The study of voter participation has generated an enormous scholarly literature, both among empirically oriented researchers interested in voter turnout in real world elections and among formal theorists who study the implications of turnout rates for candidates’ and parties’ vote-seeking strategies. Despite this, relatively little is known about the theoretical and behavioral linkages between turnout and vote choice. We observe, for example, that those with weak partisan attachments are more likely to turn out in high profile elections and vote with the tide. Likewise, in American elections Republican identifiers are more likely to both turn out and to vote for their party’s candidate than are Democratic identifiers. However, the underlying behavioral foundations that might link these turnout and vote choices are not well understood. We study these linkages using a unified model of turnout and voter choice. The model captures the reciprocal relationship between turnout and voter choice, and in doing so informs several important questions about elections. These include assessing the relative importance of policy and non-policy considerations in the calculus of voter turnout, whether elections featuring attractive candidates motivate higher turnout, and whether changes in turnout rates have partisan implications.

Unified models incorporating both turnout and vote choice have been developed by spatial modelers interested in whether the simultaneous consideration of the voter’s turnout decision and candidate choice alters vote-maximizing candidates’ policy strategies, compared with models of candidate choice alone (Riker and Ordeshook, 1968; Hinich and Ordeshook, 1970; Enelow and Hinich, 1984). While our interest is related to these studies, it is conceptually distinct. Our model is not predicated on spatial or ideological proximity alone. Rather, it also includes the behavioral predictors most closely associated with observed voter turnout, including valence factors such as competence, integrity, charisma, etc. (Stokes, 1963; Ansolabehere and Snyder, 2000; Groseclose 2001). Foremost is our goal to elucidate the linkage between turnout and voter-choice decisions and to trace its electoral implications. We do so by employing the model of turnout and voter choice also presented in Adams and Merrill (2003), but use that model to evaluate the effects of political variables on the level of turnout and partisan choice rather than the effects of abstention and other variables on candidate strategy, which is the topic of the Adams-Merrill paper. In the present paper we distinguish between abstention that results from alienation from the candidates, indifference between the candidates, and some combination of these two motivations.

The spatial voting literature distinguishes between alienation-based abstention that results when parties or candidates are too distant from a voter to justify the cost of voting, and indifference-based abstention that results when parties or candidates are too similar to justify the costs of voting. However relatively few empirical studies seek to distinguish between these two sources of abstention. Zipp (1985) and Plane and Gershtenson (2004) both estimate the likelihood of voter turnout as a function of citizen alienation and indifference, where measures of these variables are constructed from citizen self-placement and candidate placement on issue and ideological scales. The former study estimates the likelihood of voter turnout as a function of citizen alienation and indifference on several American National Election Studies (ANES) issue scales, while the latter conducts similar analysis using data obtained from the American National Election Studies, Pooled Senate Election Study. Both the Zipp and Plane and Gershtenson studies find that citizen alienation and indifference contribute to the likelihood of abstention, although they differ on the most important underlying factor, with Zipp finding that indifference is the primary source of citizen abstention in presidential elections, while Plane and Gershtenson find that alienation is the major source of abstention in mid-cycle senatorial elections.

There are at least two additional reasons why it is useful to distinguish empirically between abstention from alienation and abstention from indifference. First, this distinction may enhance our ability to forecast the electoral effects of changes in turnout rates. For instance, a widely researched question—one that has generated conflicting conclusions among political scientists—is whether there is a partisan direction to the level of turnout.1 If different types of voters abstain for different reasons—for instance if Democratic partisans tend to abstain from alienation whereas Republicans abstain from indifference, as suggested by Hinich (1978)—then the effects of increases in voter turnout on the electoral fortunes of the two parties depends on whether the increase stems from changes in citizen alienation or citizen indifference.

Second, normative political theory stresses the intrinsic value of participation, and distinguishing between alienation-based and indifference-based abstention may help forecast how changes in important political variables—such as citizens’ evaluations of the candidates’ personal qualities (competence, integrity, etc.) or the menu of policy alternatives offered by competing candidates—affects overall voter turnout. Several studies explore whether elections involving attractive candidates motivate higher turnout (Brody and Page, 1973; Hinich, 1978; Weisberg and Grofman, 1981), while scholars as diverse as Downs (1957), Schattschneider (1960), and Piven and Cloward (1988) argue that tweedledum-tweedledee politics depress turnout. Our expectations about the turnout effects associated with citizens’ candidate evaluations may depend on whether abstention is driven primarily by alienation or by indifference.

Empirically, we estimate a unified model of turnout and voter choice in the 1980—1988 U.S. presidential elections, using American National Election Study (ANES) data. We focus on these ANES studies because they are the most recent to report validated voter turnout.2 Given our interest in the relationship between turnout and voter choice, we think it is essential to have as much confidence in our dependent variables as possible. We use the estimated models to determine the proportions of the electorate in each election that abstained from alienation and that abstained from indifference. We also perform model-based simulations to gauge whether voter turnout affected the results of these elections, and to estimate how both individual-level turnout and aggregate turnout would have responded to changes in voters’ evaluations of the candidates’ personal qualities and to shifts in candidates’ policies.

We use the estimated models to answer several questions suggested by the considerations discussed above: (1) Do candidate preferences of citizens who abstain from alienation differ significantly from the candidate preferences of those who abstain from indifference?, (2) Does enhancing respondents’ evaluations of both candidates’ personal qualities (such as competence, integrity, and leadership ability) significantly decrease respondents’ tendencies to abstain from alienation but not from indifference?,3 and (3) Do citizens’ tendencies to abstain from alienation and indifference vary significantly with their perceived policy distances from the candidates? Finally, we ask whether realistic changes in citizens’ perceptions of the candidates’ policy positions would have significantly changed aggregate voter turnout.

A UNIFIED MODEL OF TURNOUT AND THE VOTE: ABSTENTION FROM ALIENATION AND ABSTENTION FROM INDIFFERENCE

The model of turnout and voter choice, developed in Adams and Merrill (2003) and Adams, Merrill, and Grofman (2005) is grounded in spatial modeling, but includes also non-policy factors such a partisanship, socio-demographic characteristics of the voters, and valence evaluations of the candidates. In its purely spatial manifestation, abstention from alienation means the voter is too ideologically distant from his or her preferred candidate to justify the cost of voting, and abstention from indifference means that the ideological distance between candidates is insufficient to justify the cost of voting. Our approach reflects this intuition, but the decision calculus is such that alienation and indifference depend on more general predictors of turnout and voter choice.

The unified turnout and vote choice model is derived by building on a specification proposed by Sanders (1998, 2001) that also reflects elements of a similar model estimated by Lacy and Burden (1999; see also Burden and Lacy, 1999). Sanders derives a unified choice model that imposes the assumption that abstention results from voter indifference between the candidates. While Lacy and Burden do not specifically present a decision calculus, their model implicitly assumes that abstention results from alienation from the candidates. Our model builds directly on the Sanders’ specification, but extends it by distinguishing these two bases for citizen abstention.

The model begins by assuming that citizens have preferences defined over three alternative behaviors on Election Day: abstaining from voting, or voting for one of two candidates. Let U i (D) and U i (R) denote citizen i’s utility for two candidates, D and R, respectively. Let T i (A) denote the citizen’s alienation threshold. The alienation threshold is an estimated value such that if it is greater than or equal to the citizen’s candidate utilities he abstains. Similarly, let T i (I) denote a non-negative indifference threshold such that if its value is greater than or equal to the citizen’s utility differential between his preferred candidate and the rival candidate, he abstains. A citizen votes if he is neither alienated nor indifferent.

Translating these principles to decision rules for voting is relatively straightforward.

$$\eqalign{&\hbox{If }[U_i (D)-U_i (R)]>T_i (I)\hbox{ and }U_i (D)>T_i (A), \hbox{ the citizen votes for candidate }D.^{4}\cr &\hbox{If }[U_i (R)-U_i (D)]>T_i (I)\hbox{ and }U_i (R)>T_i (A),\hbox{ the citizen votes for candidate }R.\cr &\hbox{If }\left\vert U_i (D)-U_i (R)\right\vert \le T_i (I)\hbox{ and/or max} [U_i (D),U_i(R)]\le T_i (A),\hbox{ then the citizen abstains}.} $$

In words, a citizen votes for a given candidate if and only if he prefers the candidate to the other candidate, his utility difference between his preferred candidate and the other candidate exceeds his indifference threshold, and his utility for his most preferred candidate exceeds his alienation threshold.

Deriving an estimable choice model with desirable statistical properties from these decision rules is more complex. The mathematics, however, leads to a unified indifference and alienation (IA) statistical model that is an extension of well-known logistic specifications. The model contains four equations, one for each of the candidate utilities and two for the alienation and indifference thresholds. In the standard notation, voter i’s utility for candidate j may be written as:

$$\eqalign{U_i (j)&=\beta _j X_{ij} +\alpha _j V_{ij} +\varepsilon _{ij}\cr &=Z_i (j)+\varepsilon _{ij},}$$
(1)

where X ij is a vector of candidate attributes as perceived by voter i, V ij is a vector of the voter’s evaluation of candidate j on policy issues, and ɛ ij is a random disturbance term. The quantities β j and α j are vectors of parameters to be estimated along with their standard errors. Z i (j) simply summarizes the measured component of voter i’s candidate evaluation.

The voter’s alienation threshold may be described similarly,

$$\eqalign{T_i (A)&=\beta_A A_i +\upvarepsilon _{iA}\cr &=Z_i (A)+\upvarepsilon _{iA},} $$
(2)

where A i is a vector of variables expected to influence whether the voter will turn out on Election Day. Again, we seek to estimate the vector of parameters β A and its standard errors.

The indifference threshold is written a bit differently. Specifically,

$$\eqalign{T_i (I)&=\hbox{exp}(\beta _I I_i)\cr &=Z_i (I),} $$
(3)

where I i is also a vector of voter attributes expected to influence the likelihood that the voter will turn out on Election Day and the β I are parameters to be estimated. Note that the definition in equation (3) constrains T i (I) to be positive. We impose this constraint because a negative indifference threshold is logically impossible, because this threshold is associated with the voter’s utility differential between her preferred candidate and the rival candidate that by definition is non-negative. The indifference threshold coefficient vector β i and its standard errors are estimable. We label the model specified by equations 1–3 the unified indifference-alienation (IA) model.

Assuming that the utility errors are distributed type-1 extreme value, the likelihood function may be written as:

$$\eqalign{P_i (D)&={\hbox{exp}(Z_i (D))\over \hbox{exp}(Z_i (D))+\hbox{exp}(Z_i (R))^{\ast}\hbox{exp}(Z_i (I))+\hbox{exp}(Z_i (A))}\cr P_i (R)&={\hbox{exp}(Z_i (R))\over \hbox{exp}(Z_i (D))^{\ast} \hbox{exp}(Z_i (I))+\hbox{exp}(Z_i (R))+\hbox{exp}(Z_i (A))}\cr P_i (Abstain)&=1-P_i (D)-P_i (R)\cr \hbox{Log-likelihood}&=\sum\limits_i\hbox{ln} [P_i(D)^{\ast}votedem+P_i (R)^{\ast} voterep+P_i (Abstain)^{\ast} abstain],} $$
(4)

where votedem, voterep, and abstain are binary variables indicating whether the ith respondent voted Democratic, voted Republican, or abstained.5 The log-likelihood function is maximized with respect to the utilities and threshold parameters. It returns estimates of the coefficient vectors and their standard errors that have the desirable statistical properties of maximum likelihood estimation.

A CONDITIONAL LOGIT MODEL FOR CITIZENS’ TURNOUT AND VOTING DECISIONS

We proceed by estimating the unified IA specification for the 1980–1988 ANES respondents.6 Our dependent variable is the respondent’s choice among three alternatives in the presidential election: vote Democratic, vote Republican, or abstain. Beginning with voter choice, we model citizens’ utilities for the candidates as a function of party identification, which is scored at 1 if the respondent identified with the candidate’s party or at zero otherwise; ideological distance to each candidate, which is measured as the squared distance between the respondent’s liberal-conservative self-placement and the candidate’s (mean perceived) ideological position; policy distance to each candidate, which is measured as the mean squared distance between the respondent’s self-placements and the candidate’s (mean perceived) positions along the policy scales included in the study; respondent assessment of candidate character, which is calculated as the average score the respondent assigned to the candidate on attributes such as intelligence, honesty, and leadership ability; race, which is scored at 1 if the respondent is black or at zero otherwise; and retrospective evaluations of the national economy, coded from −2 (much worse) to 2 (much better). These variables have been found to influence voters’ candidate preferences in prior studies of voting in presidential elections (Alvarez and Nagler, 1995, 1998).

We model citizens’ alienation and indifference thresholds as a function of several recognized predictors of turnout in U.S. presidential elections. These include respondent education, race, political efficacy and prior voting habits. Since the behavioral literature does not generally distinguish abstention due to alienation from abstention due to indifference, we do not have strong theoretical or empirical prior beliefs about which turnout predictors pertain to the alienation threshold, which pertain to the indifference threshold, and which pertain to both thresholds. Consequently, we include all predictors of turnout in both thresholds, with the exception of respondent perception of the closeness of the election. Several studies have shown that close elections, at the margins, tend to have higher turnout (Munger and Cox, 1989; see Endersby, Galatas and Rackaway, 2002, for a review.). We include this variable in the indifference threshold on the assumption that its most direct effect will be felt in the calculus involving differences in the utilities for the two candidates.7 Among variables in the alienation and indifference thresholds education is coded on a seven-point scale ranging from less than high school education to post baccalaureate degree; political efficacy, is calculated as the citizen’s mean self-placement on the ANES political efficacy scales (recoded on a scale from 0 to 1 representing low to high efficacy); previous vote, is scored at 1 if the respondent reported having voted in the previous presidential election and at zero otherwise; perceived election closeness is scored at 1 if the respondent believed the presidential election would be close and at zero otherwise.

The following equations display the resulting specifications for the respondents’ candidate utilities, their alienation thresholds, and their indifference thresholds. The D and R designators refer to the Democratic and Republican candidates, respectively:

$$\eqalign{ {U}_{i}({D})=&{b}_{1}(\hbox{Dem PID}) +{b}_{2}(\hbox{squared ideological distance between }i\hbox{ and D})\cr &+{b}_{3}(\hbox{mean squared policy distance between }i\hbox{ and D})+ {b}_4(\hbox{D. character})\cr &+{b}_{5}(\hbox{retrospective economy})+{b}_6(\hbox{race})+\upvarepsilon_{i{\rm D}}\cr =&{Z}_{i}(\hbox{D})+\upvarepsilon_{i{\rm D}}\cr {U}_{i}({R})=&{b}_7+{b}_{1}(\hbox{Rep PID})+ {b}_{2}(\hbox{squared ideological distance between }i\hbox{ and R})\cr &+{b}_{3}(\hbox{mean squared policy distance between }i\hbox{ and R})+ {b}_4(\hbox{R. character})\cr &+{b}_{8}(\hbox{retrospective economy})+\upvarepsilon_{i{\rm R}}\cr =&{Z}_{i}(\hbox{R})+\upvarepsilon_{i{\rm R}}\cr {T}_{i}({A})=&{b}_9+{b}_{10}(\hbox{race})+ {b}_{11}(\hbox{efficacy})+{b}_{12}(\hbox{previous vote})+ {b}_{13}(\hbox{education})+\upvarepsilon_{i{\rm A}}\cr =&{Z}_{i}(\hbox{A})+\upvarepsilon_{i{\rm A}}\cr {T}_{i}({I})=&\hbox{exp}[{b}_{14}+{b}_{15} (\hbox{race})+{b}_{16}(\hbox{efficacy})+{b}_{17}(\hbox{previous vote}) +{b}_{18}(\hbox{education})\cr &+{b}_{19}(\hbox{perceived election closeness})]\cr =&{Z}_{i}({I}).} $$
(5)

Note that we constrain the coefficients for partisanship, ideology, policies, and character (coefficients b 1–b 4) to be equal across candidates. This imposes the assumption that these variables are equally salient with respect to respondents’ evaluations of each candidate. For example, the effect of Democratic partisanship on one’s utility for a Democratic candidate is assumed to be the same as Republican partisanship on one’s utility for a Republican candidate.8

Table 1 reports the estimated parameters and their standard errors for the unified turnout-and-vote-choice model for the 1980–84–88 elections.9 For each election, the coefficients relating to partisanship, policy distances, and candidate character in column 1 are statistically significant at the .01 level and show the expected signs, with respondents’ utilities for the candidates increasing when they identify with the candidate’s party, when they evaluate the candidate’s character positively, and when they share the candidate’s policy positions. Blacks consistently favor the Democratic candidate, and, consistent with previous research, voters holding positive impressions of the national economy tend to favor the incumbent candidate (Fiorina, 1981; Alvarez and Nagler 1995).10

TABLE 1 Conditional Logit Equations for the Unified Indifference-Alienation Model, 1980–84–88 Presidential Elections
Full size table

With respect to respondents’ turnout decisions, the coefficients reported in columns 2–3 suggest that blacks abstain disproportionately from alienation, and that high levels of political efficacy reduce the likelihood that voters abstain from indifference. The negative values for the previous vote variable indicate that, as expected, respondents who reported voting in the previous election are less likely to abstain at the current election (from either indifference or alienation). However, after controlling for previous election turnout and political efficacy, we obtain no systematic relationship between abstention and respondents’ educational levels or their expectations of a close election.

ALIENATION AND INDIFFERENCE IN PRESIDENTIAL ELECTIONS

To estimate the relative contributions of indifference and alienation to the likelihood of abstention in the 1980–84–88 elections we use the Table 1 parameter estimates to compute the relative contribution of alienation and indifference to the aggregate likelihood of abstention. Specifically, for each respondent we calculate the probability that he or she was indifferent between the candidates but was not alienated from them; the probability that he or she was alienated but not indifferent; and the probability that he or she was both alienated and indifferent. We then use the means over all respondents of these predicted probabilities to estimate, for each election, the proportions of the electorate that fall into each category.11

Table 2a reports these predicted probabilities, which provide estimates of the relative contributions of alienation and indifference to voter abstention. These estimates suggest that both alienation and indifference depressed turnout in the 1980–84–88 elections, with alienation contributing slightly more than indifference to voter abstention. Our estimates of the proportion of ANES respondents who abstained from alienation vary between 18 and 21 percent for the these elections; the estimated proportion of respondents who abstained from indifference varies between 13 and 14 percent; the proportion that are projected to be both alienated and indifferent varies between 11 and 18 percent. These results suggest that both alienation and indifference contribute significantly to voter abstention in contemporary U.S. presidential elections.

TABLE 2 Rates of Abstention, Indifference, and Alienation in the 1980–84–88 Presidential Elections
Full size table

In Table 2b we report whether citizens who abstain from alienation differ in their candidate preferences from those who abstain from indifference. Here we compare the computed probabilities that Democratic and Republican partisans were alienated, the probabilities that they were indifferent, and their overall probabilities of abstaining from either alienation or indifference. For comparison purposes we also report projected abstention rates for independent voters. Two important conclusions emerge from these calculations. First, Democrats and Republicans do not differ dramatically in terms of their predicted probabilities of abstaining (whether from alienation or indifference). Consistent with Highton and Wolfinger’s (2001) analysis of the 1992 and 1996 ANES data, Democratic respondents were slightly, but consistently, more likely to abstain than were Republicans. The figures in Table 2b also show that Democratic and Republican identifiers do not differ significantly in their reasons for abstention. In each election Democratic partisans are slightly more likely to be both alienated and indifferent than are Republicans, and hence slightly more likely to abstain.

The simulations suggest that increases in voter turnout are unlikely to have significant partisan implications for election outcomes, regardless of whether increased turnout stems from reductions in indifference, in alienation, or in both. To confirm this conclusion we simulated the effects of increasing voter turnout under three scenarios: (a) no abstention from alienation, (b) no abstention from indifference, and (c) a “full-turnout” scenario with no abstention due to either alienation or indifference.12 These results are reported in Table 3 and reveal that the two sources of abstention have similar effects upon election outcomes, and that in each election increases in turnout would have only modestly benefited the Democratic candidate. Only under an unrealistic full-turnout scenario are the Democratic candidate’s electoral gains as high as 2–4 percentage points and are marginally statistically significant. In addition, we note that our conclusion that even full turnout would at most modestly increase the Democratic vote percentage is consistent with the conclusions reported by Martinez and Gill (2005; see figure 2) in their analyses of the 1960–64–76–84–2000 elections.

TABLE 3 Projected Election Outcomes for Hypothetical Increases in Voter Turnout, 1980–84–88 Presidential Elections
Full size table

We next ask whether the presidential candidates’ images as measured by citizens’ evaluations of candidates’ personal characteristics such as honesty, intelligence, and integrity affect the two bases for voter abstention similarly. We evaluate this question by calculating, for each election, the likelihood that ANES respondents would abstain because of alienation and indifference under three different “candidate image” scenarios: a “neutral” scenario, in which respondents’ candidate character ratings (scaled from −1 to +1) were fixed at the values recorded in the ANES studies; an “unpopular candidate” scenario, in which each respondent’s rating of each candidate was reduced by .15; and a “popular candidate” scenario, in which respondents’ ratings of each candidate was increased by .15. Our analyses of ANES respondents’ candidate evaluations suggest that this is a realistic range of scenarios to explore.13

The results reported in Table 4 support the argument that elections featuring candidates with attractive personal qualities motivate higher turnout, and do so because the respondents’ predicted aggregate probabilities of being alienated drop sharply. There is little comparable effect for indifference. In toto, these simulations suggest that the magnitude of the effects of candidate images upon aggregate turnout is on the order of 5–8 percent—i.e., that realistic variations in voters’ candidate evaluations can increase or decrease voter turnout from five to eight percentage points.

TABLE 4 The Projected Abstention Rates for Changes in NES Respondents’ Evaluations of the Candidates’ Characters
Full size table

We next evaluate the extent to which citizens’ decisions to abstain depend on their perceived policy distances from the candidates, as spatial modeling research assumes. For illustrative purposes, Fig. 1 plots the observed proportions of respondents who abstained in the 1988 ANES as a function of liberal (1) to conservative (7) self-placement, stratified by partisanship.14 We also plot the abstention probabilities projected by the model, and the mean perceived liberal-conservative locations of the candidates, Bush and Dukakis. The figures’ two most striking features are that observed and predicted probabilities for abstention match closely even when disaggregated by partisanship, and that both Democratic and Republican identifiers are least likely to abstain when their ideological self-placement is close to the mean perceived location of their party’s candidate.

FIG. 1
figure 1

Comparison of Observed and Model-Projected Abstention Rates in the 1988 American NES, by Party and by Liberal-Conservative Location. Note: The liberal-conservative positions for Bush and Dukakis represent the mean placements assigned to these candidates by the respondents in the 1988 ANES.

Full size image

Figure 2 plots the model-projected abstention probabilities of respondents stratified by whether their abstention principally owes to alienation, indifference or both of these foundations. These show that citizens’ tendencies to abstain because of alienation—but not indifference—are strongly related to perceived policy distance from the candidates. Among Republican partisans (Fig. 2a), the propensity to be alienated rises from approximately 30 percent for those who share George H. W. Bush’s ideological placement (i.e., those who self-place at 5 or 6 on the liberal-conservative scale) to over 60 percent for more liberal Republicans who are spatially distant from Bush.15 The figure also shows that conservative Republicans who prefer Bush to Dukakis on both partisan and ideological grounds are far less likely to be indifferent than are liberal Republicans. The patterns for Democratic partisans (Fig. 2b) mirror those of the Republicans: Democrats are significantly less likely to be alienated when they share Dukakis’s ideology, and projected indifference is lowest among liberal Democrats who prefer their party’s candidate on both partisan and ideological grounds. Identical computations using the 1980 and 1984 data yielded the same substantive conclusions, providing evidence that citizens’ tendencies to abstain are significantly related to their perceived policy distances from the candidates.

FIG. 2
figure 2

Model-Projected Abstention Rates in the 1988 American NES by Type of Abstention: (a) Republican Partisans, (b) Democratic Partisans.

Full size image

Finally, we ask whether realistic changes in citizens’ perceptions of the candidates’ policy positions would have significantly changed aggregate voter turnout. We simulated the effect of shifting ANES respondents’ placements of candidates’ ideological and policy positions along various dimensions. We explored four scenarios: a “convergent policy” scenario, in which respondents’ placements of the candidates’ positions along all of the policy and ideology scales were shifted .5 units closer to the mean voter position; an “actual policies” scenario, in which the candidates were fixed at their actual (perceived) positions; a “moderate polarization” scenario in which both candidates’ perceived positions along the policy and ideology scales were shifted .5 units further away from the mean voter position; an “extreme polarization” scenario in which both candidates’ perceived positions were shifted 1.0 policy unit further away from the mean voter position. Under the “extreme polarization” scenario, for instance, Dukakis’s ideological position was shifted leftward to the position 2.10 along the 1–7 Liberal-Conservative scale (Dukakis’s actual (mean perceived) position was 3.10 on the scale), while Bush’s ideological position was shifted rightward from 5.23 (his mean perceived position) to 6.23. For each scenario we computed the expected proportion of the electorate that was indifferent, the expected proportion of alienated voters, and the expected aggregate turnout rate.

Table 5 reports these simulations for 1988, although comparable analyses using the 1980 and 1984 data produced identical substantive conclusions. These suggest that changes in the candidates’ perceived policy positions would not have significantly affected aggregate voter turnout. The simulations show that moderate degrees of either policy convergence or divergence would change the expected aggregate abstention rate by less than 1 percent compared with the abstention rate for the actual candidate configuration. Furthermore, under the extreme policy polarization scenario the expected abstention rate would increase by a modest 2.4 percent.

TABLE 5 Alienation-Motivated Abstention, Indifference-Motivated Abstention, and Overall Turnout: 1988 Election
Full size table

The above conclusions appear counterintuitive given that our previous results showed that abstention is positively related to perceived policy distance from the candidates. The explanation is that in this case the candidates’ perceived policy shifts are of similar magnitude, but in opposite directions. Consequently, the net effects are minimal. To see this, consider the turnout effects associated with two polar-opposed candidate configurations, the convergent policy scenario (column 1) and the extremely polarized scenario (column 4). Compared with the convergent scenario, the stark policy choices perceived by voters in the extreme polarization scenario decrease the expected amount of aggregate voter indifference by 4.6 percentage points while increasing projected aggregate alienation by 7.5 points. This is because a large majority of NES respondents hold moderate positions that render extreme candidate positioning unattractive. We conclude that realistic changes in citizens’ perceptions of a candidates’ policy positions, if balanced by a comparable change by the other candidate, will not significantly affect aggregate voter turnout.

Conclusion

Although the distinction between abstention due to alienation and abstention due to indifference is commonplace in the spatial modeling literature, we provide the first explicit attempt to distinguish empirically between these motivations and determine the political factors that affect them using a unified model of turnout and voter choice. Our objective in this paper is to explore two questions concerning turnout and voting behavior in contemporary presidential elections: What are the relative contributions of alienation and indifference to depressing turnout in these elections, and, how does distinguishing between alienation-based and indifference-based abstention improve our ability to understand and forecast important political outcomes?

With respect to the first question, both alienation and indifference appeared to contribute substantially to depressing turnout in the 1980–84–88 presidential elections, which suggests that theories and models of turnout should incorporate both of these sources of abstention in analyses of American presidential elections. Further computations on data for 1996 and 2000 (based on unvalidated voting) support similar substantive conclusions to those we reach for 1980–84–88, which suggests that our conclusions are not confined to the 1980s.

With respect to the second question, we conclude that distinguishing between alienation-based and indifference-based abstention provides micro-foundation evidence that corroborates the argument that there is little partisan direction associated with changes in turnout, although increased turnout slightly benefits Democratic candidates. In addition, the distinction between indifference and alienation is critical for understanding how presidential candidates’ images affect aggregate turnout, and for understanding the turnout effects associated with the candidates’ policy platforms. With indifference-based abstention, presidential elections featuring attractive candidates do not motivate additional voters to participate. But with alienation-based abstention, attractive candidates should attract additional voters to the polls. Given our finding that alienation and indifference are of approximately equal importance in presidential elections, we conclude that realistic variations in the candidates’ images could alter turnout by between four and seven percentage points.

Finally, we also find that the unified indifference-alienation model illuminates the impact of voters’ policy motivations on turnout. On one hand, we conclude that abstention is largely policy-based—i.e. citizens’ individual-level tendencies to be alienated and indifferent are strongly affected by their evaluations of the candidates’ policies. This supports the perspective advanced in spatial modeling studies and suggests that such studies may be relevant to candidate strategies in real-world elections. On the other hand, we conclude that aggregate turnout does not depend significantly on citizens’ perceptions of the candidates’ policy platforms. This is because candidate configurations that leave many voters indifferent over policies (when candidates are perceived as proposing similar, centrist policies, for instance) provoke little policy alienation, whereas candidate configurations that leave few voters indifferent over policies (such as elections involving a liberal ideologue versus a conservative ideologue) provoke widespread alienation. Hence we do not find support for the oft-expressed opinion that tweedledee-tweedledum politics (i.e., convergent candidate platforms) depresses turnout. Instead, our findings suggest that tweedledee-tweedledum candidates will cause more voters to be indifferent, but fewer voters to be alienated, thereby leaving aggregate-level turnout largely unchanged.

To date, the empirical literature on turnout has largely focused on answering the question of whether citizens will choose to vote or abstain. We have shown that distinguishing why voters abstain, that is, whether abstention is driven by alienation or by indifference, is also important for understanding presidential elections.

Notes

  1. 1.

    See, for example, Tucker and Vedlitz (1986), Weisberg and Grofman (1981), Wolfinger and Rosenstone (1980), Highton and Wolfinger (2001), Grofman, Owen and Collett (1999), Teixeira (1992), Petrocik (1987), Citrin, Schickler and Sides (2003), and Martinez and Gill (2005).

  2. 2.

    We have, however, estimated this model for the 1996 and 2000 election cycles, using the self-reported turnout available from the ANES studies for those years. These results are consistent with those reported in the present study and in Adams, Merrill, and Grofman (2005). We do not estimate the model for the 1992 presidential election because of Ross Perot’s non-trivial vote share. The decision calculus presented in the following section is limited to two candidates because extending it to three candidates is significantly more complex and viable third candidates are infrequent in US elections.

  3. 3.

    The logic underlying this question is that, because abstention from alienation turns on the absolute level of citizens’ candidate evaluations, attractive candidates should motivate higher turnout; however abstention from indifference turns on citizens’ comparative evaluations of the competing candidates, and this differential is the same whether citizens evaluate both candidates positively or negatively.

  4. 4.

    Note that given that we have specified that T i (I) is non-negative, [U i (D)−U i (R)]≥ T i (I) implies that U i (D)≥ U i (R).

  5. 5.

    An important feature of the model is that, although the errors assume independence, the model does not impose the independence of irrelevant alternatives (IIA) property. This is because the denominators for the choice probabilities differ. Thus, removing an alternative changes the relative odds of the remaining choices (Sanders, 1998, p. 93).

  6. 6.

    These estimates also appear in Adams, Merrill, and Grofman (2005: Table 8.2) and those for the year 1988 appear in Adams and Merrill (2003: Table 1).

  7. 7.

    The logic for this specification is that election closeness is a proxy for the respondent’s perception of the likelihood of casting a decisive ballot, which is relevant to the instrumental decision to abstain from indifference but not to the expressive decision to abstain from alienation.

  8. 8.

    Estimation of the model without these constraints revealed no substantive differences in the parameter estimates. We retain the parameter constraints because they simplify presentation and interpretation of the estimated coefficients and simulations.

  9. 9.

    The previous vote variable was not included in the 1984 ANES and hence is omitted for this election. We implement the log-likelihood function in a STATA ado file, and estimate its parameters using the method of maximum likelihood. In each of the three elections, the likelihood function maximized without difficulty. A copy of the computer code necessary to implement the log-likelihood function is available from the authors on request.

  10. 10.

    As a diagnostic of model fit, we use the coefficients in Table 1 to predict the probabilities that each respondent voted for the Democratic candidate, voted Republican, or abstained. Using the mean values of each variable as the expected outcome, these accurately reproduce the corresponding raw frequencies in the ANES surveys.

  11. 11.

    From the estimated parameters, one may calculate the probability P i (A) that a citizen i is alienated from the candidates as:

    $$ P_i(A)={\hbox{exp}(Z_i (A))\over \hbox{exp}(Z_i (D))+\hbox{exp}(Z_i (R))+\hbox{exp}(Z_i (A))} $$

    and the probability P i (I) that i is indifferent between the candidates as:

    $$ P_i (I)=1-\left[{\hbox{exp}(Z_i (D))\over \hbox{exp}(Z_i (D))+\hbox{exp}[ Z_i (R)+Z_i (I)]}+{\hbox{exp}(Z_i (R))\over \hbox{exp}(Z_i (R))+\hbox{exp}[Z_i (D)+Z_i (I)]}\right]. $$

    Using these probabilities along with the vote probabilities from the maximized likelihood, we obtain the probability P i (A not I)=P(abstain)−P i (I) that the respondent is alienated but not indifferent, the probability P i (I not A)=P(abstain)−P i (A) that the respondent is indifferent but not alienated, and the probability P i (A and I)=P(abstain)−P i (A not I)−P i (I not A) that the respondent is indifferent and alienated, where the probability P(abstain) that the voter abstains is given in equation 4 above.

  12. 12.

    These scenarios were specified by adjusting the alienation and indifference thresholds.

  13. 13.

    The mean (scaled) character ratings that respondents assigned to the candidates in the 1980–84–88–92–96–2000 NES studies ranged from a low of .04 (for Carter in 1980) to a high of .41 (for Clinton in 1996), a difference of .37. This suggests that the range of scenarios we investigate with respect to respondents’ candidate ratings provides a reasonable estimate of the candidates’ impacts on turnout in presidential elections.

  14. 14.

    There are only five Republican partisans located at 1 on the liberal-conservative scale, so that the downturn in the empirical plot in Fig. 1a at the extreme left is not statistically significant. All other plotted points represent at least 15 respondents.

  15. 15.

    To see this, note that the proportion of alienated Republicans equals the proportion who are computed to abstain due to alienation only (the bottom strata of respondents pictured in Fig. 2a) plus those who are computed to abstain due to being both alienated and indifferent (the top strata of respondents pictured in Fig. 2a).