting at stop (representing fixed-schedule form of operation) is considered as an improvement. Relaxed waiting at home (representing dial-a-slot form of operation) is considered as a further improvement. The valuing of time deviation clearly reflects trip makers’ preference towards demand responsive services in comparison to fixed-schedule form of operation.

Among

demand

responsive

services,

dial-a-slot

is

preferred over dial-a-ride form of operation.

¾

The WTP values may be influenced by socioeconomic and trip characteristics of trip makers. Therefore, it is necessary to account for such effects while estimating WTP values. In the present work, preference heterogeneity against observed socio-economic and trip characteristics are investigated around the mean estimate(s) of random parameter(s) in RPL models. The study revealed statistically significant decomposition effect of ‘household monthly income’ and ‘present access mode’ on the mean estimates of waiting discomfort, seating discomfort and access walking distance.

¾

It

is

necessary

to

assume

suitable

distribution(s)

of

random

parameter(s) while developing RPL models. In the present work, all the RPL models are developed assuming constrained triangular (CT)

115

Chapter 5

distribution of random parameter(s). The use of CT distribution is advantageous

particularly

to

ensure

the

correct

sign

in

the

distribution of attribute values. The RPL models developed with CT distribution also indicated acceptable goodness of fit. The present work

may

encourage

advantageously

over

researchers

other

commonly

to

use

used

CT

distribution

distributions

while

developing RPL models.

¾

There is little information available in the established literature about valuing of travel attributes by rural trip makers in developing countries like India. Though context specific, the values of travel attributes as reported in the present work, are expected to be of interest to the research community.

•

It is necessary to choose an appropriate measure of effectiveness (MOE) for the selection of feeder routes and vehicle. The MOE primarily reflects the objective of providing feeder service in a quantitative manner. The selection of MOE is therefore, a policy matter.

Two alternative MOEs

namely generalized cost (GC) and passenger-km are considered. The routes and vehicle are selected to optimize the MOE (i.e. maximize the GC saving or the passenger-km served). It is shown that outcomes (i.e. routes, vehicles, etc.) are influenced by the MOE. The role of policy in terms of the selection of an appropriate MOE is demonstrated in the present work.

•

The work demonstrates that using policy instruments like ‘cross subsidy’ or ‘external subsidy’, it is possible to further optimize the MOE in the process of selecting feeder routes and vehicle. The ‘cross subsidy’ is shown as a powerful policy instrument which should be incorporated in the planning of rural feeder service. The benefit likely to be derived from external subsidy depends on the amount of external subsidy. However, a combination of external subsidy and cross subsidy is always more 116

Conclusion

beneficial than external subsidy alone. “First cross subsidy and then direct subsidy” is shown as a rational policy for optimizing the MOE in the context of rural feeder service.

•

In the absence of RP data, utility equations developed from SP data alone have been used for demand estimation. The alternative specific constants (ASCs) present in such utility equations may result biased estimation of demand. Therefore, the operational viability of the recommended routes is checked assuming 50% and 25% of the ASC values in utility equations. Accordingly, routes are classified as ‘stable’ or ‘unstable’. A similar approach may be followed in other cases where SP based models are used for demand estimation. In the present case, only the stable routes may be selected for initial operation of feeder service. Once the feeder service is operational, the ASCs may be recalibrated using the then RP database and accordingly, the service may be extended to cover more routes or bus stop influence areas.

•

The simulation of passenger movements under different forms of operation of feeder vehicles indicates that the demand responsive forms of operation are preferred over the conventional fixed-schedule form of operation. The findings may encourage a shift from the traditional fixedschedule to demand responsive form of operation in rural areas where the demand is scattered over a large geographic area. Between the two demand responsive operations investigated in the present work, the diala-slot is expected to produce more benefits to users than the dial-a-ride.

5.3. FUTURE SCOPES OF THE WORK •

In the development of RPL models, preference heterogeneity study has been carried out around the mean estimate(s) of random parameter(s). An investigation can be made on the heterogeneity effect around the variance/spread

of

random

parameter(s),

interactions in WTP values. 117

and

account

for

such

Chapter 5

•

All the RPL models have been developed assuming a constrained triangular

distribution

for

random

parameter(s).

As

triangular

distribution approximates the look of normal distribution, a constrained version of normal distribution with an expectation to get more shares in positive distribution of the values of attribute may also be investigated.

•

A linear fare structure (i.e. fare is linearly proportional to distance traveled) has been assumed while selecting feeder routes and vehicles. The selection of routes and vehicles may be attempted assuming other types of fare structure (e.g. fixed fare, telescopic fare etc.) and the fare structure optimizing the MOE may be recommended.

•

For successful operation of feeder service, it is necessary to have a proper institutional framework. The role of institutional framework becomes more pertinent if cross subsidy or direct subsidy is introduced. Further

works

are

necessary

to

develop

a

suitable

institutional

framework considering all aspects related to the operation of rural feeder service.

118

REFERENCES Adamowicz, W. L., Louviere, J. and Wiiliams M. (1994). “Combining Stated and Revealed Preference Methods for Valuing Environmental Amenities.” Journal of Environmental Economics and Management, 26: pp. 271-292. Adamowicz, W., Louviere, J. and Swait. J. (1998). “Introduction to attribute-based stated choice methods.” Final Report submitted to National Oceanic and Atmospheric Administration, U.S. Department of Commerce. Algers, S., Bergström, P., Dahlberg, M. and Johanna, L. D. (1998). “Mixed Logit Estimation of the Value of Travel Time.” Working paper series, Department of Economics, Uppsala University, Sweden. Allenby, G. and Rossi, P. (1999). “Marketing models for consumer heterogeneity.” Journal of Econometrics, 89, pp. 57-78. Alpizar, F. and Carlsson, F. (2001). “Policy Implications and Analysis of the Determinants of Travel Mode Choice: An Application of Choice Experiments to Metropolitan Costa Rica.” Working Paper in Economics 56, Department of Economics, Göteborg University. Amador, F. J., González, R. M. and Dios Ortúzar, J. D. (2005). “Preference Heterogeneity

and

Willingness

to

Pay

for

Travel

Time

Savings.”

Transportation, 32(6), pp. 627-647. Andrews, R. L., Ansari, A. and Currim, I. S. (2002). “Hierarchical Bayes versus finite mixture conjoint analysis models: A comparison of fit, prediction and partworth recovery.” Journal of Marketing Research, 39, pp. 87-98. Arne, R. H. and Felix, R. FitzRoy. (2004). “Commuting in small towns in rural areas: the case of St Andrews.” Working paper, Department of Economics, University of St Andrews.

References

Baaj, M.H. and Mahamassani, H.S. (1990). “TRUST: a LISP program for analysis of transit route configurations.” Transportation Research Record 1283, 125–135. Baaj, M.H. and Mahmassani, H.S. (1991). “An AI-based approach for transit

route

system

planning

and

design.”

Journal

of

Advanced

Transportation, 25 (2), pp. 187–210. Baaj, M.H. and Mahmassani, H.S. (1995). “Hybrid Route Generation Heuristic Algorithm for the Design of Transit Networks.” Transportation Research C, 3 (1), 31–50. Ben-Akiva, M. and Lerman, S. R. (1985). Discrete Choice Analysis: Theory and Applications to Travel Demand. MIT Press. Bennett, J. and Blamey, R. K. (2001). The Choice Modelling Approach to Environmental Valuation. Edward Elgar Publishing Limited, UK. Barros, C. P. and Proença, I. (2005). “Mixed Logit Estimation of Radical Islamic Terrorism in Europe and North America.” Journal of Conflict Resolution, Vol. 49, (2), pp. 298-314. Basu, D. and Maitra, B. (2007). “Valuing travel time and its variation level projected as traffic information on a roadside VMS board: a SP approach.” International Journal of Transport Economics, 34(2), pp. 225-243. Bateman, I.J., Langford, I.H., Munro, A., Starmer, C. and Sugden, R. (2000). “Estimating four Hicksian welfare measures for a public good: a contingent valuation investigation.” Land Economics, 76(3), pp. 355-373. Bates, J. (1982). “Stated preference technique for the analysis of transportation

behavior.”

Proceedings

of

World

Conference

of

Transportation Research, Hamburg, W. Germany, pp. 252-265. Bhat, C. R. (1995). “A heteroscedastic extreme value model of intercity travel mode choice.” Transportation Research B, 29(6): pp. 471-483. 120

References

Bhat, C. (1997). “Covariance Heterogenety in Nested Logit Models: Econometric Structure and Application to Inter City travel.” Transportation Research B, 31, pp. 11-21. Bhat, C. (1998a). “Analysis of Travel Mode and Departure Time Choice for Urban Shopping Trips.” Transportation Research B, 32: pp. 361-371. Bhat, C. R. (1998b). “Accommodating variations in responsiveness to levelof-service measures in travel mode choice modeling.” Transportation Research A, Vol. 32, pp. 495-507. Bhat,

C.R.

(2001).

“Quasi-random

maximum

simulated

likelihood

estimation of the mixed multinomial logit model.” Transportation Research B, 35, pp. 677-695. Blamey, R., Bennett, J., Louviere, J., Morrison, M. and Rolfe, J. (2000). “A test of policy labels in environmental choice modeling studies.” Ecological Economics, 32, pp. 269-286. Blamey, R. K., Bennett, J. W., Louviere, J. J., Morrison, M. D. and Rolfe, J. C.

(2002).

"Attribute

causality

in

environmental

choice

modelling."

Environmental and Resource Economics, vol. 23, no. 2, pp. 167-186. BoÈ rsch-Supan, A. (1990). “On the Compatibility of Nested Logit Models with Utility Maximization.” Journal of Econometrics, 32, pp. 371-387. Bradley, M. A. and Gunn, H. (1990). “Stated preference analysis of values of travel time in the Netherlands.” Transportation Research Record, 1285, pp. 78-89. Brownstone, D. (2000). “Discrete choice modeling for transportation, Resource Paper.” 9th International Association of Travel Behaviour Research Conference, Gold Coast, Australia, 1-7 July.

121

References

Brownstone, D. and Train, K. (1999). “Forecasting New Product Penetration with Flexible Substitution Patterns.” Journal of Econometrics, Vol. 89, pp. 109-129. Brownstone, D., Bunch, D. and Train, K. (2000). “Joint mixed logit models of

stated

and

revealed

preferences

for

alternative-fuel

vehicles.”

Transportation Research B, 34, pp. 315–338. Carlsson, F. (1999). “The Demand for Intercity Public Transport: The Case of Business Passengers.” Working Paper, Department of Economics, Göteborg University. Carlsson, F. and Martinsson, P. (2004). “Does it Matter When a Power Outage Occurs? - A Choice Experiment Study on the Willingness to Pay to Avoid

Power

Outages.”

Working

Paper,

Department

of

Economics,

Göteborg University. Carlsson, F., Frykblom, P. and Liljenstolpe, C. (2003). “Valuing Wetland Attributes: An Application of Choice Experiments.” Ecological Economics, 47, pp. 95–103. Carlsson, F., Frykblom, P. and Lagerkvist, C. J. (2004). “Consumer benefits of labels and bans on genetically modified food - An empirical analysis using Choice Experiments.” Working Paper, Department of Economics, Göteborg University. Carow, K. A. and Staten, M. E. (1999). “Debit, Credit, or Cash: Survey Evidence on Gasoline Purchases.” Journal of Economics and Business, 51: pp. 409–421. Carrier,

E.

(2003).

“Modeling

Airline

Passenger

Choice:

Passenger

Preference for Schedule in the Passenger Origin-Destination Simulator (PODS).”

Thesis Master of Science in Transportation, Massachusetts

Institute of Technology. Carson, R., Louviere, J.J., Anderson, D.A., Arabie, P., Bunch, D.S., 122

References

Hensher, D.A., Johnson, R.M., Kuhfeld, W.F., Steinberg, D., Swait, J., Timmemans, H. and Wiley, J.B. (1994). “Experimental analysis of choice.” Market. Lett., 5, pp. 351–368. Ceder,

A.

and

Wilson,

N.

H.

M.

(1986).

“Bus

network

design.”

Transportation Research Part A, 20(4), pp. 331–344. Chattopadhyay, C. (2000). “The effectiveness of McFaddens’s nested logit model in valuing amenity improvement.” Regional Science and Urban Economics, 30, pp. 23–43. Cherchi, E. and Polak, J. W. (2005). “The Assessment of User Benefits Using Discrete Choice Models: Implications of Specification Errors Under Random Taste Heterogeneity.” Prepared for presentation at 85th annual Meeting of the Transportation Research Board, Washington, D.C. Cherchi, E. and Ortu´zar, J. D. (2006). “On fitting mode specific constants in the presence of new options in RP/SP models.” Transportation Research Part A, 40, pp. 1–18. Chien, S., Yang, Z. and Hou, E. (2001). “Genetic algorithm approach for transit route planning and design.” Journal of Transportation Engineering, ASCE, 127(3), pp. 200–207. Chintagunta,

P.,

Jain,

D.

and

Vilcassim,

N.

(1991).

“Investigating

heterogeneity in brand preference in logit models for panel data.” Journal of Marketing Research, 28, pp. 417–428. Dhingra, S.L. (1980). “Simulation of routing and scheduling of city bus transit network.” Ph.D. Thesis, IIT Kanpur, India. Ding-Ming Wang (2001). “The Impacts of Policy Issues on Voting Behavior in Taiwan: A Mixed Logit Approach.” Journal of Electoral Studies, vol. 8(2): pp. 95-123. Dubois, D., Bel, G. and Llibre, M., (1979). “A set of methods in 123

References

Transportation Network synthesis and Analysis.” Journal of Operation Research Society, 30 (9), pp. 797–808. Duff-Riddell, W. R. and Bester, C. J. (2005). “Network Modeling Approach to Transit Network Design.” Journal of Urban Planning and Development, ASCE, pp. 87-97. Fan, W. and Machemehl, R.B. (2004). “Optimal transit route network design problem: Algorithms, implementations, and numerical results.” Report No. SWUTC/04/167244-1, Center for Transportation Research, University of Texas at Austin. Garrett, G. (2001). “Mixed Logit Models for Multiparty Elections.” Political Analysis, 9:1, pp. 12-49. Geok, K. and Perl, J. (1988). “Optimization of feeder bus routes and bus stop spacing.” Journal of Transportation Engineering, 114 (3), pp. 341– 354. Green P. E. and Srinivasan, V. (1990). “Conjoint analysis in consumer research: issues and outlook.” Journal of Consumer Research, 5, pp. 103123. Green, P. E. Krieger, A. M. and Wind, Y. (Jerry) (2001). ‘Thirty Years of Conjoint Analysis: Reflections and Prospects.’ INTERFACES 31:3, Part 2 of 2, May-June. pp. S56-S73. Greene, W. H. (2000). Econometric Analysis. Fifth edition, Prentice Hall Publication. Greene, W.H. and Hensher, D.A. (2003). “A latent class model for discrete choice analysis: Contrasts with mixed logit.” Transportation Research Part B, 37(8), pp. 681-698. Greene, W. H., Hensher, D. A. and John, R. (2006). “Accounting for heterogeneity in the variance of unobserved effects in mixed logit models.” 124

References

Transportation Research B, 40(1), pp. 75-92. Haaijer, E. M. (1999). “Modeling Conjoint Choice Experiments with the Probit Model.” Unpublished Ph.D thesis. Haefen, R. H. V. (2003). “Incorporating observed choice into the construction of welfare measures from random utility models.” Journal of Environmental Economics and Management, 45: pp. 145–165. Hanemann, M. (1984). “Welfare evaluations in contingent valuation experiments with discrete responses.” American Journal of Agricultural Economics, 66, 332– 341. Hanley, N., Wright, R.E. and Adamowicz, V. (1998). “Using Choice Experiments to Value the Environment.” Environmental and Resource Economics, 11(3-4), pp. 413-428. Hasselström, D. (1981). “Public transportation planning—a mathematical programming approach.” Doctoral thesis, Univ. of Gothenburg, Sweden. Hensher, D. A. (1994). “Stated Preference Analysis of Travel Choices: the State of Practice.” Transportation, 21(2), pp. 107-133. Hensher, D. A. (1998). “HEV choice models as a search engine for specification of nested logit tree structure.” Marketing Letter, 10(4),pp. 333-343. Hensher, D. A. (2000). “Service quality as a Package: What does it mean to

heterogeneous

consumers?”

Institute

of

Transport

Studies,

The

University of Sydney, Australia. Hensher, D. A. (2001a). “The Valuation of Commuter Travel Time Savings for

Car

Drivers

in

New

Zealand:

Evaluating

Alternative

Model

Specifications.” Transportation, 28, pp. 101-118. Hensher, D. A. (2001b). “Measurement of the Valuation of Travel Time

125

References

Savings.” Journal of Transport Economics and Policy, 35(1), pp. 71- 98. Hensher, D. A. (2001c). “The sensitivity of the valuation of travel time saving to the specification of unobserved effects.” Transportation Research part E, 37: pp. 129-142. Hensher D. A. (2004). “Identifying the influence of stated choice design dimensionality on willingness to pay for travel time saving.” Journal of Transport Economics and Policy, 38(3): pp. 425-446. Hensher, D. A. (2006). “Reducing sign violation for VTTS distributions through recognition of an individual’s attribute processing strategy.” Working Paper ITLS-WP-06-13, Institute of Transport and Logistics Studies, The University of Sydney. Hensher, D. A. and Greene W. H. (2001). “The Mixed Logit Model: The State of Practice and Warnings for the Unwary.” Working Paper, School of Business, The University of Sidney. Hensher, D. A., and Greene, W. H. (2002). “Specification and Estimation of Nested Logit Model: Alternative Normalization.” Transportation Research B, 36, pp.1-17. Hensher, D. A. and Greene, W. H. (2003). “The mixed logit model: The state of practice.” Transportation, 30(2), pp. 133-176. Hensher, D. A., and Sullivan, C. (2003). “Willingness to Pay for Road Curviness and Road Type.” Transportation Research Part D, Vol. 8, pp. 139-155. Hensher, D. A. and Rose, J. M. (2005). “Respondent behaviour in discrete choice modeling with a focus on the valuation of travel time savings.” Journal of Transportation and Statistics, US Department of Transportation, Vol. 8(2), pp. 17-30. Hensher, D., Shore, N., and Train, K. (2004). “Households’ willingness to 126

References

pay

for

water

service

attributes.”

Working

paper,

Department

of

Economics, University of California, Berkeley. Hensher, D. A., Rose, J. M. and Greene, W.H. (2005). Applied Choice Analysis: A Primer. Cambridge University Press. Hensher, D. A., Rose, J. and Bertoia, T. (2007). “The implications on wiliness to pay of a stochastic treatment of attribute processing stated choice studies.” Transportation Research Part E, 43, pp. 73-89. Hess, S. and Axhausen, K. W. (2004). “Checking our assumptions in value of travel time modeling: Recovering taste distributions.” CTS Working Paper, Centre for Transport Studies, Imperial College London. Hess, S., Bierlaire, M. and Polak, J. W. (2005). “Estimation of value of travel-time saving using mixed logit models.” Transportation Research Part A, 39: pp. 221-236. Holguin-Veras, J. (2002). “Revealed Preference analysis of Commercial Vehicle Choice Process.” Journal of Transportation Engineering, ASCE, 128 (4), pp. 236-346. Howarth, A., Pearce, D.W. Özdemiroglu, E., Seccombe-Hett, T., Wieringa, K., Streefkerk, C.M. and de Hallander, A.E.M. (2001). “Valuing the benefits of environmental policy: The Netherlands.” RIVM report 481505 024 (Bilthoven: RIVM). Hsu, J. and Surti, V.H. (1976). “Demand Model for bus network design.” Transportation Engineering, Journal ASCE, 102, TE3; Proceeding Paper 12309, pp. 451–460. Huber, J. (1987). “Conjoint Analysis: How We Got Here and Where We Are—an Update.” Proceedings of Sawtooth Software Conference. Huber, J. and Zwerina, K. (1996). "The importance of utility balance in efficient choice designs." Journal of Marketing Research, vol. 33 (3), pp. 127

References

307-317. Huber, J. and Train, K. E. (2001). “On the similarity of classical and Bayesian estimates of individual mean partworths.” Marketing Letters, 12, pp. 257-267. Hunt, J. D. (2001). “A stated preference analysis of sensitivities to elements of transportation and urban form.” Paper presented at the 80th Annual Meeting of the Transportation Research Board. Iles, R. (2005). Public Transport in Developing Countries. Elsevier Publication. Kamakura, W. A. and Srivastava, R. (1984). “Predicting Choice Shares Under

Conditions

of

Brand

Interdependence.”

Journal

of

Marketing

Research, 21, November, pp. 420-34. Kato, H., Kaneko, Y. and Inoue, M. (2005).

“Empirical analysis on

relationship between types of travel demand techniques and estimated user’s benefit stemming from transportation investment.” Journal of the Eastern Asia Society for Transportation Studies, Vol. 6, pp. 3937 – 3947. Khasnabis, S. and Tom, V. M. (2006). “A single-stage genetic algorithm model for transit fleet resource allocation.” Joint International Conference on Computing and Decision Making in Civil and Building Engineering, Montreal, Canada. Kish, L. (1965). Survey Sampling. Wiley, New York. Koppelman, F.S. and Wen, C.H. (1998). “Alternative nested logit models: structure, properties and estimation.” Transportation Research B, 32, (5), pp. 289-298. Koppelman, F. S., Sethi, V. and Wen, C. (2001). “Alternative Tree Logit Models: a Response to Comments by Andrew Daly on an earlier Paper of Frank Koppelman and Chieh-hua Wen.” Transportation Research B, 35, pp. 128

References

725-729. Koppleman, F. S. and Sethi, V. (2005). “Incorporating variance and covariance heterogeneity in the Generalized Nested Logit model: an application

to

modeling

long

distance

travel

choice

behavior.”

Transportation Research Part B, 39, pp. 825-853. Kroes, E. P. and Sheldon, R. J. (1988). “Stated preference methods: an introduction.” Journal of Transport Economics and Policy, Volume 22(1): pp. 11-25. Lai, K., and Wong, W. (2000). “SP approach toward driver comprehension of message formats on VMS.” Journal of Transportation Engineering, 126(3): pp. 221–227. Lahiri, K. and Gao, J. (2001). “Bayesian analysis of nested logit model by Markov chain Monte Carlo.” Working Paper, Department of Economics, State University of New York at Albany, NY. Lampkin, W. and Saalmans, P. D. (1967). “The design of routes, service frequencies and schedules for a municipal bus undertaking: a case study.” Operation Research Quarterly, 18 (4), pp. 375–397. Layton, D. F. (2000). “Random Coefficient Models for Stated Preference Surveys.” Journal of Environmental Economics and Management, 40, pp. 21-36. LIMDEP 8.0 (2005) Users’ guide for LIMDEP 8.0, Econometric Software, Inc. Lo, H. K., Yip, C. W. and Wan, Q. K. (2004). “Modeling competitive multimodal transit services: a nested logit approach.” Transportation Research Part C, 12, pp. 251–272. Louviere, J. J. (1988a). “Conjoint Analysis Modeling of Stated Preferences: A Review of Theory, Methods, Recent Developments and External Validity.” 129

References

Journal of Transport Economics and Policy, 22(1), pp. 93-119. Louviere, J. J. (1988b). “Analyzing Decision Making: Metric Conjoint Analysis.” Sage University Papers, Series No. 67, Newbury Park, CA. Louviere, J. J. and Woodworth, G. (1983). “Design and analysis of simulated

consumer

choice

for

allocation

experiments.”

Journal

of

Marketing Research, 20(4), pp. 350-367. Louviere, J. J and Timmermans, H. (1990). "A review of recent advances in decompositional preference and choice models." Journal of Economic and Social Geography, Vol. 81 No.3, pp. 214-24. Louviere, J. J., Hensher, D. A. and Swait. D. J. (2000). Stated Choice Methods. Analysis and Applications. Cambridge University Press. MacDonald, D. H., Barnes, M., Bennett, J., Morrison, M. and Michael D. Y. (2003). “What Consumers Value Regarding Water Supply Disruptions: A Discrete

Choice

Analysis.”

From

www.cmis.csiro.au/Mary.Barnes/PDF/

WTP_200305.pdf accessed in August 2006. Maitra, B., Sarkar, J. R., and Sikdar, P. K. (2002). “Road Transportation in India: Emerging Issues and Challenges.” Proc. of National Seminar on Road Transportation in India: Emerging Trends and Techniques (ROTRAN 2002), IIT Kharagpur, pp. 4.87-4.94. Maitra, B., Sarkar, J. R., and Sikdar, P. K. (2003). “Procurement and Institutional Strengthening for Road development in India”, Proc. of International Workshop and Conference on Construction Management and Materials (CONMAT 2003), IIT Kharagpur, pp. 207-215. Mandl, C. E. (1980). “Evaluation and Optimization of Urban Public Transport Networks.” European Journal of Operational Research,

6, pp.

31–56. Martins, C. L. and Pato, M. V. (1998). “Search strategies for the feeder bus 130

References

network design problem.” European Journal of Operational Research, 106 , pp. 425-440. Mccarthy, P. S. and Tay, R. S. (1998). “New vehicle consumption and fuel efficiency: a nested logit approach.” Transportation Research Part E, 34, pp. 39–51. McCulloch, R. and Rossi, P. (1994). “An exact likelihood analysis of the multinomial probit model.” Journal of Econometrics, 64, pp. 207-240. McFadden, D. (1974). “Conditional Logit Analysis of Qualitative Choice Behavior.” In P. Zarembka (ed.), Frontiers in econometrics, New York: Academic Press, pp. 105-142. McFadden, D. (1978). “Modelling the choice of residential location.” Transportation Research Record, 672, pp. 72-77. McFadden, D. L. (1981). “Econometric Models of Probabilistic Choice.’ in Structural Analysis of Discrete Data.” Manski, C.F. and McFadden, D.L. (eds.) MIT Press, Cambridge Massachusetts, pp. 198-271. McFadden, D. and Train, K. (2000). “Mixed MNL models of discrete response.” Journal of Applied Econometrics, 15, pp. 447–470. Mellisa C. Ruby., Reed Johnson, F., and Kristy E. Mathews. (1998). “Just Say No: Opt-Out Alternatives And Anglers’ Stated Preferences.” Working Paper, Triangle Economic Research, Durham. Montgomery, M. (2002). “A nested logit model of the choice of a graduate business school.” Economics of Education Review, 21, pp. 471–480. Morokoff, W., and Caflisch, R. (1995). “Quasi-Monte Carlo integration.” Journal of Computational Physics, Vol. 122, pp. 218-230. Onyango, B., Govindasamy, R. and Rodolfo M. Nayga, Jr. (2004), “Measuring U.S. Consumer Preferences for Genetically Modified Foods

131

References

Using Choice Modeling Experiments: The Role of Price, Product Benefits and Technology.” Paper prepared for presentation at the American Agricultural Economics Association Annual Meeting, Denver, Colorado, August 2004. Ortuzar, J. D. and Willumsen, L. G. (2002). Modelling Transport. 3rd edition, John Wiley & Sons, Chichester, New York, USA. Pattnaik, S. B., Mohan, S. and Tom, V. M. (1998). “Urban bus transit network

design using genetic

algorithm.” Journal

of

Transportation

Engineering, ASCE, 124(4), pp.368–375. Pels, E., Nijkamp, P. and Rietveld, P. (2000). “Airport and Airline Competition for Passengers Departing from a Large Metropolitan Area.” Journal of Urban Economics, 48, pp. 29-45. Phanikumar, C. V. and Maitra, B. (2007). “Willingness to Pay and Preference

Heterogeneity

for

Rural

Bus

Attributes.”

Journal

of

Transportation Engineering, ASCE, 133 (1), pp. 62-69. Pommerehne, W. W. (1988). “Measuring the environmental benefits: a comparison of hedonic technique and contingent valuation.” In: Welfare and Efficiency in Public Economics edited by Bos, D., Rose, M. and Seidl, C. (Berlin: Springer-Verlag), pp. 363-400. Praveen, K. S. and Rao. K. V. K. (2002). “Estimation of passenger demand on a proposed passenger water transport system.” Proceedings of Transportation Planning and Methodologies for Developing Countries, IITBombay, Mumbai. Ramirez, A. I. and Seneviratne, P. N. (1996). “Transit route design applications using GIS.”

Transportation Research Record, 1557, pp. 10–

14. Revelt, D. and Train, K. (1997). “Mixed logit with Repeated Choices: Households’

Choices

of

Appliance

Efficiency

132

Level,”

Working

Paper,

References

Department of Economics, University of California at Berkeley. Revelt, D. and Train, K. (1998). “Mixed logit with repeated choices: Households’ choices of appliance efficiency level.” Review of Economics and Statistics 80, pp. 647-657. Revelt, D. and Train, K. (2000). “Specific taste parameters and mixed logit.” Working Paper No. E00-274, Department of Economics, University of California, Berkeley. Rigby, D. and Burton, M. (2003). “Capturing Preference Heterogeneity in Stated Choice Models: A Random Parameter Logit Model of the Demand for GM Food.” Discussion paper series, School of Economic Studies, The University of Manchester. Robin Naidoo and Adamowicz, L. W. (2003). “Biodiversity and Nature based Tourism: The Potential for Sustainable Development in Uganda.” Staff

paper,

Department

of

Rural

Economy,

University

of

Alberta

Edmonton, Canada. Ruud, P. (1996). “Approximation and simulation of the multinomial probit model: An analysis of covariance matrix estimation.” Working Paper, Department of Economics, University of California, Berkeley. Ryan, M. (1999). "A role for conjoint analysis in technology assessment in health care", International Journal of Technology Assessment in Health Care, vol. 15, no. 3, pp. 443-457. Sawtooth Software (1999) The CBC/HB module for hierarchical Bayes estimation. Shih, M., Mahmassani, H. S. and Baaj, M. H. (1998). ‘‘Planning and design model

for

transit

route

network

with

coordinated

operations.’’

Transportation Research Record 1623, Transportation Research Board, Washington, D.C., pp. 16–23.

133

References

Shrivastava, P. and Dhingra, S.L. (2001). “Development of feeder routes for suburban railway stations using heuristic approach.” Journal of Transportation Engineering, ASCE, 127 (4), 334–341. Shrivastava, P. and O’Mahony, M. (2006). “A model for development of optimized feeder routes and coordinated schedules—A genetic algorithms approach.” Transport Policy, 13, pp. 413–425. Sikdar, P. K. (2002). “Pradhan Mantri Gram Sadak Yojana, For the People, By the People.” Indian Highways, June, pp 59-68. Sillano, M. and Ortuzar, J. de D. (2005). “Willingness-to-pay estimation with mixed logit models: Some new evidence.” Environmental and Planning A, Vol. 37, pp. 525-550. Silman, L. A., Brazily, Z. and Passy, U., (1974). “Planning the route system for urban buses.” Computer and Operation Research, 1, pp. 201–211. Spasovic, L. N., Boile, M. P. and Bladikas, A. K. (1994). “Bus transit service coverage for maximum profit and social welfare.” Transportation Research Record, 1451, pp. 12–22. SPSS 7.5 Users’ guide for SPSS 7.5, SPSS Inc, Chicago, USA. Street, D. J., Leonie Burgess, L. and Louviere, J. J. (2005). “Quick and easy choice sets: Constructing optimal and nearly optimal stated choice experiments. “ Intern. J. of Research in Marketing, 22, pp. 459–470. Thurstone, L. L. (1927). “A law of comparative judgment.” Psychological Review, 4: pp. 273–286. Tom, V.M. and Mohan, S. (2003). “Transit route network design using frequency coded Genetic Algorithm.” Journal of Transportation Engineering, ASCE, 129 (2), pp. 186–195. Train, K. E. (1993). “Qualitative Choice Analysis: Theory Econometrics, and

134

References

an Application to Automobile demand.” MIT Press. http://elsa.berkeley .edu/~train/books.html. Train, K. (1998). “Recreation demand models with taste differences over people.” Land Economics, pp. 230-239. Train, K. (1999). “Halton Sequences for Mixed Logits.” Working Paper, Department of Economics, University of California, Berkeley. Train, K. (2001). “A comparison of hierarchical Bayes and maximum simulated likelihood for mixed logit.” Working Paper, Department of Economics, University of California, Berkeley. Train, K. (2003). Discrete Choice Methods with Simulation. Cambridge University Press. http://elsa.berkeley.edu/~train/books.html. Train, K. and Sonnier, G. (2003) “Mixed logit with bounded distributions of partworths”, Working paper, Department of Economics, University of California, Berkeley. Train, K. and Sonnier, G. (2004). “Mixed logits with bounded distributions of correlated partworths.” In A. Alberini and R. Scarpa, Eds., Applications of Simulation methods in Environmental and Resource Economics, Kluwer Academic Publishers, Boston, M.A. Train, K. and Winston, C. (2004). “Vehicle choice behavior and the declining market of us automakers.” Working paper, Department of Economics, University of California, Berkeley. VanNes, R., Hamerslag, R. and Immer, B.H. (1988). “Design of public transport networks.” Transportation Research Record, 1202, pp. 74–83. Wei, W. and Hansen, M. (2005). “Impact of aircraft size and seat availability on airlines’ demand and market share in duopoly markets.” Transportation Research Part E, 41, pp. 315–327.

135

References

Wirasinghe, S.C. (1980). “Nearly optimal parameters for a rail feeder bus system on a rectangular grid.” Transportation Science, 14A (1), pp. 33–40. Wojcik, C. (2000). “Alternative models of demand for automobiles.” Economics Letters, 68, pp. 113–118. Yai T., Iwakura, S. and Morichi, S. (1997). “Multinomial probit with structured covariance for route choice behavior.” Transportation Research B, 31 (3): pp. 195-207. Yao, E. and Morikawa, T. (2003). “A Study on Integrated Intercity Travel Demand Model.” Conference paper, 10th International Conference on Travel Behaviour Research, Lucerne, pp. 10-15. Yates, F. (1981). Sampling methods for census and survey. 4th edn. Griffin, London. Yun, Dae-Sic, Lee, Jeong-Yeob, and Sinha, K. C. (2000). “Modeling Prework

Trip-Making

and

Home

Departure

Time

Choice.”

Journal

of

Transportation Engineering, ASCE, Vol. 126(4), pp. 308-312. Zhao, F. and Zeng, X. (2006). “Optimization of transit network layout and headway with a combined genetic algorithm and simulated annealing method.” Journal of Engineering Optimization, 38 (6), pp. 701–722.

136

ANNEXURE-A ECONOMETRIC MODELS

A.1 RANDOM UTILITY THEORY Let, the deterministic component be denoted by V and the random (disturbance) component be denoted by ε then, the Utility of an element may be expressed as following (Ben-Akiva and Lerman 1985) U=V+ε

(A.1)

The deterministic part V is again a function of the observed attributes (z) of the choice as faced by the individual, the observed socioeconomic attributes of the individual (S), and a vector of parameters (β). Then, V = V (z, S, β) A probabilistic statement can be made (due to presence of the random component) as, when an individual ‘n’ is facing a choice set, Cn, consisting of Jn choices, the choice probability of alternative ‘i' is equal to the probability that the utility of alternative ‘i’, Uin, is greater than or equal to the utilities of all other alternatives in the choice set i.e., Pn (i) = Pr (Uin ≥ Ujn, for all j є Cn) Pn (i) = Pr (Vin + εin ≥ Vjn + εjn, for all j є Cn, j ≠i)

(A.2)

The deterministic component of the utility function can be expressed as Vin = β1Xin1 +β2Xin2+…+βzXinz Where, Vin is the deterministic component of utility function β1, β2…, βz are the coefficients associated with attributes, and Xin1, Xin2,..., Xinz are the attributes describing the alternative A.2 MULTINOMIAL LOGIT MODEL The random utility model in Equation A.2 can be transformed into choice model by an assumption regarding the joint distribution about the vector of random error terms. If the random error terms are assumed to follow extreme value type I (Gumbel) distribution, and be independently and

Annexure-A

identically distributed (IID) across alternatives and cases (or observations), the multinomial (or conditional) logit (MNL) model (McFadden 1974) is obtained. In the MNL model, the probability that an individual ‘n’ chooses alternative ‘i’ can be given by (McFadden 1974; Ben-Akiva and Lerman 1985),

Pin =

eVin

(A.3)

∑e

Vin

j∈ jn

A.3 NESTED LOGIT In Nested Logit Model, the set of alternatives are divided (and sub-divided) into exclusive groups (nests) where some aspects only pertains to members of that particular group. The NL model arises as a random utility model in which the random component of utility has the generalized extreme value distribution. Each observed (or representative) component of the utility expression for an alternative (Vi for ith alternative) is defined in the term of four parts: the parameters associated with explanatory variable (β), an alternative

specific

constant

( α i ),

a

scale

parameter

(θ),

and

the

explanatory variable (x). The utility of alternative ‘i’ for individual ‘n’ is given as follows. Uni = gi(αi , βxni , ε ni ) = gi (V ni , εni ) = αi + βxni + εni

(A.4)

Var[ε ni ] = σ 2 = κ /θ 2 The scale parameter (θ) is proportional to inverse of the standard deviation (σ) of the random component in the utility expression and critical input into the set up of the NL model (Ben-Akiva and Lerman 1985; Louviere et al. 2000). Another justification for moving from MNL to NL model is to recognize the possibility that standard deviations (or variances) of the random error components in the utility expression are different across group of alternatives in the choice set. This arises because the sources of utility associated with the alternatives are not fully accommodated in Vn. The missing sources of utility may differentially impact the random components across the alternatives, resulting different variances. To accommodate the possibility of differential variances, scale parameters 138

Econometric Models

must be introduced in the utility expressions. If all scale parameters are equal, then NL model collapses to MNL. To be consitent with utility maximization, the scale parameters at highest level and the ratios of scale parametrs at each lower nest are bounnded by zero and one. The same observed set of choices emerges regardless of the (common) scaling of the utilities. Hence, the latent variance is normalized at one, not as a restriction, but of necessity for identification (Hensher and Greene 2002). Normalization is also required (Koppelman and Wen 1998; Koppelman et al. 2001; Hensher and Greene 2002) for random utility maximization. Normalization is simply the process of setting one or more scale parameters equal to unity, while allowing the other scale parameters to be estimated. The utility maximizing nested logit (UMNL) model is a special case of the Generalized Extreme Value (GEV) model (McFadden 1978; 1981) which ensures that it is consistent with utility maximization, provided that the ratio of the scale parameters is bounded appropriately. For expressing the details of three levels nested logit (NL) models, following notational expression are followed (Hensher and Greene 2002). It is useful to represent each level of NL tree by unique descriptor. The top level of the tree is repented by limbs, the middle level by a number of brances and the bottom level by a set of elemental alternatives, or twigs (Figure-A.1).

A

typical three level model has i = 1,…….,I elemental alternatives, j = 1,…….,J branch

composite

alternatives,

and

k

=

1,…….,K

limb

composite

alternatives. So, the notation i/j,k used to denote alternative i in branch j of limb k and j/k denotes branch j in limb k. Define parameter vectors in the utility functions at each level as follows: β for elemental alternatives, γ for branch composite alternatives, and δ for limb composite alternatives. The branch level composite alternative involves an aggregation of the lower level alternatives.

As discussed below, a branch specific scale parameter

µ(j|k) is associated with the lowest level of the tree.

Each elemental

alternative in the j’th branch actually has scale parameter µ′(i|jk). Since these will, of necessity, be equal for all alternatives in the same branch, the distinction by i is meaningless. As such, it may be represented as µ(j|k). The

139

Annexure-A

parameters λ(j|k) will be associated with the branch level. The inclusive value (IV) parameters at the branch level will involve the ratios λ(j|k)/µ(j|k). The IV parameters associated with the IV variable in a branch, calculated from the natural logarithm of the sum of the exponentials of the Vi expressions at the elemental alternative level directly below a branch, I| jk

IV ( j | k ) = log ∑ exp(α l| jk + β 'x l | jk )

(A.5)

l =1

have associated parameters defined as λ(j|k/µ(j|k)), but, some normalization is required.

Normalization is simply the process of setting one scale

parameters equal to unity, while estimating other scale parameters. Some analysts do this without acknowledgment of the specific normalization they have used, which makes the comparison of reported results among studies a difficult task. One approach restricts the numerator of λ(j|k)/µ(j|k) to be equal to one, while the other approaches restrict the denominator. When µ(j|k) is normalized to unity, it is called as Random Utility Model 1 (RU1), and when λ(j|k) is normalized to unity, it is called as Random Utility Model 2 (RU2). RU1 is used in the present work and discussed below. The choice probabilities for the elemental alternatives are defined as: P(i| j ,k )n =

exp[α i| jk + β ' x i| jk ] I| j ,k

∑ l =1

=

exp[α l| jk + β ' xl| jk ]

exp[α i| jk + β ' x i| jk ] exp[IV j|k ]

(A.6)

where, i|jk = elemental alternative i in branch j of limb k, I|jk = number of elemental alternatives in branch j of limb k, and the inclusive value for branch j in limb k is I| jk

IV j|k = log ∑ exp[α i| jk + β ' x i| jk ]

(A.7)

i =1

The branch level probability is P( j|k )n =

exp{λ j|k [γ ' y j|k + IV j|k ]} J|k

∑ exp{λm|k [γ ' ym|k + IVm|k ]}

=

exp{λ j|k [γ ' y j|k + IVj|k ]} exp[IVk ]

(A.8)

m=1

where, j|k = branch j in limb k, J|k = number of branches in limb k, and 140

Econometric Models

J|k

IVk = log ∑ exp{λ j|k [γ ' y j|k + IVj|k ]}

(A.9)

j =1

Finally, the limb level is defined by P(k )n =

exp{ρ k [δ ' zk + IVk ]} I

∑ exp{ρ [δ ' z n

n =1

n

=

+ IVn ]}

exp{ρk [δ ' zk + IVk ]} exp(IV )

(A.10)

where, K= number of limbs in the three level tree and K

IV = log ∑ exp{ρ k [δ ' zk + IVk ]} k =1

By law of probability, the unconditional probability of the observed choice (mode) made by an individual ‘n’ will be Pin = P(k )n P( j / k )n P(i / j ,k )n

(A.11)

Travel Limb

Branch Twig

Limb

Branch Twig

Twig

Branch Twig

Twig

Branch Twig

Twig

Figure-A.1 Typical Tree Structure for Nested Logit Model A.4 COVARIANCE HETEROGENEITY NESTED LOGIT

Allowing heterogeneity across individuals in the covariance of nested alternatives in the estimation of NL model leads to the development of Covariance Heterogeneity Nested Logit model (Bhat 1997). CHNL model in a two level allows the ‘ λ j ’, the branch level inclusive value parameters, to be function of a set of attributes, Vj, in the form

λ j * = λ j × exp[δ 'V j ]

(A.12)

δ ' is a new vector parameter to be estimated

141

Annexure-A

Since the inclusive value parameter is a scaling parameter for a common random component in the alternatives within the branch; it is equivalent to a model of heteroscedasticity. Vj may be attributes which are assumed to be same for all alternative in the branch, ‘j’ and should not contain any constant. The CHNL model considers the case in which the parameters of the inclusive values exhibit a systematic relationship with some socioeconomic characteristics of the decisions makers. Since the coefficients of the inclusive values related to the ratio of the scale parameters for the lower and upper levels determine the sensitivity of choice between the alternatives in the nested branch and the others in the tree, the CHNL model enables the explicit consideration of the role of socioeconomic variables (say, income) in determining the cross elasticity of choice. Higher values of the coefficients of the inclusive value terms imply higher cross elasticity of choice. In this context, a direct relationship between the explanatory variable in the covariance heterogeneity term would lead to increased cross elasticity.

Due to introduction of additional variables and

incorporation of the covariance structure, this model is statistically and behaviorally superior to the corresponding NL model. A.5 RANDOM PARAMETER LOGIT

Revisiting the basic utility function it can be said that, the utility associated with each alternative ‘i’, as evaluated by each individual ‘n’ is represented in discrete choice model by a utility expression as given below (Hensher et al. 2005; Greene et al. 2006) K

U in = β n X in +ε in = ∑ β nk xink +ε in

(A.13)

k =1

where, X in is the full vector of explanatory variables that are observed by the analyst, which may include attributes of alternatives, observed socioeconomic characteristics of the individuals, etc. In RPL model, the components β n and ε in are not observed by the analyst and are treated as stochastic influence in model development process.

142

Econometric Models

Individual heterogeneity of preferences are introduced through interactions between the alternative’s attributes and individual observed socioeconomic characteristics such as sex, age, family income or even trip specific characteristics such as purpose, travel conditions, etc. (Revelt and Train 1998; Ortuzar and Willumsen 2002). Within the logit context, ε in

is

conditioned that it is Independent and Identically Distributed (IID) extreme value

type-I

across

individuals

and

alternatives

(also

even

choice

situations). One way, this condition can be relaxed and can be introduced by accounting it into the utility function through

β n that may be

heteroskedastic and/or correlated across alternatives. The components β n and ε in are not observed by the analyst and are treated as stochastic influences. β n is assumed to vary across individuals both randomly and systematically with observable variables z n . If the random parameters are assumed to be uncorrelated, then the model for β n may be written as

β n = β + ∆z n + ∑1 / 2 vn = β + ∆z n + η n or β nk = β k + δ k' z n + η nk

(A.14)

where, β nk is the random coefficient for the kth attribute faced by individual ‘n’.

The

term

β + ∆z n accommodates heterogeneity in the mean of the

distribution of the random parameters. η nk

is a random term whose

distribution over individuals depends in general on underlying parameters and z n is individual specific observed data. For convenience in isolating the model components vn is defined, which is to be a primitive vector of uncorrelated random variables with known variances. The actual scale factors which provide the unknown standard deviations of the random parameters are then arrayed on the diagonal of the diagonal matrix ∑1 / 2 .

143

Annexure-A

The RPL class of models assumes a general distribution for η nk and an IID extreme value type I distribution for ε in . That is, η nk can take on different distributional forms such as Normal, Log-normal, Uniform, Triangular, Constrained-triangular, etc. Denote the joint density of [η n1 ,η n 2 ,......,η nk ] by

f (η n | Ω, z n ) , where the elements of Ω are the underlying parameters of the distribution of β n ( β , ∆, ∑) and z n is the observed data specific to the individuals such as socioeconomic and other trip specific characteristics, and

η n denotes a vector of ‘K’ random components in the set of utility function in addition to the J random elements in ε i n for all i ∈ J . For a given value of η n , the conditional probability for choice ‘i’ is multinomial logit, since the remaining error term is IID extreme value

Lin ( β n | X n ,η n ) =

exp( β n' xin ) J

∑ exp(β i =1

(A.15)

' n in

x )

Equation (A-1.15) is the simple MNL model, but with the provison that for each sampled individual additional information is defined as η n . This is where the use of the word ‘conditional’ applies and so the probability is conditional on η n (and z n ). This additional information influences the choice outcome. The unconditional choice probability is the expected value of the logit probability over all the possible values of β n , that is integrated over these values, weighted by the density of β n (Hensher and Green 2003; Train 2003). Therefore, the unconditional probability is

Pin ( X n , z n , Ω) = ∫ Lin ( β n | X n ,η n ) f (η n | z n , Ω)dη n

(A.16)

βn

Thus, the unconditional probability that individual ‘n’ will choose alternative ‘i’ given the specific characteristics of their choice set and the underlying model parameters, is equal to the expected value of the conditional probability as it ranges over the possible values of β n . The random variation

144

Econometric Models

in β n is induced by the random vector η n , hence that is the variable of integration in Equation (A.16). Models of this form are called Mixed Logit because the choice probability is a mixture of logits with f as the mixing distribution. The choice probabilities will not exhibit the IIA property. It is requited to mention that the standard deviation of an element of the

β n accommodates the presence of

unobservable preference heterogeneity in the sampled population (i.e. it allows for individuals within the sampled population to have different β n as opposed to a single β representing the entire sample population).

The unconditional choice probability as given in Equation (A.16) cannot be calculated exactly because the integral is not closed form. Instead, the probability is approximated through simulation (Brownstone and Train 1999). Maximization is then conducted on the simulated log-likelihood function (Train 2001). The integral in Equation (A.16) is approximated through simulation. For a given value of parameters, Ω , and the observed data z n a value of

β n is drawn from its distribution based on Equation

(A.14). Using this draw, the logit formula (as given in Equation A.15) for

Lin ( β n ) is calculated. This process is repeated for many draws, and the mean of the resulting

Lin ( β n ) s is taken as the approximated choice

probability

SPin ( X n , z n , Ω) =

1 R ∑ Lin (β nr | X n ,η nr ) R r =1

(A.17)

SPin is an unbiased estimator of Pin, where R is the number of replications (i.e. draws of β nr ). β nr is the rth draw and SPin is the simulated probability that an individual ‘n’ chooses alternative ‘i’. By construction, SPi is a consistent estimator of Pi for any draw R; its variation decreases as R increases. It is strictly positive for any R, so that ln(SPi) is always defined in a log-likelihood function. It is smooth (i.e. twice

145

Annexure-A

differentiable) in parameters and variables, which helps in the numerical search for the maximum of the likelihood function. The simulated probabilities can then be inserted into the log-likelihood function to get a simulated log-likelihood N

J

SSL = ∑∑ d in ln(SPin )

(A.18)

n =1 i =1

Where, d in = 1 , if individual ‘n’ chooses alternative ‘i’ and zero otherwise. Maximization is then conducted on the simulated log-likelihood function by conventional numerical methods such as quadratic hill climbing or gradient methods, etc. across a sample of n=1, 2,…., N individuals. Hence, it is called Maximum Simulated Likelihood (MSL). Suppose an individual ‘n’ is facing ‘T’ number of choice situations (t=1, 2… T), then simulated log-likelihood can be written as follows. N

SSL = ∑ ln n =1

T 1 sumrR=1 ∏ Lint ( β nr | X nt ,η nr ) R t =1

(A.19)

By the above method one could go for estimates with the mean and standard deviation/spread of parameter distributions over the population. But sometimes researchers may also be interested in getting the estimates at ‘individual-specific preferences’ by deriving the individual’s conditional distribution based (within-sample) on their known choices (i.e. prior knowledge). This happens when a researcher is willing to get ‘commonchoice-specific’ parameter estimates. These parameter estimates are strictly ‘same-choice-specific’ parameters, or the mean of the parameters of the sub-population of individuals who, when faced with the same choice situation would hence made the same choices. This is an important distinction since it is not possible to establish, for each individuals, their unique set of estimates, but rather identify a mean and a standard deviation estimates for the sub-population who make the same choice. The method to construct “individual-specific-preferences” is given below.

146

Econometric Models

For convenience, let Yn denote the observed information on choices by individual ‘n’ and X n denote all elements of xin for all

for all i ∈ J . , then

using Bayes’ Rule, one can find the conditional density for the random parameters:

H ( β n | Yn , X n , z n , Ω) =

f (Yn | β n , X n , z n , Ω) P( β n | z n , Ω) f (Yn | X n , z n , Ω)

(A.20)

In the numerator of the right-hand side, the first term gives the probability in the conditional likelihood- this is in Equation (A.15). The second term gives the probability density for the random β n given in Equation (A.14) with the assumed distribution of η n . The denominator is the unconditional choice probability – this is given by Equation (A.16). It can be noted from the above Equation that the denominator is the integral of the numerator. This result will be used to estimate the common choice-specific-parameters, utilities, choice probabilities as function of the underlying parameters of the distributions of the random parameters. The above random parameter models are less restrictive than standard conditional logit models. However, these less restrictive models should be applied cautiously. Apart from being more difficult to estimate, literature shows that the results can be rather sensitive to the distributional assumptions and the number of “smart” draws applied in the simulation (Hensher and Greene 2001). The method described above is called the classical estimation procedure. There is another method called Hierarchical Bayes

estimation

procedure,

which

has

also

undergone

remarkable

development in recent years (McCulloch and Rossi 1994; Allenby and Rossi 1999; Sawtooth Software 1999; Huber and Train 2001; Lahiri and Gao 2001; Andrews et al. 2002). By this procedure also one can estimate parameters over population and individual level. Mixed Logit (RPL) can approximate any choice model including any MNP model (McFadden and Train 2000). The reverse cannot be said i.e. the MNP model cannot approximate ML models since MNP relies critically on normal

147

Annexure-A

distributions. If a random term in utility is not normal, then ML can handle it but MNP cannot. The RPL model engenders a relatively free utility structure such that IIA is relaxed despite the presence of the IID assumption for the random components of the alternatives. That is, the RPL model disentangles IIA from IID and enables the analyst to estimate models that account for cross-correlation

among

the

alternatives.

When

the

random

taste

parameters are all zero, the exact MNL model is produced. A.5.1 Preference Heterogeneity around the Mean of a Random Parameter

Introducing an interaction between the mean estimate of the random parameter and an observed covariate is equivalent to revealing the presence or absence of preference heterogeneity around the mean parameter estimate. This is the role of the term δ k' z n in Equation (A.14). This is not the same as the standard deviation of the parameter estimate associated with a random parameter. If the interaction is not statistically significant then it can be concluded that there is an absence of preference heterogeneity around the mean on the basis of observed covariates. This does not imply that there is no preference heterogeneity around the mean, simply that, model is failed to reveal its presence. This then means that the analyst relies fully on the mean and the standard deviation of the parameter estimate,

with

the

latter

representing

all

sources

of

preference

heterogeneity (Hensher and Greene 2003; Hensher et al. 2005) in the sampled population. A.5.2 Distribution of Random Parameters

In RPL model, it is necessary to make an assumption regarding the distribution of random parameters. This assumption causes much concern in RPL model development process. In the following sections, various distributions used for the development of RPL models, are discussed.

148

Econometric Models

A.5.2.1 Normal Distribution

The Normal (Gaussian) distribution is one of the popular and commonly used distributions in RPL model. This priori assumption makes the coefficient estimate without a strict sign. The Normal distribution with density function is given by

f (ξ ) = and

it

is

1

σ 2π

−

e

defined

(ξ − µ ) 2 2σ 2

(A.21)

ξ ∈ (−∞, +∞) for

on

all

values

of

µ (mean)

and

σ (sandard deviation) . The Normal distribution is unbounded and so every real number has a positive probability of being produced as a draw; specifying a given coefficient to follow a normal distribution is thus equivalent to making a priori assumption that both positive and negative values for the coefficient exist in the population. Normal distributions have infinite tails, which would require that some individuals have implausible (near-infinite) coefficient values. A.5.2.2 Lognormal Distribution

The lognormal distribution is very popular. If a large number of random shocks, some positives, some negatives, change the size of a particular attribute, say x, in an additive fashion, the distribution of that attribute will tend to become normal as the number of shocks increases. But if these shocks act multiplicatively, changing the value of x by randomly distributed proportions instead of absolute amount, the central limit theorems applied to Y=ln(x) tend to produce a normal distribution. A variable ξ follows a Log-normal distribution if its logarithm is normally distributed. The domain of the distribution is the space of strictly positive real numbers, and with ln(ξ ) ~ N ( µ N , σ N ) , then

f (ξ ) =

1

σ N ξ 2π

e − (ln(ξ ) − µ N )

2

/ ( 2σ N2 )

(A.22)

The mean and standard distribution of the Log-normal distribution can be obtained as follows

µ ( LN ) = e

µN +

σ N2 2

149

Annexure-A

2 N

σ ( LN ) = µ ( LN ) e σ − 1 With

µ N = 0 and σ N = 1, the Lognormal distribution reduces to Gibrat’s

distribution. The Lognormal distribution has been explored in many RPL model development process. It performed well in some of the investigations (Train and Sonnier 2003). Although in some cases Lognormal is found appealing but it is limited to the non-negative domain; however it typically has a very long right-hand tail which is a disadvantage with an overestimation in standard deviation. Few researchers have seen it as being preferable to Normal distribution in the case of coefficients with a strong priori sign assumption, such as cost and time coefficients. A.5.2.3 Rayleigh Distribution

The Rayleigh distribution probability function is given by

P (ξ ) =

ξe

−

ξ2 2s2

(A.23)

s2

For ξ ∈ [0, ∞). , where s is the desired scale parameter. The mean is centered as

s×

π 2

and the standard deviation is

4 −π × s 2 . This distribution has a 2

long tail, but empirically appears much less extreme than lognormal (Hensher and Rose 2005; Hensher 2006). The earlier investigations with this distribution showed interesting results in terms of getting positive sign in value of time distribution. A.5.2.4 Uniform Distribution

The Uniform distribution is the most basic statistical distribution; it assigns constant probability to all values included in its domain of definition. A variable ξ is uniformly distributed on [a, b] then,

(A.24) The spread of the uniform distribution (i.e. the distance up and down from the mean) and the standard deviation are different. Suppose ‘s’ is the 150

Econometric Models

spread, then the standard deviation of the random variable that is uniformly distributed

between

(mean-s)

and

(mean+s)

is

σ = (Upper Limit − Lower Limit ) /(2 / 3 ) = s / 3 . Normally uniform distribution is considered for dummy parameters. But it is interesting to note that, Amador et al. (2005) obtained a better model fit considering travel time as uniformly distributed parameter. The Uniform distribution can be used for coefficients with a priori sign assumption by constraining either the lower or the upper bound to zero, leading to positive, respectively negative draws only. This distribution has rarely been used in the RPL model specification, given that it assigns equal probability to all values in its domain and thus not allow for a peak in the distribution at the population mode. However, models based on the Uniform distribution are generally very easy to estimate, such that the Uniform distribution can at least be seen as first step in the identification of coefficients for which significant random heterogeneity exists in the population. A.5.2.5 Triangular Distribution

For the triangular distribution, the density function looks like a tent: a peak in the center and dropping off linearly on both sides of the center. Let ‘c’ be the center and s be the spread. The density starts at (c-s), rises linearly to its value at ‘c’, and then drops linearly to zero again at (c+s). It is zero below (c-s) and above (c+s). The mean and mode are ‘c’. The standard deviation is the spread divided by 6 . The height of the tent at c is 1/s (such that each side of the tent has area s × (1 / s ) × (1 / 2) = 1 / 2, and both sides have area ½+1/2=1. The slope is 1/s2. Sometimes Triangular distribution can be seen as an approximation to the Normal distribution by researchers, with finite bounds, and with linearly decreasing probabilities either side of the mode. It avoids the long tails like the Normal distribution has. A.5.2.6 Johnson’s SB distribution

Johnson’s SB distribution is also being investigated in recent times (Train and Sonnier 2004). The SB distribution can be obtained as a logit-like

151

Annexure-A

transformation of the normal distribution, and with a ξ ~ N ( µ ,σ ), a draw from SB is given by

c = a + (b − a ) ×

eξ , eξ + 1

(A.24)

where the shape of the distribution depends on the choice µ and σ , and where c is bounded between a and b. The SB has a major advantage over the bounded distributions in that it can be used to approximate a number of different distributions, with bounds on both sides, and it can also replicate Beta distributions. Further more, it can be specified to be symmetrical or asymmetrical, it can have a tail to the left or the right, its density can take the shape of a fairly flat plateau with drop-offs on either side. While the SB distribution is very flexible, its use leads to a need to estimate four parameters. Furthermore, while its performance in terms of bounds is generally very good, its performance in terms of the mean and standard deviation is highly dependent on the shape of the true distribution, and in some cases, it can lead to significant bias in these measures (Hess and Axhausen 2004). A.5.2.7 Discrete Distribution

In contrast to continuous distribution (such as normal, lognormal etc.), discrete distribution is also used. Such a distribution may be viewed as a nonparametric estimator of the random distributions. Using a discrete distribution that is identical across individuals is equivalent to a latent segmentation model with the probability of belonging to a segment being only a function of constants (Louviere et al. 2000). However, allowing this probability to be a function of individual attributes is equivalent to allowing the points characterizing the nonparametric distribution to vary across individuals. Greene and Hensher (2003) contrast a latent class model with Mixed Logit.

152

ANNEXURE-B Recommended Feeder Routes and Vehicle

Table-B.1 Scenario-II: GC as MOE Bus stop Bhasra

%ASC Passenger Vehicles Passenger Served Km 100 5339 20 757 50 5088 19 718 25 4842 18 680

GC Surplus*/ Route Saving Vehicle Stability 1285 Stable 0.8 1214 Stable 2.8 1146 Stable 4 225 Stable 11.9 201 39.5 Stable 171 Stable 23

Daihara

100 50 25

1243 1147 1046

5 4 4

212 196 178

Dantun

100 50 25

3215 3053 2892

12 12 11

483 456 430

772 727 682

4.5 -3.5 4

Stable Stable Stable

Kalabani

100 50 25

1844 1759 1601

5 5 4

372 353 318

345 319 274

10.7 4.7 37.5

Stable Stable Stable

Kukai

100

940

3

289

71

Unstable

50 25

866 721

3 2

265 220

58 17

-0.3 -23.7 24.2

Unstable Unstable

Manoharpur

100 50 25

2229 2057 1887

9 9 8

497 455 414

479 433 390

0.1 -12.8 -7.7

Unstable Unstable Unstable

Salajpur

100 50 25

2290 2196 2018

6 6 5

485 461 417

482 455 402

27.5 19.7 48.9

Stable Stable Stable

Sarisa

100 50 25

645 581 455

2 2 2

238 214 167

-2 -10 -45

2.5 -27.2 -67.3

Unstable Unstable Unstable

Syamalpur

100

648

2

208

17

16.2

Unstable

50 25

583 457

2 2

187 146

-5 -37

-4.8 -56.2

Unstable Unstable

* Earning in excess of cutoff revenue in rupees

Annexure-B

Table-B.2 Scenario-III: GC as MOE Bus stop Bhasra

%ASC Passenger Vehicles Passenger Served Km 100 5292 11 774 50 5239 11 765 25 4917 11 744

GC Surplus* Route Saving /Vehicle Stability 2491 Stable 8 2462 Stable 7.6 2320 Stable 3.5 454 13.7 Stable 445 Stable 5.6 417 Stable 0

Daihara

100 50 25

Dantun

100

3170

7

494

1489

2.3

Stable

50 25

3136 3068

7 7

488 476

1469 1433

3.7 0.2

Stable Stable

Kalabani

100 50 25

1758 1634 1507

7 7 6

362 334 305

396 361 324

18 6.4 20.3

Stable Stable Stable

Khokra_2

100 50 25

945 858 767

4 4 4

216 193 170

162 141 112

2 -16.9 -27.9

Unstable Unstable Unstable

Kukai

100 50 25

830 728 625

4 4 3

276 242 208

128 98 68

9.2 -10.9 11.1

Unstable Unstable Unstable

Manoharpur 100 50 25

2229 2057 1887

9 9 8

497 455 414

479 433 390

20.1 7.2 12.3

Stable Stable Stable

Salajpur

100 50 25

2003 1862 1716

9 8 7

471 433 395

505 463 418

2.1 10.2 22

Stable Stable Stable

Syamalpur

100 50 25

613 528 439

3 3 3

199 171 142

75 45 19

3.2 -16.4 -40.5

Unstable Unstable Unstable

1216 1192 1143

3 3 3

219 215 206

* Earning in excess of cutoff revenue in rupees

154

Recommended Feeder Route and Vehicle

Table-B.3 Scenario-IV: GC as MOE

Bus stop Bhasra

Daihara

Dantun

Kalabani

Khokra_1

Khokra_2

Kukai

Manoharpur

Nachipur

Panchiyar

Salajpur

Sarisa

Syamalpur

GC Surplus* Route %ASC Passenger Vehicles Passenger Served Saving /Vehicle Stability Km 100 5276 11 771 1165 86.4 Stable 50 5110 11 746 1119 76.4 Stable 25 4816 11 699 1035 60.4 Stable 100 1210 3 218 153 70.8 Stable 50 1138 3 205 135 51.9 Stable 25 1002 3 181 99 17.1 Stable 100 3160 7 493 695 73.8 Stable 50 3056 7 475 663 67.6 Stable 25 2867 6 443 614 92.1 Stable 100 1802 5 372 337 -2.1 Unstable 50 1718 5 353 311 -9.1 Unstable 25 1561 4 317 266 23.2 Unstable 100 1605 5 428 409 -50.8 Unstable -57.8 Unstable 50 1531 5 408 381 -30.6 Unstable 25 1392 4 369 332 100 986 3 224 100 -33.8 Unstable 50 920 3 206 85 Unstable -51 25 789 2 174 46 10.7 Unstable 100 940 3 289 71 -20.3 Unstable 50 866 3 265 58 -43.7 Unstable 25 721 2 220 17 Unstable 4.2 100 2367 7 513 477 -55.7 Unstable 50 2263 7 488 444 -57.3 Unstable 25 2068 6 440 390 -42.8 Unstable 100 542 2 193 -24 -46.7 Unstable 50 479 2 170 -42 -67.5 Unstable 25 345 1 121 -75 -0.3 Unstable 100 796 3 281 40 Unstable -49 50 726 3 256 21 -65.7 Unstable 25 591 2 207 -15 -30.7 Unstable 100 2492 7 485 512 -26.2 Unstable 50 2394 6 461 485 Unstable 1 25 2213 6 419 432 Unstable -9.2 100 609 3 234 222 -98.7 Unstable 50 559 3 213 194 -106.2 Unstable 25 507 3 193 169 -117.3 Unstable 100 648 2 208 17 -3.8 Unstable 50 583 2 187 -5 -24.8 Unstable 25 457 2 146 -37 -76.2 Unstable * Earning in excess of cutoff revenue in rupees

155

Annexure-B

Table-B.4 Scenario-V: GC as MOE Bus stop Bhasra

%ASC Passenger Vehicles Passenger Served Km 100 5276 11 771 50 5110 11 746 25 4816 11 699

GC Surplus* Route Saving /Vehicle Stability 1165 106.4 Stable 1119 96.4 Stable 1035 80.4 Stable

Daihara

100 50 25

Dantun

100

3160

7

50 25

3056 2867

7 6

Kalabani

100 50 25

1678 1627 1571

7 7 6

369 356 342

816 788 757

Khokra_1

100 50 25 100

1605 1531 1392 980

5 5 4 5

428 408 369 221

409 381 332 430

-30.8 -37.8 -10.6 -100

Unstable Unstable Unstable Unstable

50 25

941 902

4 4

210 199

407 392

-79.1 -89.3

Unstable Unstable

Kukai

100 50 25

940 866 721

3 3 2

289 265 220

71 58 17

-0.3 -23.7 24.2

Unstable Unstable Unstable

Manoharpur

100 50 25

2367 2263 2068

7 7 6

513 488 440

477 444 390

-35.7 -37.3 -22.8

Unstable Unstable Unstable

Nachipur

100 50 25

542 479 345

2 2 1

193 170 121

-24 -42 -75

-26.7 -47.5 19.7

Unstable Unstable Unstable

Panchiyar

100 50 25

796 726 591

3 3 2

281 256 207

40 21 -15

Unstable Unstable Unstable

Salajpur

100 50 25 100 50 25 100 50 25

2492 2394 2213 609 559 507 648 583 457

7 6 6 3 3 3 2 2 2

485 461 419 234 213 193 208 187 146

512 485 432 222 194 169 17 -5 -37

-29 -45.7 -10.7 -6.2 21 10.8 -78.7 -86.2 -97.3 16.2 -4.8 -56.2

Khokra_2

Sarisa

Syamalpur

1210 1138 1002

3 3 3

218 205 181

153 135 99

90.8 71.9 37.1

Stable Stable Stable

493

695

93.8

Stable

475 443

663 614

87.6 Stable 112.1 Stable -59.8 Unstable -62.5 Unstable -46.6 Unstable

* Earning in excess of cutoff revenue in rupees

156

Unstable Unstable Unstable Unstable Unstable Unstable Unstable Unstable Unstable

Recommended Feeder Route and Vehicle

Table-B.5 Scenario-VI: GC as MOE Bus stop Bhasra

%ASC Passenger Vehicles Passenger Served Km 100 5276 11 771 50 5110 11 746 25 4816 11 699

GC Surplus* Route Saving /Vehicle Stability 1165 126.4 Stable 1119 116.4 Stable 1035 100.4 Stable

Daihara

100 50 25

Dantun

100

3153

11

491

1552

110.8 Stable 91.9 Stable 57.1 Stable -7.9 Unstable

50 25

3098 3038

11 11

481 471

1526 1526

-12.8 -14.3

Unstable Unstable

Kalabani

100 50 25

1678 1627 1571

7 7 6

369 356 342

816 788 757

Unstable Unstable Unstable

Khokra_1

100 50 25 100

1605 1531 1392 980

5 5 4 5

428 408 369 221

409 381 332 430

-39.8 -42.5 -26.6 -10.8 -17.8 9.4 -80

Unstable Unstable Unstable Unstable

50 25

941 902

4 4

210 199

407 392

-59.1 -69.3

Unstable Unstable

Kukai

100 50 25

940 866 721

3 3 2

289 265 220

71 58 17

19.7 -3.7 44.2

Unstable Unstable Unstable

Manoharpur

100 50 25

2367 2263 2068

7 7 6

513 488 440

477 444 390

-15.7 -17.3 -2.8

Unstable Unstable Unstable

Nachipur

100 50 25

542 479 345

2 2 1

193 170 121

-24 -42 -75

-6.7 -27.5 39.7

Unstable Unstable Unstable

Panchiyar

100 50 25

796 726 591

3 3 2

281 256 207

40 21 -15

Unstable Unstable Unstable

Salajpur

100 50 25 100 50 25 100 50 25

2277 2215 2147 609 559 507 648 583 457

10 9 9 3 3 3 2 2 2

480 462 443 234 213 193 208 187 146

1121 1087 1052 222 194 169 17 -5 -37

-9 -25.7 9.3 -58.4 -47.8 -51.1 -58.7 -66.2 -77.3 36.2 15.2 -36.2

Khokra_2

Sarisa

Syamalpur

1210 1138 1002

3 3 3

218 205 181

153 135 99

* Earning in excess of cutoff revenue in rupees

157

Unstable Unstable Unstable Unstable Unstable Unstable Unstable Unstable Unstable

Annexure-B

Table-B.6 Scenario-I: Passenger-km as MOE

Bus stop Bhasra

Daihara

Dantun

Kalabani

Manoharpur

Salajpur

Syamalpur

GC Surplus* Route %ASC Passenger Vehicles Passenger Served Saving /Vehicle Stability Km 100 5430 13 771 1188 23.4 Stable 50 5259 12 745 1141 35.7 Stable 25 4964 11 700 1058 45.8 Stable 100 1277 3 218 165 68.1 Stable 50 1203 3 205 146 Stable 50 25 1067 3 182 110 18.1 Stable 100 3273 8 493 714 21.8 Stable 50 3165 7 475 682 52.7 Stable 25 2976 7 444 633 Stable 33 100 1802 5 372 337 Unstable -2.1 50 1718 5 353 311 -9.1 Unstable 25 1561 4 317 266 23.2 Unstable 100 2293 6 513 425 16.1 Stable 50 2180 6 485 393 Stable 8.4 25 1976 5 436 340 30.8 Stable 100 2290 6 485 482 Stable 7.5 50 2196 6 461 455 -0.3 Stable 25 2018 5 417 402 28.9 Stable 100 642 2 208 9 11.6 Unstable 50 576 2 187 -12 -11.8 Unstable 25 447 2 145 -42 -65.5 Unstable * Earning in excess of cutoff revenue in rupees

158

Recommended Feeder Route and Vehicle

Table-B.7 Scenario-II: Passenger-km as MOE

GC Surplus* Route %ASC Passenger Vehicles Passenger Served Saving /Vehicle Stability Km Bhasra 100 5430 13 771 1188 43.4 Stable 50 5259 12 745 1141 55.7 Stable 25 4964 11 700 1058 65.8 Stable 100 1277 3 218 165 Daihara 88.1 Stable 50 1203 3 205 146 Stable 70 25 1067 3 182 110 38.1 Stable 100 3273 8 493 714 Dantun 41.8 Stable 50 3165 7 475 682 72.7 Stable 25 2976 7 444 633 Stable 53 100 1902 5 372 354 Unstable Kalabani -0.1 50 1817 5 353 327 -4.8 Unstable 25 1658 5 318 283 -19.5 Unstable 100 940 3 289 71 Kukai -0.3 Unstable 50 866 3 265 58 -23.7 Unstable 25 721 2 220 17 24.2 Unstable Manoharpur 100 2293 6 513 425 36.1 Stable 50 2180 6 485 393 28.4 Stable 25 1976 5 436 340 50.8 Stable Salajpur 100 2290 6 485 482 27.5 Stable 50 2196 6 461 455 19.7 Stable 25 2018 5 417 402 48.9 Stable 100 645 2 238 -2 Unstable Sarisa 2.5 50 581 2 214 -10 -27.2 Unstable 25 455 2 167 -45 -67.3 Unstable 100 650 2 208 10 Syamalpur 16.1 Unstable 50 583 2 187 -11 -5.3 Unstable 25 453 2 145 -42 -56.1 Unstable

Bus stop

* Earning in excess of cutoff revenue in rupees

159

Annexure-B

Table-B.8 Scenario-III: Passenger-km as MOE %ASC Passenger Vehicles Passenger Served Km Bhasra 100 5430 13 771 50 5259 12 745 25 4964 11 700 100 1277 3 218 Daihara 50 1203 3 205 25 1067 3 182 100 3273 8 493 Dantun 50 3165 7 475 25 2976 7 444 100 1902 5 372 Kalabani 50 1817 5 353 25 1658 5 318 Khokra_2 100 986 3 224 50 920 3 206 25 789 2 174 Kukai 100 940 3 289 50 866 3 265 25 721 2 220 Manoharpur 100 2293 6 513 50 2180 6 485 25 1976 5 436 100 2639 7 485 Salajpur 50 2538 7 461 25 2355 6 419 100 645 2 238 Sarisa 50 581 2 214 25 455 2 167 Syamalpur 100 660 2 208 50 593 2 187 25 462 2 145

Bus stop

GC Surplus* Route Saving /Vehicle Stability 1188 63.4 Stable 1141 75.7 Stable 1058 85.8 Stable 165 108.1 Stable 146 Stable 90 110 58.1 Stable 714 61.8 Stable 682 92.7 Stable 633 Stable 73 354 19.9 Unstable 327 15.2 Unstable 283 Unstable 0.5 6.2 100 Unstable 85 Unstable -11 46 50.7 Unstable 71 19.7 Unstable 58 -3.7 Unstable 17 44.2 Unstable 425 56.1 Stable 393 48.4 Stable 340 70.8 Stable 535 Stable 0.7 507 -3.7 Stable 454 21.1 Stable -2 22.5 Unstable -10 -7.2 Unstable -45 -47.3 Unstable 12 Unstable 2.3 -10 Unstable -14.9 -40 -59.7 Unstable

* Earning in excess of cutoff revenue in rupees

160

Recommended Feeder Route and Vehicle

Table-B.9 Scenario-IV: Passenger-km as MOE

Bus stop Bhasra

Daihara

Dantun

Kalabani

Khokra_1

Khokra_2

Kukai

Manoharpur

Nachipur

Panchiyar

Salajpur

Sarisa

Syamalpur

GC Surplus* Route %ASC Passenger Vehicles Passenger Served Saving /Vehicle Stability Km 100 5276 11 771 1165 86.4 Stable 50 5110 11 746 1119 76.4 Stable 25 4816 11 699 1035 60.4 Stable 100 1277 3 218 165 68.1 Stable 50 1203 3 205 146 Stable 50 25 1067 3 182 110 18.1 Stable 100 3212 7 493 703 72.9 Stable 50 3106 7 475 671 Stable 67 25 2916 7 444 622 Stable 46 100 1802 5 372 337 -2.1 Stable 50 1718 5 353 311 -9.1 Stable 25 1561 4 317 266 23.2 Stable 100 1605 5 428 409 -50.8 Unstable 50 1531 5 408 381 -57.8 Unstable 25 1392 4 369 332 -30.6 Unstable 100 986 3 224 100 -33.8 Unstable 50 920 3 206 85 Unstable -51 25 789 2 174 46 10.7 Unstable 100 940 3 289 71 -20.3 Unstable 50 866 3 265 58 -43.7 Unstable 25 721 2 220 17 Unstable 4.2 100 2367 7 513 477 -55.7 Unstable 50 2263 7 488 444 -57.3 Unstable 25 2068 6 440 390 -42.8 Unstable 100 542 2 193 -24 -46.7 Unstable 50 479 2 170 -42 -67.5 Unstable 25 345 1 121 -75 -0.3 Unstable 100 796 3 281 40 Unstable -49 50 726 3 256 21 Unstable -65.7 25 591 2 207 -15 -30.7 Unstable 100 2492 7 485 512 -26.2 Stable 50 2394 6 461 485 Stable 1 25 2213 6 419 432 -9.2 Stable 100 600 3 238 -7 -135.5 Unstable 50 540 2 214 -15 -114.4 Unstable 25 423 2 166 -47 -140.8 Unstable 100 642 2 208 9 11.6 Unstable 50 576 2 187 -12 -11.8 Unstable 25 447 2 145 -42 -65.5 Unstable * Earning in excess of cutoff revenue in rupees

161

Annexure-B

Table-B.10 Scenario-V: Passenger-km as MOE Bus stop Bhasra

Daihara

Dantun

Kalabani

Khokra_1

Khokra_2

Kukai

Manoharpur

Nachipur

Panchiyar

Salajpur

Sarisa

Syamalpur

%ASC 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25 100 50 25

Passenger Km 5276 5110 4816 1277 1203 1067 3273 3165 2976 1902 1817 1658 1770 1692 1552 986 920 789 940 866 721 2367 2263 2068 585 520 381 796 726 591 2639 2538 2355 600 540 423 710 642 522

Vehicles Passenger Served

11 11 11 3 3 3 8 7 7 5 5 5 6 6 5 3 3 2 3 3 2 7 7 6 2 2 2 3 3 2 7 7 6 3 2 2 3 2 2

771 746 699 218 205 182 493 475 444 372 353 318 428 408 371 224 206 174 289 265 220 513 488 440 193 170 123 281 256 207 485 461 419 238 214 166 208 188 152

GC Surplus* Route Saving /Vehicle Stability 1165 106.4 Stable 1119 96.4 Stable 1035 80.4 Stable 165 88.1 Stable 146 Stable 70 110 38.1 Stable 714 41.8 Stable 682 72.7 Stable 633 Stable 53 354 Unstable -0.1 327 -4.8 Unstable 283 -19.5 Unstable 436 -94.8 Unstable 407 -97.5 Unstable 365 -91.8 Unstable 100 -13.8 Unstable 85 Unstable -31 46 30.7 Unstable 71 -0.3 Unstable 58 -23.7 Unstable 17 24.2 Unstable 477 -35.7 Unstable 444 -37.3 Unstable 390 -22.8 Unstable -18 -47.6 Unstable -36 -64.1 Unstable -71 -103.1 Unstable 40 Unstable -29 21 -45.7 Unstable -15 -10.7 Unstable 535 -19.3 Unstable 507 -23.7 Unstable 454 Unstable 1.1 -7 -115.5 Unstable -15 -94.4 Unstable -47 -120.8 Unstable 26 -80.8 Unstable 3 -27.8 Unstable -18 -78.4 Unstable

* Earning in excess of cutoff revenue in rupees

162

Recommended Feeder Route and Vehicle

Table-B.11 Scenario-VI: Passenger-km as MOE

GC Surplus* Route %ASC Passenger Vehicles Passenger Served Saving /Vehicle Stability Km Bhasra 100 5430 13 771 1188 63.4 Stable 50 5259 12 745 1141 75.7 Stable 25 4964 11 700 1058 85.8 Stable 100 1277 3 218 165 Daihara 108.1 Stable 50 1203 3 205 146 Stable 90 25 1067 3 182 110 58.1 Stable 100 3273 8 493 714 Dantun 61.8 Stable 50 3165 7 475 682 92.7 Stable 25 2976 7 444 633 Stable 73 100 1952 6 372 361 Unstable Kalabani -23 50 1865 5 353 335 Unstable 6.9 25 1706 5 319 290 Unstable -5.9 100 1770 6 428 436 Khokra_1 -74.8 Unstable 50 1692 6 408 407 -77.5 Unstable 25 1552 5 371 365 -71.8 Unstable 100 986 3 224 100 Unstable Khokra_2 6.2 50 920 3 206 85 Unstable -11 25 789 2 174 46 50.7 Unstable 100 973 3 289 77 Kukai -29.3 Unstable 50 898 3 266 312 -50.5 Unstable 25 752 3 221 312 -70.4 Unstable Manoharpur 100 2367 7 513 477 -15.7 Unstable 50 2263 7 488 444 -17.3 Unstable 25 2068 6 440 390 -2.8 Unstable Nachipur 100 603 2 193 -15 -54.3 Unstable 50 537 2 171 -33 -66.8 Unstable 25 396 2 123 -69 -97.8 Unstable 100 832 3 281 45 Panchiyar -65.9 Unstable 50 760 3 256 26 -77.2 Unstable 25 629 3 211 -1 -103.9 Unstable Salajpur 100 2639 7 485 535 Stable 0.7 50 2538 7 461 507 -3.7 Stable 25 2355 6 419 454 21.1 Stable Sarisa 100 610 3 238 -3 -121 Unstable 50 550 3 214 -11 -136.5 Unstable 25 433 2 166 -44 -128 Unstable Syamalpur 100 710 3 208 26 -60.8 Unstable 50 642 2 188 3 -7.8 Unstable 25 522 2 152 -18 -58.4 Unstable

Bus stop

* Earning in excess of cutoff revenue in rupees

163

Resume Present Address Sudhanshu Sekhar Das Department of Civil Engineering I.I.T. Kharagpur-721302 Permanent Address Sudhanshu Sekhar Das Newcolony, Apratibindha, Bhadrak-756100 Phone : +91-6784-240838(O); +91-9437544200 (Cell) +Email- [email protected], [email protected]

Educational Qualification Doctor of Philosophy (PhD) – Transportation Engineering (July, 2003- January, 2008) –Thesis Submitted. Indian Institute of Technology Kharagpur, Kharagpur -721 302, India. Master in Town and Regional Planning (M.T.R.P) - (July,1996- Dec, 1997) Bengal Engineering College (Deemed University), Presently Known as Bengal Engineering and Science University, Shibpore, Howrah -711 103, India. Bachelor of Engineering (B.E) - Civil Engineering (1986-1990) College of Engineering and Technology Orrisa University of Agriculture Technology, Bhubaneswar, Orissa. Professional Experience and Responsibility 1. Assistant Professor -2006 till date Responsibility Academic: TeachingEngineering engineering, Surveying

drawing,

Transportation

2. Sr. Lecturer - 1999 – 2006, Bhadrak Institute of Engineering & Technology Responsibility a. Academic: Teaching- Engineering engineering, Surveying b. Administrative: Development officer

drawing,

Transportation

i. Supervision of new civil construction work ii. Campus Maintenance 3. Lecturer - 1993 – 1999, Bhadrak Institute of Engineering & Technology a. Academic: Teaching- Engineering drawing, Transportation engineering, Surveying b. Administrative: In charge of ISTE sponsored Environmental Citizenship Cell. Expertise in Software - ARC GIS-9, ARC VIEW, GEOMEDIA, LIMDEP PUBLICATION Published •

Das, S. S. and Maitra, B., (2007) ‘A Covariance Heterogeneity Nested Logit Model for Choice of Rural Feeder Service to Bus Stop’ Indian Highway, IRC, December, pp. 31-39.

•

Das, S. S., Maitra, B. and Boltze, M., (2007)‘Valuing Attributes of Rural Feeder Service to Bus Stop’ Indian Highway, IRC, May, pp. 9-16.

•

Das, S. S., and Maitra, B. (2005) ‘An Approach for Planning of Passenger Transportation System in Rural India’ National Conference on Advances in Road Transportation Feb12-13 2005 NIT Rourkela.

•

Das, S. S. and Maitra, B., (2007) ‘Effect of Tree Structure on Nested Logit Model for Choice of Rural Feeder Service to Bus Stop’ International Conference on Civil Engineering in the New Millennium: Opportunities and Challenges, January12-13 2007 BESU Howrah.

•

Das, S. S. and Maitra, B. (2007) ‘An Application of Stated Choice Approach for Valuing Attributes of Rural Feeder Service to Bus Stop ’ National Conference on Recent Advances in Civil Engineering March 1-2, 2007 CET Bhubaneswar.

Communicated • Das, S. S, Maitra, B. Manfred Boltze, ‘Valuing Travel Attributes of Rural Feeder Service to Bus Stop: a Comparison of Different Logit Model Specifications.” communicated to ASCE

Professional Training Short term Course 1. “Advances in civil engineering materials” held from 28/12/98 to 08/01/1999 at NIT Jalandhar. 2. “Advanced Road Infrastructure Planning and Development” held from 17/7/2002 to 23/7/2002 at I.I.T. Kharagpur. 3. “Soft Computing Tools in Civil Engineering” held from 10/11/2003 to 16/11/2003 at I.I.T. Kharagpur. 4. “Effective Teaching Skills” held from 31/1/2005 to 05/2/2005 at I.I.T. Kharagpur. 5. “Erosion and Sedimentation of River Beds” held from 21/11/2005 to 26/11/2005 at I.I.T. Kharagpur. 6. “Soft Computing Tools in Civil Engineering (SCTICE 2006)” held from 13/11/2006 to 19/11/2006 at I.I.T. Kharagpur. 7. “Advanced Technology for water and Wastewater Treatment” held from 20/11/2006 to 25/11/2006 at I.I.T. Kharagpur 8. “Introduction to ArcGIs-9” from 20/12/2004 to 24/12/2004 by ESRI India at New Delhi. Membership of Professional Bodies/Societies ISTE

LM-18383

Research/Consultancy Project 1 ƒ ƒ ƒ ƒ ƒ ƒ ƒ

Period : 2006 Name of Employer : During Ph. D * Position held : NA Name of Project : Planning of Public Transportation System in the City of Buraida, Saudi Arabia Funded by : Govt. of Saudi Arabia Cost of the project : Not Known Task performed (Describe in detail): Estimating willingness to pay values, Mode choice modeling, and Demand estimation.

2. ƒ ƒ ƒ ƒ

Period Name of Employer Position held Name of Project

ƒ

Funded by

: : : :

2004-2005 During Ph. D * NA Transportation study for Increase in Traffic Level on Vidyasagar Setu, Kolkata, India : The Hooghly River Bridge Commissioners, Kolkata

ƒ ƒ

Cost of the project : 7 lakhs Task performed : Planning traffic surveys, organizing stated preference surveys, modeling route choice behavior, estimation of willingness to pay values, demand modeling, studying the toll sensitivity.

ƒ ƒ ƒ ƒ

Period : 2001-2002 Name of Employer : ZSS Position held : Coordinator Name of Project : Condition survey and cost estimate for health centers of Bhadrak District, Orissa India Funded by : DFID Cost of the project : 5 lakhs Task performed : Condition Survey of the Health Centers, Preparation of the detail estimate with cost analysis of renovation as per OPWD.

3

ƒ ƒ ƒ

Areas of Interest: •

Rural transit planning.

•

Mass transportation planning, management and operation

•

Behavioural models in transportation and demand estimate

•

Transport Economics

•

Environmental aspects associated with traffic

•

Application of GIS in transportation and database managements