What is a correct vote?

Determining the quality or ‘correctness’ of a vote decision is no easy task. Who is to decide what is correct? Should we impose a particular set of values upon the judgement? Following Dahl, Lau and Redlawsk adopt a theoretical middle ground by defining correctness based on the ‘fully informed’ interests of individual voters. As Dahl writes, ‘a person's interest or good is whatever that person would choose with fullest attainable understanding of the experiences resulting from that choice and its most relevant alternatives’.Footnote ⁸ In any modern-day national political campaign, voters are inundated with far more information than they can possibly process. This in turn impedes access to the information that is actually relevant to the dimensions of judgement the voter actually cares about. Almost by necessity, then, most voters are choosing with less (and often with a lot less) than full information. Thus Lau and Redlawsk define a correct vote decision as one that is the same as the choice that would have been made under conditions of full information.

Lau and Redlawsk present two measures of correct voting.Footnote ⁹ The first is only available in a controlled experimental setting in which ‘all alternatives in the choice set’ and ‘all dimensions of judgement’ can be carefully controlled, and complete information can be provided to voters after they have made their decisions; the voters then can determine for themselves if, had they known everything they know now, they would have changed their decision. These restrictions of course limit the usefulness of this first measure of correct voting.

Lau and Redlawsk's second measure, which is closely related to the first empirically, can be estimated with fairly standard survey data in which voters report their own values and interests. The researcher then applies objective criteria to determine which candidate is the voter's correct choice, given those values and interests. For example, most citizens hold a party identification that can be easily matched to the parties and candidates competing in an election. Respondents tell us which political issues they care about (at least enough to report a position), which can then be compared to the actual stands of candidates or parties (where ‘actual’ is determined by political experts) on those policies. People tell us how much they care about an incumbent's job performance (again by answering survey questions), which can be used to weigh expert judgements about the incumbent's actual job performance in an overall vote calculus. In some surveys, respondents report whether they ‘identify with’ or ‘feel close to’ different social groups that are widely believed (again, by experts) to benefit from one party or another being in power. They may also assess the competence or integrity of party leaders. These different evaluative considerations can be combined into overall ‘utility’ scores associated with each party or candidate. The correct vote is for the party/candidate with the highest total. Lau and Redlawsk call this second measure of correct voting normative naïve – naïve in that it is based on the voter's own preferences and determination of what criteria of judgement are important, and normative because (1) it is based on experts’ judgements of which party or candidate is closest to the voter's expressed preferences and (2) the same criteria of judgement must be applied to all candidates or parties in the choice set. We build upon this second procedure here.

First, however, it is worth considering more carefully what ‘fully informed preferences’ actually are, even in theory, and whether there are any boundaries to the type of alternatives or dimensions of judgement that should be considered in determining a correct vote, especially when the number of alternatives (particularly in systems that allow for write-in candidates or parties) and potential dimensions of judgement are almost infinite. If correct voting allows citizens to determine for themselves what dimensions of judgement they care about, then ‘fully informed’ must apply only to those particular criteria. We would not expect people to be true experts on any of those judgement criteria, only that they would know (at most) what the ‘reasonably informed citizen’ would know: that they would be aware of the official policies or public statements from party leaders about the topics (even if they are not a major part of the current election campaign), and that they would be aware of any approximate consensus among elite opinion leaders as to significant differences (or lack thereof) between the parties or candidates on those topics. If voters prioritize multiple considerations by relative importance, then fully informed preferences should somehow reflect that rank ordering as well.

As to boundaries, we might begin by limiting consideration to those aspects of society over which government has some control (governments should not be held accountable for the weather, for example), and by focusing on past government performance (to the extent this can be objectively determined) or factors that could influence future performance, such as policy proposals that are meant to address the important issues of the day. We want to include both retrospective and prospective concerns in any definition of correct voting, because they have both been normatively associated with democratic representations.Footnote ¹⁰ Purely idiosyncratic personal concerns (fixing the pothole in the street at the end of my driveway) are excluded on normative grounds; there has to be some larger social (though not necessarily societal) aspect to the consideration.Footnote ¹¹ Thus social identifications with a class, ethnicity or religion that are associated with a party or leader – and that might reasonably be expected to benefit all members of the group if that party or candidate were elected – should be considered because they affect us, not just me. The most obvious and directly relevant such social identification would be with a political party. In almost every country, political parties represent historical social divisions and current and past policy performance reflecting those divisions, and as such are very relevant to determining a correct vote. Aspects of political leaders’ personality that could likely influence government performance, such as their intelligence, experience, competence or integrity, are very legitimate considerations for correct voting; other ‘personal’ factors (such as a leader's attractiveness) would not be relevant if they do not directly influence government performance – even though such factors might sometimes be very important determinants of actual vote choices. So-called strategic concerns about the electability of a particular party or candidate would also be very relevant, because no party can influence government if they cannot get elected.Footnote ¹²

In practice, of course, the alternatives in the choice set and dimensions of judgement that can be used to try to estimate correct voting will be limited to the alternatives and dimensions a (survey) researcher has chosen to ask about. While such compromises are inevitable in any shift from the conceptual to the manifest level, this particular compromise does not strike us as unusually troubling. Given what we know about the typical citizen's civic knowledge and political interest, we doubt that most citizens ever consider voting for any but the most prominent parties and candidates who are discussed in the media, and we strongly suspect that the major dimensions of judgement for almost all voters will again be those that have been primed and framed by elite discussion. These are exactly the parties and dimensions of judgement that are most likely to be included in high-quality election surveys, and thus available for analysis. No measure will ever capture every consideration of every voter, but the type of information one gets from the typical high-quality election survey should include the major considerations of most voters.

Before grappling with those more practical issues, we want to be very clear about our purpose. Correct voting is not meant to describe what voters actually do; there is a huge voting behaviour literature that addresses that question. Correct voting is a normative criterion representing what voters ‘ought’ to do, given their own values and relative priorities. We take these – at least for the election immediately at hand – as exogenous and fixed. Our question is: how many voters chose the party or candidate that best represents their own expressed values and priorities? The very essence of democracy is undermined if large numbers of voters fail to do so.

What should predict correct voting?

Democracy works best when citizens are interested in political matters, knowledgeable about constitutional rules and regulations that give responsibility for specific political outcomes to different political actors, attentive to the performance of people and parties in government, and aware of historical and regional comparisons to put those performances in proper context. Such ideal democratic citizens would also be very likely to have the information and perspective to vote correctly. Unfortunately, because of limited cognitive resources, very few people in any democracy have the time and energy to develop this type of knowledge about (and enduring interest in) politics. Home, family, children and work all regularly make more immediate and compelling demands on our time.

Following Payne, Bettman and Johnson and Lau and Redlawsk,Footnote ¹³ we believe that all decision makers are typically guided by two chief motivations: the desire to make a good decision, and the desire to make an easy decision. In mass politics, where the stakes are rarely very high and the probability that one's vote will change the outcome of the election is essentially zero, ease is almost always the dominant motivation. Thus anything that makes the choice easier for cognitively limited decision makers should increase the probability of a correct vote, while anything that makes the choice more difficult should decrease correct voting. Similarly, anything that would make consequences of the election more salient to voters, factors that should increase motivation, should increase the probability of a correct vote. Our ultimate goal is to present a multilevel model that combines institutional- and individual-level predictors into a single analytic framework.

Method

Lau et al.Footnote ³² were able to compare levels of correct voting across recent US presidential elections because the American National Election Study (ANES) survey data includes a large ‘core’ of items that were asked in an identical manner across election years. Although not all items were asked in every election year, enough of them were that Lau et al. did not hesitate to make cross-election comparisons. While high-quality survey data are certainly available from national elections across numerous other advanced democracies around the world, making a case for even conceptual similarity of these questions across different national election studies becomes much more problematic.

Fortunately, the CSES data is designed to overcome this difficulty by establishing a common set of questions that is asked in an identical manner in election surveys cross nationally. We examined every Module 1 and Module 2 survey in the CSES in which, as in presidential elections in the United States, the head of government was being elected. The major trade-off in using the CSES data is that there is a relatively small number of questions asked in its core. While Lau and Redlawsk were able to employ an average of twenty-three criteria from the ANES data to calculate correct voting in the United States, we have far fewer available from the CSES core – essentially only three. Still, there are solid estimates of retrospective performance evaluations,Footnote ³³ policy (ideological) considerationsFootnote ³⁴ and basic historical party-based attachmentsFootnote ³⁵ from the CSES data that we hope will be enough to reliably estimate correct voting. If we had to limit ourselves to only three considerations, it would be these three.

Together, these considerations provide a mix of instrumental and more expressive benefits or ‘utility’ to citizens. Correct voting makes no attempt to distinguish between different types of utility, or to tell people what criteria should be important. If a voter indicates that he or she cares about a given criterion (usually by answering survey questions about it), and that criterion can be objectively applied to all of the different alternatives under consideration, we will include it as a standard for determining a correct vote.

1. (1) Retrospective performance was measured in Module 1 by combining evaluations of the current state of the country's economy and judgements about change in the economy over the past year. For Module 2, these questions were modified to refer to how well the government handled the problem the voter ‘cared the most about’. Favourable evaluations were assumed to impact positively on evaluations of the incumbent candidate or party (and in the case of governing coalitions, to impact positively on all members of that coalition), and to impact negatively on evaluations of all challengers. Negative evaluations were assumed to have the opposite effects.Footnote ³⁶ Thus following Lau and Redlawsk's normative naive procedures, this criterion combines objective or normative components with respondents’ own evaluations and opinions into a criterion for determining correct voting.

2. (2) Respondents placed themselves on an 11-point left-right ideology scale, which we used as a measure of ideological self-identification that, in the absence of more specific questions, we use as a proxy for prospective policy-based considerations. The survey researchers provided expert judgements about where the political parties stand on the left-right scale, and we combined these expert judgements with respondents’ own ideological self-placement to construct ideological closeness scores to each of the major political parties.Footnote ³⁷ Again, this criterion combines both objective (expert judgement) and subjective (self-placement) evaluations.

3. (3) Respondents were asked whether they usually think of themselves as close to any particular political party in their country, and if so to state how close they felt to each party.Footnote ³⁸ Party identifications are usually learned from parents, often reflect basic social divisions within a society and are primarily expressive in nature. This identification is completely subjective – that is, the closeness evaluations come entirely from each individual respondent, and it is easy to match them to the different parties or candidates competing in an election.

These three variables are all standard predictors in the vote choice literature, although some authors see performance evaluations as mostly partisan rationalizations and others think that party identification, in many multiparty systems, is little more than a restatement of the vote choice. If these reservations applied to large numbers of our survey respondents, then multiparty systems would be positively (though artificially) related to correct voting. As will soon be seen, however, the results are just the opposite. Thus while we certainly reject the overly simplistic view of party identification as a permanent ‘unmovable mover’ of political attitudes and behaviours that is never influenced by political events or past behaviour, we nonetheless believe that for most voters in the run-up to an election, party identification is reasonably exogenous and plays both a direct and often an indirect causal role in determining a vote decision.

Each of these three measures was re-scaled to range from 0 to 1. We then combined the party, ideology and performance judgements into summary ‘utility scores’ associated with each of the major political parties or candidates running in the national election.Footnote ³⁹ We do this in two slightly different ways, first treating each of these three criteria equally, then using a weighting scheme that treats each criterion of judgement somewhat differently, depending on both subjective individual and aggregate-level differences between respondents and countries (see Appendix A for details). The weighting scheme, consistent with various ‘heterogeneous’ models of vote choice,Footnote ⁴⁰ assumes that party, economic or ideological considerations should be more important, for some voters or in some electoral or institutional contexts, than others. Notice, however, that before any weighted or unweighted mean utilities are calculated, voters tell us which considerations of judgement are important to them by their answers to the survey questions. If a respondent says they do not feel close to any political party (48.8 per cent), or they do not place themselves on the left-right ideology scale (16.4 per cent), or they cannot make a performance evaluation about the incumbent party (3.0 per cent), that criterion does not factor into determinations of which party is correct. If respondents answer any of those questions, however, the criterion is applied equally to every party or candidate in the choice set. We assume that respondents should have voted for the alternative associated with the highest utility. We then compare our judgement of which candidate is correct with the respondent's reported vote choice to calculate overall rates of correct voting in an election.Footnote ⁴¹

Let us be clear in distinguishing between correct voting and ‘sincere voting’ – that is, voting for the candidate you like best.Footnote ⁴² We use the term ‘utility’ rather than ‘evaluation’ to help keep our meanings distinct. From our perspective, voters could choose the party or candidate they ‘like’ the most and still vote incorrectly. Of course, they might also rationally choose to vote for a less preferred party if they believed that party had a better chance of actually winning a seat. Unfortunately we do not have the survey questions to include such ‘strategic’ concerns in our determination of which candidate is correct. We address the relationship between ‘strategic voting’ and correct voting more fully below.

We have put a great deal of effort into assuring ourselves that these three considerations do a surprisingly good job of faithfully representing the original concept introduced by Lau and Redlawsk. We reserve the details for online Appendix B, but briefly we have located the full national election studies for nineteen of the sixty-nine elections studied here, and operationalized correct voting with the much larger number of potentially relevant considerations (on average, just under twenty) within each individual country. This comparison produces two major results. First, relative to a measure of correct voting based on the much larger number of items available from the full national election studies conducted in these countries, the measure available in the CSES data overestimates levels of correct voting by an average of 10 per cent.Footnote ⁴³ Secondly, the relative rankings of these nineteen elections produced by the two different measures strongly agree with each other (r = 0.77, p < 0.001) and the individual-level variables that predict them are very similar across the two different operationalizations. For the purposes of this article, to identify the individual- and institutional-level predictors of correct voting, the latter finding is far more important. As we discuss in the online appendix, any measure based on twenty considerations has to be more reliable than a measure based on only three considerations, and all else equal we would always prefer the more reliable measure. But all else is not equal. The more reliable measure is only available to us for nineteen elections from nine countries, which greatly limits the breadth of institutional variation that can be invoked to explain cross-national differences in observed levels of correct voting. We could expand the number of elections by adding election years outside those included in the CSES data, but could not greatly expand the number of countries under consideration. We therefore now analyse the CSES data to take advantage of this greater institutional-level variation, while acknowledging that measurement of the dependent variable is less than ideal and that, for this reason, our findings are preliminary.

Online Appendix C provides detailed information on scale construction and data sources for aggregate-level predictors. Because the scale of the individual-level survey variables is arbitrary, we have recoded them to have a 1-point range. However since most of the higher-level variables are familiar and widely available to researchers in the discipline, we therefore keep their original scale (see Online Appendix Table C1). With the standard rule of thumb that one should have at least ten cases per predictor, we should include at most seven higher-order predictors.Footnote ⁴⁴ All analyses presented below conform to these guidelines.

We test these hypotheses with a non-linear multilevel model using the program HLM6,Footnote ⁴⁵ in which the individual-level predictors are all nested within elections. The dependent variable, correct voting, is dichotomous, and our hierarchical model involves a logit link. We employ robust standard errors. We report results from the weighted version of correct voting because it seems more likely to represent the way people actually make such decisions, but repeat all analyses with the unweighted variant as a check on the robustness of the results. There are 72,328 Level 1 survey respondents in our analyses, divided across sixty-nine elections in thirty-three different democracies (Level 2).Footnote ⁴⁶

Before presenting the results we want to address one further issue. Our standard procedure treats all incorrect votes as if they were equally wrong, but with three or more parties in the choice set, it is easy to imagine that one could vote for a party that by our calculations has only slightly less utility than one's ideal choice, or vote for one that has considerably less utility. Somehow we would like to capture this ‘degree of error’ variation, but in practice this proves to be impossible to do convincingly. The basic problem is that ‘utility’ is a joint function of the individual voter's values and beliefs and the objective placement of the available parties on the criteria of judgement. This variation means that even voters in the same election do not face the same difficulty of choice. The correct choice is a unique calculation for every individual voter. Thus while there is always a best choice (the correct vote) and a worst choice for every voter (and with more than two parties, often choices that fall somewhere in between), making the ‘distance’ between best and worst somehow quantifiable, and thereby comparable across respondents, is virtually impossible. We therefore limit consideration to a dichotomous dependent variable that indicates whether voters chose the party or candidate that was best for them. Distinguishing between the egregiousness of the errors made by different incorrect voters is a topic for future research.