The impact of the repeat-voting-habit persistence phenomenon on the probability of voting in presidential elections.

Cebula, Richard J. ; Durden, Garey C. ; Gaynor, Patricia E. 等

1. Introduction

The purpose of this study is to develop and test a probability model of individual voting behavior under the assumption that, when voting for the Office of the President, all potential voters are affected at the margin by the expected costs and benefits of voting. Additionally, some potential voters will have higher voting probabilities that are independent of such considerations. These voters may be independently motivated by external and/or internal factors that are captured under the umbrella of what is referred to here as the "repeat-voting-habit persistence phenomenon." The latter might reasonably be interpreted so as to include the impact of "social conditioning." According to Tollison and Willett (1973), unless social conditioning is accounted for, empirical voting models will not be completely specified and estimates of turnout on individual voting probabilities will be biased. Specifically, Tollison and Willett (1973, p. 61) argue that:

   Where social conditioning has affected individuals' response
   patterns in a significant manner, for instance, by instilling in
   individuals a sense of duty to vote, then to be useful, explanatory
   models must take this into account, even if this comes at the cost
   of the disciplinary pureness of the resultant model.

Interestingly, some years after this work by Tollison and Willett (1973), Knack (1992) provided a perspective and empirical findings that seem to smoothly interface with the notion of social conditioning. In particular, in the Knack study, voting is regarded as a collective action problem, one that is overcome by means of solidarity and purposive selective incentives. Specifically, Knack finds empirical support for the hypothesis that enforcement of voting norms through social sanctions significantly augments voter participation. Within this framework, the decline in voter turnout in the United States in recent decades is interpreted in terms of a weakening of social ties, which adversely affects the socialization and enforcement of those norms responsible for generating civic participation. The latter argument is echoed frequently in the well-known nonempirical work of Putnam (2000, pp. 31-3).

To date, very little empirical work has attempted to measure how social conditioning, or a related pllenomenon such as that found in Knack (1992), might influence general election voting behavior. One exception is the recent study by Matsusaka and Palda (1999). Although the primary focus of the Matsusaka and Palda (1999) study is the relative lack of explanatory power for empirical voting models, the effect of internal, unobserved factors on voting is briefly analyzed. A basic assumption of Matsusaka and Palda (1999) is that individuals who voted in a previous election have a higher probability of voting in a current election. Underlying this assumption is a corollary, namely, that consecutive-election voters are motivated by socially conditioned internal factors that are not controlled for in standard Downs (1957) rational interest models. To correct this shortcoming, the Matsusaka and Palda (1999) model includes a dummy variable, whose value equals 1 if the respondent voted in the immediately previous election and whose value is 0 otherwise. The variable is always very statistically significant, suggesting that consecutive-election voters are different from those who are not consecutive-election voters.

The present study seeks to extend and refine the procedures introduced by Matsusaka and Palda (1999). To begin with, we use data on individual voting in two consecutive Presidential elections. Adopting this data set, we then integrate the "repeat-voting-habit persistence phenomenon" (or simply, REPVOTHAB) into the empirical analysis. This hypothesis argues that a factor or variety of factors such as (i) social conditioning and/or (ii) some logic-based or education-based reasoning involving recognition of the need to vote in order to preserve democracy and freedom and/or (iii) family-instilled values or observed family voting behaviors, and/or (iv) perhaps even a simple psychologically generated "habit" of "mechanically" voting for President per se combine to result in a situation in which those persons who have voted in a given Presidential election are more likely to vote in the following one, ceteris paribus. Furthermore, the present study empirically demonstrates that, although the exact processes cannot be easily identified, the REPVOTHAB phenomenon (which we argue does include social conditioning and/or related phenomena) appears to systematically affect voting among the respondents in the data sample. These results can be interpreted as supporting the Tollison and Willett (1973) view (and presumably that of Knack [1992]) that unless the repeat-voter-habit phenomenon (inclusive of social conditioning) is accounted for, flawed estimates of voter turnout and of individual voting probabilities will be the likely result.

Section 2 of this study provides a brief survey of the related empirical literature and establishes the basis for the variables used in the empirical analysis. Data limitations prevent incorporation of all previously used variables, but the most consistently statistically significant influences are accounted for. In section 3, we modify the rational voter model to include the repeat-voting-habit persistence phenomenon. In section 4, we create a measure for the REPVOTHAB variable and incorporate this variable into a regression using the overall sample of those who were potential voters in 1984. In section 5, we stratify the data and empirically demonstrate that white males who voted in the 1980 Presidential election had a much higher probability of voting in 1984 than those who did not vote in the 1980 Presidential election. This is partly due to differences in mean values and in coefficient estimates between the two subgroups. Using the decomposition technique developed by Oaxaca (1973) and Blinder (1973) for use in wage and income discrimination studies, we isolate those portions of the total probability differential that are attributable to mean and coefficient differences. After this adjustment, a substantial probability differential remains. We attribute this probability differential to the components of the REPVOTHAB phenomenon, which includes (among other things) differing social conditioning and differing internal voting motivations as between those who did and did not vote in the 1980 Presidential election. A summary of the results and their implications are presented in section 6.

2. A Review of Empirical Studies of Voting Behavior

As Tullock (2006, p. 41) observes, since the modern introduction of rational interest as a primary voting motivation (i.e., the rational voter model [Downs 1957; Black 1958; Buchanan and Tullock 1962]), general election voting has been extensively tested using a wide variety of variables, methodologies, and data. The common assumption among these studies is that politicians and constituents maximize a political utility function, which incorporates both the expected costs and expected benefits of voting. Empirical results are rather mixed, but the Downs (1957) rational interest theory of voting is largely supported.

Early studies focused on the marginal effects of closeness, income, age, race, gender, concurrent elections (such as in the Senate or House of Representatives), and other factors on the probability that an individual will or will not vote or that average voter turnout will increase or decrease. Examples of important works include the study of individual voting by Ashenfelter and Kelly (1975), cross-section studies of turnout by Tollison and Willett (1973) and Silberman and Durden (1975), and the excellent early survey paper by Foster (1984), which summarizes the literature on empirical voting models up to that point. An outstanding, somewhat more recent contribution to the individual voting literature that is pertinent to the present study is that by Knack (1992), as is summarized above.

Somewhat more recent work by Durden and Gaynor (1987) tried to bring the empirical analysis up to date and to extend or refine the basic model. Even more recent refinements and extensions include new analyses of the following: (i) the effect of closeness on voter turnout (Kirchgassner and Zu Himmern 1997; Grofman, Collett, and Griffin 1998; Cebula 2001), with all three of these studies suggesting that closeness is more influential than previously indicated; (ii) the determination of how weather influences turnout, with Knack (1994) finding little or no influence and no particular "bad weather benefit" to Republicans; (iii) the impact of motor voting laws, where Knack (1995), Franklin and Grier (1997), and Knack and White (1998) find strong evidence that states with such laws experience larger turnout in both voting and registration; (iv) the historical impact of voting restrictions, with Heckelman (1995) finding that poll taxes and secret balloting were very effective in preventing targeted groups from voting; (v) reducing information costs, in which Matsusaka (1995) presents an information-based theory, which successfully incorporates the standard determinants from the Downs (1957) rational interest model and finds that if information shows the candidate to be what the voter thinks he (she) should be, then voter turnout increases; (vi) the low explanatory power of most empirical voting models, and suggestions that better results may require the identification of additional cost and benefit variables, higher levels of aggregation, or time-series analysis (Matsusaka and Palda 1999); and (vii) "expressive voting," in which, according to Copeland and Laband (2002) and Cebula (2004), voting allows voters to express their feelings/emotions on one or more issues or candidates.

3. Incorporation of the Effects of the Repeat-Voting Habit Phenomenon on the Internal Motivation to Vote

Some years ago, Riker and Ordeshook (1968) argued that since perceived marginal costs are high and expected marginal benefits relatively small, purely consumption and/or investment-based economic motives do not sufficiently explain the act of voting. Distributed among a voting population, then, there logically must exist a repeat-voting habit, a phenomenon that includes a socially conditioned sense of civic duty or sense of responsibility to vote or not vote (Tollison and Willett 1973; Knack 1992). The socially conditioned or otherwise enhanced voting proclivity may exist, and it apparently is more strongly felt by some potential voters than others. These internal, unobserved influences presumably affect individual voting choices in a way that is independent of the rational voter model as it typically is interpreted. Thus, REPVOTHAB is a complement to the rational voter model.

The rational voter model suggests that an individual will make the voting decision after evaluating a cost-benefit ratio, such as:

P(EB)/C, (1)

where P is the probability that one's vote will determine an election outcome, EB is a vector of expected consumption and investment benefits to be forthcoming if the voter's candidate wins, and C is the opportunity cost of voting, which may be interpreted to include lost time and/or direct expenditures (gasoline, cab fare, lost wages). But this specification is clearly insufficient because the probability that one's vote will influence an outcome is extremely small even in very close elections. Moreover, any thinking/circumspect voter will know this to be true.

We suggest that an arguably more realistic expression of the cost-benefit ratio is given by the following:

(P(EB) + IB)/C, (2)

where IB represents the internal benefits one obtains from the very act of voting such that IB reflects all of the manifestations included within the repeat-voting-habit persistence phenomenon as described above. The magnitude of IB is not dependent upon the success or failure of the voter's candidate but is determined, for each potential voter, by the level of influence of the REPVOTHAB phenomenon, inclusive of any socially conditioned internal motivation for that individual. Given that (P(EB) + IB)/C provides a reasonable expression of the voting cost-benefit ratio, if IB is sufficiently large, ceteris paribus, then P(EB) can be inconsequential, and one will still vote and, alternatively, if IB is small, one may abstain from voting.

The underlying assumption in this reasoning is that if internal voting motivations differ, then potential voters can be stratified into two classes. The first class includes those who appear to have a strong inclination to vote because of the influence of (the strength of the reaction to) the repeat-voter-habit persistence phenomenon and what it encompasses, with instrumental and observable consumption-based factors held constant, and the second class includes those who appear not to possess this characteristic (i.e., to not be thusly influenced). Estimation results from each of the subsamples may then be used to compute voting probabilities to determine whether there are significant differences. If there are, then further analysis can be performed (Blinder 1973; Oaxaca 1973) to determine what portion of the total probability of voting differentials may be attributed to differences in variable means, variable coefficients, and "internal voting motivations." Given the existence of differences in internal motivations to vote (i.e., differential responsiveness to the REPVOTHAB phenomenon), models that do not account for these motivations and the resulting internal benefits, will produce biased voting probabilities (Tollison and Willett 1973; Knack 1992) and lower levels of explanatory power (Matsusaka and Palda 1999).

In the next section of this study, we create a measure of the REPVOTHAB variable. In the subsequent section of the study, we provide the empirical findings. In this empirical analysis, we stratify the data, perform empirical analyses, and use the results to determine whether two classes of potential voters in fact do exist. If so, then the probability of voting between the two classes will differ and (using the Oaxaca-Blinder decomposition technique mentioned earlier) this differential can be separated into a portion due to mean differences and a portion due to differences among coefficient estimates. We expect that the decomposition procedure will leave a portion of the probability differential unexplained and argue that this unexplained differential is due to internal voting motivations reflected in the repeat-voting habit persistence phenomenon. Given that this is a reasonable assumption, we then create a measure for internal motivations and test this variable in a regression that employs the full data set.

4. Creation of the REPVOTHAB Variable

The nature of the REPVOTHAB variable created for use in full sample regressions warrants some explanation. For simplicity of exposition and for consistency with the well-known related studies by Tollison and Willett (1973) and Knack (1992), we may at times use the phrase "social conditioning" interchangeably with the phrase "REPVOTHAB phenomenon;" although, the latter is, in fact, understood for purposes of this study to be realistically clearly more inclusive in scope than "social conditioning." In other words, the REPVOTHAB phenomenon is expressly represented as including a host of potentially difficult-to-quantify although, easy-to-describe factors, including social conditioning per se as presented in Tollison and Willett (1973). That said, it is observed that the Matsusaka and Palda (1999) study attempted to account for "social conditioning" (which they label "citizen duty") by using a dummy variable with value equal to 1 if the respondent voted in the previous election, and a value equal to 0 otherwise. The variable is always highly significant statistically but, because the same or similar influences will have affected voting at the margin in both elections, this technique may produce inaccurate results (Kau and Rubin 1979; Hird 1993; Burkey and Durden 1998).

The latter three studies faced a problem very similar to the one faced here. Each of the these works is concerned with accounting for differences in Congressional voting patterns among legislators, which were not captured by standard measures such as campaign contributions and constituent characteristics. Specifically, they require a measure of a given legislator's "own ideology," which can then be incorporated into models for estimating the determinants of Congressional voting on particular types of special interest legislation. To properly measure a Congressperson's own ideology, they required an ideology variable that has been purged of general constituent and special interest influences. To create this variable, ratings from such ideology-based organizations as the Americans for Democratic Action and the League of Conservative Voters were regressed on a set of independent variables that were designed to capture the marginal effects of external influences on how a legislator is rated by the various ideology-based groups. Given that external influences have been controlled for, residuals from these equations are assumed to measure the effect of a legislator's own internal ideological leanings. For example, suppose that a legislator's expected score by the League of Conservative Voters is low, but the actual score is relatively high. Then the corresponding residual will also be high, and this condition is taken to be evidence of an internally held "own" ideological leaning. The internal ideology variables thusly created turned out to be highly useful in empirical regressions for all three studies, improving the estimation results and suggesting that legislator preferences can be isolated and approximated.

Creation of the legislators' own ideology variable is admittedly imperfect and subject to the criticism that what the variable actually measures cannot be known with certainty. What the previous studies did may be considered valid (or at least useful), however, because the existence of an ideological leaning on certain issues seems reasonable and because their use of the manufactured "own ideology" variable improved the specification of their models. The variable created and used here to measure the effects of internal motivations is subject to similar concerns but also may be valid for similar reasons. Specifically, it seems quite reasonable to believe that there is a class of voters that is motivated not only by perceived costs and benefits, but also by internal motivating factors (included under the REPVOTHAB "umbrella"), which have been recognized but never adequately accounted for in estimating models. As will be shown, use of the internal motivation variable created here (REPVOTHAB) is consistently significant and correctly signed, and appears to improve model specification.

The procedure used in the Kau and Rubin (1979), Hird (1993), and Burkey and Durden (1998) studies provides a blueprint for constructing the measure of REPVOTHAB variable to be used here. A logit equation is estimated, with the variable "Voted in 1980" equal to 1 if the respondent voted in 1980 and "Voted in 1980" equal to 0 otherwise, treated as the dependent variable. Independent variables are the same as those used in the stratified-data regressions, namely, South, West, Standard Metropolitan Statistical Area (SMSA), Married, North Central, High School education, College 3 years, College 4 years or more, Government employment, home ownership (Home Owner), Income, Age, and Unemployed.

The effects of the independent variables on the probability of voting should be generally the same between the 1980 and 1984 Presidential elections. For example, if education increases, the effect is expected to be positive with respect to both. However, one potential drawback is that individuals in the sample will not have maintained exactly the same socioeconomic characteristics. Some respondents may have more education, higher income, become married or divorced, retired, and so forth. These problems cannot be entirely eliminated but, since the 198-1984 time period is relatively short, the probability of substantial bias would seem modest. We have attempted nevertheless to lessen biases by restricting the data set to those individuals who are aged 25-65 in 1984 (21-61 in 1980). The sample thus includes for both years only primarily nonretired persons, and persons with a four-year college education potential.

5. Formal Empirical Analysis

In this section of the study, in Tables 2, 3a, and 3b, we provide the results of four logit regressions. All four empirical estimations are based on a sample of white males who were potential voters in the 1984 Presidential election. These data were obtained from the U.S. Census Bureau (1984). Although not provided here in order to save considerable space and in order to focus more directly on the measurement of self-interest and the repeat-voter-habit phenomenon variable, the findings for females are entirely consistent with those for males. Unfortunately, we were unable to find Census Bureau surveys for more recent Presidential elections than that for 1984, which included the critical question of whether the individual voted in the previous Presidential election; therefore, the data for 1984 were logically the best option to adopt in order to pursue the goals of this study. In any event, the explanatory variables adopted (see Table 1 for precise definitions of the explanatory variables) are similar to those employed in a great many studies and have been justified in a range of published works (see the literature listed in the References of this study).

Table 2, columns 2 and 3, is a regression using the full sample and with the value of the dependent variable equal to 1 if the respondent voted in 1980, and equal to 0 otherwise. This estimate provides the residual values for the REPVOTHAB variable and is the means by which, in principle following the procedure in Kau and Rubin (1979), Hird (1993), and Burkey and Durden (1998), we attempt to capture the effect of unmeasured internal influences on individual voting behavior (i.e., the factors impounded in the repeat-voting-habit persistence phenomenon variable, inclusive of social conditioning). Table 2, columns 4 and 5, is a regression of the full sample that includes the REPVOTHAB variable. The dependent variable is again dichotomous with a value of 1 if the respondent voted in 1984, and a value equal to 0 otherwise. To obtain the results shown in Tables 3a and 3b, data are stratified into two subsamples of white males who were potential voters in 1984. The first subsample (results shown in Table 3a) consists of white males who did not vote in 1980, and the second subsample (results shown in Table 3b) consists of those who did vote in 1980. This framework in Tables 3a and 3b allows us to compare relative influences of socioeconomic conditions on the two groups to determine whether they are different and, if so, to attempt to measure the nature and extent of those differences in terms of relative probabilities of voting.

Based on the Downs (1957) rational voter model, our expectation is that the probability of voting in 1984 will be affected (at the margins) by perceived costs and benefits for all variations of the empirical model and that this will be demonstrated via the four regressions summarized. Further, our purpose is to identify any evidence that strongly suggests that there is a measurable difference in the probability of voting in 1984 for those who are known to have voted in 1980 versus those who are known to have not voted in 1980. This higher voting probability would derive at least partly, we believe, from the components of the repeat-voting-habit persistence phenomenon, which we argue includes socially conditioned, internally felt motivations.

We now discuss generally the results of the regressions, focusing on the reliability of the REPVOTHAB variable, differences in marginal influences of socioeconomic conditions between those who are consecutive voters and those who are not, and on measurement of the differences in the probability of voting between consecutive and nonconsecutive voters.

There are no surprises with respect to the influences of the most important commonly used independent variables. In the estimations in Table 2, being a homeowner increases the probability of voting, as do age, income, government employment, higher levels of educational attainment (High School, College 3, and College 4), residence in the North Central region, and status as married. In both estimates, being unemployed and residence in the South both exercise negative effects on voter turnout. SMSA location exercises a negative impact on voter turnout according to Table 2, columns 2 and 3, but although negative, it is not quite significant at the 10% level in Table 2, columns 4 and 5. West region location is insignificant in Table 2.

The variable of primary interest in this study is the REPVOTHAB variable. As shown in columns 4 and 5 of Table 2, the estimated coefficient on the REPVOTHAB is positive and statistically significant at far beyond the 1% level. Indeed, the t-statistic on this variable is by far the highest of any in the estimates, +35.2. This result suggests, as hypothesized in the present study, that the components of the REPVOTHAB phenomenon are very powerful motivators for voter participation. Accordingly, it is clear that models are incomplete when they do not control for internal expressive motivations included under the REPVOTHAB umbrella (i.e., for motives such as social conditioning, which are not directly associated with costs and benefits as typically considered).

To determine whether the means and coefficients are statistically different between subsamples (those who did not vote in 1980 as compared with those who did), we conducted z-tests for the means and Chow tests for the coefficients. These results are provided in Tables 3a and 3b. The z-tests suggest that respondents who voted in 1980 are more likely to live in non-South regions, to be married and own their own homes, and to have considerably more education, as compared with those who did not vote in 1980. They are also on average older and more affluent. The Chow tests involved creating a dummy variable with a value of 1 if the respondent voted in 1980, and a value of 0 otherwise, and using this dummy variable to create interaction terms for each variable in the whole-sample estimation. The coefficients that appear to be statistically different are indicated with an asterisk (*). The comparisons of coefficients suggest that several are different at the 5% significance level, including College 4+, Home Owner, Income, and Age, as well as the intercept. The estimated coefficients on variables West, Married, Government, and Unemployed appear different at the 10% significance level.

We now compute probabilities for each equation, with the expectation that there will be a significant difference between them. We then use the Oaxaca-Blinder decomposition technique to see how much of that differential is due to means, how much is due to coefficient differences, and whether an unexplained portion remains that might logically be attributable to REPVOTHAB motivations not accounted for by costs and benefits as perceived by respondents and as proxied by the specified set of independent variables.

The computation of probabilities is accomplished using the following equation:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. (3)

In Equation (3), [b.sub.0] is the intercept, [b.sub.1], ..., [b.sub.n], are the coefficients, and [x.sub.1], ..., [x.sub.n] is the vector of independent variables. The respective probabilities are 0.693 for those who voted in 1980 and 0.306 for those who did not vote in 1980, a differential of 0.387. To determine the portion of this differential that is due to differences in coefficients, one addresses the following question: How much would the probability of voting in 1984 increase if the coefficients from the "did not vote in 1980" equation were replaced with coefficients from the "did vote in 1980" equation, other things (means) equal? The answer is that the probability estimate would increase from 0.306 to 0.353, so that the portion that is due to coefficient differences is given by 0.353 - 0.306 = 0.047.

To determine the portion that is due to differences in means, the question is: How much would the probability of voting in 1984 increase if the means from the "did not vote in 1980" equation were replaced with means from the "did vote in 1980 equation," other things (coefficients) equal? The answer here is that the probability estimate would increase from 0.306 to 0.541, so that the portion that is due to means differentials is given by 0.541 - 0.306 = 0.235. Together, coefficient and mean differences account for 0.047 + 0.235 = 0.282, or about 73% (=0.282/0.387) of the total probability differential, leaving an unexplained residual of 0.105 or 27% (=0.105/0.387).

These results imply that if the two subgroups were, in the aggregate, exactly alike with respect both to average characteristics and responses to marginal changes in explanatory variables, those who did not vote in 1980 would have a mean voting probability of 0.588, while those who did vote in 1980 would have a mean voting probability of 0.693. It seems logical and reasonable to then infer that the latter group is very likely subject to the repeat-voting-habit persistence phenomenon (inclusive of socially conditioned, internal motivations to vote) and that this proclivity is demonstrated/manifested in the form of consecutive Presidential-election cycle voting. If so, then models that include consecutive and nonconsecutive voters but do not account for the repeat-voting-habit persistence phenomenon will produce seriously biased voting probability estimates.

6. Summary and Conclusions

This study extends the well-known rational voter model to include a composite measure to capture the residual effects of internal, psychological, and sociological motives not previously accounted for in empirical studies of Presidential election voting. These motives are captured in this study under the umbrella of the repeat-voting-habit persistence phenomenon, which arguably includes "social conditioning" (Tollison and Willett 1973; Knack 1992) or "internal motivation" and may to at least some extent reflect a sense of duty or sense of civic duty to vote, as well as a simple "habit" of voting, among several other possible factors (refer to the Introduction of this study). Estimations using U.S. Census Bureau (1984) data from the 1984 Presidential elections suggest that previously unmeasured internal motives, which we capture in a variable called REP VOTHA B, may exert a powerful influence on individual voting behavior. When the data are stratified into subsamples of those who exhibit the REPVOTHAB motivation and those who do not, the mean voting probability for the former is 0.693 and for the latter is 0.306. Of this differential, .047 is due to differences in coefficients, and 0.235 is due to differences in variable means (where 0.047 + 0.235 = 0.282 or 73% of the entire differential). The balance of the differential is 0.105 (i.e., 27% of the entire differential), which reflects the impact of the repeat-voting-habit persistence phenomenon (REPVOTHAB), which is argued here to include "social conditioning" as considered in Tollison and Willett (1973) and arguably in Knack (1992). These results suggest that if REPVOTHAB is not accounted for, then the probability of voting among those who do exhibit this characteristic will be understated and the probability of voting will be overstated among those who do not exhibit this characteristic.

Received April 2007; accepted November 2007.

References

Ashenfelter, O., and S. Kelly. 1975. Determinants of participation in Congressional elections. Journal of Law and Economics 18:695-733.

Black, D. 1958. The theory of committees and elections. Cambridge, UK: Cambridge University Press.

Blinder, A. 1973. Wage discrimination: Reduced form and structural estimates. Journal of Human Resources 8:436-55.

Burkey, M. L., and G. C. Durden. 1998. The political economy of clean air legislation: An analysis of voting in the U.S. Senate of the 1990 Clean Air Act. Environmental and Resource Economics 11:119-34.

Buchanan, J., and G. Tullock. 1962. The calculus of consent. Ann Arbor, MI: University of Michigan Press. Cebula, Richard J. 2001. The Electoral College and voter participation: Evidence on two hypotheses. Atlantic Economic Journal 29:304-10.

Cebula, Richard J. 2004. Expressiveness and voting: Alternative evidence. Atlantic Economic Journal 32:216-21.

Copeland, Cassandra, and David Laband. 2002. Expressive voting. Public Choice 110:351-63.

Downs, Anthony. 1957. An economic theory of democracy. New York: Harper and Row.

Durden, Garey C., and Patricia E. Gaynor. 1987. The rational behavior theory of voting participation: Evidence from the 1970 and 1980 elections. Public Choice 53:231-42.

Foster, Carroll B. 1984. The performance of rational voter models in recent Presidential elections. American Political Science Review 78:678-90.

Franklin, David P., and Edward Grier. 1997. Effects of motor voter legislation: Voter turnout, registration, and partisan trends in the 1992 Presidential election. American Politics Quarterly 25:104-17.

Grofman, Bernard, Charles Collett, and Robert Griffin. 1998. Analyzing the turnout-competition link with aggregate cross-section data. Public Choice 95:233-46.

Heckelman, Jac C. 1995. The effect of the secret ballot on voter turnout rates. Public Choice 82:107-24.

Hird, John A. 1993. Congressional voting on superfund: Self-interest or ideology? Public Choice 77:333-57.

Kau, James B., and Paul H. Rubin. 1979. Self-interest, ideology, and logrolling in Congressional voting. Journal of Law and Economics 22:365-84.

Kirchgassner, Gebhard, and A. Meyer Zu Himmern. 1997. Expected closeness and turnout: An empirical analysis for the German general elections, 1983 1994. Public Choice 91:3-25.

Knack, Stephen. 1992. Civic norms, social sanctions, and voter turnout. Rationality and Society 4:133-56.

Knack, Stephen. 1994. Does rain help the Republicans? Theory and evidence on turnout and the vote. Public Choice 79:187-209.

Knack, Stephen. 1995. Does motor voter work? Evidence from state-level data. The Journal of Politics 57:796-811.

Knack, Stephen, and James White. 1998. Did states' motor voter programs help the Democrats? American Politics Quarterly 26:344-66.

Matsusaka, John G. 1995. Explaining voter turnout patterns: An information theory. Public Choice 84:91-117.

Matsusaka, John G., and Filip Palda. 1999. Voter turnout: How much can we explain? Public Choice 98:431-46.

Oaxaca, Ronald L. 1973. Male-female wage differentials in urban labor markets. International Economic Review 14:693-709.

Putnam, Robert. 2000. Bowling alone. New York: Simon & Schuster.

Riker, William H., and Peter C. Ordeshook. 1968. A theory of the calculus of voting. American Political Science Review 62:25-42.

Silberman, Jonathan, and Garey C. Durden. 1975. The rational behavior theory of voter participation: The evidence from Congressional elections. Public Choice 23:101-8.

Tollison, Robert D., and Thomas D. Willett. 1973. Some simple economics of voting and not voting. Public Choice 16:59-71.

Tullock, Gordon. 2006. Some thoughts on the voting process. Atlantic Economic Journal 34:41-6.

U.S. Census Bureau. 1984. March 1984 supplement to the current population survey. Washington, DC: U.S. Government Printing Office.

Richard J. Cebula, * Garey C. Durden, ([dagger]) and Patricia E. Gaynor ([double dagger])

* Economics Department, Armstrong Atlantic State University, 11935 Abercorn Street, Savannah, GA 31419, USA; E-mail richard.cebula@armstrong.edu; corresponding author.

([dagger]) Economics Department, Appalachian State University, Boone, NC 28608, USA.

([double dagger]) Economics Department, Appalachian State University, Boone, NC 28608, USA. The authors are indebted to the editor and two referees whose suggestions greatly improved this study.

Table 1. The Independent Variables

1.  Age: Categorical age groupings used in regressions: 25-31 = 1; 32
    38 = 2; 39-45 = 3; 46-52 = 4; 53-59 = 5; 60-65 = 6.
2.  Income: a categorical variable for family income levels: 1-14.
3.  Education: Dummies = 1, = 0 otherwise for high school, college
    1-3 years, college 4+ years.
4.  Married: Dummy = 1, = 0 otherwise.
5.  Own home: Dummy -- 1, = 0 otherwise, a measure of wealth.
6.  Unemployed: Dummy = 1, = 0 otherwise.
7.  Government: Dummy = 1 if employed by government, = 0 otherwise.
8.  SMSA: Dummy = 1 if residing in an SMSA, = 0 otherwise.
9.  Region: Dummies: South = 1, = 0 otherwise; West = 1, =
    0 otherwise; North Central = 1, = 0 otherwise; reference =
    Northeast.
10. REPVOTHAB: residuals from the weighted voted in the 1980 regression
    equation, as a measure of internal motivation, social conditioning,
    habit persistence, and/or other internal factors influencing
    voting.

Table 2. Determinants of Voting in 1980 and 1984

                          1980                       1984

Variable        Coefficient    t-Value     Coefficient    t-Value

South              -0.0911     -2.22 **       -0.2171    -4.38 ***
West                0.0036      0.08           0.0134     0.26
SMSA               -0.1418     -4.68 ***      -0.0596    -1.65 *
Married             0.3442      8.98 ***       0.2999     6.86 ***
North Central       0.2446      5.78 ***       0.2333     4.62 ***
High school         0.7412     15.22 ***       0.8750    15.90 ***
College 3           1.3249     21.03 ***       1.5516    22.49 ***
College 4           1.9586     27.40 ***       2.3287    29.16 ***
Government          0.5139     10.10 ***       0.6660    10.91 ***
Own home            0.4506     12.01 ***       0.7788    14.71 ***
Income              0.0675     13.78 ***       0.0715    12.31 ***
Age                 0.3502     26.71 ***       0.3384    21.99 ***
Unemployed         -0.1518     -2.20 **       -0.3010    -3.54 ***
REPVOTHAB           --            --           0.6646    35.20 ***
Intercept          -2.5864    -14.93 ***      -2.9066    -6.28 ***
DF               8907                      11564
% CC (a)           70.6                       79.0

(a) Percent correctly classified.

*, **, and *** indicate statistical significance at the 10%, 5%,
and 1% levels, respectively.

Table 3a. Subsample of Respondents Who Did Not Vote in 1980

Variable        Coefficient    t-Value    Mean    SD    z-Test

South             -0.0847 *   -1.16       0.28   0.45   -3.80 *
West               0.1203 *    1.54       0.24   0.43    2.56 *
SMSA              -0.0292     -0.55       0.54   0.50   -1.11
Married            0.0401 *    0.68       0.64   0.48    5.93 *
North Central      0.1028      1.33       0.23   0.42    2.59 *
High school        0.4308      3.82 ***   0.35   0.48   -8.46 *
College 3          0.9255      5.51 ***   0.20   0.40    2.72 *
College 4+         1.6025 *    7.61 ***   0.17   0.38   22.28 *
Government         0.2989 *    3.17 ***   0.12   0.32   16.88 *
Own home           0.1991 *    3.32 ***   0.59   0.49    9.28 *
Income             0.0348      3.55 ***   6.03   3.85   12.90 *
Age                0.0474 *    2.64 ***   3.73   1.62    9.54 *
Unemployed        -0.0554 *   -0.49       0.10   0.31   -6.22 *
Intercept         -2.2142 *   -5.73 ***
DF              4698
% CC (a)          70

(a) Percent correctly classified.

* indicates a 10% or better level of statistical significance for
differences in coefficients and means (z-test values). ** and
*** indicate statistical significance at the 5% and 1% levels,
respectively, with respect to the t-values.

Table 3b. Subsample of Respondents Who Did Vote in 1980

Variable        Coefficient    t-Value    Mean    SD    z-Test

South             -0.2537 *   -2.87 ***   0.25   0.43   -3.80 *
West              -0.0547 *   -0.59       0.26   0.44    2.56 *
SMSA              -0.0599     -0.09       0.53   0.50   -1.11
Married            0.2754 *    3.12 ***   0.69   0.46    5.93 *
North Central      0.1928      2.15 **    0.25   0.43    2.59 *
High school        0.6691      6.11 ***   0.28   0.45   -8.46 *
College 3          1.0490      7.99 ***   0.22   0.41    2.72 *
College 4+         1.5621 *   10.89 ***   0.35   0.48   22.28 *
Government         0.6175 *    6.35 ***   0.24   0.42   16.88 *
Own home           1.0739 *    9.17 ***   0.67   0.47    9.28 *
Income             0.0491      4.79 ***   7.20   3.80   12.90 *
Age                0.3060 *   10.77 ***   4.18   1.69    9.54 *
Unemployed        -0.5162 *    2.79 ***   0.07   0.26   -6.22 *
Intercept         -2.7748 *   -6.04 ***
DF              7661
% CC (a)          76.1

(a) Percent correctly classified.

* indicates a 10% or better level of statistical significance for
differences in coefficients and means (z-test values). ** and
*** indicate statistical significance at the 5% and 1% levels,
respectively, with respect to the t-values.