Voting, Networks And Communication

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

1

Voting, Networks, and Communication Arnold B. Urken 1 Abstract—This paper compares the pre-modern legacy experience of voting systems with the development of voting systems in mass democratic elections, calling attention to technological opportunities and challenges for innovation. In both cases, voting is seen as a network communications process in which a voting system contains a method for expressing voting information, transmitting votes across a centralized or distributed network, and collecting and counting the votes to produce a collective outcome. Legacy systems, still used today, include a rich communications environment. Traditional systems integrate redundant network communication with opportunities for individual initiative to assure that votes are reliably recorded and counted. Analog and digital counting of votes is also used to provide efficient and effective processing of voting information for simple choices in relatively small groups. The operation of these systems is outlined to highlight their flexibility in processing voting information. The transition from legacy to modern voting systems occurred without integrating the development of a science of voting methods with the creation of machines for collecting and counting votes. Analytical insight into the properties of voting methods can be traced back to Pliny the Younger in the Roman Republic, but knowledge was developed episodically by lone scholars until the golden age of mathematical modeling of voting processes in the 18th century French Academy of Sciences. Robert’s Rules of Order (RRO), developed more than 50 years later, did not incorporate any of this knowledge. Although RRO instructs us to develop by-laws for managing information in the choice of a voting method, it does not—contrary to popular opinion—provide any guidance about how to do this. While the study of voting methods has found niches in academia, it remained separate from the technological impetus to develop speedy vote counts, which was seen as a defense against error and corruption. Initially free of government regulation and led by entrepreneurs, voting machine development was normalized with versions of voluntary federal voting standards, which have become less voluntary with the introduction of federal money under the HAVA (Help America Vote Act). The institutionalized split between the study of the expression of voting information and engineering of systems for collecting and counting votes is explored by analyzing what the seminal theory of the Marquis de Condorcet and his contemporaries. Then some properties of voting methods are reviewed to point out forensic implications for modern election administration, including a way of gaining time in election recounts to meet legal deadlines. The paper ends by briefly presenting proposals for expanding the scope of individual citizen voting rights in computer networks to begin discussion of a model voluntary standard and to challenge system designers to take account of individual voter needs in an imaginative, creative way. Index Terms—History of voting technology, invention of voting machines, scientific development of voting theory, voting rights.

1. Introduction Amid news about failed voting machines, allegations of stolen elections, and controversy about whether we should depend on election machines at all or stick with paper ballots, it is useful to consider an historical perspective. Historically, pre-modern experience with voting systems includes a rich systems legacy for combining the expression of voting information with vote collection and counting, a tradition that has not been fully appreciated. Traditional systems integrate redundant network communication with opportunities for individual initiative to assure that votes are reliably recorded and counted. They also make use of analog and digital measurement to count votes and provide guidance about the processing of voting information. Legacy systems continue to provide efficient and effective processing of voting information for simple choices in relatively small groups. 1

Manuscript received February 20, 2009. A. B. Urken is a Professor of Social Science with Stevens Institute of Technology, Hoboken, NJ 07030, USA, phone: 520-820-5128; fax: 928-569-6316, e-mail: [email protected].

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

2

However developers of modern voting systems face the challenge of integrating knowledge about the expression of preferences and judgments with modern technology for collecting and counting votes in mass electorates. What can we learn from traditional, legacy systems about how to analyze and deal with this challenge? This paper suggests that revisiting the origins of voting systems and the role of individual rights in communication can provide new ways of defining and resolving problems in computer-mediated voting. Section 2 analyzes the properties of traditional systems that use votes to express preferences and judgments and then collect them to create collective outcomes. Section 3 relates the operation of these legacy voting systems to the evolution of the scientific study of voting methods and the emphasis on rapid and accurate collection and counting of votes. Section 4 describes why it is important to take account of voting methods in dealing with problems of election administration and illustrates how voting theory can be used to improve election management. And Section 5 outlines a voter-rights oriented approach to promoting voting system innovation. This approach would explicitly recognize individual voter rights in network communication to create a voluntary model standard that can be used to manage technological change and promote system innovation.

2, Traditional Voting Networks Voting is a network communications process in which agents, humans or nodes, interact with each other by sending voting data to represent information about preferences or judgments to form a collective outcome. “Voting” is a process that has been used to explain behavior in natural systems, including bees making collective decisions about the location for a new hive. [1] Normally, in human elections and even in collective choice mechanisms in computer science and computer engineering, voting data are sent to a central host to be “fused” or counted to determine the outcome. In these situations, voters are the “clients” and the central host is the “server.” In client-server (CS) network configurations, information sharing is usually limited to having the central host, or server, compute a collective outcome and inform the voters—or perhaps the entire network, including non-voters—about the results. In contrast, if agents vote by sending their votes to every other voter participating in a collective decision, each voter functions both as an agent that sends votes as well as a host that collects votes and computes a collective outcome. In these situations, voters are considered to be “peers” or equals, so the network would be operating in a “peer-to-peer” (P2P) configuration. In this mode, information sharing includes knowledge about the input as well as the output of a voting process [2 3]. At first glance, it may seem impractical or unfeasible for human agents to use P2P systems, much less simultaneous use of CS and P2P voting mechanisms. But traditional or “legacy” schemes associated with procedural guidance for managing voting information found in manuals such as Robert’s Rules of Order (RRO) actually combine CS and P2P communication when voice votes (VV), a show of hands (SOH) or division of the whole (DOW) are employed [4]. Each of these three ways of articulating voting information has its strengths and weaknesses as a form of communication. Voice voting is quick and the quality and loudness of sound convey information about intensity, but depending on the size of the group and the acoustic attributes of the voting venue, the collective result can be noisy and difficult to interpret. A show hands is also quick and arm angle, height, and speed of movement express passion, though, depending on group size and venue, getting a correct count may not be straightforward. And if all of voters

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

3

move to separate parts of the voting venue or stand up in place to express their preference or judgment, movement styles and utterances can flavor the communications process, yet the collective outcome may still not be immediately obvious. RRO provides guidance about how to process these ways of communicating voting information based on a rich sensory environment. Yet, despite common belief to the contrary, RRO— depending on the version or edition—includes very little advice about how to choose voting methods, particularly those that allow more complex expression of preferences and judgments. Instead, it cautions members to devise systems for managing the different types of information created in a voting process. RRO recommends creating by-laws to manage the choice and use of a voting system, but does not explain precisely how to do this. However RRO includes a system of voter rights and redundant communication to assure the integrity of voting processes by assuring that votes are recorded and counted correctly. Usually, by-laws in human organizations allow a single voter or a small set of voters to request that voting results be scrutinized. Although such requests are not always automatically honored, it is routine to actively audit a voting outcome in various ways. If VV produces the initial vote, the results may be scrutinized by taking another VV or using SOH or DOW to check the results. Of course, while all three vote communication methods produce “votes” that can be counted simultaneously in CS and P2P modes, the counting processes may be constrained by group size and voting venue. In small groups, regardless of venue, P2P processing of VV data may allow matching of individuals and their votes, while such matching may be limited for larger groups in more noisy venues. Group size and venue can also affect the possibility of P2P matching of voters and votes with SOH and DOW systems, though the less transient voting data produced by SOH and DOW processes may compensate to facilitate matching. It is important to appreciate that VV, SOH, or DOW communication methods all permit vote counting to be done in analog or digital mode. Even though recent manuals of procedure remove options for individual intervention to scrutinize vote counting, a strength of legacy systems for vote counting is that a vote counter (or counters) in CS (and P2P) network mode can make a reliable inference about the collective outcome without taking the time to receive (read) and count each vote. Viewed from this perspective, a gestalt or composite perception of a collective outcome can be used to determine if voting support for one choice is greater than another. In an instant, vote counters can receive voting data and infer that the sound of “Aye” is greater than the sound of “Nay,” that the number of hands shown for “Aye” is greater than number of hands shown for “Nay,” or that the size of the “Aye” group is larger than the “Nay” group. When initial votes with VV, SOH, or DOW produce close or indecisive collective outcomes that cannot be resolved with analog methods, voting results can be scrutinized by repeating a vote with the same method, using one of the other two methods, or resorting to marking and counting ballots. When collective choices are repeated, voters can change their votes or avoid public detection when VV or SOH are used. Moreover, in such cases, voters may practice evasion or deception such as standing in the DOW so that they are hard to identify or mouthing “Aye” in a faint voice. But such behaviors are usually not feasible with DOW voting. And when SOH results are scrutinized with counting of individual hands, changing votes can still be feasible, though changes may still be detected—if not collectively recorded. In some homogeneous cultures, the potential noise of collective VV is avoided by having an election conducted so that—as in some Swiss cantons—voters individually announce their votes in public [5]. In such cases, open presentation of a choice is not considered to be a loss of

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

4

individual privacy, but an opportunity to gain social recognition for one’s contribution to a public good. 3. Voting Theory and Voting Systems A strength of traditional voting networks is their use and management of redundant communication about voter preferences and judgments that incorporates voter rights and initiative. A limitation of these networks is the narrow scope of guidance provided for voting methods other than one person, one vote and majority or plurality rule. This restriction is understandable given that RRO was compiled for small organizations to help them reach consensus about practical problems. This compilation distills hundreds of years of experience about making laws for governmental and non-governmental operations. Although the best known manuals of “parliamentary procedure” were developed over time in British legislative practice and were reinvented by thinkers such as Jefferson [6], Robert’s rendition is distinctive for two reasons. First, RRO synthesizes learning about voting from the trial-and-error experience of a broad range of governmental and non-governmental organizations. RRO, adopted by military and civilian institutions in the US, became a general cultural standard and reference for organizational practice. And, second, RRO obscured the importance of choosing a voting method and establishment of by-laws for managing information in a voting process. This routinization neutralized any impetus to integrate scientific knowledge about the properties of voting methods. For instance, minimally, RRO might have stipulated that by-laws take account of problems in reaching consensus and analyze them with current knowledge about the properties of voting methods. But this type of knowledge was not part of Robert’s frame of reference. His goal was to provide a manual of practice to improve meeting management. Moreover, even in the middle of the 19th century, theoretical knowledge about voting was still fragmentary and not part of an established academic curriculum and traditions. For hundreds of years, analysts of voting processes worked in isolation, inventing and reinventing models, and rediscovering voting system methods and properties without producing cumulative, empirical knowledge that could be used in choosing a voting system [7]. Although the “first golden age” of voting theory in the 18th century French Academy of Sciences (FAS) occurred more than 50 years before the first edition of RRO was published [8], there was no cross-fertilization of ideas between these streams of analysis. Because this historical opportunity for integrating systems for expressing and processing voting information was lost, technological inventions and innovations have been limited by the lack of a framework for connecting different methods for expressing votes and judgments and the systems used to collect voting information and compute collective outcomes. The experience of Thomas Jefferson illustrates why the expression of preferences and judgments and the counting of votes were not integrated during the golden age. Jefferson’s scientific bent led him to think mathematically about how to apportion seats in a legislature, but his practical interests and educational background left him ill-equipped to comprehend the significance of the development of voting theory going in Paris during the first golden age of voting theory [6]. Even though Jefferson served as Minister to France, followed scientific trends, and corresponded with scientists such as the Marquis de Condorcet—Secretary of the French Academy of Sciences—about intellectual matters including standards for human rights, neither Condorcet nor Jefferson made a connection between their scientific interests and an explicit concern with voting rights 2 beyond a general commitment to equality, a radical view at the time. Jefferson owned a

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

5

copy of Condorcet’s best know scientific work on voting, Essai sur l'application de l'analyse à la probabilité des decisions rendues à la pluralité des voix [the 1785 Essai] [9], but there is no direct or indirect evidence that Jefferson ever read either the mainly textual summary of the ideas in the Preliminary Part, much less the mathematical exposition of the argument in the main body of work. Indeed, Condorcet wrote the Preliminary Part to get across his ideas without burdening his scientific colleagues, who, for different reasons, could not be expected to understand the framework of analysis used in Condorcet’s mathematical theorizing [6, 10]. Almost without exception, before the first golden age of voting theory, analysts of voting methods including Pliny the Younger, Raymond Lull, and Nicolas Cusanus [7] never imagined a problem with the actual collection of votes, regardless of how they were communicated. None of these theorists grappled with the definition of majority rule, which, at one point, was known as sanior et majors pars—the right/healthy and dominant part [11]. In medieval monasteries, the results of internal elections were often manipulated by superiors who set aside the majority vote winner in favor of the “right” choice. When participants in elections could find no consensus about how to determine the right choice and complaints began to undermine the authority of electoral outcomes, the sanior was dropped in favor of finding the choice that received a majority of votes. Collecting and computing of votes was never an issue. In the Institute of France (which succeeded the French Academy of Sciences after the French Revolution), scientists first discovered that a voting system might not produce expected information even though all the votes were collected. They discovered that a majority winner might not be produced when they used Borda’s proposed method of voting, a method of voting by ranks. This problem occurred because fellow scientists were not ranking all the candidates in the hope of decreasing the likelihood that their least preferred choices would garner enough votes to win a majority. As Daunou [13] noted, with unranked candidates assigned zero points, no majority winner was produced and academic elections were often delayed for weeks or months. When confronted with this fact, Borda allegedly replied, philosophically, that his system was designed for “honest” (i.e., sincere) voters [7]. No one dreamed that voting information would not be reliably collected and counted. Although the legacy of voting theory from the golden age was forgotten for decades during the 18th century, its heritage established a conventional set up for voting analysis that typically assumed that votes are communicated by marking ballots and that the votes would—almost automatically--be reliably counted. Later theorists (e.g,, Dodgson) independently rediscovered classical theoretical voting results or (e.gs. Guilbaud, Granger and Nanson) read publications from the first golden age and extended the ideas without addressing the collection and counting of votes [see 7]. Yet it was not until Black’s [14] rediscovery of Condorcet’s work and the study of problems of majority rule by Arrow [15] and others that theorizing about voting methods gained an academic niche in social science research. By this time, however, the intellectual split between the study of voting methods and the administration of elections was intensified by an emphasis on formalization and mathematical sophistication in voting theory. Indeed, the term “social choice,” which has become the standard term of scientific art, was invented by Arrow to describe the process underlying the “paradox of voting,” the problem of deriving a rational or transitive collective outcome from sets of individual voting preferences that are rational or transitive. The actual collection and counting of the votes was not considered as part of the analysis. Compared to Arrow and today’s voting theorists, Condorcet was more connected with traditional forms of voting communication (VV, SOH, and DOW) found in RRO [10]. Condorcet, speaking for his colleagues as Secretary of the French Academy of Sciences, observed that VV was a noisy

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

6

form of communication inconsistent with considered deliberation and careful analysis of voting information, a conclusion consistent with the lessons of voting for God in the early history of Christianity—experience that Condorcet was aware of [10, 16]. Moreover, Condorcet and his colleagues seemed to consider voting by ballot as a modern technological improvement over earlier practices. Before him Nicolas Cusanus [in 7]described paper balloting as a recent innovation in Church elections.. While the French Academy of Sciences had a quasi-official manual of voting practice regarding the election and promotion of members of the Academy, it focused exclusively on rules for voting by ballot, élections au scrutin, and did not mention, much less analyze, other forms of voting communication [17]. Implicitly, the Academy assumed that digital collection and counting of votes. Whether plurality, Condorcet, or Borda voting rules were used, each person’s votes were counted in sequence. Condorcet regarded SOH and DOW [9] as communications systems that are inconsistent with the practice of independent, autonomous decision making and prevention of cabales.. Moreover, Condorcet explicitly argued for the use of voting methods that would maximize the group probability of making a correct collective choice. Ironically, Condorcet, an atheist, may have derived the idea of a “correct” choice from his knowledge of traditional practices in election of Popes, where Cardinals pray for divine guidance before casting their vote for the correct choice. Developing what is now known as “cognitive political theory” [18], Condorcet developed a moral and physical model of integrity in voting networks. Although Condorcet— like Jefferson—believed in the ideal of democratic equality, he realized that some voters are more likely to make competent or reliable decisions. So to maximize group performance, Condorcet recommended educating citizens to improve the average competence, the independent variable in what Black labeled the “jury theorem” [6]. Condorcet’s analysis of voting established a voting grammar based on the semantics (rules governing the meaning of expressing information with votes) and syntax (rules for regulating the order of processing of voting data) to form a collective outcome [10]. For Condorcet, the integration of voting semantics and syntax was part of his concern with the integrity of voting system. He not only enabled us to see that voting integrity includes moral and physical components, but he also introduced the idea of conditionally normative advice about the choice of a voting system. This idea evaluates systems by investigating the feasibility of producing a hypothetical collective outcome given a set of semantic and syntactical rules. Although Condorcet and his colleagues—as well as contemporary voting theorists—disagreed—and continue to disagree—about voting system goals and desirable properties, voting theory has grown by posing (and trying to explain) conjectures, propositions, and problems about voting processes. So even when analysts disagree about how voting systems ought to work, all theorists can all benefit from theoretical and empirical analysis about how these systems actually or possibly work. For such analysis can reveal unappreciated areas of agreement as well as uncover new questions and avenues for research. By today’s standards, Condorcet’s concept of voting integrity may seem to be of purely antiquarian interest, yet the questions he raised are not. Unfortunately, Condorcet may be most widely known for the Esquisse d'un tableau historique des progrès de l'esprit humain [19], a work that has been characterized as the “utopian,” deathbed work of a frustrated scientist [20]. However, more recent scholarship has shown that Condorcet’s oeuvre, while fragmented and incomplete, is more subtle and complex than historians, philosophers, and (most) voting theorists have realized. Indeed, there is some evidence to suggest that in his “scientific” debate about voting systems with Borda and fellow scientists in the Academy, Condorcet did not follow his own admonition to pursue theoretical ideas boldly and critically. Nevertheless, his concern with the improving the “competence” of voters seems germane in contemporary culture, where

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

7

information overload, not 18th century France’s limited access to information and learning, challenges voter abilities to render competent decisions. Moreover, the Marquis’ work on a new Constitution for the French Republic highlighted the importance of structuring voting systems (agendas and methods for representing preferences and judgments) to elicit reliable information [7, 10]. Although historical experience has shown us that ideologists on the political right and left might use “competence” as an issue to subvert democratic participation, Condorcet, enlightenment optimist that he was, did not base his outlook solely on naiveté or audacious hope, but on the growth of human knowledge and creativity [10]. Condorcet raised—but did not live long enough to systematically examine—the issue of the systemic integrity or completeness of a voting system in his fragmentary oeuvre and politicized debates about the choice of voting system within the French Academy of Sciences (FAS) [21, 22]. The debate between Condorcet and Borda is the most frequently cited dispute. Condorcet was a reformer, Borda a staunch, absolute monarchist; Condorcet valued theoretical imagination, Borda was computationally adept in dealing with empirical questions [6, 10]. As FAS Secretary in control of Academy publications, Condorcet delayed open debate about Borda’s voting ideas. A poignant indication of Condorcet’s antagonism toward Borda’s work is the fact that Condorcet attacks Borda’s proposal because it omits critical information about the inputs and outputs of a voting process. Regarding voting inputs, as Iain McLean [23] has explained, Condorcet pointed out that Borda voting violates what is now known as the “independence of irrelevant alternatives” (IIA), an attribute of individual voter preferences that violated the principles of statistical independence and decision maker autonomy that are promoted in the 1785 Essai and other works. Similarly, Condorcet pointed out that Borda’s method might select the least preferred choice, an unthinkable possibility for a theorist committed to designing systems that would maximize the likelihood that a voting process would produce the “best” choice. If, as with many important thinkers, Condorcet raised more questions than he answered, his paradigm (or metaphysical research program) has enabled subsequent theorists to investigate these questions [10]. Certainly, we now appreciate that, depending on the choice of decision objective, IIA may or may not be a critical condition. We also understand that voting methods can yield different—sometimes inconsistent—collective outcomes. And we appreciate that voting methods communicate selectively about voting information to allow us to learn from complex relationships in voting data. So voting systems are tools for answering different questions about the same data; the answers, themselves, may be partially or entirely consistent or inconsistent. For example, plurality voting and Borda scoring may both yield the same plurality winner, but the complete collective rank-orderings produced by these systems could be different. The legacy of the first golden age of voting analysis does not instruct us about integrating the voting inputs, communication, and outputs. It does not even provide a set of systems requirements for doing this. However, it does draw our attention to the importance of finding the truth in computing voting results. In contemporary culture, public discussion of different ways of interpreting election data is rare. In fact, outside the community of academics interested in voting theory and practice, abstract analysis of voting methods can produce a palpable sense of discomfort. Indeed, in some cases, such analysis is regarded as subversive because it challenges an accepted standard for establishing consensus and legitimacy. Some professional societies (such as the IEEE and the American Mathematical Society) have debated and adopted new voting methods—such as approval voting [24], but these debates implicitly presume that all fellow mathematicians or electrical engineers share an equal, high competence, a presumption that Condorcet—and Borda—would have attributed to their 18th century scientific colleagues. In public elections, we seem to operate on the presumption that

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

8

voters behave as if they have an equally-likely chance of making a correct choice or, if not, that the average individual competence is high enough to assure the production of reliable choices. An important aspect of this heritage is the passionate search for the truth in measurement. In this respect, despite his passion for the truth, Condorcet’s theoretical work includes no experiments and few examples, so it does not provide much guidance in ascertaining moral or physical truth. Indeed, at the end of his 1785 Essai, Condorcet reminds the reader that his 300 page exposition is but a sketch of the outline of a theory of voting presented to spur others to develop experimental knowledge. Nevertheless, Condorcet is meticulous about avoiding experimentation that produces what he called physicaille, or trivial experiments. He realized that it is easier to know when theories fail than when they succeed. Good theory contributes to the development of experimental knowledge by providing insight into the limiting conditions of a model and by stating predictions that can be falsified. So when counting votes, it is often easier to discern voting results that are not internally consistent than it is to show that collective outcomes are positively consistent with the information processed by the voting system [10]. 4. Voting Theory and Election Administration The cultural differences between developers of scientific knowledge about voting methods and leaders in the art of collecting and counting votes has been exacerbated by the emergence of mass elections in modern nation-states. Analysts of voting found disciplinary niches in academia and industry to continue to investigate the properties of voting methods. And engineers and inventors have focused on building systems that would improve the accuracy and speed of vote processing in large electorates. Increasing specialization in both groups has left us without intellectual resources to grapple with the problem of integrating the expression of voting information and computation of voting results. Although analysts from computer science and other disciplines have begun to critique the systems used in election machines, their analyses and demonstrations, however trenchant, have not been widely accepted by election managers, who are governed by their own experience in elections. Demonstrating possible threats to election systems and critiquing the limitations of software tools for building them is not easy to communicate to practitioners who are skeptical about the value of abstract analysis and experimental results divorced from their experience [25]. Indeed, leading researchers in formal methods for software security face a similar problem in explaining the value of mathematical methodology to their colleagues without an operational example that would demonstrate the merits of their approach [27, 28]. The current emphasis on rapid and accurate processing of votes is an outgrowth of the dominance of an entrepreneurial approach to building election machines. In the 19th century, although inventors patented and proposed use of electromagnetic technology to speed up legislative votes, their proposals was out of touch with political cultures in which the slow, public and controllable casting of votes was favored by political leaders. Consequently, their inventions were not considered for public elections. But inventors of machines such as Jacob Myers, motivated by electoral corruption, built mechanical computing devices for collecting and counting votes [29]. The expression of voter preference on these machines was limited to binary choices or selection of a subset of candidates. Since there were no governmental or industrial standards set for these devices, it required entrepreneurial marketing of the lever machine to induce cities and states to adopt the machine for public elections. The adoption of technologies for building election machines in the US follows a pattern of borrowing technologies originally developed for non-election applications. These developments tended to opportunistic, responding to the successes of the day with the latest and greatest

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

9

communications innovations. Myers, a mechanical engineer, vowed that his system would prevent corruption by keeping ballots out of the hands of corrupt politicians. Joseph Harris and William Rouverol, inspired by the success of the Hollerith card in industrial data processing, created a punch card machine to create fast and accurate processing of votes. Phone devices with touch pad interfaces have also been used outside the US. And more recently, digital computer scanning and direct entry (via mouse, keyboard, and touch screen) technology has been developed to create better systems for entering votes (and, in some cases, verifying the entries). All of these technologies have been limited by a lack of specifications for what we expect an election system to do. In the 19th century, the allure of mechanical engineering was impressive and marketers were able to push speed and perceived accuracy. But experience in contested elections revealed that lever machines could produce inconsistent voting results due to degradation of mechanical parts and could be manipulated if the machines were not properly secured and managed. Punch card machines were considered problematic before the notorious hanging chad problem in the 2000 US Presidential election. Some states had banned them, though Rouverol, one of the inventors, maintained that the chad problem occurred because vendors used inferior materials in the machine and ballots [30]. Phone technology, scanning technology, direct entry technology still have viable industrial markets in which individuals can communicate their choices, but the general problem of end-to-end processing of voting information that was a highlight of legacy systems for expressing and counting votes has not been resolved. Early inventors of voting machines were neither guided nor limited by social standards except for what the market would allow. As the market became crowded with competition between lever, punch card, scanning and other technologies, specifications standards for machines became a recurring issue. Before 1990, when the US Federal Election Commission drafted its first voluntary election standards guidelines, states relied on a single industrial expert, Robert Naegele, to define engineering requirements, to test machines to determine if the requirements were met, and to certify the machines as reliable so that they could be placed on state lists of approved voting machines that could be adopted by counties. Naegele was instrumental in institutionalizing professional testing of election systems by creating Independent Testing Laboratories (ITA’s) to enforce engineering standards. Naegele’s work as a election systems consultant set a conventional industrial standard for software reliability (he thought “black box” testing, not “white box” testing provided sufficient reliability) and transparency (testing protocols and test results were not made public and were considered proprietary information to be shared only by the tester and vendor) [31, 32, 33, 34]. Election vendors have been criticized for not building better election systems, but they have been limited by the lack of clear standards and predictable markets to guide them in making upfront investments in systems. The uncertainty of county purchasing stymied vendors in most states, except for a few such as Oklahoma and Minnesota that set uniform state standards [34]. The Help America Vote Act (HAVA) changed the marketing of election machines in line with the recommendation of a MIT-CalTech study that election systems be modularized so that market forces could gradually implement marginal, incremental improvements in the overall system [35]. This recommendation presumed that annual state and local budgets would be available to create incentives for industrial innovation. But investment in election systems was not a high priority in states and counties. HAVA money, available for election management and machine purchase, enabled the federal government to set mandatory standards, taking discretion away from the states that chose not to set their own system specifications and manage testing. Management of federal regulation was shifted to the National Institute of Standards and Technology (NIST), which revamped the management of ITAs, upgraded standards, and made ITA testing protocols and results a matter of public record, enabling open discussion and debate.

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

10

Although incremental engineering improvements have occurred, they do not take account of the implications of voting methods for the processing of voting information. Ironically, the issue was raised in 2000 by a panel of technical experts charged with making recommendations to a Presidential commission about the development of Internet voting [36]. Their recommendations included requiring any election machine vendor to provide plug-and-play software functionality for alternative voting methods. The recommendation did not include a list of methods to be included. This proposal did not take account of the complex social implications of adopting new voting methods, something that would be obvious to parliamentary practitioners. Changing voting methods can create uncertainty about political competition, gauging of political support and strategic planning of political campaigns. Election managers would also be concerned about uncertainty associated with system change, budgetary constraints, and career paths. Whatever the relative salience of these concerns, the Commission omitted the voting method recommendation from their report to the President Commission on issues to be studied. Advocates of voting method reform tend to highlight the benefits of changing voting methods without taking account of the administrative implications of implementing system change. In principle, if programming specifications existed for implementing alternative voting methods, administrative management of systemic change might be accommodated. But election officials tend to be oriented toward the operational realities and experience of running elections with whatever voting system is mandated by law. Within this context, they have little incentive to be concerned with voting possibilities that they have not seen. For example, at workshops that consider cyber and other threats to election systems, managers have been skeptical about malicious manipulations carried out in experimental laboratory settings. Voting theorists know that system rules governing the expression and aggregation of preferences and judgments should be assumed to be neutral. Yet voting method reformers tend to be singleminded, unaware of the problems of integrating voting methods in new election machines. What can election system designers and managers learn from voting theory about managing election data? Certainly, knowing about the properties of voting methods is important in setting the requirements for election systems. For example, in the US, plurality voting creates individual rankings in which one choice receives a vote and the other choices receive zero votes. This pattern is typically used to audit elections to control “undervoting,” “overvoting,” and voter fraud (including cyberattacks on election databases). If voting methods are used that allow voters to express more differentiated preferences, new approaches to auditing and fraud prevention would have to be devised. For instance, approval voting (AV), Borda voting (BV), Condorcet scoring (ConS), and Instant Runoff Voting (IRV) (also known as proportional voting), all make use of more complicated rankings of choices. In all these systems, the one vote to one voter relationship that is normally used to track voter participation must be modified. Moreover, in monitoring changes in computer databases during an election, auditing software would have to be designed to detect and report more possibilities for malicious and inadvertent error. For approval voting, voters do not normally record their individual ranking, but cast one vote for each approved choice to communicate information about an implicit, unstated set of rankings. This voting information shows which choices are approved and which ones are not approved. And the winner is the choice that gains the most votes. In AV systems, voters can cast up to N votes, where N is the number of choices. But voters might also approve one choice, so rules governing the interpretation of voter intent would have to be revised to take account of these possibilities. Also,

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

11

collective outcomes under AV can be tied for plurality, majority, and unanimous aggregation rules. Under certain conditions, there is some evidence that the probability of a tie would be up to five times more likely under AV than it would be under plurality voting systems. When voter preferences are evenly distributed across the spectrum of choices, tied outcomes are likely to include more than the two choices associated with ties in plurality voting systems. Such complex ties could create greater impetus to litigate tied outcomes, to demand recounts by hand, or to seek new elections. Certainly, states that allow ties to be “randomly broken” might consider alternative rules for resolving indecisive outcomes to avoid dissention, litigation, and loss of legitimacy. BV, ConS and IRV systems all involve similar implications for counting votes. All three operate on ranked data, respectively, to assign and aggregate points assigned based on ranked position, compute the number of times that each choice is rated higher than every other choice to find a winner, or sort voter preferences until a candidate wins a majority of first-place votes. This relatively complex voting data requires that auditing software be able to detect and monitor changes in databases to preclude inadvertent or malicious manipulation of voting information. Indeed, this capability is particularly important for IRV systems in which individual rankings are transformed to search for a majority winner.

Although voting methods pose challenges for the management of voting information, a knowledge of voting theory can also be used to devise new tools to manage elections. For example, one of the challenges of election management is to meet legal deadlines for processing votes. Often, electronic voting data must be recomputed to assure reliable results. In some cases, statutes allow collected votes to hand sampled and counted to audit the results, with consistency among three samples accepted as evidence of validity. In other cases, a complete machine recount may be mandated. Both of these cases pose a potential tradeoff between information reliability and vote processing time. Recent work in error-resilient data fusion [37, 38] has shown that it is possible to predict how many samples to draw to have high confidence that the sampled voting results will be the same as the voting outcome that would be found if all the votes were counted. To predict the reliability of the sampling of votes, two factors, the number of voters and the distribution of preferences (as revealed by votes), must be known. Then the size and number samples can be found to manage the tradeoff between information reliability and vote processing time. For instance, when hand sampling is used, instead of using legal standards for sampling that rely on what seems intuitively reasonable, sampling can be conducted based on error-resilient principles that predict how valid a sampling strategy can be expected to be. If data were sorted by voting district, the actual hand sample could come closer to meeting the sample size for assuring a reliable inference about the voting results. When a complete machine recount is required, this sampling technique can be used to reduce computation time. For instance, if a machine processes 10,000 ballots an hour for an election with 500,000 voters, it would take 50 hours to recompute the results. If, with errorresilient techniques, the same results could be computed with 400,000 ballots, saving 10 hours and freeing up computational and human resources that could be used to expedite the processing of official voting results. This early interpretation of a recount might used tentatively to begin organizing before all the votes have been recounted to carry out post-election actions. If experiments establish the empirical validity of this analysis, the practice could become part of normal administrative practice.

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

12

5. Toward a Voter Bill of Rights Model of Electronic Voting State rules for regulating elections are described in legislative statutes, manuals of practice, and court opinions. These regulations deal with machine recounts and hand counts of based on the assumption of plurality voting. These guidelines are irrelevant for voting situations in which voters can select more than choice from a list or, as explained above, where expressions of ranks are used to produce voting information. For example, in an IRV election, what would one do if last possible round of successive reprocessing of rankings did not produce a majority winner? Would it be acceptable to reprocess the original individual rankings to search for a Borda winner or a Condorcet winner? This action would save the costs of holding a new election, but would it end up producing protracted conflict? Clearly, protracted litigation can undermine the integrity and legitimacy of a voting systems as much as irregularities in vote counts. Operational requirements for running elections always involve a tension between speed and accuracy. But sacrificing accuracy for speed does not assure integrity or legitimacy. For transparency requires that vote counting assure—as much as possible—that the results are not dishonest or fraudulent. Reengineering systems for expressing and collecting voting information might extend the scope of available means for assuring reliability. If there were a model standard for individual voter rights, a voluntary Voting Bill of Rights for citizens, the model could serve as a benchmark for assessing, discussing, and enabling policy changes to resuscitate the role of individuality in expressing voting information and counting votes? At first glance, a voluntary standard seems like it might reintroduce policy fragmentation and economic and technological inertia characteristic of pre-HAVA voting standards. But voluntary standards are a normal part of the infrastructure for engineering systems used in modern products and services. Standards that are defined, studied, and negotiated until a consensus is achieved. Despite roadblocks involving proprietary systems, agreements can be reached that sort out complex issues and recognize the value of multiple ways of achieving a desired functionality The goal of this section is to stimulate discussion of the issues that must be addressed in a model standard of voting rights in a networked world, not to draft the model itself. Legacy voting systems operate well in small groups where voter identity, intent, communication, and vote counting can be managed. In mass elections, the scope of voter rights is limited by a mismatch between the remnants of restrictions on the individual expression and communication of voting information and the lack of new technology. New markets for voting information could emerge if voting systems were designed by building systems that allowed voters to use network technology to communicate with existing or new technologies. Right now, it is technologically feasible for a voter to photograph a marked ballot in a voting booth and email the ballot to friends for fun, transmit it to political parties to help in targeting their get-out-the vote drives, or send it to individuals or groups that have struck deals to trade votes or pay money in exchange for votes. What if, before touching a screen or marking a ballot, the voter deliberated by exchanging ballot information with other voters in other booths in the same or different voting jurisdictions? What if real-time “Nader trading” were coordinated by a distributed network connections to affect the outcome of a Presidential election? What if voters delayed elections by tying up the network in online deliberation? Although there is no public evidence that these techniques have been used, the mere possibility that such “unthinkable” things might happen might be used to impose regrettable countermeasures. By anticipating the unthinkable, we can avert draconian reactions in

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

13

or out of courts and develop pro-active approaches based on a balanced analysis of options and the pros and cons of policy choices. The following proposed options for individual voting rights can serve as a baseline for inventing and exploring policy choices so that technology is pushed to meet voter needs by becoming imaginative and innovative. The order in which they are presented does imply any sense of priority. Voters should have the right to Actively audit that their votes are properly recorded and counted. Votes are at least as important as money. Banks and financial organizations enable individuals to use redundant means (electronic and paper) to verify monetary transactions. Touch screen devices have not been well engineered to enable voters to verify input by reading a printed ballot and/or hearing playback. New technologies that enable imaging of ballots and digital counting open up the possibility of verifying that votes are not counted improperly. This can be done by using new software tools based on open code to count all of the votes (by precinct or for all voters) or—with randomly assigned ballot identifications—to verify that a single vote is counted as intended. Control dissemination of personal voting information. The gradual integration of computer systems for registering voters and the processing their votes can enable individuals to opt in or opt out of information sharing. In most states, rules created by political parties allow information about individual voter registration and pattern of electoral participation to be sold at nominal cost. Electronic information systems should include a specifications requirement to allow opt-in or optout profiles. These data structures could be expanded to allow voters to expand the scope of shared information in a voluntary and informed way. Such profiles would help define and gain acceptance for the use of partial privacy in network communication. Participate in elections that last for more than one day. The tradition of holding elections on a single day is a holdover from agricultural culture in which the beginning and duration of elections was constrained by economics, transportation and communication. Allowing voters to spend a week of online deliberation and voting would enable voters to become more informed. As a consequence, electoral strategy could move away from spending large amounts of money to condition voters to behave as they would in a supermarket and make it possible for voters to change their minds. Communicate voting information traditionally expressed by voice and gesture. Although voice and gesture would demand more bandwidth than is currently used in computer-mediated voting, enabling these capabilities would enrich voting information and allow differentiation of intensity of preference, enable voluntary partial or complete voter identification, and open up the possibility of using analog and digital methods to compute collective outcomes from such data. Enter comments in electronic ballots. Comments—text, voice, or video—would be useful for electoral winners to help them gauge the strengths and weaknesses of public support. During an election that lasted more than one day, comments would enable voters to use network communication to assess why other voters agree or disagree with their preferences and judgments and to make informed judgments about changing or trading their votes. This practice could neutralize distorting, manipulative polls by enabling voters to learn the reasons behind polling numbers [39].

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

14

Allow their ballots to include advertising to defray the costs of elections. As an extension of partial privacy and partial anonymity, voters can opt in to make a contribution to cost management. This participation might be expanded to enable governments to function as information vendors that match partial privacy profiles with government databases. For instance, governments could finance elections by bidding out the matched profiles of voters who volunteer to participate in marketing studies. The scope and richness of this marketing data would make it very valuable. Use private vendors to collect and manage their votes. Personal services for making sure that votes are recorded and counted properly when they are aggregated at a final counting point would allow votes to be treated as importantly as money. The cost of such services needs to be investigated, but evaluations should take account of registration, equipment investment and management, management infrastructure, legal, maintenance and other costs in comparing market and non-market operations. This service might be expanded to enable a new class of private agents—similar to the “protective agents” to provide services for voters such as proxy services in deciding how to cast votes or trade them [40]. Discussion of such options could revive consideration of ideas about “fungible voting” as a mechanism for promoting stable socioeconomic development [41,42]. Although market forces and the modularization of election machines have produced changes and incremental improvements for voters, progress in the voting interface for expressing and recording votes, a main focus of current research, has not provided end-to-end transparency in the electoral process. Investigating and debating this list of voter rights can contribute to the refinement of long-term requirements to guide government and industry in nurturing, certifying, and managing the evolution of voting machine vendors. In a rudimentary sense, we humans act as if we are computers. We take input data and process it to produce new information. The integration of voting expression and vote counting in legacy voting systems emerged from human trial and error over many centuries. In an increasingly pervasive networked world of computing marked by uncertainty and crises, we should be more proactive in growing our knowledge about integrating voting methods and communication in computer networks. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8]

T.D. Seeley and P. K. Visscher, “Quorum sensing during nest-site selection by honey bee swarms,” Behav. Ecol. and Sociobiol, vol. 56, pp. 594-601, 2004. A.B. Urken, “Social choice theory and distributed decision making”, in R. Allen, ed., Proceed. of the Internat. Conf. on Off. Info. Sys., Palo Alto: IEEE/ACM, 1988. B. Parhami. “A taxonomy of voting schemes for data fusion and dependable computation”, Reliab. Eng. and Syst. Safety, vol. 52, 139-151, 1996. H.M. Robert, Robert’s Rules of Order. 9th Edition, New York: Scott Foresman, 1990. B. Barber, Strong Democracy. Berkeley: University of California Press, 2004. A.B. Urken, “The Condorcet-Jefferson connection and the origins of social choice theory,” Pub. Ch., 1991. I. McLean and A. B. Urken, (eds.) Classics of Social Choice, Ann Arbor: University of Michigan Press, 1995. I. McLean, 'The first golden age of social choice 1784--1803', in W. Barnett, H. Moulin, M. Salles, and N. Schofield eds, Social choice, welfare, and ethics: proceedings of the 8th International Symposium in Economic Theory and Econometrics (Cambridge: Cambridge University Press, pp.13-33, 1995..

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited [9]

[10] [11] [12] [13] [14] 15] [16] [17] [18] [19] [20] [21]

[22]

[23] [24] [25]

[26] [27] [28]

[29]

[30] [31] [32]

15

J.A. N. Condorcet, Marquis de, Essai sur l'application de l'analyse à la Probabilité des Décisions Rendues à la Pluralité des Voix. Paris: l'Imprimerie Royale, 1785. [facsimile edition, New York: Chelsea Publishers, 1975.] A.B. Urken, “Grokking Condorcet’s 1785 Essai,” Electronic Journ@l for the History of Probability and Statistics, vol. 4, 1, June, 2008. M.L. Moulin, “Sanior et major pars: Note sur l’évolution des techniques électorales dans les orders religieux du Vie au XIII siècle,” Revue historique de droit français et étranger, 1958. I. McLean and A.B. Urken, “Did Jefferson or Madison understand Condorcet’s social choice theory?,” Pub. Ch., 73: 445–458, 1992. P.C. Daunou, “A paper on elections by ballot,” in [7]. D. Black, The Theory Of Committees and Elections. Cambridge: Cambridge University Press, 1958 K. Arrow, Social Choice and Individual Values. New Haven: Yale University Press, 1951. R. MacMullen, Voting about God in Early Church Councils. New Haven: Yale University Press, 2006. J. Hellot, Annotated Rules of the French Academy of Sciences. Paris: French Academy of Sciences, 1753. C. List and R. E. Goodin “Epistemic democracy: generalizing the Condorcet jury theorem,” The J. of Polit. Philos., vol. 9, No. 3, pp.277 – 306, Dec 2002. J.A.N. Marquis de Condorcet, Esquisse d'un tableau historique des progrès de l'esprit humain, Paris: Arago Edition, 1849. G.H. Sabine, A History of Political Theory, New York: Holt, Rinehart and Winston, 1961. J.P. Schandeler et P. Crépel et le groupe Condorcet, Condorcet, Tableau historique des progrès de l’esprit humain Projets, Esquisse, Fragments et Notes (1772-1794), Paris: INED, 2004. I. McLean, and A.B. Urken, ‘La reception des oeuvres de Condorcet sur le choix social 1794-1803: Lhuilier, Morales, et Daunou’ in A.-M. Chouillet et P. Crépel ed., Condorcet: Homme des Lumières et de la Révolution (Fontenay-St. Cloud, ENS Editions), pp. 147-60, 1997. I. McLean, “The reasonableness of independence: a conversation from Condorcet and Borda to Arrow and Saari,” Nuffield College Politics Working Paper 2003-W6 S.J. Brams and P.C. Fishburn, “Approval voting in scientific and engineering societies, Gp. Dec. and Negot., Vol. 1, 1, April 1992. D.W. Jones, “Connecting work on threat analysis to the real world,” Threat Analyses for Voting System Categories, Wkshp. on Rat. Vot. Meths., George Washington University, 2006. F.S. Schneider, G. Morrisett, and R. Harper, A Language-Based Approach to Security. Berlin/Heidelberg: Springer, 2001. S.K. Chin, C. Irvine, D. Frincke, An Information Security Education Initiative for Engineering and Computer Science, Monterey: Naval Postgraduate School, 1997. G.C. Necula, S. McPeak, and W. Weimer, “CCured: type-safe retrofitting of legacy code,” Annual Symposium on Principles of Programming Languages, Proceedings of the 29th ACM SIGPLAN-SIGACT symposium on Principles of programming languages. D.W.Jones, Technologists as political reformers: lessons from the early history of voting machines, Presented at the Soc. for the Hist. of Tech. Annual Meeting, Las Vegas, October 13, 2006. C. Bagley, “The last lone inventor, “ Salon.com, June, 23, 2004. R. Dugger, “Counting votes,” The New Yorker, Nov. 7, 1988. R.G. Saltman, “Rules, politics and policy,” Election Law Journal August 1, 2007, 6(3): 312-321.

DRAFT for IEEE Special Issue on Electronic Voting—Comments Invited

16

[33] R.G. Saltman, The History And Politics Of Voting Technology: In Quest Of Integrity And Public Confidence. Basingstoke: Palgrave Macmillan, 2008. [34] B. Harris, Black Box Voting: Ballot-Tampering in the 21st Century, http://www.blackboxvoting.org. [35] Committee on House Administration, “ Caltech-MIT Voting Technology Project: Findings and Recommendations,” Jul 12, 2001. [36] C.D. Mote Jr., “Report of the national workshop on Internet voting: issues and research agenda,” ACM Internat. Conf. Proced. Ser.; Vol. 128. Proced. of the 2000 Ann. Natl. Conf. on Digit. Gov., pp. 1-59. [37] A.B. Urken, “Time, error, and collective decision system support,” in Proced. of the Internat. Conf. on Telcomm. Systs., Monterey, October, 2003. [38] A.B. Urken, “Using collective decision system support to manage error in wireless sensor fusion,” Intl. Conf. on Info. Fusion, Philadelphia, Jul. 25-29, 2005. [39] S. Herbst, Numbered Voices: How Opinion Polling Has Shaped American Politics. Chicago: University of Chicago Press, 1995. [40] R. Nozick, Anarchy, State, and Utopia. New York: Basic Books, 1974. [41] J.S. Coleman, “Political Money,” Amer. Polit. Sci. Rev., 67: 131 – 45, 1973. [42] J.S. Coleman, Power and the Structure of Society. New York: W. W. Norton, 1974.