Candidate Character Traits and Voting in Presidential Elections

There is general agreement in the voting behavior literature that voter perceptions of the personal qualities of the candidates are important short-term forces that influence voting in presidential elections. The authors of the American Voter found that voters in the 1952 and 1956 presidential elections provided many positive and negative comments about the personal traits of the presidential candidates in those elections (Campbell et al. 1960, 54–8). Other models of voting behavior have similarly viewed perceptions of candidate personal traits as important short-term forces, along with issue orientations and retrospective evaluations of the performance of the incumbent administration (Miller and Shanks 1996). These short-term forces are strongly influenced by party identification, ideology, and general policy dispositions, which are farther up the funnel of causality. A number of studies have analyzed survey data on presidential elections and have found that perceptions of traits affect the vote in presidential elections (Bartels 2002; Doherty and Gimple 1997; Funk 1999; Gant 1983; Goren 2002; Hayes 2005; Keeter 1987; Kenney and Rice 1988; Kinder 1986; Marcus 1982; Miller, Wattenberg, and Malanchuk 1986; Sullivan et al. 1990). These studies have used data from different years, have employed different measures of trait assessment, and have relied on different analysis methods, yet they come to the same general conclusion. Moreover, these studies share some basic theoretical perspectives that guide this study.

A number of specific traits have been identified in the existing research as important in presidential elections. The ones most commonly cited in the existing research are: (a) integrity, which refers to qualities such as honesty, morality, and trustworthiness; (b) leadership, which refers to being a strong leader, decisive, and inspiring; (c) competence, which refers to experience, knowledge, and intelligence; and (d) empathy, which refers to compassion and caring about people (Funk 1999; Goren 2002; Kinder 1986; McCann 1990; Miller and Shanks 1996). Some scholars identify only three of the above four dimensions (Conover 1981; Nimmo and Savage 1976). Others feel that personal traits can be defined by just two dimensions, one that combines leadership and competence and another that combines integrity and empathy (Funk 1996, Marcus 1982). Despite these differences, there is a general consensus that leadership, competence, integrity, and empathy cover the domain of relevant trait assessments, and this study employs that theoretical framework.

Measuring Presidential Character Traits in 2004

This study uses the closed items on candidate traits in the 2004 American National Election Study (ANES) survey. There were seven pairs of items in the survey. Each item asked the respondent whether he or she felt that a certain trait characterized a candidate extremely well, quite well, not very well, or not at all. Although most researchers have treated this set of possible responses as a four-point scale, there is reason to question that approach. The difference between saying that a trait characterizes an individual “extremely well” versus “quite well” seems theoretically small. Many people might use the two terms almost interchangeably. Furthermore, while “not very well” is not as negative as “not at all,” the two terms seem close to each other, and both seem far from “quite well” or “extremely well.” Some empirical investigation of these items supports the above theoretical claim. If we examine the presidential vote for each response category for each item, holding party identification constant, we find that there generally are only small differences in candidate choice between the “extremely well” and “quite well” groups and between the “not very well” and the “not at all” groups, but the difference between the “quite well” and “not very well” groups generally is relatively large. Furthermore, if we examine how party identification relates to these responses, we find an interesting pattern. Individuals who judge the candidate of their party negatively on a trait are much more likely to select “not very well” than “not at all.” Similarly, those who judge the candidate of the opposing party positively on a trait are more likely to select “quite well” than “extremely well.” For the above reasons, this study treats responses to each trait item as dichotomous (extremely or quite well versus not very well or not at all).

Table 1 contains responses to the seven trait item pairs.¹

As we can see, Bush was judged more positively on the two leadership items, where he had a 15 to 20 percentage point lead over Kerry. Kerry came out better on the two competence items, where he had a 20 to 25 point lead over Bush, and on the empathy item, where he had a 10 point lead. Neither candidate had a significant advantage on integrity.

Five of the seven trait item pairs asked in 2004 were worded in a positive direction, meaning that respondents were asked in each case whether a candidate possessed a desirable trait. Two of the item pairs were worded in the negative; respondents were asked whether the candidates were dishonest and indecisive, not whether they were honest and decisive. This appears to have introduced some error into the responses, a fact which is suggested by the interrelationships among the items. For both candidates, being a strong leader and being decisive (specifically, not being indecisive) correlate less than we would expect, given that these are supposed to be highly related aspects of a more general leadership dimension.²

The correlation is particularly weak for Bush, but even for Kerry, the “strong leader” item correlates more strongly with being compassionate and with being moral than it does with being decisive, certainly not what we would expect. A very similar pattern exists for being moral and being honest (specifically, not being dishonest). These two items do not correlate as strongly as we would expect, and for both candidates being honest correlates more strongly with being compassionate than with being moral.

A plausible explanation of these patterns is that some random error was introduced by wording the questions in a negative fashion. Some respondents may have become confused by the negative wording, which led them to provide a response opposite of what they intended. It appears that the error is more likely to be random than systematic because the mean scores for the two items in question fit our prior expectations. Bush is substantially ahead on being decisive, which corresponds to his advantage on being a strong leader. On being honest, the two candidates are judged about even overall, which matches overall assessments of their morality. Thus, it seems that there were errors in both directions, leaving mean scores as representative of true overall sentiment but weakening the correlations between the negatively worded items and other variables.

In order to measure respondent assessments of candidate traits, an index was formed using all seven trait items.³

The index is the sum of responses favorable to the Republican candidate (i.e., positive assessments of Bush and negative assessments of Kerry) minus the sum of responses favorable to the Democratic candidate. Thus, a respondent who judged Bush positively on each of the seven traits and who judged Kerry negatively on each of the traits would receive a score of 14. A respondent with exactly the opposite set of responses would have a score of −14, and a respondent with an evenly balanced set of responses would have a score of 0. The index combines responses for both candidates on the assumption that what influences the vote decision is the relative standing of the two candidates in the voter's mind—i.e., to what extent does the voter perceive one candidate to be superior to the other in personal traits?

Table 2 contains the distribution of responses to the candidate trait index. The most significant aspect of this distribution is the very even balance of scores. About 47% of the respondents had a net perception that was favorable to Bush, whereas about 48% were more favorable on balance to Kerry.⁴

The story in the popular media may have been that voters perceived Bush as superior to Kerry in personal terms, but these data do not support such a conclusion. Bush had a clear advantage on leadership, but Kerry was stronger on other trait aspects, and the overall assessments were very even for the two candidates.

The Effect of Candidate Trait Perceptions on the Vote

In order to analyze the effect that perceptions of candidate traits had on the vote in 2004, a logistic regression of the presidential vote was used, with a number of predictor variables besides the candidate trait index:

(a) Party identification and ideology were included as fairly stable orientations that influence a range of attitudes.⁵

(b) Three measures of issue orientations were included in the model. An index of economic and social welfare issues combined responses to several highly related questions regarding the role of the government in the economy.⁶

An index of moral issues combined responses to questions on abortion and gay marriage, two items that are strongly correlated.⁷ An index of defense issues combined responses to questions about the appropriate level of defense spending and about the use of force versus diplomacy in international relations; these two items also were highly correlated.⁸ There are a number of policy issue items that are not included in the three issue indices (e.g., items dealing with capital punishment, gun control, school vouchers, and deficit reduction). These items did not cluster in a fashion that allowed for the construction of an index. However, the number of issues that are included, plus the fact that ideology also is included, suggests that the omission of some issues should not present a problem for this analysis, as the concern here is not with the effect of any specific issue on the vote.

(c) Two measures of approval of the performance of the Bush administration were included. One measured approval of Bush's handling of the economy.⁹

The other measured approval of Bush's handling of national security by combining responses to three related questions that asked about Bush's handing of foreign affairs, the war on terror, and the war in Iraq.¹⁰ The measure of approval of Bush's handling of national security undoubtedly combines both performance appraisals and policy orientations more so than does the item dealing with Bush's handling of the economy, but for the purpose of this analysis it is not important to disentangle those components.

The conceptual framework of this study assumes that party identification, ideology, issue orientations, and performance evaluations are causally prior to assessments of candidate traits, an approach that will yield conservative estimates of the effect of candidate traits on the vote. This conceptual framework is similar to that employed by Miller and Shanks (1996), although their framework differs somewhat in the specific variables that are included.

The results of this analysis are presented in Table 3. Model 1 uses all of the variables discussed above. Three of these variables (ideology, orientations on moral issues, and orientations on defense issues) turn out to be statistically insignificant in the model. Of course, while ideology has no direct effect on the vote, it has indirect effects through its effects on other variables in the model. Similarly, attitudes on defense issues have no direct effect, but they are related to approval of Bush's handling of national security. The variables that have a significant direct impact indicate that the vote was influenced by party identification, issue orientations (particularly those involving economic issues and those related to the war in Iraq), performance evaluations (regarding both the economy and foreign policy), and candidate traits. Furthermore, the direct effect of candidate traits is substantial, even with all of the controls. Model 2 uses only the variables that were statistically significant in Model 1; the results are very similar, but the coefficients change slightly. For both models, the coefficient for the candidate trait index is very strong relative to its standard error. In fact, the trait index has the highest Wald statistic of all the variables in the model.

Interpreting the coefficients in a logistic regression is easier if we translate them into probabilities of voting given certain conditions. Table 4 estimates the probability of voting for Kerry, given a certain score on the candidate trait index, for three different groups. Average Democrats are defined as having scores on the other variables in the model that are about equal to the mean scores for all Democrats on these variables. For example, the average Democrat is a weak Democrat who is moderately left of center on economic issues (a score of 3.3 on a 7 point scale), and about halfway between disapproving and strongly disapproving of Bush's handling of the economy and of national security. Average independents and average Republicans are similarly defined as holding mean scores for their partisan group on the model variables.

As we can see, changes in the candidate trait index have a substantial effect on the probability of voting for Kerry among independents. An average independent who judged the candidates to be exactly even overall in personal traits had a 53% likelihood of voting for Kerry, whereas one who had an index score of 3 (which would be only slightly favorable to Bush) had just a 36% likelihood of voting for Kerry. Partisans were less influenced by their perceptions of the personal traits of the candidates unless they were substantially favorable to the candidate of the opposite party. For example, the average Democrat was still highly likely to vote for Kerry even if he or she rated Kerry as no better (or worse) than Bush on personal traits. Of course, the average Democrat disapproved of Bush's handling of the economy and national security, which provided sufficient reasons to vote for Kerry. A similar pattern exists for Republicans. Significant defection rates do not appear until the trait index is considerably favorable to the candidate of the opposite party, something that did not occur with great frequency. When partisans did defect in their presidential vote, it appears that unfavorable net assessments of the personal traits of their party's candidate played a substantial role in that decision, but defection rates among both Democrats and Republicans in 2004 were relatively low overall.

Having established that perceptions of candidates' personal traits were important in influencing the vote in 2004, we may now ask which particular traits were most important. This is an especially interesting question because the advantage of Bush or Kerry varied across traits. If leadership traits were foremost in the minds of voters, that would advantage Bush, for example. In order to investigate this topic, a logistic regression of the vote was run, using the same variables in Model 2 of Table 3, except that the candidate trait index is replaced with four specific trait variables. The four trait variables measure the relative advantage of Bush versus Kerry on the following traits: (a) providing strong leadership; (b) being knowledgeable; (c) caring about people like the voter (referred to as compassionate); and (d) being a moral person.

These specific traits were chosen because they tap the four theoretical dimensions identified earlier (i.e., leadership, competence, empathy, and integrity) and because each is worded in a positive direction, thereby avoiding the possible problems with negatively worded items that have been discussed previously. Thus, the item asking whether the candidate was a strong leader, not the item asking about decisiveness, is used to measure leadership. Since Kerry's alleged indecisiveness was a central aspect of the campaign, it might have been desirable to use the item that focused on that trait, but the negative wording of the item makes it a weaker measure of the leadership dimension.¹¹

The results of the analysis are presented in Table 5. The first column gives the results for all voters. The next two columns give the results for Democrats and Republicans, respectively. Examining each partisan group separately is desirable because some scholars have found that Democrats and Republicans give different weights to different candidate traits (Goren 2002; Hayes 2005). Among all voters, the most important trait is compassion, followed by being knowledgeable and moral. Surprisingly, leadership has the lowest coefficient of all of the four traits, and the coefficient is not statistically significant. Popular accounts may stress that many voters chose Bush over Kerry because of doubts about Kerry's leadership or confidence in Bush's leadership, but these data do not indicate that variations in the relative assessment of Bush and Kerry as strong leaders had much effect on the vote once other factors are taken into account.¹²

Also surprising is the strength of the relationship between the vote and assessments of the relative compassion of the two candidates. Most researchers have not found this trait to be the most important one, although there are some exceptions (Bartels 2002; Funk 1996; Funk 1999).

There are clear differences between Democrats and Republicans in their use of candidate traits in the vote decision. Among Democrats, competence and empathy played a strong role. Democrats who voted for Bush came disproportionately from those who did not see Kerry as substantially superior to Bush on being knowledgeable or on caring about people. Variations in the relative assessment of the leadership abilities of the two candidates were far less related to candidate choice among Democratic voters. Among Republicans, leadership, empathy, and possibly morality had a strong impact on the vote (the coefficient for perceived morality is statistically insignificant, but it has a fairly strong absolute value). Republicans who failed to vote for Bush came disproportionately from those who did not give Bush a large advantage on the leadership dimension. This is consistent with the findings of other scholars, who argue that Republicans tend to put more weight on leadership qualities (Goren 2002; Hayes 2005). Surprisingly, the coefficient for the variable that measures how much the candidates care about people is twice as large for Republicans as it is for Democrats. The strong impact that assessments about the relative compassion of the candidates have on the vote among Republicans is not what we expect. Democrats are supposed to be more responsive to this consideration (Goren 2002; Hayes 2005).

This analysis of the data from the 2004 ANES survey does not support the conclusion that Bush won the presidential election because of a decided advantage on character traits. Voter perceptions of candidate personal qualities did have a substantial influence on the vote. The more favorable that the voter's net evaluation of candidate personal traits was to Bush, the more likely the voter was to cast a ballot for Bush. However, these assessments were not the source of Bush's victory, in the sense that he had a sizable net advantage on this factor. Overall assessments of the two candidates were extremely even. Bush had an advantage on leadership, but Kerry had an advantage on competency and compassion. Moreover, the trait that Bush did best on, leadership, was not the most important one in the vote decision. While leadership was more important to Republicans than to Democrats, even among Republicans, assessments of empathy were more potent influences on candidate choice than were assessments of leadership.

Some Caveats

Two caveats to these conclusions should be stated. First of all, the assessments of character traits were measured in the pre-election ANES interviews, which were conducted in September and October. Voter assessments of these traits could have changed by Election Day, so perhaps Bush had a sizable advantage on this dimension when voters actually cast their ballots. However, this seems unlikely. The Swift Boat ads and the Republican convention occurred before the pre-election interviews, so most of the charges about Kerry's lack of presidential character had already been aired by the time that interviewing began for the 2004 ANES. It does not appear that negative assessments of Kerry's character grew during September and October. Public opinion polls showed that Kerry was further behind Bush in mid-September trial heats than he was in late-October trial heats. An average of four major polls showed Bush ahead by six points in mid-September and by just two points in late October.¹³

If anything, there was some slight movement toward Kerry between Labor Day and Election Day, so it is unlikely that perceptions of his personal qualities, relative to those of Bush, deteriorated during the same time period. Furthermore, public opinion polls indicate that Kerry did considerably better than Bush in the first debate and as well in the next two debates (Abramson, Aldrich, and Rhode 2006, 40–2; Conley 2005, 126–7; Thomas 2004, 147–64). The debates were the most important events during October that might have altered perceptions of the personal traits of the candidates. This suggests that respondents who were interviewed in September would have been unlikely to have lowered their assessment of Kerry relative to Bush by early November.

Further evidence that these data do not misrepresent the situation on Election Day can be provided. If we divide the ANES respondents into September and October interviewees, the latter group is no less favorable toward Kerry in its assessment of the personal qualities of the two candidates.¹⁴

Specifically, the September interviewees had a mean trait index score of −0.21 and the October interviewees had a score of −0.29 (slightly more favorable to Kerry but statistically not different from the September score). We noticed in Table 4 that independents were much more influenced in their voting by their assessments of candidate traits, so it might be interesting to see if this group drifted toward Bush during the campaign. This does not appear to be the case. The mean trait index score for all independents interviewed in September was −1.82, and for those interviewed in October it was −1.85; both scores are slightly favorable to Kerry.¹⁵

The results of a late October Gallup poll that contained a few items dealing with candidates' personal traits support the validity of the ANES data.¹⁶

The wording of the items in the Gallup poll differed from the wording employed in the ANES survey, so the results are not precisely comparable. In the Gallup poll, respondents were asked whether a specific trait applied more to Bush or to Kerry; respondents were not given the option of saying that the trait applied about equally to both or that neither candidate possessed the trait (although a few respondents nevertheless volunteered such responses). The Gallup poll found that Bush had a sizable advantage on leadership (57 to 38%) and that Kerry was seen as more intelligent (49 to 35%) and as better when it came to caring about people (51 to 43%), conclusions that basically agree with the findings of this study. Furthermore, the Gallup poll also asked whether each candidate had “the personality and leadership qualities that a president should have.” About 58% of the respondents said that Bush had such qualities, and about 55% felt that Kerry possessed such qualities, a small and statistically insignificant difference.

A second caveat is that these data contain no measure of general warmth or likableness. If, as some pundits claimed, voters viewed Bush as the more likeable person, the one with whom they would prefer to have a beer (or, in Bush's case, a soft drink), that might have provided Bush with an advantage in personal traits. While this is a possibility, it would seem that perceptions of candidate warmth would correlate highly with perceptions of some other traits, most likely compassion and trust. If a voter viewed one candidate as more likable than the other, it would seem odd if that voter felt that the more likable candidate was less trustworthy or less compassionate. People usually do not like individuals whom they see as untrustworthy or self-centered. Since Kerry was viewed as equal to Bush in trust and superior in compassion, any warmth advantage for Bush was probably small. Furthermore, Kinder et al. (1980) found that voters regarded candidate warmth as a relatively unimportant aspect of the ideal presidential candidate. Finally, the Gallup poll data cited above indicated that Bush had at best a slight overall advantage on personal traits. If warmth (or some other unmeasured trait) was both important to voters and one on which Bush had a decided advantage, then the Gallup respondents should have been more favorable in their overall assessment of the personal qualities of Bush relative to Kerry.