\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\2\\\\\\\\\\\\\\\ OVERDUE swag: . 25¢ per W per in; RETURNIQQ LIQRARY MATERytLS: Place in book return to remove charge from circuhtion records ,3“; i raw-as: ram 30 2904 ‘ .195 4.51” APR 0 8 2005 m .1. 7’ ‘ :U H 03 N O f 0-) 5 VOTER REGISTRATION AND TURNOUT IN THE UNITED STATES, 1952-1976 By Wiiiiam Austin Boyd A DISSERTATION Submitted to Michigan State University in partiai fquiIIment of the requirements for the degree of DOCTOR OF PHILSOPHY Department of Politica] Science 1980 ABSTRACT VOTER REGISTRATION AND TURNOUT IN THE UNITED STATES, 1952-1976 By William A. Boyd The question of the relationship between voter registration and turnout is one of concern to scholars of political science as well as politicians. This research considers two well-known models of voter turnout, generalizes the models to include the citizen's decision to register, and finally tests these models using multiple regression analysis on state-level data over the time period l952-l976. The calculus of voting introduced by Riker and Ordeshook (1968) and the minimax regret approach to voter turnout developed by Ferejohn and Fiorina (1974) constitute two models of voter turnout which are familiar to many political scientists. These models are extended in chapter 2 to include the citizen's registration decision, and the aggre- gate (state-level) and individual«level implications of this generali- zation are explored analytically. The models are examined in both their standard cross-sectional form and in the context of dynamic, first- difference equations. The construction of suitable measures for concepts is the primary t0pic of chapter 3. The principal difficulty is to construct proxy variables at the state level corresponding to the individual-levelcon- cepts of expected candidate differential, the costs of voting, and citi- zen duty. This attitudinal simulation is only partially successful. The test of the theories indicates that neither the calculus of voting nor minimax regret are fully adequate theories of voter regis- tration and turnout in the U.S. This becomes especially apparent in the test of the dynamic models. In the cross-sectional testing of the theories, the minimax regret decisionmaking model best fits the data. Other conclusions are that same day voter registration should increase voter registration by about seven percent and long-term voter turnout levels by about five percent. Literacy tests seem to have had no significant impact on levels of voter registration or turnout. ACKNOWLEDGEMENTS Many people contributed to the completion of this study in their training, encouragement and understanding, without which this work would have suffered enormously. Professor John Aldrich provided my intro- duction to the beauty and power of formal analysis. His detailed comments and criticisms of early drafts of this work provided me with guidance and minimized the numter of blind alleys down which I doubtless would have stumbled. Professor Charles Ostrom's incisive questioning of the formal structure of the model development provided the inspiration for the full formalization presented in chapter 2. Professor Paul Abramson's expert grasp of the voting literature as well as of the mechanics of survey research proved to be of tremendous assistance in my survey analysis, and his entreaties not to take the research too seriously helped me to keep this project in perspective. My experience working as research assistant for Professor Robert Jackman allowed me to see first-hand the process of research design and data analysis. I can think of no one individual who could have benefitted my scholarly growth more than Bob Jackman. His encouragement to complete this work has taught me that the best research is that which is completed. In addition to the professional obligations which I owe my friends and colleagues, I am deeply indebted to my family. My daughter, Jennifer, will be happy to know that "the imitation" is done. By her frequent questioning of that fact, she provided a strong motivation for ii completing this work. My wife, Rhonda, typed with painstaking effort both the initial and final drafts of this thesis. Her constant presence, as well as her confidence and encouragement made this work much better than I expected it could be. For this reason, I dedicate this thesis to her. TABLE OF CONTENTS LIST OF TABLES vi LIST OF FIGURES vii CHAPTER l. VOTER REGISTRATION AND TURNOUT: THE SALIENT ISSUES 1 l.l Introduction 1 1.2 The Contemporary Setting 2 1.3 Variables Pertaining to Voter Turnout 4 1.4 Socioeconomic Variables and Voter Turnout 5 1.5 Attitudinal Variables and Voter Turnout 9 1.6 The Electoral Context and Voter Turnout 12 1.7 Legal Impediments and Voter Turnout 15 1.8 Variables Pertaining to Voter Registration 17 1.9 Summary 20 CHAPTER 2. TWO FORMAL MODELS OF REGISTRATION AND VOTING 24 2.1 Introduction 24 2.2 Theoretical Primitives and Early Development 24 2.3 Two Formal Models of Voter Turnout 28 2.4 Some Considerations of Model Selection 31 2.5 Some Individual-Level Extensions of the Calculus of Voting 34 2.6 An Extension of the Minimax Regret Model to Include the Registration Decision 37 2.7 The Calculi of Registration and Voting 44 2.8 The Calculi of Registration and Voting: A General Specification 51 2.9 An Aggregate Specification of the Minimax Regret Model 60 2.10 Summary 61 CHAPTER 3. A STATE LEVEL SIMULATION OF CANDIDATE DIFFERENTIALS. VOTING COSTS AND DUTY 65 3.1 Introduction 65 3.2 Approaches to Attitudinal Simulation 66 3.3 Assumptions of the Simulation Model 71 3.4 The Theoretical and Methodological Implications of Aggregation 73 3.5 The Effects of Aggregation on Variables in This Study 78 3.6 A Simulation of Candidate Differential, Information Cost and Citizen Duty 83 3.7 Validation of the Simulation: Expectations 94 3.8 Validation of the Simulation: Results 101 3.9 Summary 109 iv CHAPTER 4. A TEST OF TWO FORMAL THEORIES OF REGISTRATION AND TURNOUT .1 Introduction .2 Operational Specifications of the Aggregate Hypotheses .3 Estimation and Interpretation of the State Registration and Turnout Equations .4 Estimation and Interpretation of the Dynamic Registration and Turnout Equations 4.5 Summary -> #h-b CHAPTER 5. VOTER RETISTRATION AND TURNOUT: SOME EXTENSIONS AND CLARIFICATIONS 5.1 Introduction 5.2 The Place of These Findings in the Literature of Voting Behavior 5.3 Summary APPENDIX BIBLIOGRAPHY 114 114 115 118 136 148 152 152 159 163 168 2.1 2.2 2.3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 LIST OF TABLES A Derived Decision Table of Utilities Incorporating Registration Regret Matrices From Derived Table of Utilities A Minimax Approach to Purging and Voting Relative Weights Applied in the Simulation of Mean Candidate Differential Relative Weights Applied in the Simulation of Level of Political Information Relative Weights Applied in the Simulation of Political Efficacy Mean Survey Scores for Political Efficacy and Interest in Five States Simulated Scores on Political Efficacy for Five States Simulation Scores on Level of Political Information for Five States Simulation Scores on Interest for Five States Cross-Sectional Registration Results, 1952-76 (Calculus of Voting) Cross-Sectional Registration Results, 1952-76 (Minimax Regret) Cross-Sectional Turnout Results, 1952-76 (Calculus of Voting) Cross-Sectional Turnout Results, 1952-76 (Minimax Regret) The Evaluation of Hypothesis 4.1 Dynamic Registration Results, 1960-76 (Calculus of Voting) Dynamic Registration Results, 1960-76 (Minimax Regret) Dynamic Turnout Results, 1960-76 (Calculus of Voting) Dynamic Turnout Results, 1960-76 (Minimax Regret) vi 40 41 49 88 9O 93 103 103 106 106 122 124 126 127 132 140 142 144 146 1.1 2.1 LIST OF FIGURES A Summary Representation of Variables Pertaining to Registration and Turnout A Hypothetical Decision Tree vii 22 52 CHAPTER 1 VOTER REGISTRATION AND TURNOUT: THE SALIENT ISSUES 1.1 Introduction The principal orientation for this research emerged as the result of the inability of the literature in political science to answer what seems to be a very basic question: What is the relationship between voter registration and voter turnout? This presents a particular problem for those interested in examining the effect of state voter registration provisions on turnout, but it also concerns scholars interested in the nature of the sequential decision process involved in the voter's choices to register or not, to vote or not, and the candidate to whom he gives his vote. Any attempt to systematically explore this area naturally leads to a number of sub-issues. For example, a basic decision must be made about the time frame of such a study, the appropriate "level of ana- lysis" for evaluating the relationship, and, of course, the relevant variables and theoretical orientation with which one begins the exam- ination of the problem. It is signifiCant, perhaps, that questions about registration and turnout have been recognized as of public interest, but the principal aim of this work is to suggest one method for the analysis of the relationship. In so doing, the primary purpose of this chapter is to introduce those variables which will be considered in the analysis which follows. The second chapter of this work will provide the theo- retical basis for examining the questions set out, and the analysis will be contained in the third and fourth chapters. Finally, in the fifth 2 chapter I will speculate on the extent to which the results obtained in the previous two chapters might be sensitive to the particular theo- retical orientation developed in chapter 2, and other methods for examining the problem will be discussed at this point. 1.2 The Contemporary Setting For the past several Congresses, proponents of registration reform have argued that steps must be taken to ease registration requirements in order to increase voter turnout. As a normative goal, high voter turnout has been criticized by some social scientists (Berelson et a1, 1954; Almond and Verba, 1965), but among politicians and the public at large, 'tt seems to find general acceptance. Although numerous proposals have been offered to increase registration, that which has received most attention has been "same day voter registration". The nature of this proposal is to permit qualified voters to register at polling places on the day of an election, provided they can present acceptable proof of identity and residence. Funding would be provided by the national government to individual state governments to pay for the implementation of these new registration procedures (Congressional Quarterly Neek1y_ Report, May 14, 1977). This constitutes only the latest attempt of many to encourage higher levels of voter participation by making the process of registration less cumbersome. The most notable earlier attempt to ease registration procedures involved the debate over postal registration (U.S. Congress, 1971; U.S. Congress, 1972; Smolka, 1975; Phillips and Blackman, 1975). Two questions are really involved in this debate over registration provisions: Will the changes in legal procedures for registration actually lead to increased voter registration? and, Will increases in 2 chapter I will speculate on the extent to which the results obtained in the previous two chapters might be sensitive to the particular theo- retical orientation developed in chapter 2, and other methods for examining the problem will be discussed at this point. 1.2 The Contemporary Setting For the past several Congresses, proponents of registration reform have argued that steps must be taken to ease registration requirements in order to increase voter turnout. As a normative goal, high voter turnout has been criticized by some social scientists (Berelson et al, 1954; Almond and Verba, 1965), but among politicians and the public at large, it seems to find general acceptance. Although numerous proposals have been offered to increase registration, that which has received most attention has been "same day voter registration". The nature of this proposal is to permit qualified voters to register at polling places on the day of an election, provided they can present acceptable proof of identity and residence. Funding would be provided by the national government to individual state governments to pay for the implementation of these new registration procedures (Congressional Quarterly Weekly_ Report, May 14, 1977). This constitutes only the latest attempt of many to encourage higher levels of voter participation by making the process of registration less cumbersome. The most notable earlier attempt to ease registration procedures involved the debate over postal registration (U.S. Congress, 1971; U.S. Congress, 1972; Smolka, 1975; Phillips and Blackman, 1975). Two questions are really involved in this debate over registration provisions: Will the changes in legal procedures for registration actually lead to increased voter registration? and, Hill increases in 3 voter registration be reflected in higher voter turnout in general elections? These issues have formed the basis of the policy-oriented controversy. Broadly opposing reforms in the procedures of voter regis- tration, Phillips and Blackman (1975, p. 2) argue that "registration requirements gg_ggt_appear to be the drag on voter turnout that federal action advocates maintain. Public alienation, boredom, or inertia seem more important." This conclusion is supported by Smolka (1975, p. 83) in his evaluation of the impact of mail registration on turnout. lkasuggests that "no appreciable increase in voter turnout was related to mail regis- tration. Persons who registered by mail during 1974 voted in approxi- mately the same percentages as other registered voters." Further he notes an additional objection to the p1an--the possibility of voter fraud. These arguments are frequently echoed by opponents of the legislation in the House and Senate. The National Municipal League (1973), on the other hand, sees registration reform as a cornerstone in its plan to increase political involvement in the electorate. The major premise of the Model Election System presented here is that the problem of non-voting in America is directly related to the machinery states have created for registering voters and administering elections. The Model is designed to provide states with guidelines to restructure—tFETr administrative systems so that citizens have greater access to the electoral process. (p. 1). Interestingly enough, their work is premised on research undertaken by Kelley, Ayres and Bowen (1967). In their well-known paper, these authors suggest that "registration requirements are a more effective deterrent to voting than anything that normally operates to deter citizens from voting once they have registered." (p. 362) Somewhat less equivocal in this debate are Rosenstone and Wolfinger (1978, p. 41), who, on the basis of their 1972 study, conclude that 4 "given the results that would be achieved by the relatively minor changes...we are confident that establishing a European-type registration system would, at the least, increase voter turnout by substantially more than nine percent." This figure is quite similar to one offered by Vice- President Walter Mondale in a speech in Milwaukee in support of the national election day registration bill (Congressional Quarter1y_Weekly Regggt, May 14, 1977). 9 . Rather than jump directly into the policy debate, I will attempt instead to explore the nature of the registration-turnout relationship. The first step in this process involves a delineation of the principal independent variables of concern. Since voter turnout provides a more developed body of literature, this seems a good place to start. 1.3 Variables Pertaining to Voter Turnout A most impressive feature of the literature on voter turnout is that the set of variables considered to be important has changed little in the half-century since the scientific study of voter turnout began. An example of this is the work by Merriam and Gosnell (1924) and Gosnell (1927; 1930). In this pioneering research, these scholars from the Uni- versity of Chicago isolated such socioeconomic traits as sex, age, race, and nativity as correlates of voter turnout. Moreover, Merriam and Gosnell (1924) spent considerable effort in examining non-voting due to "general indifference" and "disgust with politics". Later researchers would emphasize these psychological factors and term them, respectively, the voter's "sense of citizen duty" and "political efficacy". Further, Merriam and Gosnell emphasized early on the disenfranchising effects of legal restrictions on voting, and Gosnell (1927; 1930) was an early advocate for easing these barriers. Finally, it might be noted that in 5 Gosnell (1930) party competition and electoral closeness were first recognized as major political variables influencing voter turnout. This early literature on turnout, therefore, identified four sets of relevant variables: certain socioeconomic characteristics, attitudinal and psychological variables, legal restrictions on registration and voting, and the state of electoral competition. Since this early work, only one additional variable has been generally accepted as an important influence on turnout. This is referred to by Downs (1957) as the citi- zen's "differential candidate evaluation". This term refers to the expected difference in utility (or benefit) the individual citizen will receive if one candidate, rather than another, is elected to a particular office. Clearly this is an election-specific variable which involves for each individual a comparison of the alternative candidates. Having set out a reasonably well-defined set of causal variables, let us examine them more closely. 1.4 Socioeconomic Variables and Voter Turnout Until recently, one could reasonably argue that the role social and economic variables play in the determination of voter turnout was becoming increasingly circumscribed. At least in the survey-based studies, with the possible exception of the respondent's level of edu- cation, socioeconomic variables have been supplanted by such attitudinal variables as political efficacy and party identification. A primary reeason for the denegration of socioeconomic explanations was the diffi- culty of integrating such variables into a solidly grounded theory of voter turnout. Ironically, this failure of the socioeconomic variables to have a straightforward theoretical interpretation has resulted in renewed interest in them. 6 The best example of this stems from a controversial article by Frey (1971). Observing that people with high incomes tend to parti- cipate more in politics, Frey wonders how this observed relationship might be explained. His attempted explanation is couched solely in terms of the relative economic positions of the high and low income citizens. This viewpoint met with immediate objections (Fraser, 1972; Russell, 1972). Nittman (1975, p. 738) recounts the major issues quite clearly: (1) Higher income people can afford to take time off for voting; but their opportunity cost is higher since voting is so labor intensive. (2) Higher income people have more to lose (by being taxed), but then low income peole have more to gain from government expenditure since they would not be taxed as much as the rich. (3) The educated know more about the election but if they know so much they would know it is irrational to vote. This is the sort of muddle from which a renewed interest in the inter- pretation of the socioeconomic relationships has sprung. In a most incisive observation of this theoretical confusion, Tollison and Willette (1973, p. 59) suggest that One of the problems in this literature is the intermingling of two questions about voting and the consequent failure to distinguish what hypothesis about voting behavior is being specified. In one case the issue is raised as to Why individuals vote, especially in large polities, or what factors explain the absolute turnout of voters in a given election. In another case the related, but distinct, empirical issue of why do greater proportions of high income than low income individuals vote is raised. There also has been controversy over what role variables that are not strictly economic in nature (sometimes termed 'sociological') play in economic models of the voting decision. Further, the effect of information on the incentive to vote has, to our knowledge, not been fully clarified in the economic theory of democracy. This lack of theoretical specificity has also been noted by Nittman (1975) and more generally by Przeworski and Soares (1971). Certainly the failure to rigorously specify an appropriately narrow hypothesis has led to much confusion regarding the significance of this observed 7 relationship between income and turnout. This problem clearly extends to many other of the "relevant" socioeconomic variables as well. It is not my purpose in this review to repeat all the findings linking certain socioeconomic variables to patterns of voting behavior and especially voter turnout. These are readily summarized elsewhere (Campbell et a1, 1960; Verba and Nie, 1972; Berelson et a1, 1944; Lazarsfeld et a1, 1954). Rather I hope to point out certain problems in the analysis of voter turnout which, while not confined to socioeconomic variables, seem quite often to accompany them. Certainly this should provide a more constructive approach for establishing a firm groundwork in my research. Besides the frequent ambiguity of the underlying theoretical rela- tionships underscored above, several problems have tended to plague the analyst employing socioeconomic data. First, one might note the problem of examining seriatim a series of bivariate relationships between various socioeconomic variables and turnout or other forms of voter participation. Such is the approach taken by Glaser (1958) and by Verba and Nie (1972) in much of their research. Both works are characterized by a series of bivariate relationships where the same variation in participation is again and again partitioned by variables which are doubtless highly related. This does not permit a fair assessment of the "true" impact of a given variable as shared variation is passed from one independent variable in one bivariate relationship to another. A sounder approach is seen in Matthews and Prothro's (1966) study of southern blacks. To the extent their analysis would permit, they attempted to control for the effects of a series of variables before making claims about the inde- pendent effects of one on voter turnout or registration. This more 8 complete specification of the relationship provides a fairer method for assessing the impact of each variable on turnout. A second problem which is not so easily resolved concerns the strong relationships which typically exist between socioeconomic vari- ables. Common examples include high correlations between education and income, and race and income. This collinearity can become so pernicious that it is often difficult to determine the relative importance of the individual variables with respect to turnout. The problem is serious enough that Gordon (1968), in his well-known article on multicollinearity and problems of interference, chose collinearity among socioeconomic variables as a primary illustration. This does not necessarily mean that ng_inferences can be drawn from a properly specified socioeconomic model of voter turnout, but it does make inference difficult since different socioeconomic variables do not generally provide much new information because of collinearity. This makes it incumbent on the researcher to clarify the nature of the information he is trying to recover in the socioeconomic approach. The most common assumption linking socioeconomic characteristics to voting is stated in Verba and Nie (1972, p. 135). Here a linkage is drawn between socioeconomic status and civic attitudes. The assumption of this approach is that political attitudes arise out of one's sociali- zation experiences, and that these experiences can be captured to some extent by socioeconomic indicators. From their work, two conclusions can be drawn. First, it is apparent that the overall linkage is relatively weak (r = .39). This suggests the possibility of other causal agents affecting attitude formation. If one adopts the perspective of the 'VZvniel of causality" introduced in The American Voter (Ch. 2). the SEt 9 of potentially relevant variables is large indeed. Secondly, an inde- pendent relationship between SES and voting (controlling for attitude) is unexplained according to Verba and Nie (1972). This could be re- flective of (1) an incomplete specification or, similarly, (2) that SES indicators are imperfectly (or incorrectly) linked to turnout only through political attitudes when they might operate through other vari- ables as well. An acid test of the linkage might be seen in the ability of these SES variables to account for attitudinal changes over time. The evidence of this is mixed (e.g., Converse, 1972, pp. 327-29; Burnham, 1968, p. 21; Scammon and Nattenberg, 1970, pp. 39-40), but the predomi- nant view in light of recent research is that SES variables are unable to account for current trends in political attitudes (House and Mason, 1975). Thus a direct linkage between SES and civic orientations remains a controversial topic. 1.5 Attitudinal Variables and Voter Turnout Perhaps the most important contribution made by Campbell, Converse, Miller and Stokes has been their continuing emphasis on the importance of political attitudes. Rather than begin with more remote causes of voting behavior, these authors (1960, p. 35) propose to focus initially on more proximate factors. "Until we can command an array of attitudes that relates highly to the vote, we can sort out these intentions only by inference. Furthermore, once such an array is at hand, it provides us with a set of empirical priorities to guide research deeper into the funnel (of causality)." Foreshadowing a great deal of future controversy over the explanatory ability of these political attitudes (Repass, 1971; Crewe, 1974), the Michigan authors note in the same passage that ’Weasurement close to the behavior runs the risk of including values that 10 are determined by the event we are trying to predict--that is, the vote decision. To the degree that this occurs, some elements of a system of supposed independent variables may in fact be effects rather than causes." Nevertheless, it is to their credit and our benefit that Camp- bell chose to focus predominantly on attitudinal variables. A further strength of The American Voter lies in its authors' understanding that long term political attitudes alone might not suffice to explain a particular electoral phenomenon (e.g., voter turnout). Rather these attitudes might fruitfully be seen working ig_conjunction with.other types of variables. This is, of course, not to say that attitudinal variables were not emphasized as being of foremost importance among the set of causal agents. Thus, the Michigan authors introduced a further controversy--a determination of the relative contribution of each variable (or set of variables) in accounting for voter turnout. Before moving in this direction, however, let us clarify what types of attitu- dinal and psychological variables are suggested as germane to the voter's decision process. The most crucial attitudes affecting turnout, according to Ihg_ American Voter, are the individual's intensity of partisan preference, level of interest in the campaign, concern over the electoral outcome, sense of political efficacy and sense of citizen duty. Of these atti- tudes, level of interest and concern over the electoral outcome certainly seem more geared to a specific electoral context, although Lane (1959) notes that persons interested in and concerned about one particular election tend to be interested and concerned, more generally, in all elections, and conversely. Nonetheless, Campbell et al suggest that: 11 our measures of interest and concern over the election outcome refer explicitly to the election at hand. As such, they are likely to catch important short-term fluctuations of the individual's political involvement. These measures may tap more enduring orientations to politics as well. (pp. 103-104) Thus "concern" and "interest" appear to be comprised of both long- and short-term attitudinal orientations, according to the Michigan authors. Particular emphasis, however, is given in this literature to voter turnout not as a decision which is resolved at every election. Rather it is primarily a long-term orientation to politics. Inquiry into the determinants of voting turnout is less a search for psychological forms that determine a decision made anew in each campaign than it is a search for the attitude correlates of voting and non-voting from which these modes of behavior have emerged and by which they are presently supported....(S)ome of the dimensions of attitude that are most helpful in accounting for turnout appear to have the character of orientations to politics much more than they do the character of forces acting on a present decision. (Campbell et a1, 1960, p. 93) Ths most significant of these long term attitudes are one's strength of political involvement and sense of political efficacy and citizen duty. The political involvement attitude is measured by the strength of parti~ san identification in the SRC questionnaire, while the efficacy and citi- zen duty variables are measures of somewhat more abstract attitudes. Efficacy basically refers to the degree to which "the affairs of govern- ment can be understood and influenced by individual citizens" (Campbell et a1, 1960, p. 104). Citizen duty is a reflection of the citizen's sense of obligation to vote regardless of the electoral context. Camp- bell et a1 claim that the questions used to ascertain these attitudes formed "cumulative" (e.g., Guttman) scales, but this may no longer be the case.1 Although most of the analysis performed relating to turnout in Ihg_ immerican Voter involves the sort of bivariate association pattern 12 discussed earlier, one result is of particular interest. In their dis- cussion Campbell et a1, find that a linear combination of the attitudes results in a strong multivariate relationship between their components and voter turnout. The most interesting feature of this relationship is that among those respondents at the lowest end of the attitude scale, twenty-two percent voted regardless. The reason for my interest in this is that in this respect a paral- lel can be found in the attitudinal and "economic" orientation to turnout. A frequently made generalization about voter turnout in a "rational" electorate is that since the benefits one anticipates from voting after discounting the value of the vote are quite unlikely to overcome even negligible voting costs,2 for the vast majority of the electorate voting is irrational if a strict investment-oriented approach to the problem is taken. Tullock (1967) provides a strong example to this effect. Never- theless citizens do vote. Thus this problem of underpredicting turnout is shared by both approaches to voter turnout. 1.6 The Electoral Context and Voter Turnout Despite the emphasis in The American Voter on the importance of long-term attitudes on voter turnout, a strength of their work is that short-term election specific factors are also recognized as important determinants. Campbell (1966, p. 41) suggests that in particular the context of a specific election is important in attracting "peripheral" voters into the electoral arena. "These short term forces move the turn- out by adding stimulation to the underlying level of political interest of the electorate, and they more the partisanship of the vote from a baseline of 'standing committments' to one or the other of the two parties." 13 The types of election-specific variables most frequently mentioned are electoral closeness and the expected candidate differential. Electoral closeness is important because where a contest is close, the chance of one's vote deciding the election is much greater (Riker and Ordeshook, 1973). A further reason for close races stimulating turnout may be that they receive more media attention. A well-known feature of media election coverage is the tendency of the media to emphasize the "horse race" aspect of the contest (Patterson and McClure, 1976). This type of stimulation certainly increases voter interest and awareness. If, however, one was indifferent between the candidates running, voting would be a meaningless exercise. For this reason, it is hypothesized that the greater the difference one sees between the candidates (or the more attractive one's preferred candidate is when compared with his opponent) the more value an individual vote will have. Key (1949, p. 523) brings these points together in his seminal work, Southern Politics: A general rule of politics is that the closer the vote between two candidates is expected to be, the greater turnout will be. Observation of southern politics suggests that the theory needs to be amended to take into account the fact that unless potential voters feel deeply about the issues, they will display a vast indifference no matter how close a race is predicted. Although there is widespread agreement that the electoral context is an important determinant of voting turnout, there is substantial dis- agreement about the relative importance of the attitudinal (or con- sumption) aspect of turnout and its contextual (or investment) aspect. These divisions are broadly reflected in divergent theoretical orien- tations, although both sides recognize the significance of the entire group of variables. An example might clarify this point. 14 One of the questions addressed in Campbell's (1966) study of electoral "surge and decline" is why turnout during high stimulus general elections exceeds the most modest turnout of congressional mid- term elections. Related to this question is another: Why are upsurges in voter turnout generally associated with increased votes for one party while the other remains more or less constant? Campbell explains these patterns by distinguishing "core" voters who have a standing committment both of party support and turnout from "peripheral" voters who are drawn into high stimulus elections by context-specific factors. In particular, Campbell (1966, p. 44) suggests: The partisan surge which characterizes a high-stimulus election consists of two components: (1) those peripheral voters for whom the stimulus of highly differentiated party-candidate alternatives provides the needed impetus to move them to the polls and who, depending on the strength of their party identification, are swung toward the ticket of the advantaged party, and (2) those core voters who are drawn from their normal position as Indepen- dents or identifiers with the disadvantaged party to the candidate of the party which is advantaged by the political circumstances of the moment. Moreover, the factor which Campbell stresses as of overwhelming short- term impact is the citizen's expected party (or candidate) differential. On the other hand, however, the clear majority of voters-—core voters-- are affected only slightly by the effects of electoral context. Thus long-term political attitudes constitute the more important determinant of voter turnout. In The American Voter, a similar argument is made except that the contextual variable emphasized is electoral closeness. At the other end of the spectrum, most investment-oriented or eco- nomic approaches to voter turnout rely on the current context as a deter- minant of turnout. Although Downs (1957) speaks of the "long run parti- cipation values" of voting and of the citizen's delight at affirming satisfaction with the political system, he treats the decision whether or 15 not to vote as basically a cost-benefit calculation specifically depen- dent on such contextual factors as the size of the electorate, the expected candidate differential, and the costs of information. Subse- quent work by Riker and Ordeshook (1968) formally incorporated antici- pated electoral closeness into this decision formula. In addition, they explicitly incorporated a variable which represents the positive pay off an individual receives for complying with the ethic of voting. The theoretical emphasis in this literature is primarily on the contextual elements, but, in general, the empirical results have provided little comfort for this orientation.4 1.7 Legal Impediments and Voter Turnout In its 3.929113: the President's Comission on Registration and Voter Participation (1963, p. 464) echoed many of the findings of the research just examined. "By far the most important psychological factor affecting an individual's decision to vote is his identification with a political party." Nevertheless, the Report_stressed that legal procedures for registration constitute a "major reason" for low voter participation. A useful history of the major changes in these legal procedures is con- tained in Dye (1973, pp. 74-80) and Rosenstone and Nolfinger (1978). There is, however, some confusion in this literature which centers about the interesting relationship between voter registration and turnout. Clearly most of the legal variables which are addressed in the 3gpg[t_ impact on voter turnout only insofar as they keep people from registering. Should one assume that such variables as residency requirements and literacy tests affect most directly voter registration or turnout? In this discussion I will focus only on legal provisions which clearly 16 affect voter turnout directly. Remaining legal variables will be dis— cussed as determinants of voter registration levels. Those variables most frequently mentioned which impact directly on turnout are state provisions for (l) absentee voting, (2) hours of poll operation, and (3) time off from work (with pay) to vote. Absentee voting has long been recognized by most states as an important democratic right. In 1950, for example, only two states had no legal provisions for absentee voting (Pennsylvania and South Carolina). Some state laws were more or less restrictive, but by the mid-1950's all states had legis- lative provisions for absentee voting. In the past, many voting groups have argued that the hours of operation of polling places should be extended so that workers who could not take off from work to vote would be permitted to exercise this right in the evening after work. To some extent localities have acted independently of state law to keep polls open longer, but, in general, there has been little systematic variation over time in the hours of poll operation. At least 32 states, during the 1960's, kept the polls open 12 or more hours, but empirical evidence of a direct linkage between the number of hours of poll operation and turnout has never been directly made. According to Zinzer and Hausafus (1978), 30 states have provisions for allowing employees time off from their jobs to vote. Of these 30, 24 require that time off be given with_pgy, Zinzer and Hausafus studied 301 congressional districts during the 1972 congressional midterm elections and concluded that "time off with pay, if implemented nationally, would raise turnout by approximately 4.2 percentage points" (p. 25). The rationale of this increase is essentially that provision for time off 17 with pay would decrease the costs of voting for those individuals who were unable to vote at other times. What can be seen from this brief review of legal impediments to voting is that (1) most legal variables operate only indirectly on turn- out through voter registration levels, (2) in some instances, there is little inter-state variance in legal provisions, and (3) little academic attention has been given to the evaluationmof the impact of these vari- ables. This is reflected in Downs' (1957) treatment of voting costs. He treats these as a combination of the fixed costs of voting and the infor- mation costs of gathering, processing and evaluating information about the candidates. The fixed costs of voting could include lost salary due to taking time off from the job or the cost of returning to one's home in order to vote if absentee balloting were impossible. But the major emphasis in this literature has been on information costs and how a rational voter will attempt to reduce these costs. 1.8 Variables Pertaining to Voter Registration Only rarely has the literature in political science treated voter registration as a dependent variable of interest. A more frequent inter- pretation of the act of registration is an independent variable which "predicts" turnout. Kelley, Ayres, and Bowen (1967, p. 362) note, for example, that "in the full sample of 104 cities, 78 percent of the vari- ation in the percentage of the population of voting age that voted could be accounted for by variations in the percentage of the population that was registered to vote." Similarly, Tollison, Crain and Pautler (1975) found a strong relationship between registration and turnout. This literature does not, however, give us a good indication of the variables 18 which impact on a citizen's decision to register. Such guidance as does exist is provided by Kelley, Ayres and Bowen (1967, p. 361): Registration is hardly ever represented as anything but a purely instrumental act: one registers in order to vote. One would suppose, therefore, that most of the factors that encourage or discourage voting also encourage or discourage registering to vote. Since the linkage between registration and turnout is crucial in the theoretical development which follows, it deserves careful scrutiny. First, one should establish whether it is supported by empirical evi- dence. Some evidence to this effect is supplied by Kelley, Ayres and Bowen themselves. Using socioeconomic variables, contextual variables, and state legal provisions, Kelley, Ayres and Bowen (1967) found that some of each of these groups of variables were important determinants of voter registration levels. Less direct evidence supporting this is found in the article by Tollison et a1. As was previously mentioned, they included registration as an independent variable "predicting" turnout levels. Interestingly enough, in their study contextual variables also included as independent variables are largely insignificant. If regis- tration and voting are both the result of contextual (as well as other) variables, then the inclusion of registration levels in their work might have simply "controlled" out the effect of the included contextual vari- ables. Thus some evidence exists in support of the Kelley assertion. A second point which should be made here is that it seems reasonable to assume that most legal obstacles to voting affect registration levels and only indirectly voter turnout. These variables deserve some attention here. In addition to absentee provisions and the inaccessability of the polls to many individuals, the President's Commission (1963) specifically focussed on the disenfranchising effects of residency requirements. 19 These effects have come about largely because of increasing mobility by the American people. For example, in 1960 it is estimated that residency requirements disenfranchised eight million citizens. Moreover, in the 1952 SRC survey, fully 12 percent of nonvoters cited residency require- ments as their reason for not voting. Other major legal barriers to voting include the existence of literacy tests and poll taxes-~especially in the‘South. The disen- franchising effect of these laws is well known, and is reported in some detail in Key (1949) and Matthews and Prothro (1963; 1966). Key, noting the overt discriminatory nature of many of these provisions, writes (p. 563) "every local registration officer is a law unto himself in deter- mining the citizen's possession of literacy, understanding, and other qualifications. No state agency knows what these officials do or how they interpret the requirements." Key argues further that the poll tax was probably not as invidious and discriminatory as many have suggested since it also had the effect of disenfranchising poor whites. Poll taxes were prohibited in federal elections by the 24th Amendment in 1964. and were abolished altogether by the Supreme Court two years later in Harper vs. Virginia State Board of Education.5 Likewise, literacy tests were sus- pended in much of the South by the Voting Rights Act of 1965, and they were nationally abolished by the Voting Rights Acts Amendments of 1970. Other legal variables which might be mentioned briefly include the state provisions for purging registration rolls and the existence of facilitative structures such as neighborhood registration, and deputy registrars. Certainly the frequency of re-registration required by the states should be one factor influencing the ease of voting. Further, Rosenstone and Wolfinger (1978) note that in some areas citizens must go 20 to a county seat in order to register. In others, there exist local offices where this might be accomplished with less inconvenience. Taken together, there seems to be good reason to expect these vari- ables to influence voter registration totals in the years of this study. Further, there is sufficient variation over time to examine the effects of changes in the past in order to evaluate the degree to which future changes in the law might facilitate turnout. 1.9 Summary In this chapter, I have set out many of the important independent variables which impact on the voter's decision to register and vote. This list is certainly not exhaustive of all variables which have been mentioned in the voting literature--consider, for example, Neisberg's (1977) argument that low voter turnout may be due to an overall "satis- faction" with the system--but it does include those variables which pro- vide the dominant themes in this literature: socioeconomic variables, long-term attitudinal orientations to the political system, election- specific contextual effects, and the barriers of state legal provisions. From this review of the literature, two points deserve special at- tention. First, as one of three related decisions which must be made in order to vote, the decision to register has received far less attention than either voter turnout or direction of the vote. Where it has been explicitly taken into account, voter registration is often treated as a predetermined variable used to explain some other phenomenon, such as voter turnout. As the first of a series of decisions the voter (or non- voter) must make, the registration decision has been largely ignored. A second point of potential importance is that, in our concern for understanding voter turnout, many legal impediments to registration have 21 been modeled as direct effects when, instead, their effects are indirect. The pattern of relationships developed in this chapter is presented in Figure 1.1. Note that no relationship between registration and turnout is posited. The reason for this omission is Egg-that none should exist. but its nature is not yet known. Should the registration decision prove to be a perfect linear combination of turnout (i.e., they are statistic- ally indistinguishable) then no serious harm would result from such a misspecification. Since this relationship is, however, unknown at present, it will form the principal concern of the theoretical development which follows. Only after setting out the nature of this relationship between registration and turnout can the significance of this second point be fully realized. Costs of Registration 22 Socioeconomic Variables Electoral Context 4 .3 V V >Registration Turnout A A /7 Costs of Voting Attitudinal Variables FIGURE 2.1 A SUMMARY REPRESENTATION OF VARIABLES PERTAINING TO REGISTRATION AND TURNOUT 23 FOOTNOTES FOR CHAPTER 1 1See, for example, Campbell, Gurin and Miller (1954), Appendices A and B. Recent results by Balch (1974) suggest that efficacy scores may no longer be cumulative. 21h Tullock's (1967. pp. 110-111) example, a candidate differential of even $10,000 cannot overcome voting costs of $1.00 because of the discount term he applies. 3One should note the dependence of this argument on those developed in Converse (1966). More recent amendments and qualifications to this Eheory have been suggested in Arsenau and Wolfinger (1973) and Kernell 1977 . 4Barry (1970), for example, argues that all the explanatory power of the economic model lies in the "consumption" or citizen duty term. See Aldrich (1976) for an alternative interpretation. 5383 U.S. 663 (1966). CHAPTER 2 TWO FORMAL MODELS OF REGISTRATION AND VOTING 2.1 Introduction Discussion up to this point has emphasized the general inability of the literature dealing with voter turnout to subsume the registration act into a tight theoretical framework. In this chapter, the decision whether or not to register will be incorporated into two models of voter turnout which will be explored in some detail later--Riker and Ordeshook's (1968; 1973) "calculus of voting" and Ferejohn and Fiorina's (1974; 1975) minimax regret model of voter turnout.' This exploration will occur at two distinct levels of analysis. First, following common practice, these models of turnout will be extended at the individual level of analysis. For example, below, Ferejohn and Fiorina's minimax regret decision process will be extended to incorporate the registration decision. Following this, the impli- cations of these individual level results will be explored in aggregate. From this aggregate pattern of development will emerge a new dynamic model which explicitly derives the nature of the registration-turnout linkage. Aspects of both the individual level and aggregate results will subsequently be subjected to empirical analysis in chapter 4. 2.2 Theoretical Primitives and Early Development The orientation for the theoretical development in this work arises from an "economic" approach to voting. By this, I mean that the acts of registering and going to the polls constitute decisions where an 24 25 individual chooses one alternative rather than another if he believes the net benefit of the selected alternative exceeds the net benefit of its alternative. For example, it is assumed that an individual deciding whether or not to go to the polls selects that alternative which results in his greatest satisfaction, or, conversely, minimizes his discomfort. This selection of the most preferred alternative constitutes rational behavior. Let us formalize this notion for the sake of clarity. First, let us define a set of mutually exclusive and exhaustive future states of the world, and let us define this set, 0, such that O = {01...Om} 2.1 At some future time one of these possible states will obtain. Secondly, it is assumed that each individual in society can order these states such that every state of the world is related to others through the relations of preference and/or indifference. For example, if we consider Oj and Ok’ then we can say that either Oj is preferred or is indifferent to 0k, or we can say that 0k is preferred or is indifferent 1 to 0j' In addition to this, it is assumed that the ordering thus ob- tained is transitive. This means that if one prefers Oj to Ok’ and Ok to Ot’ then one must also prefer Oj to 0t in binary comparison. Next we must define a set of alternative actions: a = {a1...a } 2.2 It is assumed that individuals believe that the alternative which is selected will have some effect on the future state of the world which obtains. If the selection of some alternative, ak, leads invariably to 26 a future state, Oj, then the individual is making his decisions under conditions of Certainty. If “k will produce outcome Oj with some known probability, then the individual is deciding under conditions of risk,2 If we define a utility function, U, which associates with each alternative a mumber which Riker and Ordeshook (1973, l4n) call a "measure of preference", then we can also speak of a citizen's expected utility function for an alternative. EU(a]) = P1a] + P20,2 + ... + Pnan 2.3 where Pi is the probability that the selection of a] will result in 01..3 That individual who selects the alternative which has the highest ex- pected utility behaves rationally. In his seminal work, An_Economic Theory Qf_Democracy, Anthony Downs first applied a similar form of logic to the problem of voter turnout. Although his definition of rational behavior differed somewhat from that offered here,4 Downs concerned himself with several of the variables introduced in the last chapter. First he focussed on the expected party differential. This represents the difference to an indi- vidual (in units of utility) if one candiate rather than another is elected. We will define B = |EU(a1) - EU(az)| 2.4 where it is understood thata1 and a2, the alternatives, represent alternative candidates.5 The value of the vote is therefore a function of B. But there are other concerns as well. Downs points out that voting itself is a costly act. Treating the costs of voting and registration together, Downs notes (1957, p. 265): 27 Heretofore we have assumed that voting is a costless act, but this assumption is self-contradicting because every act takes time. In fact, time is the principal cost of voting: time to register, to discover what parties are running, to deliberate, to go to the polls, and to mark the ballot. Since time is a scarce resource, voting is inherently costly. From this, we see that the voting decision can be seen as a cost-benefit problem. If the costs of voting are denoted CV, then the value of the vote, V, is equal to benefits less costs, or: V=B-Cv 2.5 A further consideration to this problem, however, must be introduced. A citizen who decides to vote is not guaranteed that his preferred candi- date will win. Therefore B must be somehow discounted. Downs intro- duced a discount factor which has been interpreted as the probability that an individual's vote is pivotal in either creating or breaking a tie between the candidates (Aranson, 1972). This term, denoted P, is applied to discount the value of B--a recognition that the individual voter is only one in a potentially large electorate. Thus equation 2.5 must be modified: V=PB-C 2.6 V Finally, although not often recognized, Downs (1957, pp. 267-268) suggests that in addition to these factors, people vote because of civic orientations: 1. Rational men in a democracy are motivated to some extent by a sense of social responsibility relatively independent of their own short-run gains and losses. 2. If we view such social responsibility as one part of the return from voting, it is possible that the cost of voting is outweighed by its returns for some but not all rational men. 28 From this, it is clear that the value of a vote is a combination of election specific factors and costs and benefits which do not depend on the success or failure of one's preferred candidate. Downs' theory of voter turnout can now be fully specified with a new term, 0, repre- senting a positive utility for satisfying one's social responsibility. V = PB - Cv + D 2.7 This result forms the basis for subsequent development by Riker and Ordeshook, and Ferejohn and Fiorina. 2.3 Two Formal Models of Voter Turnout In his original work, Downs (1957) had partially formalized his understanding of voter turnout, but subsequent work by Tullock (1967) and Riker and Ordeshook (1968) led to a more complete specification of Downs' ideas. Tullock's work emphasized the investment aspect of voting even more than Downs' original work, insofar as a positive payoff for fulfilling one's obligations as a citizen was ignored. It was also Tullock who made most forcefully the point that given a strict invest- ment orientation, voting was for the most part an exercise in irration- ality. This is because even small costs would outweigh one's expected gains after the value of the vote was discounted. Not satisfied with Tullock's specification, Riker and Ordeshook (1968, 1973) reintroduced the "D" term and also extended Downs' work by reinterpreting the discount term, P. P, Riker and Ordeshook argue, should reflect more than the size of the electorate. Another variable also influences the value of an individual's vote--the closeness of the election. This is clearly evident where the race is so close that a tie is expected. If this is the anticipated result before I cast my vote, 29 then my vote will "decide" the contest and therefore no discount need be applied to the vote at all. Further, because of propagandizing or misrepresentation of some sort, voters also feel that their vote is more important than it may objectively be. For this reason, the discounted value of the vote is necessarily subjective. Thus, according to Riker and Ordeshook, the decision problem faced by the potential voter is that of equation 2.7 with the provision that (l) the estimate of P is a subjective estimate and (2) B is not neces- sarily restricted to a calculation based on politically relevant vari- ables--it merely reflects the total benefit one would feel if his most preferred candidate were elected.6 A final point about Riker and Ordeshook's (1968) work should also be noted. They suggest that we may be underpredicting voter turnout in economic models by ignoring the fact that we vote for a series of offices in one election--from President down to school board members, for example. This raises the possibility that the PB interaction termishould pertain to all levels of elections, as should some costs of voting which vary from office to office (for example, the costs of gathering infor- mation about the candidates). This point has not been explored in the past, but it will enter into the theoretical development which follows. Following the work of Riker and Ordeshook, two concerns have been predominant. First, can one really even estimate the appropriate dis- count value for one's vote? If this is not possible, then it might be the case that instead of employing the calculus of voting, some alter- native decision mode might be preferred. Secondly, if, as early evi- dence indicated, much of the empirical support for the calculus of voting depended on the inclusion of the "D" term, one might wonder if 30 the economic model really "explains" at all. This argUment was most forcefully put by Brian Barry (1970, p. 16): Riker and Ordeshook say ...that people vote because they derive satisfaction from voting for reasons entirely divorced from the hope that it will bring about desired results. This may well be true but it does not leave any scope for an economic model to come between the premises and the phenomenon to be explained. Instead, the question shifts back to: 'why do some people have this kind of motivation more strongly than others'? These two points form the basis of Ferejohn and Fiorina's (1974, 1975) departure from the calculus of voting. First, since we can no longer estimate a value for P, we must operate in a decision environment characterized by uncertainty. Here one cannot judge which future state of the world is likely to obtain, thus one selects an alternative knowing only what the consequences will be 1f_any future state obtains. In such an environment, Ferejohn and Fiorina suggest that rational indi- viduals will employ a decision making strategy called minimax regret.8 The second alteration to the calculus of voting was to eliminate the controversial "D" term in order to evaluate the explanatory power of the minimax regret model without reference to it. From a comparative theoretical evaluation of the calculus of voting and minimax regret models, it was clear that in general, the minimax regret model allowed more individuals to vote than the calculus of voting. The compelling logic of voting for some individuals is most appropriately seen in the regret one would feel if one's preferred candidate lost by only one vote. To summarize briefly, we have under consideration two alternative models of voter turnout--the calculus of voting, developed by Riker and Ordeshook and Ferejohn and Fiorina's minimax regret model. The prin- cipal difference between these models, as Aldrich (1976) has pointed out, is whether or not "there is a 'P' in participation". At a more 31 serious level, the question is one of identifying the appropriate decision making environment under which individuals decide whether or not to vote. 2.4 Some Considerations of Model Selection As indicated, this work focuses exclusively on two formal theories of voter turnout. Since this is not an exhaustive listing of all theories of voter turnout, some justification for the selection of these two models is clearly in order. The initial reason for selecting these models was that both are strongly rooted in theoretical development. By this, I mean that the conditions under which the individual citizen, i, is expected to vote are derived formally from a basic set of assump- tions, and that considerations other than simply "goodness of fit" have been emphasized in the development of this literature. Thus a solid theoretical framework underlies much of the subsequent development. A corollary of this is that whereas some difference of opinion may arise as to the operationalization of each model, particularly at the aggre- gate level, the basic primitives of each model are incontrovertible. A further strength of these formal models is that they are more interesting than other approaches. As deductive models, if the assump- tions of the models are correct, then, of necessity, so are the con- clusions which are deduced from these assumptions. This allows for the potential discovery of findings which are non-obvious. It means further that, like physical scientists who are able to predict events which have not been empirically observed, political scientists employing similar methods may be able to discover linkages which had hitherto goneeunrecog- nized. Employing methods of theoretical development based only on empir- ical observation does not permit the derivation of non-obvious findings 32 if theoretical statements do not go beyond offering explanations for "discovered" relationships. Formal models, while considering empirical findings, emphasize moving beyond conjectural theorizing to a systematic exposition and testing of such conjectures. What, then, are the drawbacks of formal analysis? Two are most frequently mentioned: (1) the inability of the "average" man to perform the calculuations suggested by many formal theories and (2)°the omission of much detail from these rather simplistic formal theories. Let us see how serious these drawbacks are. The first problem of the complexity of calculation is a serious concern. For an individual actually to address the formal decision problem posed by minimax regret would be most unusual. Friedman (1953) suggests that what is important is not necessarily that men actually perform such intricate calculations. Rather he suggests that a theory posits simply that individuals act "as if" they perform these exercises. In other words, according to Friedman's point of view, the utility of a theoretical orientation lies not in the plausibility of its assumptions but in its empirical applicability. The primary difficulty with such a response to critics is that true (i.e., empirically valid) conclusions may follow from false premises. This difficulty, therefore, requires that the premises have some plausibility. With respect to this study, I assume that citizens see registration and voting in the context of a decision which involves a comparison of costs and benefits of alter- native courses of action. The different approaches to this cost-benefit calculation suggested by Riker and Ordeshook and Ferejohn and Fiorina merely constitute two plausible methods by which costs and benefits may be compared. The plausibility of such approaches pertains not 33 to the specific nature of each calculation, but to the decision environ- ment posited by each model. It seems reasonable to argue that if citizens see the decisions with which they are faced as cost-benefit problems, then their internal decision making processes may_approximate either of these two approaches. Another concern voiced over formal models is that they frequently oversimplify complex phenomena. A classic example of this is Downs' (1957) definition of a political party as a "team" of men all having the same preference schedule over alternatives. Similarly, many object to the notion that parties, candidates or individuals can adequately be modeled as being concerned with a single goal (Knight, 1921; Conrad, 1970). Again, Friedman would respond that the power of a theory lies not in its simplicity or complexity, but in its empirical relevance. My feeling is that such a view is too narrow. One must look again at the costs and benefits of such simplification. Surely in most contexts individuals pursue multiple goals. But such complexity necessitates some (possibly arbitrary) weighting of goals on the part of the re- searcher if he is to determine the expectations of a theory when goals come into conflict. For example, if we agree that political parties pursue numerous goals such as winning office, maintaining ideological cohesion, and promoting membership growth, then our theory of parties will be intellectually sterile unless we stipulate the relative import- ance of these goals when conflicts arise. Simplistic theories sacri- fice depth for analytical insight. Much of the power of such formal theories derives from simplifying assumptions. As such, the decision which is implicitly made by the researcher is that greater analytical rigor is worth the costs of simplicity. It should be noted also that 34 all explanations of political phenomena involve some degree of simpli- fication. Thus, unless one confronts the staunchest followers of Winch (1958), the true debate is not over whether or not to simplify, but on the degree of simplification which is desirable. The primary orientation of this work derives from a view of conduct of scientific inquiry suggested by Richard Rudner (1966, p. 11): We are all familiar with the view that it is not the busi- ness of science to collect unrelated, haphazard, disconnected bits of information; that it is an ideal of science to give an or anized account of the universe--to connect, to fit together in relations of subsumption, the statements embodying the know- ledge that has been acquired. Such organization is a necessary condition for the accomplishment of two of science's chief functions, explanation and prediction. But the sort of systematic relatedness exemplified among the statements of scientific:theories is deductive relatedness. Accordingly, to the extent that a theory has been fully articulated in some formulation, it will achieve an explicit deductive development and interrelationship of the statements it encompasses. The idea behind this research is to work with existing theory and its deductive extensions, and to contribute a new theory of voter regis- tration in order to explore the interrelationships between the regis- tration and turnout decisions. As such, it is hoped that this work will contribute to a more general understanding of these aspects of political participation. 2.5 Some Individual-Level Extensions of the Calculus of Voting The starting point of this analysis is the assumption that each citizen employs the calculus of voting in deciding whether or not to vote. That is, the citizen will vote in some election at time t, if and only if Vt > O in the following calculation: V = P B - C t t t + D 2.8 vt t where PtBt represents the interaction of the individual's subjectively estimated probability of affecting the outcome and the expected candi- date differential, C , represents the costs of voting, and D repre- vt t sents the citizen's feeling of civic obligation. The t subscripts emphasize that all these estimates are obtained at time t. As was noted earlier, although their argument has been largely ignored in subsequent analyses of the calculus of voting, Riker and Ordeshook (1968) suggest that both Pt and Bt can be viewed as vectors over the set of elections at t. For convenience, I define the vector, Pt, p”, p2t”°pnt' and the vector Bt, b”, b2t"'bnt over the set8, of elective offices at t. By assumption, the ith citizen, if N, the set of all eligible (i.e., eligible to register) voters, is able to associate with each office, e168 , a particular pit-20 and bitzo' Furthermore, following Tullock (1967, pp. 110-114), I partition the costs of voting, Cvt’ into the fixed costs, Cft’ which are entirely independent of the ith office, and information costs which pertain to the costs of gathering th elective office. These are de- and analyzing information about the i noted cit' Fixed costs include such factors as transportation costs to go the polls, costs, if any, of leaving work to vote, and the like. Information costs pertain strictly to one office, which implies that these costs vary for each individual office while the fixed costs of voting do not. Riker and Ordeshook (1968) suggest that in the voter's calculation of his expected utility, the PtBtproduct is a scalar which is the sum of the component elements of the Pt and Btvectors. Thus, they suggest for k elective offices, e], e2...ek: 36 P.B b b t t = p1t 1t + p2tb2t + "‘ I pkt 2'9 kt Allowing for the distinction between fixed and variable costs, and de- fining Vt to be the citizen's expected utility of voting, the Riker- Ordeshook specification is: k Vt =j§](pjtbjt - Cijt) - Cft + Dt 2.10 If we define the subset, TEEI, of m elective offices which meet the condition: \/ e .1” T5:( (pjtbjt- cm)>0 2.11 then a necessary and sufficient condition for turnout in the Riker- Ordeshook model is that Maggi JtJtc1.J.,c)-cft+pt>o 2.12 The implications which may be drawn from this statement are fruitful in understanding other aspects of voter turnout as well. For example, it is clear that for the voter to vote in all k elections it must be the case that: -Vj€ap.b.>c.jt and;(p.b )>C -D 213 jt jt j- _] jt bjt Cijt ft t ' A corollary to this is that if we can establish that the necessary and sufficient conditions fOr voting have been met, we can predict roll-off9 by examining the set of ei offices and by observing for which offices pitbit> cit and for which offices this is not the case. This raises an important empirical issue, however. Consider the following case where ten offices are being contested (e.g., j = 1,2...10) and the numerical entries represent the expression (pjbj - Cij) for individuals a and b. 37 10 = 999339533 2 °b"'=]0 a (10 O O O 0 O O O O O) j=1(pJ J C13) 2.14 10 = sasssasss 'b-"°= b ( 1 o o o o o o o o 0) jEl(pJ J C13) 10 The Riker-Ordeshook model assumes these cases to be the same, although a good case could doubtless be made that they are in no way equivalent. The employment of the roll-off argument as a behavioral manifestation of this difference is not satisfactory, as an equally extreme example could be constructed with all entries positive. If we are forced to reject the notion that utility can be summed across offices, then the sufficiency condition for turnout becomes much more restrictive: 3318: (p31)j - cm.) > c - D 2.15 f While this formulation still recognizes Riker and Ordeshook's contri- bution in acknowledging multiple levels of office, it denys the appropri- ateness of summing utilities. Although expression 2.15 has been more frequently employed in empirical analyses of voter turnout (reflecting the standard concern with a single office), in chapter 4 I will focus on several offices, thereby assuming the conditions for voting derived in expression 2.12. 2.6 An Extension of the Minimax Regret Model to Include the Registration DeEision In this section, I take the minimax regret (MMR) model developed by Ferejohn and Fiorina (1974, 1975) and offer a straightforward extension. The theoretical model thus developed is able to account for (1) voting for one's preferred candidate, (2) registering, but abstaining, and (3) failing to register. The implications of the MMR model are then compared 38 with those of the calculus of voting. Throughout this discussion, how- ever, one should note that this extension pertains to a single election where those confronting this problem must pptp_register and vote. Thus this calculation would not apply for those persons who need not register. The impetus for this extension was suggested in an article by Aldrich (1976). His specification incorporates two significant new de- velopments into the Ferejohn and Fiorina model. First, by including the registration act in the decisional framework, Aldrich generalized the MMR model to include a wider range of voting-related activity. A further contribution he makes is the inclusion of registration and opportunity costs associated with this more general model. Thus Aldrich identifies three types of costs: c], the costs of voting; c2, the costs of registration; and c3, the opportunity costs associated with failing to register and subsequently discovering that one wanted to vote for his preferred candidate. It should be pointed out that these opportunity costs differ from those typically discussed by Niemi (1976) and others where they refer primarily to the costs associated with voting instead of doing something else. Here opportunity costs refer to the sense of disappointment which could arise from being unable to vote when, at the time of the election, this would be the preferred alternative. An unfortunate drawback of Aldrich's analysis is that it does not permit one to identify the circumstances under which each alternative action will be undertaken. Instead he states the condition (1976, p.721) that one votes if (c1 + c2 + c3) < %. Fortunately it is possible to carry Aldrich's analysis one step further. This is accomplished by noting that one's opportunity cost is really the "regret" one finds when Prohibited from voting by virtue of not having registered when such is 39 the desired action. In Table 2.1, below, this would occur when the out- come is tied (53) or where one's preferred candidate loses by one vote (S4). The "regret", or opportunity cost, in either case is 8 - c1 - c2. Table 2.1 is obtained by substituting this expression for c3 in Aldrich's 10 result. From this, regret matrices can be calculated. Two matrices are necessary because if c2< % - g1, R,A is the greatest entry in columhs,S], S2, and SS, whereas if this is not the case, NR is the highest utility payoff. These regret matrices are displayed in Table~2.2. Careful evaluation of the regret matrices leads to the following conclusions: 1. If Case A obtains, one votes for his preferred candidate (V1) if c1 < .25; otherwise he registers but abstains (R,A). 2. If Case B obtains, the condition for voting (v ) is that c + c < .375 (c + c has a theoretical maximum of .5); otherwise he doesInot Fegister (NR). Thus a relatively straightforward extension of Aldrich's theoretical development leads to a specification of the conditions under which the voter will vote for his preferred candidate, register and abstain, and not register. Substantively, expression 1 above suggests that when Case A obtains, the citizen should vote for his most preferred candidate if the costs of voting are less than one-fourth the utility of having his most preferred candidate in office. If the voting costs are greater than that, the citizen is expected to register but abstain from voting. Case B suggests that the citizen will vote for his most preferred candidate if the sum of the voting and registration costs is less than .375. This condition is somewhat less restrictive than that of Case A in that the costs of voting can exceed one-third the utility of electing one's pre- ferred candidate and voting is still possible. If this condition is not met, then the citizen does not register and, therefore, will not vote. 4O .mopocu m.:m~wu_u mwsa mcwpcsou pzozupz nmcwesmpmv «cm mmsoUpzo pmpucmpon m>pum=msxm wcm m>wm=~oxw a—Fmaazs mmmgpn .w u a?“ a we can .o u mamcwucmu vwccmwmcq commmp one we xu___u: ms» ._ u m:_::wz mpmcwucmu eattmcaca m.m=o to »u___p= we» pogo umszmma wt 8“ .Aomu .a .mem_v cottu_< acct umpaee N> mpmcwucmo Nu . Fe - Nu - Po - M No . Fu . P No 1 Po - F No - Fo . P umccwemcm cow muo> P> muo> mco muo> mco w_nmp_m>< cash mcoz am «90> 8:0 mm wuo> mco xm gush mcoz Am mmuwogu memo; mucuwccmo memo; mumuwucmo mew: mamupucmu mew: oumwwccmu umccmemcm coccmwmca owe umccmemcm cmccmwmcm n=u~co2 mg» to mwumum= co mmeouuzo mFQPmmoa ozo~h~mua < —.N u4m k - El_ 2 Possible Outcomes or "States of the World" Choices Available S1 S2 S3 S4 S5 V2 2c1+2c2-% 2c1+2c2 l 8 2c1+2c2-% NR 0 O l-2c1-2c2 1-2c1-2c2 O Outcome: V1 if c1 + c2 .375, otherwise NR 42 As was the case with Ferejohn and Fiorina, and Aldrich, voting against one's preferred candidate (V2) in the two-candidate, plurality election modeled here is always dominated by other alternatives. In contrast to the MMR model, the implications of the expected utility model can also be drawn from Table 2.1. With the calculus of voting, it is assumed that some probability, pi, is linked to each state such that pi 2,0 and .g 1 EU(VI) > EU(E,A) iff: pi = 1. Then it becomes apparent that l [:p1(l-c1-c2) + p2(l-c1-c2) + p3(1-c1-c2) + 94(8-c1-c2) + (l-pl-pz-p3-p4) <-c1-c2X] > [:p1(1-c2) + p2(l-c2) + 93(8-c2) + p4(-c2) + (l-p1-p2-p3-p4) (-c2):l which reduces to the condition that: 8(93 + p4) > c1 2.16 It is interesting to note that with one qualification this is the same expected utility result derived by Ferejohn and Fiorina (1974) in their examination of the choice between voting and abstention. The quali- fication is that Ferejohn and Fiorina implicitly lump the costs of registration with those of voting to achieve their C. Here these are separate. In this sense, therefore, my result is less restrictive than theirs. Other results may be summarized briefly: EU(V1) > EU(NR) iff: a + %(p3 + p4) > c1 + c 2.17 2 EU(R,A) > EU(NR) iff: e - fl_> c2 2.18 2 43 In equation 2.17, note that if c1 + c2 reaches its maximum (8). voting is impossible because the conditions of the inequality cannot be satis- fied. The third result is also of interest as this condition is iden- tical to that discriminating the two regret matrices of Table 2.2 In Table 2.2, we find the result in equation 2.18 mirrored in that where EU(R,A) > EU(NR), the R,A row is the row where S1, S2, and S are 5 entries of minimum regret. Conversely, if the inequality is reversed (as is the case in Case B) NR proves to have minimum regret in these columns. To continue the comparisons of MMR with the expected utility result, consider the condition for voting under MMR where c2 < 8 - c]. If c1 < .25, what does this suggest in terms of the expected utility model? Substituting .25 for c1 in equation 2.16 above implies that p3 + p4 (the probabilities of creating or breaking a tie for one's pre- ferred candidate) must be greater than .5. This is hardly an encour- aging result for expected utility maximizers. It illustrates further Ferejohn and Fiorina's contention about how the true p must be inflated by the individual to make voting a seemingly rational result under the expected utility model. Should the utility maximizer object to this specification of the issue and insist on entering 0 into the equation, then he is still susceptible to the familiar criticism that l'all the action is in the 0 term".. Should Case B obtain, the condition for voting under MMR is that c1 + c2 < .375. Inputting this into equation 2.17 above suggests again that p3 + p4 > .5, and the same criticisms as those for equation 2.16 apply. Thus it is apparent that if p3 + p4 > .5, the individual should invariably vote (although he may vote at lower sums of p3 and p4 if the costs are lower). It seems clear that, like Ferejohn 44 and Fiorina (1974) and Aldrich (1976), this analysis also suggests that the MMR model is more facilitative of turnout than the voting calculus. Why is this? In his discussion, Barnett (1973) links the minimax process of decision making to a particular Bayesian prior distribution. Specifi- cally, he suggests (p. 228) that the minimax decision rule "is the Bayes' decision rule with respect to the 'least favorable prior distri- .pppipp' for O (the parameter of interest); that is, the one which has the highest risk." (emphasis in original) Thus it is the inherent pessimism of the minimax decision process (in selecting the "best" of a bad lot) which lends itself to this tendency to be more accomodating in allowing citizens to vote. Ferejohn and Fiorina (1974, p. 535) depict the minimax regret decision maker as one who "if asked why he voted... might reply 'My God, what if I didn't vote and my preferred candidate lost by one vote? I'd feel like killing myself.'" The logic of voting irusuch a case is compelling. Were probabilities relevant to the decision pro- cess, this rather unstable individual might choose a less vigorous acti- vity on election day, such as gardening or sailboat racing. 2.7 The Calculi of Registration and Voting Because this discussion is concerned with registration, R, and voting, V, at both the individual and aggregate levels, I adopt the notational convention that the absence of an i subscript denotes the aggregate meaning. For example, Rt is taken to mean percentage of citizens of voting age who are registered at time t, while Rit repre- sents the value, in utiles, which is the outcome of this decision problem: 45 R. = P B 1t it C ' + D it ‘ Cr11; ' Vlt it 2°19 where the meanings of Pt’ Bt’ CVt and 0t are as defined in equation 2.8, where Crt refers to the costs of voting pertaining directly to regis- tration, and where the i subscripts indicate this is an individual decision problem. In the aggregate, we simply note that Rt = P B - c - c + o 2120 t t rt vt t Following a similar logic, expression 2.8 can be written in the aggregate as: V = P B - C + Dt 2.21 As is readily apparent, the difference in the voter's registration and turnout decisions entails the costs of registration, Cr' Before pursuing this further, let us recall some other interesting features of the relationship between registration and turnout. This may affect the models presented here in that the perceived utility of voting, and particularly the PB interactions may differ over this short period of time. Presumably, if one were concerned with the probability that PB > Cv - D, then the point estimate of this probability as well as the kurtosis associated with the distribution might vary between the time of registration and voting. One would suppose that as election time approached the distribution might narrow progressively toward a point in much the same fashion as a consistent estimator in the limit (Kmenta, 1971).11 This analysis begins initially with the opposite assumption, for the sake of simplicity, and it is subsequently relaxed to clarify its implications. Specifically, in the initial specification it is 46 assumed that all variables are identically perceived at registration and at the time of the voting decision. This initial assumption leads directly to the interesting result that everyone who actually registers at time t is also expected to vote since registration, R, is equal to the expected utility of voting less the costs of registration. Thus, the initial utility of voting at t must be higher than the utility of registering unless Crt = O, in which case Rt = Vt, and again regis- tration implies voting. Another feature of the relationship between registration and voting is essential to my argument. This is that in most states, registration at t is a necessary prerequisite to voting at t, but that not voting at t-l mpy_pp_mpy_ppt affect one's registation status at t. This is dependent on the state's provisions for purging from the electoral rolls due to nonvoting. By state law, citizens who have registered in the past may be deleted from the registration listing if they fail to exercise the franchise in a specified period of time. The length of time varies from state to state, but most states do have purging provisions. This suggests that an important analytical distinction might be made among those individuals who are registered at t--those who physically approach the registrar and enroll and those who are "pre-registered" by virtue of having registered at some point in the past and having also avoided being purged. The distinction thus suggested is: Rt = Rpt + Ret 2.22 where Rpt designates those who are "pre-registered" at time t and Ret denotes those just entering (or reentering) the ranks of the registered at t. If the previous assumption regarding identical assessments of the 47 electoral context holds, then we would expect the entire group Ret to vote after registering, whereas this implication does not necessarily follow for the Rpt group, as their registration decision was precipitated by a different electoral context. Thus an asymmetry is introduced which will have important implications in the relationship between turnout and registration. Furthermore, for the Rpt group, no registration costs are incurred whereas Ret must incur these costs if they wish to vote. This serves to reinforce the thesis that Ret will vote at a higher rate than Rpt' This distinction suggests that under some circumstances, it may benefit a potential voter in the R t group to vote even if he has little P interest in any of the e1 races at t if failure to vote will result in being purged (and, if he votes in the future, he will be forced to incur future costs of registration, Cr+k) ppg_if_it is normally the case that his vote value V. will exceed zero. Specifically, an implication of 1t+k the relationship between turnout and registration is that a citizen may actually vote in an election at t even if the conditions for voting in 2.8 are not met, if some more general conditions are satisfied. These con- ditions are that: A Vit+k> 0 and IV“) - Cirt+k< O 2.23 where "hats" emphasize that future values are estimates. A numerical example might clarify this point. Assume the following payoff schedules: 48 93.5.5.1. 2552.2. vit = - .2 - .15 Em = .3 3 Qit+k = '4 2 In Case 1, the calculus of voting as developed here would suggest that a rational citizen would abstain at t and vote at t+k. His net utility for so doing would be 0 + .1 =.1. His condition, however, might be improved since the conditions set out in 2.23 are met. Specifically, if he votes at t, then he will incur a short-term loss (-.2), but his net payoff over two elections will be -.2 + .4 = .2. This strategy clearly dominates the first because no costs of registration are incurred at t+k. In the second instance, the static theory of the calculus of voting would imply nonvoting at t and t+k. By viewing the voting calculus in a more dynamic context, the individual who votes twice will be better off (.05 compared with O). A possible objection to this method of analysis may be that the individual citizen may not know if his failure to vote at t will result in his name being taken from the registration rolls. This suggests that instead of operating under full information, this aspect of the voter's decision problem may be uncertain. Employing e1 minimax decision strategy.12 the individual in this position is faced with the problem in Table 2.3 below. Here it is clear that if condition 2.23 is met, then one should vote at t. If we use the values assigned in Case 1, above, then the security level for the voting alternative is -.2. This result 49 TABLE 2.3 A MINIMAX APPROACH TO PURGING AND VOTING Future State: For Abstention, Citizen i will be: Strategies Purged Not Purged Vote for Preferred Candidate at t Vit - O Vit - 0 Abstain at t O - Cirt+k 50 obtains regardless of future state. On the other hand, if failure to vote results in being purged, the payoff is -.3. Otherwise it is zero. Thus the security level for this alternative is -.3. Since -.2 > -.3, the rational individual would vote under conditions of uncertainty if the conditions in 2.23 were satisfied. This is an important result since it shows that under conditions of uncertainty many more fieople may vote than if they had knowledge of their states' purging provisions. Further, these results suggest that voting models in the past may have underpredicted voter turnout because they failed to recognize the dynamic nature of elections. Instead, confined by available data, they focussed more on static formulations. It should also be pointed out that many of these findings do not accord with common understanding. Here we have seen that high costs of registration may insure higher (not lower!) voter turnout, at least among those already registered. Finally, one should note that the results reported here depend on certain assumptions, the strongest of which is that citizens are "future-oriented"--they consider future payoffs and not just short term losses. Such an assumption could clearly not apply to all citizens. In fact, an orientation toward the future is a crucial indicator employed by Banfield (1970) in distinguishing upper class orientations from lower class. Interestingly, this makes possible a new answer to the question "Why do higher income citizens vote more?" If we can assume some linkage between higher income and future-orientedness, then we have seen in this section that citizens employing a longer time perspective will vote at higher rates. Thus in response to the questions about participation among higher income voters, the response suggested by this discussion is 51 that they may employ a different process for evaluating the costs and benefits of voting--one defined dynamically over time. 2.8 The Calculi of Registration and Voting: A General Specification Let us return to the initial aggregate specifications of the regis- tration and turnout equations (2.20 and 2.21) and see how the distinctions made in the last section pertain to a general model. Recall that our initial assumption of invariant perception between registration and turnout implies that everyone who registers at t will also vote at t. How should the asymmetric relationship between R and V alluded to earlier manifest itself over time? One way of examining this effect is presented in Figure 2.1. In this figure, we see that there are many paths by which one can vote at t. For example, the voter may have chosen just to th election, which is what we would expect of young register for the t voters just entering the electorate and voting for the first time. Or, alternatively, a voter may have registered just once, say at t-3, and having voted in the past elections need not consider the current costs of registration.) Note also that the possibility register, not vote at t-3 (and subsequently) is disallowed because of the implication that to register at t implies voting at t. From this diagram we can see that this assumption allows us to simplify the presentation by disregarding one entire branch of the tree. The dashed lines in the figure display the possibility that failure to vote in one (or even two) elections does not necessarily imply that registration costs must be absorbed in the next election if one wishes to vote. Indeed this feature of electoral law remains the prerogative of the individual states, and is a feature of considerable importance in subsequent model development. If one were 52 Numb onmHuuo 4<8Hhm=poe>= < _.N mm=wHu p F-p N-» m-p eotpu< to ms_e .>z .mzmp mc_mc=a macaw op 5:5 mm_u___atmmoa acmmmcams mmcwp umgmmo mao>111> muo> muoz-->z ape) can“ .tmtm_mmm-->m ”xdx 53 able to further complicate the tree diagram by including all the elections over the oldest voter's lifetime, we would be able to account for all paths linking registration and voting. In fact this is unnecessary. To see why, let us assume that the initial specification above is reflective of a "state of nature" where everyone wishing to vote must register.13 In actuality, everyone is in this position at some point in his life. The purpose of this simple assumption is to allow us to examine the asym- metric effects of registration and turnout over time. For example, at t+1, these models can be specified as: Rt+l vt + w(APB - AC - Aer,”1 + ADt+]) 2.24 t+1 vt+1 V t+1 Rt + (APBt+l " ACvt+1 + A01m) 2-25 where A is a simple difference operator (i.e., C (C vt+1 = vt+1 ' Cvt))' In the registration equation, the coefficient w is a binary variable taking on a value of one if the quantity in parenthesis is positive and zero otherwise. Substantively, if the quantity in parenthesis is posi- tive, this indicates that the overall electoral context at t+1 is more conducive to turnout than at t. This in turn suggests that some portion of the voting age population who did not register and vote at t should be motivated to take this step at t+1. If, on the other hand, the electoral context at t+1 is less conducive to turnout, this reflects itself in an overall decline in turnout since the coefficient w is not present in the turnout equation, but it has no effect on registration since it would imply a counterfactual "de-registration" of those regis- tered at t. Note also that the distinction between Rp and Re is main- tained in this specification in that the Vt voters are already'registered, and that the remainder of expression 2.24 pertains to the effect of the 54 current election of luring previous nonregistrants to the registrar's office (i.e., Re)' This model is still not reflective of electoral reality for two reasons. First, it does not allow for the incorporation of peg potential voters and registrants into the electoral process over time. In addition, in recent years migration patterns have caused substantial discreprencies in the voter registration rolls. Similarly, old voters die, thus leaving the electoral arena. These demographic changes in the electorate can be incorporated straightforwardly into the model. Young voters find them- selves precisely in the "state of nature" described earlier. By assump- tion they are all able to assess the current electoral situation (e.g., they are able to calculate PBt+]. C , and so on), and therefore the vt+1 group means exist for these variables.1 In some cases, these means might be the same as those for the electorate as a whole. For example, the overall perception of electoral closeness should be invariant with respect to age. On the other hand, factors such as decision costs should be relatively higher for the neophyte than for the electorate as a whole. The net gain or loss of voters from a state due to citizen mobility is incorporated in the turnout equation by a constant, 5. This constant is negative when there is a net out-migration and positive when the state is experiencing significant in-migration. The death of a certain percentage of the electorate between elections can be straightforwardly incorporated by multiplying Vt by a constant, A, where 0.: 1.5.1. Incorporating these changes into the turnout model results in a modified specification: V t+1 = ARt + g + (APB + 5t+1(PB C yt+l' yvt+l + Dyt+l) 55 Here<§+1 refers to the percentage of the eligible electorate which is new, and the y subscripts emphasize that the final expression is calcu- lated only over the group of new entrants and may or may not be related to this quantity for the electorate as a whole. The registration equation is similarly modified: R = v_ + w(APB t+1 t t+1 ' AC vt+1 rt+l + 5t+l(PB - C - C yt+l yvt+l yrt+1 + Dyt+l) Note that neither 5 nor 1 affect the registration equation as they do not affect the size of the election rolls immediately. In registration, changes are reflected more slowly and over time by means of the Vt in the registration equation. Thus the first difficulty with the initial speci- fication of the model is easily remedied. A second problem remains. The incorporation of state provisions for purging the registration rolls constitutes the second difficulty with equations 2.24 and 2.25. Discussion in the last section of the paper emphasized that purging provisions combined with high voting costs may encourage more persons to vote than otherwise would, but it is equally clear that its effect on registration can be only to decrease the size of the electoral list. For this reason, it must constitute a negative effect on registration. One would expect, ceteris paribus, that the percentage of citizens who are purged for failing to vote at t would vary with the length of time between purges. For instance, in a state where failure to vote in one election results in being purged, we would expect to find a significantly higher proportion of citizens purged than in a state where purging occurred, say, every ten years. What is unclear is how the results pre- sented in 2.23 and Table 2.3 affect this expectation. For example, with 56 frequent purging and high costs of registration, the curvilinear result suggested above might alter to the extent that ppy_purging provisions would reduce the registration rolls by approximately the same amount. Thus the functional form of the impact purging provisions have on regis- tration levels is difficult to determine theoretically. There are several ways to approach this problem. The simplest method would entail the use of dummy variables, X], X2...Xn_], which would, in conjunction with the intercept, represent the n possible purging rules. This method of inferring the functional form of the relationship, however, would involve a high cost which is manifested in a serious loss in degrees of freedom. A second method involves estimating the separate effects of purging by examining the composition of those Voting over time. Defining N t+k to be those voters voting for the first time in election t+k. we could write the following series of equations: vt = vt 2.28 Vt+1 = B11Vt I B12Nt+1 2'29 vt+2 = 821% l B22Nt+2 T B33(Vt ‘ B11Vt) l B24Nt+1 2'30 ‘ (321 + 82311 ' B11))Vt + B22Nt+2 + B24"t+1 vt+3 = B31Vt I B32”t+3 I B33(Vt ‘ (823(Vt ' 311Vt))) 2'31 1 B34(Nt+1 ’ 824Nt+1) + 835Nt+2 l B35Nt+1 (531 T 833 ' 833822 + 833823811)Vt l (834 I 835 ' 634824)Nt+1 + B35Nt+2 + B36Nt+3 57 The terms on the right-hand side of 2.31 may be straightforwardly inter- preted as: 1. B31Vt = the number of voters who have voted in all elections. 2. B32Nt+3 = new voters at t+3. 3. 833(Vt - (823(Vt - 811Vt))) = the number of voters at Vt who dld not vote at Vt+1 or Vt+2but.who voted at Vt+3' 4. 834(Nt+1 - 824Nt+1) = the number of t+1 new voters who did not vote at t+2, but who voted at t+3. 5. 835Nt+2 = the number of new voters at t+2 voting again at t+3. 6. 836Nt+l = the number of new voters at t+1 who have voted at t+2 and t+3. When the terms are collected and the equations simplified, it is evident that the general specification of this system can be characterized as: 'k * * vt+n - 50vt + B1Nt+1 + ... + BnNt+n 2.32 * In this case, the Bi coefficients represent the sum of the double sub- ]4 For example, if n = 2, scripted coefficients in the system above. which is frequently the case, then B; = 821 + 823(1 - 81]). Clearly it is the case that some information of interest is lost by estimating the single equation representation of this system, but an estimate of purging effects after failing to vote in n elections can easily be calculated as (Vt - Bzvt).15 This quantity can then be entered into the original equation to modify the estimate of registered voters at t+n. The final method for incorporating the effects of purging into equation 2.27 is by positing (assuming) a particular functional form of 58 the relationship. This method has the strong advantage of simplicity, but its primary disadvantage is that the assumed form may imperfectly approximate the true empirical relationship. Nevertheless, the empirical work presented in chapter 4 is based on the assumption that the time- purging effects on registration can be approximated by the curvilinear relationship: 1 PG Flv 2 .33 t+1 = B t where PG is the percentage of those registered at t who are purged, t+1 n represents the number of years of nonvoting after which a citizen is purged from the registration rolls, and B is a regression parameter to be estimated empirically. Because there are alternative methods for incorporating this aspect of purging into the aggregate registration model, let PG represent the effects of purging on the size of the registration rolls, without reference to any of the methods described above. Thus, equation 2.27 can now be rewritten in final form as: R = V - PG + w(APB - AC 2.34 t+1 t t+1 vt+1 ' ACrt+1 + ADt+1) + 5(PB - c C D yt+l yvt+l + yrt+l + yt+l) These aggregate specifications of the calculi of registration and voting contained in 2.26 and 2.34 form the basis for some of the empirical work contained in chapter 4. Before concluding this discussion, one final question remains unex- plored: What would be the implications of relaxing the assumption that for those persons registering at t, the estimates of the electoral context costs and benefits are identical at times of registration and voting. As 59 was noted earlier, this assumption implies that for any individual, i, R > 0 implies Vi > O. The implication of this finding to the aggre- it t gate model is that Rt = Vt' This in turn means that these terms can be used interchangeably in the aggregate modeling process. Should this assumption be incorrect, as is likely the case, then Rt # Vt’ and signi- ficantly different results should obtain, depending on the specification. This is easily tested, but what sorts of differences should be expected if the initial assumption i§_incorrect? Given differential perception, perhaps the most likely possibility is that the individual voter associates some opportunity cost with failing to register. This means simply that the citizen who fails to register may find on election day that he would indeed like to vote. Returning to the earlier example where the citizen's probability distri- bution about his point estimate that PB > C-D, it may be the case that he will register if there is even a five percent chance that PB > C-D. The particular confidence level chosen by the individual would be dependent on such factors as his risk-acceptance or avoidance as well as the amount of opportunity cost he is willing to incur. In any event, it will cer- tainly be the case that Rt > Vt“ With reference to the turnout equation, this suggests that the specification for turnout including Vt as a lagged endogenous variable should perform much better than using Rt’ as Rt is no longer a guarantor of voting at t. A final note about the liklihood that perceptions of the same event differ is that the extent of the difference in perception is most likely a function of the length of time between the registration and turnout decisions. Thus one would expect that the less time between registration and turnout, the more similar would be the estimates of the relevant 60 variables. This means that distortion or misperception should be related to the length of time between the last day of registration and the day of the election. Since the recent trend has been in the direction of shortening the closing dates, we should expect a greater difference in the past if Rt is incorrectly used than in subsequent years. This is a testable proposition. 2.9 An Aggregate Specification of the Minimax Regret Model The most difficult aspect of generalizing the minimax regret model toaggregate voting implications involves preserving the decision environ- ment in which the citizen operates. Following Aldrich (1976), I focus on the "P" term in my attempt to distinguish between the calculus of voting and minimax regret models. Further, I include the voter's sense of social responsibility in pptp_the calculus of voting and minimax regret. Thus the only difference in my treatment at the aggregate level of the calculus of voting and minimax regret centers around the hnportance of P. Therefore, static representations of the minimax regret regis- tration and voting equations are: R = B - c - c + D 2.35 2.36 The aggregate dynamic expectations of the minimax regret model are ident- ical to 2.26 and 3.34 except for the missing P term, since the same logic applies to the development of these models. Rt+1 = vt - PG + w(AB Ac - AC 2.37 t+1 ' vt+1 rt+l + ADt+1) + 6(B - C C yt+l yvt+l + yrt+l + Dyt+l) 61 v = ARt + g + (A - AC Dt+1) 2.38 t+1 Bt+1 vt+1 l A + 5(B - C D yt+l yvt+l + yt+l) 2.10 Summary In this chapter a number of interesting results have been derived which attest to the desirability of deductive models. Many more propo- sitions have been derived in this chapter than can be tested in the remainder of this work. Some of these results may constitute a basis for future research. The most significant findings deserve a brief summary here. Not surprisingly, there are additional individual level insights into the calculus of voting and minimax regret models which are possible. A testable theory of roll-off is one proposition deriving from a straight- forward extension of the calculus of voting. Further, I have generalized in Tables 2.1 and 2.2 the standard minimax regret decision problem where there are two candidates to include the citizen's decision to register. Like Ferejohn and Fiorina (1974), I have compared the results of minimax regret favorably to the calculus of voting in that the citizen is more likely to vote in this decisional context. The crucial linkage, I would argue, between registration and turn- out is represented in Figure l and in equation 2.22. This distinction between those who are "pre-registered" by having voted in the past and those who must register in order to vote is extremely useful, for it leads to a wealth of derivative hypotheses. For example, since everyone is not motivated to register by the current electoral context (some being already registered) one would expect the empirical connection between current electoral variables and registration percentages to be more1weakly 62 related than that between current electoral context and turnout per- centages--a decision which j§_made by everyone at a particular election. This distinction also led to the nonobvious finding that under the conditions given in 2.23 higher voting costs may lead to increased voter turnout. It was demonstrated that future-oriented voters who are un- certain of their registration status if they fail to vote should vote under conditions of uncertainty if the conditions of 2.23 are met. If citizens are ignorant of the fact that their names may be purged, how- ever, these results would not apply. Finally, this distinction between those already registered and those who must re-register was maintained in the aggregate development of both the calculus of voting and the minimax regret models. This difference, therefore, provides the basis for two dynamic theories of voter turnout based on two different decision environments. 63 FOOTNOTES FOR CHAPTER 2 1This observation that every outcome can be connected to every other through some ordering relation, R, is discussed in detail in Sen (1970). 2For a discussion in greater depth on the different possible decision environments, see Luce and Raiffa (1957, Ch. 2). 3Technically, Luce and Raififa (1957, p. 29) define a cardinal utility function with respect to lotteries of outcomes: "If a person imposes a transitive preference relation.2 over a set of lotteries and if to each lottery L there is assigned a number u(L) such that the magnitudes of the numbers reflect the preference, i .e. , u(L) > u(L' ) if and only if L > L', then we say there exists a utility function u over the lotteries. If, in addition, the utility function has the property that u[:qL, (1 - q)L':] = qu(L) + (l - q)u(L'), for all probabilities q and lotteries l and L', then we say the utility function is linear." This latter property is known as the expected utility hypothesis. 4Downs (1957) insists that in order to avoid tautology the utility component must be stipulated to contain only politically relevant factors--hense his denial of the rationality of a woman voting as her husband instructs. In this work, utility is defined without regard for substantive concern, but the operationalization of utility in chapters 3 and 4 injects a substantive interpretation to utility calculations. 5Or, alternatively, a] and Oz could represent political parties. 6See footnote 4. 7This point of view, however, puts a great strain on even the basic Downsian model if carried to its logical conclusion. For example, shouldn't the theory explain as well the source of individual differences in candidate differentials? The approach taken here is to view one's sense of civic obligation as a positive pay off for voting independent of electoral outcome. 8Regret matrices are calculated in the following manner: First identify the column entry in the pay off matrix with the highest utility. This represents, then, the payoff with zero "regret". The regret for other column entries is calculated by subtracting these entries from the one with zero regret. When this is completed for all columns, the maximum regret for each row is identified. This represents the worst possible outcome should that alternative be selected. When all row maxima have been identified, the individual chooses that alternative which provides him the minimum maximum regret. For further discussion, see Luce and Raiffa (1957, Ch. 13.). 64 9Burnham (1965, p. 9)suggests that"'Roll-off' measures the tendency of the electorate to vote for 'prestige' offices but not for lower offices on the same ballot and at the same election." This notion is expanded here to mean the act of voting for some offices and not voting for others on the same ballot. An interesting sidelight is Burnham's recognition that roll-off tends to be associated with "a growing public indifference to elections for administrative offices..." which is interpretable in this context as implying that bi = O for many lesser offices. 10See footnote 8. HKmenta (1971, p. 165) defines an estimator, G, to be consistent if plhne = O. In this context, the meaning would imply that fpr each term in equation 2.21, say B, the probability distribution of B would collapse about 8 as the time of the election neared. 12For a discussion of minimax strategies, see Luce and Raiffa (1957, App. 2 and Ch. 4). 13Alternatively, one can assume that the initial node of the tree begins at the first election after registration is introduced in a state or nation. '4Note that a: = 1. 'sihis estimate will likely be high in empirical estimation since this analysis focuses on only three offices. It is certainly possible, for example, that a citizen might not vote for Governor, Senator or President and would nevertheless preserve his registration status by participating in other state or local races which are often held at scattered times. CHAPTER 3 A STATE-LEVEL SIMULATION OF CANDIDATE DIFFERENTIALS. VOTING COSTS AND DUTY 3.1 Introduction In the previous chapter, two essentially individual-level models of the voting process were modified by examining the implications of indi- vidual decision making at a higher level of aggregation. As has been shown, this constitutes an interesting new perspective from which some- what different implications of the models may be drawn. In addition, because of the indeterminacy of the individual-level attempts to evaluate these models (Aldrich, 1976), a different empirical perspective might prove instructive. Nevertheless, such an aggregate-level test of the theories provides the researcher with some serious difficulties. It is the content of these difficulties that constitutes the focus of this chapter. The problem addressed in this chapter is basically one of measure- ment. Since no state scores are readily available to use as measures of 8, CV and 0, some instruments for these concepts must be obtained if the theoretical development presented in the last chapter is to be tested empirically. As Leege and Francis (1974, p. 42) point out, however, itis never the theory itself which is tested, but "through empirical operations we test 'test theory'". The crucial point here is that evaluation of a theoretical statement is highly dependent on the rules of correspondence linking concept to measure. To this end, this chapter will concern itself with the simulation and validation of these unobserved variables. 65 66 The basic research strategy employed is first, to establish individual-level linkages from the CPS American National Election Studies and then to extrapolate on the basis of these findings to aggregate simu- lations of the basic theoretical concepts. This involves four steps: 1. To identify the relevant concepts. This has been done in chapters 1 and 2. 2. To operationalize these concepts at the individual-level on the basis of available data. Sources of data are described in the Appendix. 3. To relate these concepts to indicators which are available at both the individual and aggregate levels. 4. To construct and examine aggregate instrumental variables which are then interpreted as proxies for the original con— cept in the later analysis. To simulate state scores, each individual level of interest variable will be regressed on a series of demographic and socioeconomic variables which are linked to a particular concept. Some of these linkages were drawn in the last chapter. This strategy as well as other possible strategies of simulation are the topic of the following section. 3.2 Approaches to Attitudinal Simulation Simulation has been widely used in political science as both a teaching and research tool.1 Frequently it has been employed as a method of analysis designed to ascertain the implications of a complex mathe- matical structure. For example, Pool, Abelson, and Popkin (1965, p. 2) suggest that: Simulation is a technique that is appropriate when so many variables are simultaneously in Operation that simpler methods of calculation fail. It is a technique used when the equations that describe a system involve stochastic variables, or discon- tinuous variables, or such a complex of variables as to make an analytic solution impractical or impossible. Clearly this is not the principal concern here. Richard E. Dawson (1962, 67 p. 3) defines simulation in a manner which more closely approximates the underlying dynamic of the specific research task at hand: (S)imu1ation...refers to the construction and manipulation of an operating model, that model being a physical or symbolic repre- sentation of all or some aspects of a social or psychological pro- cess. Simulation, for the social scientist, is the building of an operating model of an individual or group process and experimenting on this replication by manipulating its variables and their inter- relationships. This is a more accurate portrait of attitudinal simulation because the researcher interested in state-level simulation must first be concerned with at least a part of the underlying dynamic for the development of the attitude or variable of concern. Thus, for example, this analysis pro- ceeds on the assumption that particular citizen attitudes are the result of the social and economic context within which the individual citizen operates. This linkage can be evaluated at the individual level, and to the extent that the assumptions of this developmental model are met, it can be generalized to estimate average scores for the aggregates of indi- viduals in particular states. One of the first attempts to simulate attitudes in this manner was made by Ronald Weber (1971). Working from the Simulmatics approach of constructing "voter types", Weber concerned himself with attempting to simulate and validate policy preferences in the states on issues ranging from teacher unionization to gun control. The strategy common to him and Pool, Abelson and Popkin was to isolate policy preferences within parti- cular voter types. One voter type, for example, might be a white male, protestant, Southerner living in an urban area. Pool et al employed some 480 "voter types“, and Weber's voter types ranged in number from 23 to 960. Once these patterns were identified, past information about their 68 policy concerns or voting behavior were used to forecast expected future behavior for each type. A more precise statement of the "voter type" method of attitude simulation is that for any measure of attitude, preference Rk for a particular voter type, k, is defined to be: 3.1 where A is the national preference, Si is the mean preference of a sub- group defined by one characteristic, and m is the number of character- istics defining a voter type. State-level preferences, Vj’ for state, are then computed on the basis of the proportion of each voter type in the state. y vj = k§1ijRk. 3.2 where ij refers to the proportion of voter type k in state j. Although this model has received little critical comment as a simu- lation methodology, Seidman (1975) does introduce some questions about the appropriateness of this methodology in particular, and of the possi— bilities of attitude simulation more generally. Seidman's critique of the Weber methodology (p. 336) is particularly compelling: It appears that Weber and his associates have calculated the S.'s incorrectly. For any subgroup, they have simply calculated the mean attitude in a national survey. The difference between this mean and the mean calculated over all respondents is taken to be the predispositional effect of being a member of that sub- group. But this ignores the influence of the subgroup's mean of the distribution of its members among the other subgroups. For example, for the subgroup of Catholics, if Catholics are disproportionately urban, the measure will attribute to Catholi- cism some of the predispositional effects of urbanness, so for urban Catholics there is a kind of double counting. In short, what this criticism implies is that the Weber model is 69 misspecified in the sense that other relevant factors should be "con- trolled for" in attempting to assess the impact of a particular demo- graphic variable (i.e., Catholicism). Seidman also criticizes the specification of the Weber model on other counts. First he notes that all relevant variables may not be included in the simulation. For example, if a simple regional variable (i.e., South-Nonsouth) is employed in the analysis along with a set of other demographic variables, Weber must assume that the only relevant attitudinal variance pitpjp_the South (or Nonsouth) is attributable pplely_to the demographic variable considered. Again, this criticism rests on the possibility that a simple construct from demographic vari- ables alone might ignore other important, but unincluded, variables. Of these two problems Seidman identifies, the latter is clearly the more problematic. The strategy employed in this chapter does not completely avoid this difficulty. But it is hoped that if the theoretical linkages of the developmental models presented here are sufficiently strong, aspects of model misspecification to which Seidman alludes will be of only minor concern.2 But what of the original problem? Seidman suggests that one pos- sible solution of the failure to provide appropriate controls is to adopt a multiple regression framework. This advice is taken by Erikson (1978) in his attempt to replicate the Miller-Stokes representation study. He notes that in that study, constituency opinion was represented by quite small samples within the districts (in some cases less than 10 respond- ents). Erikson adopted the alternative approach of attitude simulation in Congressional districts. To do this he adopted Seidman's suggestion and, in a multiple regression framework, regressed a set of individual 70 level demographic variables on policy preferences using data from the ICPSR. The independent variables he used were also available as aggre- gate indicators from various sources. From the individual-level re- gression results obtained, Erikson computed aggregate district scores based on the relative weights (standardized betas) of each individual- level variable. He used standardized betas rather than unstandardized regression coefficients because of the different metrics involved in the aggregate variables. The simulated variables, then, were the weighted sums of individual-level regression results. Clearly, the possibility remains that a particular attitude, or policy preference, might not be the result only of demographic factors. Other unincluded variables are clearly of relevance. For example, psycho- logists emphasize the proximity of other attitudes as important causal agents in attitude formation (Oskamp, 1978). This is clearly the major emphasis in the research design of the "funnel of causality" (Campbell, Converse, Miller and Stokes, 1960, Ch. 2). Although the variables to be explained are essentially behavioral (i.e., turnout and direction of vote), primary emphasis is accorded to the developmental relationship between political attitudes. Other variables which are unmeasured in the Erikson study mightalso be of concern. One variable given prominent attention in state studies is political culture. Dye (1973) notes, for example, that West Virginia has always proved a deviant case in that turnout levels are consistently higher in that state than any "objective" model of voter turnout would indicate. He suggests that the political culture of the state is such that one's civic obligations are simply taken more seriously. There isani uneasy circularity in much of the political culture literature (i.e., do 71 we explain all deviant cases in terms of political culture in much the same way as "national character" was once invoked as a critical variable in comparative analysis?). Since political culture is rarely operation- alized and included as a relevant variable, it is difficult to interpret. Nevertheless, it could suggest that similar variables have differential impact in different states. It could also suggest some sort of contextual model of voting over time where at t . the point of original franchise- O ment, turnout was high, and the time-dynamic of contextual effects has kept the state at higher levels of turnout, but no differences in re- gression slopes are observed. Despite these difficulties and uncertainties with the regression framework, it is clearly superior to the "voter types" methodology in that it at least provides for controls among those variables which are employed. For this reason, this simulation of state variables will follow the suggestion of Seidman, and the example of Erikson. Because the explanatory variables are constrained to socioeconomic data, however, one must recognize the restrictiveness of the model specification. 3.3 Assumptions of the Simulation Model If we define Y1 as the vector of attitudes of individuals, Xi as the n x k matrix of individual-level attributes which are related to Y1, B as a k x 1 vector of regression parameters and U1 as an 1 x n vector of disturbances, then the assumptions of the individual-level attitude simu- lation are: Y1 = X18 + U1 3.3 Esome extent, racial character. Thus, it appears that proxy variables for citizen duty are available which should evidence sufficient between-state variation to be of value. 83 To summarize, it is expected that Cr and P are grouped optimally. Although the other variables are grouped in less than optimal ways, they should all be of sufficient variation to provide us with some indication of their relative importance. One factor which will be of obvious importance is the variance of each of the non-optimally grouped variables. To the extent that the aggregate variance is decreased relative to the individual level variance, simply because of the grouping criterion, one can expect the effects of these variables in the macrospecification to be underestimated. The next task of this analysis will be to see how successfully information costs, candidate differential, and citizen duty can be simulated at the state level. 3.6 A Simulation of Candidate Differential, Information Cost and Citizen Duty In this section, I will present the results of a state-level simu- lation of several concepts taken from the calculus of voting and minimax regret models which were developed in chapters 1 and 2. These are pre- dicted mean scores for state candidate differentials, information costs and citizen duty (or social responsibility). As indicated above, the technique employed is to regress a measure of the concept at the indi- vidual level on a series of independent demographic and socioeconomic variables also available in survey data. The standardized regression coefficients thus obtained are employed as weights which are applied to analogous standardized state-level variables in order to predict mean state scores. These predicted values should display the properties of standard normal variates--having a mean of zero and a standard deviation of one. This process, when applied to survey data from 1952-1976, 84 should yield a time-series of state-level estimates for each of the simulated variables. Besides the serious possibility of incomplete model specification, two problems arise in the context of such simulation. The first is the selection of the appropriate operational indicator of the concept in the survey data. In the cases of candidate differential and sense of social responsibility, several indicators have been employed in earlier empiri- cal work. On the other hand, information costs have never been measured in tests of the calculus of voting or minimax regret models. Where several indicators are available, the strategy employed was to regress each indicator on the same set of independent variables. If these variables can be linked at a conceptual level to the dependent variables (i.e., candidate differential), then this procedure can be seen as an 8 In other words, it is expected that exercise in construct validation. the relationship between independent variables and dependent variables representing the same general concept should be approximately the same. Where information costs are concerned, I have attempted to create a crude indicator of the concept following the rationale that some attempt, how- ever primitive, to evaluate the merits of the theoretical argument is better than none. Specifics of each operationalization and simulation are discussed below. The second difficulty pertains to the process of standardizing the state-level independent variables. If variables are standardized cross- sectionally (i.e., for a particular year, a mean and standard deviation among the fifty states are identified) then over-time trends in these independent variables are likely to be obscured, since for each year the mean of the variables should be approximately equal to zero. This 85 detrending could have serious implications in testing equations 2.26, 2.34, 2.37 and 2.38, since these equations involve differences over time. This is a severe problem for which no optimal solution is clearly ap- parent. The approach adopted here to preserve the trends over time is to standardize the variables longitudinally. Thus instead of standardizing some variable, say income, for the year 1952, I have grouped all state income scores for the entire time series. From this grouped data I have located a mean and standard deviation, and variables have been standardized according to these parameters. Where trends have existed in the original data, therefore, they have not been removed. The difficulty with this procedure is that the weights which are applied to these vari- ables are determined in a strictly cross-sectional analysis. To summarize briefly, the procedure of the simulation involves four steps: 1. To regress at the individual level a measure of each concept on a set of individual demographic and socioeconomic traits (race, region, education, etc.). Data employed in this step include the CPS/SRC National Election Study data for 1952, 1956, 1960, 1964, 1966, 1968, 1970, 1972, 1974 and 1976. 2. From these results, standardized regression coefficients are obtained which will be used as "weights" in the simulation. 3. Aggregate state demographic and socioeconomic data corres- ponding to the individual regressands are then standardized longitudinally. State scores on each variable are computed by re4arranging the data set so that cases are "state years". For example, the initial case is "Connecticut, 1952" (following standard ICPSR state codes). Sources for these data are con- tained in the Appendix. 4. The standardized regression coefficients from Step 2 are now employed as weights on the standardized scores obtained in Step 3. For example, in state j for 1952, the vector of standardized regression coefficients for 1952, B . is multi- plied by the set of 1952 aggregate standardized géate scores for 3’ 252‘ 86 X = 2 B 3.16 j52 j52 52 Clearly if Zégg is a row vector of magnitude k, and if B is a column vec also of magnitude k, then X. , the "simalated score" for state j in 1952, will be a scalar?2 This operation is repeated for all j (all 50 states) for the years of the study (listed in Step 1). Let us ruwv consider the simulation of candidate differentials. Six different measures of the concept were considered--strength of party iden- tification, concern over the electoral outcome, interest in the election, candidate and party likes and dislikes, differential candidate feeling thermometer scores, and differentials computed over the seven-point issue scale questions. None of these measures was available for the entire period of this study with the exception of strength of party identifi- cation. The existence of a single continuous measure over time would seem to argue in favor of its superiOrity assuming it provided a good representation of the concept. In his well-known work, however, RePass (1971) showed that a respondent's political partisanship did not neces- sarily reflect his issue partisanship. That is, while political partisan- ship is certainly a component of B, it is not a complete representation. With the growing emphasis in the voting literature on issue-voting and the declining importance of partisanship,9 one might be hesitant to simulate partisanship as a measure of B. A similar logic could pertain to issue scales. It is certainly a tenuous proposition to assume that (1) the issue-oriented questions reflect the concerns of the majority of voters and (2) these issue con- cerns, even if important, constitute the sole components of differential candidate evaluation. Clearly partisanship, issues, and specific candi- date traits such as leadership all constitute aspects of the citizen's evaluative process.10 87 For this reason, one might look for a measure not dependent on only one specific aspect of candidate evaluation. This leaves us with interest, concern, and feeling thermometer differences. Of these three, interest is available for all years except 1974. Where "interest" and "concern" are both available, the overall results are quite similar, except that goodness of fit was generally slightly better for interest.H In addition to this, the pattern of relationships between interest and the independent variables is quite similar over time (see Table 3.1).12 The column of Rz's provides evidence that the simulation model of candidate differentials is underspecified, but some comfort may be de- rived from the stability of the more important coefficient estimates over time. Clearly older, more educated and wealthier respondents have higher differentials. The effects of other variables are much less and in some cases (i.e., religion) appear to be idiosyncratic. The overall fit of the simulation model compares favorably with the results presented by Erikson (1978). The method by which I have measured information costs is quite in- direct. I assume that, ceteris paribus, the greater the respondent's level of political information, the lower his costs of gathering and pro- cessing information. This would suggest, for example, that someone who is politically aware, either through formal or informal education, would have to incur fewer costs in arriving at a voting decision than a citizen who knows little or nothing of the world of politics. This necessitates a view of information costs which treats those who enjoy keeping up with current events as expending no utility in this enterprise. Instead, since this j§_enjoyable (or perhaps it is a manifestation of a sense of social obligation) it is not a cost to these people to stay informed. 88 .chcmga< ozu =_ umnwgommv ago mwugzom memo mmeo; czo mega czo on; mucmucoqmmg among Low «so vmvoo m? opamwgm> m:_m:o: och .mpwszco: so; oLmN .mawgz cow mco .mcoppmmaa mpwum mamm_ ucwoaum mg» op mmmcoamms mp; :o woman Aonmpv so» one umuou mw cowm_pmm umcoum mpamwgm> msosouozuwc m m? mung mmumnwucmu so Aesopv mmwugma mg» on acmccoammg use we mmmcmmopu wgu no woman m_ mczmmms as» mum, can «map .cowyowpm acmcgzu ms» cw umummgmucw mw ucmccoammc on» we: so cwsumnz mw umm: mpnawgm> ucmucwamc ms“ .mmpnmwgm> acmucoamucw uwum__ mzu co meucmgmmwwc wumu_ucmu mo mgammma a mcvmmmgomc gmumm umcwmuno mpcmpuwmmmoo :opmmmgmog umNchwucmum one ma;m_m3 m>pampmmm .mm_3gm;uo ocm~ .guaom ww mco umuou m? :owmmm we mgsmmms on» .uowgmn mnummmp on» go; .wso; an mcw>wp mucmccoammg Low :0 mgmucmg com oLmN new .mm_3gm;uo ogmu .owposumo mo. cpp. poo.u nmo. opo.n ooo.u ooo. ooo.n omp.u ouop mo. Pmo. mmo. moo. oNo.u oNo. omp. omp. mm—.u emop mo. coo. omo. mNo.u mNo.n «NP. mmw. —¢P. eeo.u mmop mo. mNo.u ceo. omo. PNo.u pop. mpm. —m~. emo.n ouop oo. Foo. omo.n ovo. eoo. omo. mum. “so. m—o.u moop Np. epo.u mpo. ..... omo.u omp. oom. omp. oeo.u coop op. opo. mpo. oNo.u Nm~.u moo. mum. ooo. omo.u coop pp. moo.u oN_.u ..... mpo.u nP—. NmN. oep. Foo. oomp oo. coo. coo.a ..... ¢mo. omp. mop. mop. Npo.u omop my. ooo. ovo.u moo.u Noo.n mmp. mom. mop. moo. «mop mm onP~k<4mm F.m u4o ommgp .P.m mpnap a“ ago: as“ a? umnwcumou .mmpnmwgm> ucmucmamucw umpm_P one co :owumsgowcw qu_uwpoa oo Pm>mp m.u:mucoamoc ms» mo mgammms m mcwmmmcmmg cmpwm vmcvmpno mu:m_u_$$mou cowmmmcmmc om~wugmucmum «so musmwmzm mp. who. pmo. ~mo.- oNo.u sup. mum. ooo.u oop.u womop NF. who. muo. moo.u moo. mop. emm. opp. Poo.u mwmmp mp. mmo.u lllll ooo.u Neo.u mop. mom. ooo.- Neo.u mummp op. “mo. ..... mmo. moo.u on". opp. omo.u oo_.n wouop up. mus. meo.n mmo. omo. moo. Pom. omo. omo. swoop om. ooo.: ooo.u lulu: ooo. cup. use. on—. mmo. cooop «F. mpp. omo.u upo. who. oNo.u Nmm. Noo.u Npo. neoop mo. meow. Nmo.u lllll poo. mop. omo. mop. omo. uooop ep. omo. moo. ..... opp. cop. ocm. oep. omo. uommp em. omp. Foo.n omo. ooo. mop. com. oNo. pup. swoop mm onhm4 onhHH<4mm 91 .mewpgee Le meueewecee seen we ppm; we w_emewc eueeeF epeee pceeeeemec one news: so meweem memmw newee-w we sense: we» mw epeewce> peeeeeeee echo .eewuesgewew Peewuwpee we Pm>ew m.u=ee:eemeg esp we mueswume e>wpeeneem m.geze_>eeuew egg mw epeewce> «emceeeee eswe .mewpcee me» «nope mexwpmwe Le mexwp mw; mewegemeg useeceemeg mg» we eemmecexe meewewee we sense: may mw epeewce> pceeeeeee egwe .meueewecee Pawneeewmece one peeee mmxwpmwe Le mexw— mw; newesemeg wceeeeemeg one an eemmegexe mcewcwee we geese: esp mw epeewge> acmeceeee expo 92 ...among the factors that contribute to citizen duty are the following satisfactions: 1. The satisfaction of complying with the ethic of voting which, if the citizen is at all socialized into the democratic tra- dition, is significant. 2. The satisfaction from affirming allegiance to the political system: for many people this is the.main rationale for voting. 3. The satisfaction from affirming a partisan preference. 4. The satisfaction of deciding, going to the polls, and so on. These items are usually regarded as costs, but for those who enjoy the act of informing themselves for the decision, these supposed costs are actually benefits. 5. The satisfaction of affirming one's efficacy in the political system: the theory of democracy asserts that individuals and the act of voting are meaningful, and for most people the only chance to show that they and their actions are in fact mean- ingful is in the voting booth. Most frequently employed at the individual level, the citizen duty score has a major disadvantage in the fact that the marginals of the four com- ponent questions are highly skewed.14 An empirical investigation showed, further, that trust scores could not be adequately simulated with the variables available. Further, those results which were obtained suggested that trust results were not consistent with either of the other two oper- ationalizations or even with itself over time. For example, while both duty and efficacy scores showed education to be consistently the most important variable, the trust results fluctuated from significance to insignificance, displaying a randomness even in the sign of the education coefficient. Thus trust was not seriously considered as a measure for citizen duty. The remaining two variables-~duty and efficacy--both showed con- sistency with each other and over time. Since the efficacy results evi- denced a marginally better goodness of fit as well as a pattern of conti- nuity, this was selected as the variable to be simulated. 93 .~.m mpneh e» epez one :w eeewgemee me wee meweewge> emegw .mepeewce> “ceeceeeeew eeumww en» ce xeeewwwe peewuwpee we exemeee e mcwmmecmeg Loewe eecweuee muemwewwweee :ewmmecmeg ee~wegeeeeum ego mucmwez e>wuepeze 0.. wep.- New. Pmo. owe. mew. _mo. mmm.- ewe? AN. mmu. «NP. eeo.- ewe. mwo.- 4mm. mew.- ekmp m_. wee. _m_. moo.- mmo. Nee. mew. eoe.- -m_ NN. owe. mmm. ~N¢.- ems. m¢_. cop. “mo.- ewm_ em. eeo.- mm_. Noe. mmo. New. oeN. moo.- wea_ 4N. omo.- «op. ..... ¥ «me. New. _mm. emo. com. FN. weo.- . mwo. “mo.- _Ne. 55o. Fem. o_o. eemp ~_. mmo.- mop.- ----- Nee. Pop. Nam. emo. coop AN. Ame. ewe. ..... owe. ~_F. «mm. _mo. emmp _N. m~_.- emo.- _~e.- ewe. emu. com. ~m_. «new we wummnm zowwwwwoe ezemao: zoeemm mzoezo zoewomw<4mm m.m m4m L Ne w w w w w wwa wwa ewe ezeewzew r Aw2weo> we wwewwwewo we-Nwa .weewwwx zowweeewwwwe wezewewww-wwoee P. e 59: 123 .weewueewe Feweeeewweee Lew wcewww>eee eewueeuwwmme weweeew wwwxe eeeeu ww — eeeee ww eweewee> wwew .uweu xeeeeuwew .waee cwv euee mcwwepoe .ewwzeegue eeeN e .Eepwxw :ewpeeuwweee we weep we» we eeewwma e we eweewee> wweu .Awomwv we we xe__ex onesew—eee .Aueeseeweeee xeceewwee Seawews Eeswxes esuo eewe> ewegmw; esp wueepwee eweewee> wwcp .euepw we» :_ maceEeeweeee zeeeewwee weewee> ecu wo e .w_e>mp poweuwwe new eueeee .eaepw we» we wpeeseeweeee xeeeewwee we saw me» ww eweewee> mew .eewpeeuwwmee :e wueesegweeee xeeeewmee we wpeewwe we» we weeweee e wee .wewpwwueuwuu age weweeueeeee cw wewuwuceeo .1 em.e ....... Awe.e-o Ape._ o AFN. o Awe. v Lee. o Awe. o ww._ we. ..... -- eee.w - wee.e ww.ew we.e ew.ew N_.ee e_.ewe weep ww.w ....... Lew.e-o Awe. o Lee.e-o Lew.e o Aew.e-v ...... ew.w we. ---”--- eew.~ - wee. we.ee~- .e.ww we.ew_- ------ ~w.e~w eee_ L L L e Ne w e N e e e e >e wee ewe wzeewzee AV e.e.eeeeo _.e wowee 124 ------- ....... Ae_. o Awe. -v Awe.e o Awe. v we. .............. wwe. w_.we_- we.ew_ ww.ww we._ee wee. Lew.e o ....... Lew. -o Awe. o Aww.w e A_e._ o ew. wee. ------- wwe. - ww.w ee.ww_ we.ww ww.www eeee Aww. o Awe._-o ------- Lee. -o Aw..e-o Aww._ e we. ewe. eew.ee- ....... ee.e - ww.wew- we.eee ww.wwe www_ Lew.e-o Aee. o Lew.e-e Awe.w o Awe.w-o eww.w e ww. wew.w- www.w_ wee. - ee.ewe ww.wew- ew.ewee ww.wew wwe_ Awe. -o Aww. -o Lew. -o Awe. .o Aww. -o Aw_.w o ww. wwe.e- ew_.ew- eee. - ew.we ew.we - ew.eww ee.wew ewe_ eww. -o eww.w o Aww. -e e_e._ o Aww. -o Aww. e ww. wew. - ewe.e wwe. - we.ww_ ee.we - w..ee ww.wwe ewe, Aww..-o ewe. e Ape._ v Lee. o Aew.w-o whww.we we. Twww. - wee._ wee. ew.ww ee.eee- ww.www ew._ww wwwe we wee wee Pee e >e w wzewwzee “wweeww xeze2wzo N.¢ m4meee coweeeuwwmee wwweeew uwwxe eeege we w eeeee we ewewwew> mwswe .Eeewww cewueeuwwmme we weep we» we eeewwee e ww ewewwee> weep .Awomwv we we wewpex mcwzewweee 125 . .Aueeseeweeee weeeewwee seawews Eeswxwe eeev eewe> “macaw; we» muemwwme ewewwee> weep .eeeuw we» cw wpceseeweeee weeeewwee weewee> esp woe .w_e>ew peweuwwe eee wuceee .eueuw esp aw wuceeeeweeee weeeewwee we saw esp ww eweewew> mew .cewewepwwmee ce weeeseeweeee weeeewwee we weeewwe one we eeeweee e wee .wewuwweeewlu mew weweeuceeee cw wewewucweo k. ew.w ...... - eww.e-e Lew.e e Lew. e Awe. -e Lew. e ww._ ew. ------- mew.w - eww.e we.ee ww._w - ew.e_w ew.ewe wee, ew.w ------- eew.w-e Awe. -e eww.w-e eww.e e Lew. .e ew.w we. ------- mwe.w - ewe. - ew.w_w- .w.ww. ee.ww - we.wwe eeee e we wee wee Pee e >e w wzeewzee A.e.eeeeo w.e eewew 126 .wewuwweepwuu wee wewegpeeewe cw wowewucweo .1 we.w .ee.. e .ew. -e .ww.we .ww..-e ..w. e ...e ww. ee.ew.- ee.ee - ww.ew ew..w- we.we. ew.www wee. ew.e .ew.w-o .ee. e .ww. e ..w. e ..... - ww... ww. ww.eew- ww.we we... ee.ew ------ ew.ewe ewe. we.w .ee.wee .ee.w e ..e.we .ew. e .ew.we ww.w ww. ww.ww.- ee.ew .e.ew ww.e ...ew. ww.wee wee. .e.e .ww.w-e .we.e e .we. e .w..-. ------ w..w. ew. w..we.- ww.wew we.w.- ew.w.- ------ ew.w.e ewe. ee.w .we..-e .ew. e .we. e .ew.-. .ew.we ew.w ew. e..ww.- ew.ew we.w. we.ww- ww.www ww.www wwe. ee.e .ew.w-e .ww.w e .ee... .ww. e ...... ww.e. ww. ee.wew- w..www ew.ew ww..w ------ .e.wwe wwe. ee.w .we..-e .ww.w e .ee.we .we. e ..w.we ww.ww we. .ee..w.- ee.eew ew.ew ew... ww.www ww.eew ewe. ee.w .e.. -e .ww.w-e .ww.we_ .ww... .ew.we we.w. ww. ew.w. - ww.e.w- ww.we ww.ww w..wee ee.wew ewe. we.w .we.w-e .ee. e .we.we .we..e .ww.ee .w.e. ee. ww.www- we.e e..ew. ew.ww ew.w.e. ww.wwe wwe. we.w .ew..-e .ww.w-e .ww.we .ew. e e.we.we we.w. .w. w..ee.- ww.www- ew.we. ew.e. w..www ew.wew www. e we e >e ewe wwe ewe wzewwzee Aezewee ee weeeeeeee omsmmow .mheomum woozmoh 4 ew.www we.wwww ee.ww..- we.eww .w.ee.w.- ww.wee - ww.wew ww.eww ew .w.. -e .we. e .ew. -e .we.. e .w... -e .ww. -e .ew. e .ww. e ww..ee - ew.ww ee.wew - ww.www ww.ew.w - w..eww - ew.ww. ew.www ewew ew.w-. .w.. -e .3. ..e .ww. e ..w. -e :w. -e 3e... .w.. -e ew.wwww- we..ew - ew.ewe.- ww.wew e...we. - ww.wee.- ww.eew we.ww. - wwew wwe. ewe. wwe. ewe. wwe. wwe. ewe. ewe. ..e.eeeee w.e wew<. 142 .ew. -e .ew. e ..w. e .ew. e .ww.. -e .... -e .w.. e .ww. -e .e wee.e. - www.w eww. e.w. eee.ewew - ew.. - e... ew... - . ee .ww. -e ..w. e .w.. -e .we.. e ..e. -e .ww. e .we.. e .ew. -e e ww.wwew- we.www. ew.www - ww.wew. ew.wwww. - we.weww ee.eew. ew.wew - ew .we. e .we. -e .... e .ww. -e .ww. e .ew.. -e .ww. -e ..w. e we.www ee.www.- ee.e.w ee.wwe.- w..ewew. we.wewe- ew.w.w.- ww.eww. we .ww. e .ww.. -e .we. -e .ew.. -e .we. -e .w... e .we.w -e .ww.. -e ww.www ww.eww - ww.ww - ww.www - ew.www - we.wew ww.wee - ww..ww - we .ee. e ........ . ....... -------- ---------- -------- ................ . ...... -- -------- -------- ---------- -------- -------- -- ...... we ewe ee .em. e ........ wwe. -e .wm. e .ww. e .ew. -e .ee. -e ..w. -e .e e ewe w ....... ew. - ww. eew ww. ew. - ewe - ee. - e .w.. e .ww. -e .we. e .ww. -e .ew. e .ww.. e .we.. -e .w... e > ew.wew w..ww. - ww.ew we.ww. - ww.wew .w.eew ww.www - w..www ee .ee. -e .we.. e .... -e ..ww.. e .ew. e .we.w -e .ww. e .e.. e .w ... - we.wew ww..w - ww.www ew.ww.. we.wwe - we..e we.e we .ww.w -e .we. e .... -e .ww. -e .ew. e .ew. e .ww. -e ..ee. -e ew.. - .e. we. - ee. - w.. we. we., - ww.ee - we wwe. ewe. wwe. ewe. wwe. wwe. ewe. ewe. .weeeee xez.z.ze ww-ewe. .w.eewwe ee.eeeww.eee e.z ee wweweeeee ww-wwe. .w.ewwee wwezee. eezezew 145 .wewewwuwew1u mew wmwezeceeww cw weweweceeo k. ew.w. ww... ww.w. e... e..w. w.... w.w. w..w. ww.. we.w ew... we.w ew.w w.... we.w. ww.e. e we. ew. ew. ew. ww. ew. we. we. we w..eee ee.wew ww.wwe we.wwe w..wwe ew.wwe ww.wew ww.eww eeeeweee .ww. -e .ee.w -. .we. e .w.. -e .we.. -e .ww.. e .ww.. -e .ww. -e we.eww. - .e.wwww- we.ee ew..ew - ew..ww.- ww.ewew ew.wwew- e..eww.- ww .ew. -e .w.. -e .ew.. e .we.w e .ew. -e .e..w -e .ww. e .ee.w -e e .e.w.w. - ww.wwe - ww.wew ww.ewee ww.wwe.- ee.weew- ww..ww ww ..ee- ew .e.. e .we. -e ..w. -e .ww. e .e.. e .ew.. e .ew. e .ee. -e ew.wew ew.we - ww.ew - ww..we we..w ww.eww .w.eew - .w.ww. - ewew wwe. ewe. wwe. ewe. wwe. wwe. ewe. ewe. ..e.eweee w.e wewew 146 .wewuwwueaw1e mew wewespweeee cw wewuwuceeo .4. ew.w ww.w ew.w w.w w..w w..w ...w e..w ew.. w..w ww... ew.. w..w. .e.w. ww.w w..w e ww. ew. ww. ww. ww. ew. ww. ww. we ww.wwe .w.www w...we ew.www ww.eww ww.wwe ...eww we.wew eeeeweee .ew. -e .w... -e .ww. e ..e. e .ww. -e .ww.w e .we. -. .ww. e ww.wew - ww.wee.- ww..ww ww.wwe ww.wwe - ww.wwww .e.wew - ew.wwe ww .ww. -e .w... e e..w. e .we.. e .we.. -e .ee.w -. .we. e .we.w -e > ww.ew.w- ....www ww..ww ww.wwwe we.wwww- we.wwew- we.eww we.wwew- ew .e... e ..e.. e .ee. e wee. 1. .w..w e .ew.w e .ww.. e .w.. -. .w..wew ww.eww. ww.ww ww.ww.w- ww.wwew w..weww ww.we.. ww.eww - ww .ww.. -e .ww. e .ew.w e .ww. -e .we. -. .ew.e -. .ww. e .w... -e we.www - ww ww. ww.eww we..ww - ew.ww. - ww.wwe - ew.ee ww..ww - we .we.. e .we.. -e wwe..-. .ww. -e ..we. e .we. e .ee.w e .ww. -e e we.w.e we eww - ee.ww.- ew.www - ww www ww..e .w.www ew..w. - we wwe.. -. .ww. e .ee. -. .ww. e .we. -e .ww. -e .we. -e .ew. -e we..ww - w. .w .e.ww - we.w.w ew... V- .w.ww - we.ee - ww.ww. - we .w... -e .ww. e .ww.. e .ww. e ..w. e .ew. -e .we.. -. .ew. -e we ew.w - .e. ew.. ew.w ee. ew. - we.w - w.. - .ww. -e ..w.. -. .ew.w-e .ew. -e .ww.w -e .w... -. .ww.w -. ..we.. -e . ww.. - ee.. - ww.w - ww. - ew.. - ww. - we.w - w... - w wwe. ewe. wwe. ewe. wwe. wwe. ewe. ewe. o.¢ mem EU(az). Conversely, if EU(01) < EU(aZ), the citizen votes for “2’ and if EU(G1) = EU(aZ), the citizen abstains from voting because he is indifferent between the candidates.1 A generalization to n candi- dates is provided by McKelvey and Ordeshook (1972). Now it is possible to see voter participation in terms of a sequential decision process involving thr§g_decisions--registration, turnout, and direction of the vote. Using the cross-sectional calculus of voting model in a two-candidate race as an example, we can write 154 Rit = PitBit ' Civt ' Cirt + Dit 5-1 Vit = PitBit ' Civt 1 Bit 5'2 CDit = EUi(O1) - EUi(a2) 5.3 where lCDitl = Bit“ Curiously, in the system of equations above, CDit = 0 does not necessarily imply either Vit 5_O or Rit 5_O. Thus, "abstention due to indifference" is not a deducible consequence of the system of equations presented here. If Rt and Vt both exceeded zero when CDit equalled zero, the individual would flip a coin to decide between the candidates--but he would vote. In short, CDi = O is t neither a necessary nor a sufficient condition for non-voting. This has implications for empirical research. Although most scholars today accept that voting has some consumption aspect associated with it, a clear test would be to look at turnout patterns for those who have zero candidate differentials. If this segment of the population ggg§_vote, the validity of equation 5.2 is supported. On the other hand, if these citizens do not vote, then a more desirable theoretical speci- fication would be one where abstention due to indifference shgglg_be a deducible consequence of the system. This could be accomplished by either eliminating the Bit term altogether, or by altering the specifi- cation Of 5.1 and/or 5.2 as indicated in 5.1a and 5.2a below: Rit PitBit ‘ Civt ' Cirt + BitDit 5"a V P it ’ itBit ‘ Civt + BitDit 5°2a The absence of systematic empirical investigation Of the relationship between Bt and 0t has resulted in an ignorance of the proper 155 specification, although the overall empirical significance of the Dt terms seems well established. For this reason the model analyzed was the standard calculus of voting model. The important point here is that the citizen's direction of voting--CDit--is easily incorporated into the decision process outlined in chapter 2. An important assumption of this analysis has been that citizens evaluate candidates on the basis of their positions on the issues. This is discounted by the citizen's estimate that his vote will make a difference. Tullock (1967) introduces another discount term--the citi- zen's estimate of the accuracy Of his judgement about Bit‘ This new term, referred to as "A" by Tullock, could be quite important in explaining the declining weight which was associated with the PBP term observed in chap- ter 4 in the calculi of voting and registration. Popkin et a1 (1976, p. 794) stress a similar theme: Competence is a relevant dimension of candidate evaluation for these reasons: (1) The candidate's competence directly affects the probability of his being able to deliver output from the system once he is elected. (2) Much of what both the President and Congress do involves the general management of the country. Since the voter has only limited information he may vote for a candidate who seems capable of managing the country even if the candidate is not the "closest" to his specific issue preferences. (3) Finally, numerous problems will emerge during a candidate's term Of office that he will have to solve but that neither he nor the voters can anticipate on election day. Competence in unfamiliar areas may be inferred from the perceived competence of the candi- date in other areas. In this sense, the candidates' assurances that they have "leadership" and "experience" and that they can "handle the tough issues" are attempts to assure the electorate that they can deliver on the issues which are important to individual citizens. Similarly, challengers' attempts to 156 demonstrate how the incumbent has not kept his promises--how he has "not kept faith with the American people“--are strategic maneuvers designed to subvert presidential support. In this context, it is interesting to note that forceful presi- dential actions--particularly in the international arena--seem inevitably to be accompanied by gains in popular support for the president (Mueller, 1970). Although this undoubtedly can be explained in many different ways, an interpretation consistent with the argument presented here is that such decisive action increases the incumbent's perceived competency and thereby 2 If one could document trends in the his chances of winning re-election. citizen's feelings that candidates "deliver“ on their promises, and if these trends paralleled the decline of regression coefficients in Tables 4.1 and 4.3, then the inclusion Of a competency term would be given additional support. Although it does not correspond perfectly to this concept, one item in the “trust in government" scale dgg§_provide an approximation. This item asks whether the respondent "feel[s:lthat almost all Of the people running the government are smart people who usually know what they ar§_doing, or [goes he] think that quite a few of them don't seem t9_know what they are doing?" The cynicism Of the electorate as reported in Miller (1974) and in subsequent national election surveys is clearly on the rise. In 1964, 27.4% of the sample expressed the view that the people running government didn't seem to know what they were doing. By 1974, this percentage had doubled to 45.3%.3 This pattern differs to a certain extent from the reported regression coefficients in Table 4.3, but it seems clear that such a linkage might provide a valuable key for future research in this area. 157 Besides being outside the mainstream of the voting literature, this research is atypical of other research in this area since it relies strictly on the economic approach to voting. If we can simplify the approaches to voting into three categories-~sociological, social-psycho- logical, and economic--then one might question how the results Obtained here pertain to those of the other approaches to voter turnout and registration. One of the desirable features of any scientific research is that it be cumulative--by this I mean that it relates to and builds on other results in the field. A lack of cumulative growth has been a persistent problem in the study of voting behavior. The research of the economic and the social-psychologicalapproaches with respect to voter turnout have one common feature which may prove important in the future synthesis of these approaches under a larger covering law--the problem Of underprediction. Much of the evolution of thought in the economic approach to voter turnout has focused on this problem. Minimax regret models of voter turnout permit higher levels of voter turnout than the calculus of voting because one's impact on the electoral outcome is assumed to be "unknowable". Similarly the emphasis on the "D" term has been on its role in allowing higher levels Of voter participation. In chapter 2, I showed that in a future-oriented electorate where condition 2.23 Obtained, the calculus of voting model would permit higher levels of voting than had heretofore been recognized. This increase in turnout is higher still if voters are ignorant Of the periodicity Of state purging provisions. Thus much progress has been made in the ability Of the economic voter turnout models to deal with the problem of underprediction. 158 Much less progress has been made in the social-psychological approach. Although Campbell, Converse, Miller and Stokes (1960, p. 107) found a strong monotonically increasing relationship between voter turnout and intensity Of political involvement--an index representing the combined effects of such attitudinal variables as interest, inten- sity of partisan preference, sense Of political efficacy, sense of citizen duty, and concern over the electoral outcome--it was nonetheless noted that among those with the lowest index score, twenty-two percent voted anyway. When asked their reason for voting, the dominant response documented the effects of friends and neighbors. This raises a question about the incorporation Of the individual's social and political context into the models Of voter turnout. A limitation of the models employed here is that the primitive terms comprising the models are taken as "givens". That is, there is no attempt here to suggest why it is that some individuals are less interested in the election than others. The only exception to this was in the simulation Of attitudes where it was necessarily assumed that the attitudes of citizens are the result of their socio-economic background. The line of argument developed in Campbell et a1 (1960) suggests that the specification of the origins of preference including an expli- citly contextual variable would be a step in the right direction. This would mean that the attitudinal implications of education, income, and similar variables should be more carefully assessed by looking not only at the individual but also at the referent groups--especially those with which the individual exchanges information--with which the indi- vidual identifies. This would suggest, for example, that a high income 159 individual living in a poor neighborhood could be expected to behave differently from his counterpart living among the well-Off. An interesting study would be to examine individuals who operate in an environment which is atypical. One would expect, if the contextual hypothesis is correct, higher voter turnout and registration levels among blacks living in predominantly upper class white neighborhoods than among other blacks. Or, as Niemi (1976) suggests, if thfire exists a norm of voting, regardless Of the demographic composition Of the district, the individual coming into this environment could be "infected" by this dominant orientation (Boulding, 1962). The formal models as they now are stated, depend wholly on the assumption Of methodological indi- vidualism. More realistic approaches may move toward relaxing this assumption. 5-3 59111952 The goal of this research has been to explore the nature of the linkage between registration and turnout. In so doing, it was assumed that the registration and voting acts constituted separate, yet inter- dependent decisions. The theoretical development in the second chapter utilizes this assumption and derives two distinct models of this decision making process--based on the extant literature dealing with the calculus of voting and minimax regret. From this theoretical development, a number Of interesting speculations and hypotheses were advanced. The most original of these was the discovery that some citizens may vote even if Vit-510 if the conditions stated in 2.23 are met. Should the theories prove to be correct, this finding would constitute a significant gain in 160 our understanding of voting--a gain realized only through the linkage of the registration and turnout decisions. Unfortunately the empirical results do much to undermine the optimism of the theoretical chapter. First, in chapter 3 it is clear that severe measurement problems are present in the simulated variables, and the dynamic variables based on simulated scores are invalid. More- over, the process of estimating over groupings (states) rather than individuals carries with it its own set of difficulties. Even so, the results of the fourth chapter cast doubt on the entire theoretical frame- work set out in chapter 2. While anomolous coefficient estimates could be "explained away" by poor measurement and high collinearity in the state models, the dynamic model results undeniably contradict the pre- dictions of the model. Using the laws of deductive logic, it is clear that since the dynamic model is deduced from the same primitives as the state model, the failure of this aspect Of the model implies that the entire theoretical framework is deficient. As such, some changes are required in the basic primitives of the model. Some possible changes have been the subject of section 5.2 of this chapter. Although it is unfortunate that the major goal of this study remains unattained--the results do not support the theoreticaldevelopment of chapter 2--some lesser conclusions deserve some attention at this point. First, for those scholars who continue to work within the frame- work of the calculus of voting or the minimax regret models, this research would indicate a clear preference for minimax regret in the cross-sectional analysis of voter turnout and registration. It might sound inconsistent to say in one paragraph that neither model is satis- factory and in the next to recommend one model over another. In an ideal 161 world this might be the case, and my results clearly demand a fresh approach to the problem. But in the meantime one must work with the tools available. Of the available economic models, the results of chapter 4 would favor minimax regret. A second conclusion pertains to the question of the effects of registration provisions on voter turnout. My results indicate that turnout increases on the order of five percentage points might be expected were same day voter registration laws enacted in all the states. This differs from the results of other studies in that the anticipated increase in voter turnout due to same day voter registration is less by almost half. Nevertheless, policy makers interested in increasing voter turnout might expect an increase on the order of 5% were such provisions enacted. The only remaining course which would increase turnout would be to return to a system of permanent voter registration, but this has received no serious attention by policy makers. Finally, it should be noted that a negative result can constitute a contribution to the discipline-~although it is doubtless less a con- tribution than was anticipated at the outset. While most scholars would agree that political science remains in a pre-paradigmatic stage, it is nonetheless important to note instances where accepted models fail to perform adequately. Such an instance certainly occurs when the models considered here are cast as dynamic processes over time. In this sense, therefore, this research establishes a benchmark beyond which more advanced models of registration and turnout processes may be assessed. 162 FOOTNOTES FOR CHAPTER 5 1Page (1976) makes the point, however, that if candidate positions are represented by distributions of points, the expected utility model may not adequately represent the decision making process of some indi- viduals. For example, were an individual given a choice between one candidate whose position on an issue could be represented by a proba- bility distribution with mean, u. and an alternative candidate adopting a position at the point u, this individual might not be indifferent between the lottery (the first candidate) and the certainty option (candidate two). Thus some notion of risk-acceptance or -aversion should be considered. 2This discussion, however, assumes that the firmer position will be more accepted by the voters. In some theoretical evidence to the contrary, Shepsle (1972) develops a theory where ambiguous strategies can, under some circumstances, defeat a candidate whose position is represented by a single point. See also, Page (1976). 3There may be problems of validity with this item, however, as respondents may feel that "smart" means either "wise" or "unscrupu- lously crafty". APPENDIX The purpose of this appendix to provide a description of the data employed in tne analysis contained in chapters 3 and 4 and to make APPENDIX explicit the variables employed in the attitudinal simulation. Many of the data in this thesis are contained in collections supplied by the Inter-University Consortium for Political and Social Research. interpretation of these data, it is acknowledged that without these data this analysis would have been impossible. While they bear no responsibility for either the analysis or Data sets from the ICPSR which are employed in this analysis include the following: 1952 1956 1960 1962 1964 1966 1968 1970 1972 1974 American National Election American National Election American National Election weighted data American National Election American National Election cross-sect1on sample American National Election American National Election cross-section sample American National Election cross-section sample American National Election American National Election weighted data 1976 American National Election Study: Study: Study: Stugy: Study: Study: Study: Study: Study: Study: Study: weighted data 163 1899 1762 respondents respondents 1181 respondents, 1279 respondents 1571 respondents, 1291 respondents 1557 respondents, 1580 respondents, 2705 respondents 1575 respondents, 2248 respondents, 164 State-Level Congressional, Gubernatorial and Senatorial Election Data for the United States, 1824-1972: 51 cases State-Level Presidential Election Data for the United States, 1824-1972: 51 cases In addition to these sources, original state-level data were collected from the following sources: Barone, Michael; Ujifusa, Grant; and Matthews, Douglas. The Almanac Of American Politics, 1974, 1976,9and 1978. New York: E. P. Dutton, 1974-1978. Source for updated state election returns and some state registration totals. Council of State Governments. The Book Of the States, Vols. VIII-XXI. Chicago: The Council of State Governments, 1950-1978. Source for most registration totals and state registration provisions. Johnson, Douglas W.; Picard, Paul R.; and Wuinn, Bernard. Churches and Church Membership in the United States: 1971. Washington: Glen- mary Research Center, 1971. One source Of percent Catholic in state population. ’ National Council of Churches. Churches and Church Membership in the Uni- ted States: An Enumeration by Counties, States and Regions. New York: National Council of Churches, 1957. A source Of percent Catholic in state population. Reitman, A., and Davidson, R. B. The Election Process: Voting Laws and Procedures. Dobbs Ferry, N.Y.: Oceana Publications, 1972. Source Of'recent state registration provisions. Scammon, Richard M., and McGillivray, Alice V. America Votes 12. Washington: Congressional Quarterly, 1977. Source of vote totals updating Burnham state-level studies. Smith, Constance. Voting and Election Laws. New York: Oceana Publi- cations, 1960. Source of early state registration provisions. U.S. Bureau of the Census. Statistical Abstract of the United States: 1950-1977. Washington: 1950-1977. The source fOr 511 state- level demographic data except percent Catholic. United States Commission on Civil Rights. The Voting Rights Act: Ten Years After. Washington: The Commission on Civil Rights, 1975. Source for registration figures for southern states. 165 Some specific coding decisions might be noted at this point. Following Kelley, Ayres, and Bowen (1967), the "registration system" variable was coded: Annual Cumulative Poll Tax 10 Annual Noncumulative Poll Tax 25 Annual Personal Registration 34 Permanent Registration System 100 Again figllowing Kelley, Ayres, and Bowen (1967), the literacy test variable is coded: Literacy Tests in South 20 Literacy Tests Outside South 80 NO Literacy Test 100 This coding scheme is an attempt to approximate the discriminatory application of literacy tests which has been particularly noted in the South. The figure for voter turnout is calculated by using the highest vote total figure for Presidential, Gubernatorial, or Senatorial elections divided by the eligible voting population. As such it provides a mini- mal estimate of the number of voters who cast a ballot. The variables employed in the attitudinal simulation deserve special attention. The dependent variables in the survey-based research are political efficacy, level of political information, and interest in the current campaign. The efficacy items are available for all the years reported in Table 3.3. This index is constructed by summiwg"efficacious" responses to the identified efficacy items in the SCR/CPS National Election Studies. No distinction is made here between "internal" and "external" efficacy. The "interest" variable is a score having three values--"very" interested, "somewhat" interested and I'not at all" interested in the 166 current campaign. In 1974 and 1976, I used an issue-based measure which relies on the 7-point issue scales. This value for some individual, k, is computed as follows: Bk = 1/N Q I(in ‘ eli) ' (in ‘ @2i)| i=1 where in is individual k's most preferred position on issue i, 011 is the Democratic candidate's (1976) or party's (1974) perceived location on issue i, and 921 is the perceived position of the Republican candidate (1976) or party (1974) on issue i. N refers to the number of seven- point issue scales on which the individual could locate himself as well as both candidates or parties. Finally the level of information variable refers generally to the number of candidate and party likes and dislikes the respondent expresses when asked these open-ended questions. For 1966, this variable is the interviewer's subjectively estimated estimate of the respondents' level of political information. The relationship between the survey-based independent variables and state variables are as follows: Survey Measure State Measure Binary variable: white-nonwhite Percent black Respondent's education Percent literate Respondent's family income Ranked state median family income Respondent's age Median state age Binary variable: home owner- Percent owner occupied not owner housing units Binary variable: Catholic- Percent Catholic not Catholic Urbanization Population density Binary variable: South—non-South South-non-South Where state figures were unavailable, they were interpolated or extra- polated from existing data. Southern states include Virginia, Alabama, 167 Arkansas, Florida, Georgia, Louisiana, Mississippi, North Carolina, South Carolina, and Texas. Any further questions about specific coding decisions should be addressed to the author: William A. Boyd Department of Political Science Emory University Atlanta, Georgia 30307 BIBLIOGRAPHY BIBLIOGRAPHY Books Abramson, Paul R. The Political Socialization of Black Americans. New York: Free Press, 1977. Almond, Gabriel A., and Verba, Sidney. The Civic Culture: Political Attitudes and Democracy in Five Nations. Boston: Little, Brown & CO., 1965. Banfield, Edward C. The Unheavenly City. Boston: Little, Brown & CO., 1970. Barnett, Vic. Comgarative Statistical Inference. London: John Wiley & Sons, 1973. Barry, Brian. Sociologists, EconomiSts and Democracy. London: Macmillan & CO., 1970. Berelson, Bernard R.; Lazarsfeld, Paul F.; and McPhee, William N. Voting. Chicago: University of Chicago Press, 1954. Black, Merle; Kovenock, David M.; and Reynolds, William 0. Political Attitudes in the Nation and the States. Chapel Hill, N.C.: Institute for Research in Social Science, 1974. Boulding, Kenneth E. Conflict and Defense A General Theory. New York: Harper & Row, 1962. Campbell, Angus; Converse, Philip E.; Miller, Warren E.; and Stokes, Donald E. The American Voter. New York: John Wiley & Sons, 1960. Campbell, Angus; Converse, Philip E.; Miller, Warren E.; and Stokes, Donald E. Elections and the Political Order. New York: John Wiley 8 Sons, 1966. Campbell, Angus; Gurin, Gerald; and Miller, Warren E. The Voter Decides. Evanston, 111.: Row, Peterson & CO., 1954. Coplin, William 0., ed. Simulation in the Study of Politics. Chicago: Markham Publishing CO., 1968. Downs, Anthony. An Economic Theory of Democracy. New York: Harper & Row, 1957. 168 169 Dye, Thomas R. Politics in States and Communities. 2nd ed. Englewood Cliffs, N.J.: Prentice-Hall, 1973. Evans, Rowland, and Novak, Robert. Lyndon B. Johnson: The Exercise of Power. New York: The New American Library, 1966. Friedman, Milton, ed. Essays in Positive Economics. Chicago: Univer- sity of Chicago Press, 1953. Gosnell, Harold F. Getting_0ut the Vote: An Experiment in the Simula- tion of Voting. Chicago: University of Chicago Press, 1927. . Why Europe Votes. Chicago: University of Chicago Press, 1930. Guetzkow, Harold, ed. Simulation in Social Science: Readings. Engle- wood Cliffs, N.J.: Prentice-Hall, 1962. Guetzkow, Harold; Alger, Chadwick F.; Brody, Richard A.; Noel, Robert C.; and Snyder, Richard. Simulation in International Relations: Developments for Research and Teaching. Englewood Cliffs, N.J.: Prentice-Hall,i1963. Hannan, Michael T. Aggregation and Disaggregation in Sociology. Lexington: Lexington Books, 1971. Johnston, John. Econometric Methods. 2nd ed. New York: McGraw-Hill, 1972. Key, V. 0.. Jr. Southern Politics in State and Nation. New York: Alfred A. Knopf, 1949. Kmenta, Jan. Elements of Econometrics. New York: Macmillan CO., 1971. Knight, Frank H. Risk, Uncertainty,yand Profit. Boston: Houghton Mifflin, 1921. Kovenock, David M., and Prothro, James W. Explainingythe Vote: Presi- dential Choices in the Nation and States, 1968. Part II: Presidential Choices in Individuai States. Chapel Hill, N.C.: Institute for Research in Sociai Science, 1973. Lane, Robert E. Political Life. Glencoe, 111.: Free Press, 1959. Lazarsfeld, Paul F.; Berelson, Bernard R.; and Gaudet, Hazel. Ihg_ Peoples Choice. 3rd ed. New York: Columbia University Press, 1944. Leege, David C., and Francis, Wayne L. Political Research: Design, Measurement and Analysis. New York: Basic Books, 1974. Luce, R. Duncan, and Raiffa, Howard. Games and Decisions. New York: John Wiley & Sons, 1957. 170 Matthews, Donald R., and Prothro, James S. Negroes and the New Southern Politics. New York: Harcourt, Brace & World, 1966. Merriam, Charles E., and Gosnell, Harold F. Non-Voting; Causes and Methods of Control. Chicago: University Of Chicago Press, 1924. Miller, Warren E.; Miller, Arthur H.; and Schneider, Edward J. American National Studies Data Sourcebook. Cambridge, Mass.: Harvard Uni- versity Press, 1980. Morris, Milton 0. The Politics of Black America. New York: Harper & Row, 1975. O Nie, Norman H.; Verba, Sidney; and Petrocik, John R. The Changing American Voter. Cambridge, Mass.: Harvard UniversityTPress, 1976. Oskamp, Stuart. Attitudes and Opinions. Englewood Cliffs, N.J.: Prentice-Hali, 1978. Patterson, Thomas E., and McClure, Robert D. The Unseeing Eye: The Myth of Television Power in National Politics. New York: Putnam‘s, 1976. Philips, K. P., and Blackman, P. H. .Electoral Reform and Voter Parti- cipation. Washington: American Enterprise Institute, 1975. Pindyck, Robert S., and Rubinfeld, Daniel L. Econometric Models and Economic Forecasts. New York: McGraw-Hill, 1976. Pool, Ithiel de Sola; Abelson, Robert P.; and Popkin, Samuel. Candidates, Issues and Strategies: A Computer Simulation of the 1960 and 1964 Presidential Elections. Cambridge, Mass.: M.I.T. Press, 1965. Riker, William H., and Ordeshook, Peter C. An Introduction to Positive Political Theory. Englewood Cliffs, N.J.: Prentice-Hall, 1973. Rudner, Richard S. Philosophy of Social Science. Englewood Cliffs, N.J.: Prentice-Hall, 1966. Scammon, Richard M., and Wattenberg, Ben J. The Real Majority. New York: Coward, McCann & Geoghegan, 1970. Sen, Amartya. Collective Choice and Social Welfare. San Francisco: Holden-Day, 1970. Smolka, Richard G. Registering Voters by Mail: The Maryland and New Jersey Experience. Washington: American Enterprise Institute, 1975. Theil, Henri. Principles Of Econometrics. New York: John Wiley & Sons, 1971. 171 Tullock, Gordon. Toward a Mathematics of Politics. Ann Arbor: Univer- sity of Michigan Press, 1964. U.S. Congress. House. Committee on Post Office and Civil Service. The Concept of National Voter Registration. HR 5400. 92nd Congress, 1st Session (Feb.-Ju1y 1972). U.S. Congress. Senate. Committee on Post Office and Civil Service. Hearings on Voter Registration. 92nd Congress, 1st Session (October 1971). Verba, Sidney, and Nie, Normal H. Participation in America. Political Democracy and Social Equality. New York: Harper & Row, 1972. Weber, Ronald E. Public Policy Preferences in the States. Bloomington, Ind.: Institute of Public Administration, Indiana University,l97l. Winch, Peter. The Idea of a Social Science. New York: Humanities Press, 1958. Wolfinger, Raymond E., and Rosenstone, Steven J. Who Votes? New Haven, Conn.: Yale University Press, 1980. Articles Achen, Christopher H. "Measuring Representation: Perils Of the Cor- relation Coefficient." American Journal of Political Science 21 (November 1977): 805-15. . "Measuring Representation." American Journal of Political Science 22 (August 1978): 475-510. Aldrich, John H. "Candidate Support Functions in the 1968 Election: An Empirical Application of the Spatial Models." Public Choice 22 (Summer 1975): 1-22. "Some Problems in Testing Two Rational Models Of Partici- pation." American Journal of Political Science 20 (November 1976): 713-33. Balch, George I. "Multiple Indicators in Survey Research: The Concept 'Sente of Political Efficacy'." Political Methodolggy 1 (Spring 1974 : 1-43. Burnham, Walter D. "The Changing Shape Of the American Political Universe." American Political Science Review 59 (March 1965): 7'28. "Election 1968--The Abortive Landslide." Transaction 6 _——(_December 1968): 18-24. 172 Cain, G. G., and Watts, H. W. “Problems in Making Policy Inferences From the Coleman Report." American Sociological Review 35 (1970): 228-42. Campbell, Angus. "Surge and Decline: A Study Of Electoral Change." In Elections and the Political Order, pp. 40-62. Edited by Angus Campbell, Philip E. Converse, Warren E. Miller and Donald E. Stokes. New York: John Wiley and Sons, Inc., 1966. Congressional Quarterly Weekly_Report. "Partisanship Growing on Voter Registration.“ (May 14, 1977): 909-11. Conrad, Thomas. "Rationality and Political Science: A Critical Analysis of the Consumer-Choice Model." Polity 2 (Summer 1970): 479-93. Converse, Philip E. "Change in the American Electorate." In The Human Meaning of Social Change, pp. 263-337. Edited by Angus Campbell and Philip E. Converse. New York: Russell Sage, 1972. . "Information Flow and the Stability of Partisan Attitudes." In Elections and the Political Order, pp. 136-57. Edited by Angus Campbeii, Philip E. Converse, Warren E. Miller and Donald E; Stokes. New York: John Wiley and Sons, Inc., 1966. . "The Nature of Belief SyStems in Mass Publics." In Ideology andypiscontent,pp. 206-61. Edited by David Apter. New York: The Free Press, 1964. Cramer, J. S. "Efficient Grouping, Regression and Correlation in Engel Curve Analysis." Journal Of the American Statistical Association 59 (March 1964): 233-50. Crewe, Ivor. "00 Butler and Stokes Really Explain Political Change in Britain?" European Journal Of Political Research 2 (1974): 47-92. Dawson, Richard 0. "Simulation in the Social Sciences." In Simulation in Social Science: Readings, pp. 1-15. Edited by Harold Guetzkow. Englewood Cliffs, N.J.: rentice-Hall, 1962. Duncan, Otis D. "Partials, Partitions, and Paths." In Sociological Methodology 1970, pp. 38-47. Edited by E. F. Borgatta and G. W. Bohrnstedt. San Francisco: Jossey-Bass, 1970. Erikson, Robert S. "Constituency Opinion and Congressional Behavior: A Reexamination of the Miller-Stokes Representation Data." American Journal of Political Science 22 (August 1978): 511-35. Ferejohn, John A., and Fiorina, Morris P. "The Paradox Of Not Voting: A Decision Theoretic Analysis." American Political Science Review 68 (June 1974): 525-36. 173 Ferejohn, John A., and Fiorina, Morris P. "Closeness Counts Only in Horseshoes and Dancing." American Political Science Review 69 (September 1975): 920-25. Fraser, John. "Why do High Income People Participate More in Politics? The Wrong Answer." Public Choice 13 (Fall 1972): 115-18. Frey, Bruno S. "Why 00 High Income People Participate More in Politics?" Public Choice 11 (Fall 1971): 101-5. Glaser, William A. "Intention and Voting Turnout." American Political Science Review 52 (December 1958): 1030-40. Goldberg, Arthur S. "Social Determinism and Rationality as Bases Of Party Identification." American Political Science Review 63 (March 1969): 5-25. Gordon, Robert. "Issues in Multiple Regression Analysis." American Journal of Sociology 73 (March 1968): 592-616. Hanushek, Eric A.; Jackson, John E.; and Kain, John F. "Model Specifi- cation, Use of Aggregate Data, and the Ecological Correlation Fallacy." Political Methodology 1 (Winter 1974): 89-107. House, James S., and Mason, William M. "Political Alienation in America, 1952-68." American Sociological Review 40 (April 1975): 123-47. Kelley, Stanley; Ayres, Richard E.; and Bowen, William G. "Registration and Voting: Putting First Things First." American PoliticalScience Review 61 (June 1967): 359-79. Kernell, Samuel. "Presidential Popularity and Negative Voting: An Alternative Explanation of the Midterm Congressional Decline of the President's Party." American Political Science Review 71 (March 1977): 44-66. Key, V. 0. Jr., and Munger, Frank. "Social Determinism and Electoral Decision: The Case Of Indiana." In American Voting Behavior, pp. 281-99. Edited by Eugene Burdick and Arthur J. Brodbeck. Glencoe, 111.: Free Press, 1959. McKelvey, Richard D. "Policy Related Voting and Electoral Equilibrium." Econometrica 43 (September-November 1975): 815—43. McKelvey, Richard D., and Ordeshook, Peter C. "A General Theory of the Calculus of Voting." In Mathematical Applications in Political Science VI, pp. 32-78. Edited by James F. Herndon and Joseph L Bernd. Charlottesville, Va.: The University Press of Virginia, 1972. 174 Matthews, Donald R., and Prothro, James W. "Social and Economic Factors and Negro Registration in the South." American Political Science Review 57 (March 1963): 24-44. Miller, Arthur H. "Political Issues and Trust in Government: 1964-1970." American Political Science Review 68 (September 1974): 951-72. Miller, Arthur H.; Miller, Warren E.; Raine, Alden S.; and Brown, Thad A. "A Majority Party in Disarray: Policy Polarization in the 1972 Election." American Political Science Review 70 (September 1976): 753-78. Mueller, John E. “Presidential Popularity From Truman to Johnson." American Political Science Review 64 (March 1970): 18-34. Niemi, Richard G. "Costs of Voting and Nonvoting." Public Choice 27 (Fall 1976): 115-19. Page, Benjamin 1. "The Theory of Political Ambiguity.“ American Political Science Review 70 (September 1976): 742-52. .________. "Elections and Social Choice: The State Of the Evidence.” American Journal Of Political Science 21 (August 1977): 639-68. Popkin, Samuel; Gorman, John W.; Phillips, Charles; and Smith, Jeffrey A. "Comment: What Have You Done for Me Lately? Toward an Investment Theory of Voting." American Political Science Review 70 (September 1976): 779-805. President's Commission on Registration and Voting Participation. "Psycho- logical and Administrative Barriers tO Voting." In Political Opinion and Behavior; Essays and Studies. 2nd ed., pp. 462-67. Edited by Edward C. Dreyer and W. A. Rosenbaum. Belmont, Calif.: Wadsworth Publishing CO., 1970. Przeworski, Adam, and Soares, Glaucio A. D. "Theories in Search Of a Curve: A Contextual Interpretation Of Left Vote." American Polit- ical Science Review 65 (March 1971): 51-68. RePass, David E. "Issue Salience and Party Choice." American Political Science Review 65 (March 1971): 389-400. . "Comment: Political Methodologies in Disarray: Some Alter- native Interpretations Of the 1972 Election." American Political Science Review 70 (September 1976): 814-31. Riker, William H., and Ordeshook, Peter C. "A Theory of the Calculus Of Voting." American Political Science Review 62 (March 1968): 25-43. Robinson, William S. "Ecological Correlations and the Behavior of Indi- viduals." American Sociological Review 15 (June 1950): 351-57. 175 Rosenstone, Steven J., and Wolfinger, Raymond E. "The Effect of Regis- tration Laws on Voter Turnout." American Political Science Review 72 (March 1978): 22-45. Russell, Keith P. "Political Participation and Income Level." Public Choice 13 (Fall 1972): 113-14. Seidman, David. "Simulation of Public Opinion: A Caveat." Public Opinion Quarterly 39 (Fall 1975): 331-42. Shepsle, Kenneth. "The Strategy of Ambiguity: Uncertainty and Electoral Competition." American Political Science Review 66 (June 1972): 555-68. Sullivan, John L.; Piereson, James E.; and Marcus, George E. "Ideo- logical Constraint in the Mass Public: A Methodological Critique and Some New Findings.“ American Journal of Political Science 22 (May 1978): 233-49. Tollison, Robert; Crain, Mark; and Pautler, Paul. "Information and Voting: An Empirical Note." Public Choice 24 (Winter 1975): 43-90 Tollison, Robert,and Willett, T. 0. "Some Simple Economics Of Voting and Not Voting." Public Choice 16 (Fall 1973): 59-71. Ward, Joe H. Jr. "Partitioning of Variance and Contribution or Impor- tance of a Variable: A Visit of a Graduate Seminar." American Educational Research Journal 6 (1969): 467-74. Wittman, Donald. "Comment." The Journal Of Law and Economics. (December 1975): 735-41. Zellner, Arnold. "An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias." In Readings in Econo- metric Theory, pp. 167-87. Edited by J. Malcolm Dowling andiFred R. Glahe. Boulder: Colorado Associated University Press, 1970. Unpublished Sources Abramson, Paul R. "The Decline of Electoral Participation in America: Political Attitudes and Behavioral Change." Paper presented at the Annual Meeting of the Western Political Science Association, San Francisco, Calif., 27-28 March 1980. Aranson, Peter. "A Theory Of the Calculus of Voting for Alternative Three-Contestant Election Systems." Ph.D. dissertation, University of Rochester, 1972. 176 Arsenau, Robert B., and Wolfinger, Raymond E. "Voting Behavior in Congressional Elections." Paper presented at the Annual Meeting of the American Political Science Association, New Orleans, La.- 1973. Weisberg, Herbert F. "Rational Abstention Due to Satisfaction: A Revised Look at the Effect of Candidate Evaluations on Turnout." Paper presented at the Annual Meeting of the Midwest Political Science Association, Chicago, 111., 21-22 April 1977. Zinser, James E., and Hausafus, Kurt F. "The Economics of Time-Off to Vote Laws." Paper presented at the Annual Meeting of the Midwest Political Science Association, Chicago, 111., 20-22 April 1978. "I11111111111111“