Do Voters Want Candidates They Like or Candidates They Agree With?Affect vs. Cognition in Voter Decision Making

Do Voters Want Candidates They Like or Candidates They Agree With? Affect vs. Cognition in Voter Decision Making

David P. Redlawsk University of Iowa Iowa City, Iowa 52242 [email protected] Richard R. Lau Rutgers University New Brunswick, NJ 08903 [email protected]

March 15, 2003

Abstract All candidates try to paint as favorable a picture of themselves as possible, and when voters find a candidate who agrees with them on the issues, is smart, trustworthy, and generally good looking and likeable to boot, the vote choice is quite easy. But what happens when voters are faced with flawed candidates, who are either good looking and likeable, but hold issue positions a voter generally disagrees with, or hold all the right policy stands but look like Richard Nixon with a 4 o-clock shadow, and act like Bob Dole before they invented Viagra? Both conventional wisdom and some previous research would seem to suggest that when push comes to shove, image generally trumps substance. We revisit this question in the context of a mock presidential election campaign where in both the primary and the general election, voters faced a choice between candidates flawed in these two complementary ways. We do not simply assume that voters are simply passive recipients of whatever information candidates would like the voters to see. Instead, a dynamic process tracing methodology allows voters to expose themselves to varying amounts of issue- and image-related information about all candidates. Voters in our experiment generally preferred the dislikeable candidate they agree with on the issues over the much more likeable candidate they disagreed with on the issues. In both the primary and general election campaigns, the type of information voters chose to expose themselves to had a major impact on the vote decision.

Paper prepared for delivery at the annual meeting of the Midwest Political Science Association, Chicago, April 3-6, 2003. An earlier version of this paper was presented at the Shambaugh Conference on Affect and Cognition in Political Action, University of Iowa, Iowa City, IA, March 6-9, 2003

1

Do Voters Want Candidates They Like or Candidates They Agree With? Affect vs. Cognition in Voter Decision Making *

The importance of a candidate’s physical image in presenting him or herself has been understood by candidates and their handlers probably for as long as campaigns have existed. Common phrases such as “clothes make the man [woman]” and “looks can be deceiving” tell us much about how human beings are impacted by the visual (Dion, Berscheid, and Walster, 1972.) McGuiness (1988), in his groundbreaking book about the Nixon campaign, describes how the advertising executives hired by Nixon thought long and hard about how his visual image would come across on television, knowing that they needed to create a “new Nixon” in order to win in 1968. Why an emphasis on the “person” rather than the “issues”? Probably because the advertising executives were convinced it would work in a time when more and more people got their political information from television. Since that time, of course, the rise of candidate-centered campaigning has been well documented (Wattenberg, 1992), as party labels and perhaps issues have become less important and the individual candidate moreso. It may be that in 2000, Vice-president Al Gore was at a distinct disadvantage to Texas Governor George Bush, given the general consensus that although quite knowledgeable on the issues – sort of a “policy wonk” – Gore appeared wooden and often less than likeable, while Bush despite his apparent limited grasp of many issue details, came across as warm and approachable. Certainly in 1980 relatively few voters shared the “Reagan agenda” of a strong policy shift rightward, yet he won that election convincingly, and the next by a landslide. Candidate image and personality clearly play a role, perhaps even the leading role, when voters evaluate candidates (Ottati, 1990; Riggle, Ottati, Wyer, Kuklinski, and Scharz, 1992.) Yet, while campaign consultants know in their gut that image matters, a surprisingly limited amount of political science research has attempted to address directly the question of how substantial a role image plays when compared to other information voters may acquire. That voters respond affectively as well as cognitively towards candidates is not seriously doubted (Granberg and Brown, 1989; Glaser and Salovey, 1998, Redlawsk, 2002.) While there may be some disagreement about the mechanisms of affective response in politics (Marcus and MacKuen, 1993; Marcus, Neuman, and MacKuen. 2000; Lodge and Taber, 2000) there is no doubt that politics is about feeling at least as much as it is about thinking. Voters potentially have a wide variety of information that may be considered when evaluating candidates. Issue positions, group endorsements, the candidate’s visual image, personality, and experience are all clearly important factors in voter decision making (Lau and Redlawsk, 1997; 2001.) If all of these considerations aligned themselve s consistently in some election, the choice would be easy. But we doubt this happens very frequently outside of the movies, where the hero is always smart, trustworthy, holds all the right issue positions, and looks like Robert Redford. More often, a cand idate might have “good” positions on issues, but appear to have limited relevant experience. Another candidate might appear quite attractive physically, but not be supported by the groups with which a voter identifies. Further, some of the information may trigger more thoughtful cognitive processing (issue positions, group endorsements), while other types of information might trigger first a strong emotional or affective reaction. There is reason to believe that visual images and *

Support for data collection and analyses came from National Science Foundation grants SBR-9411162 to the first author and SBR-9321236 to the second author. Redlawsk gratefully acknowledges support from the Obermann Center for Advanced Studies at the University of Iowa.

2

personality traits fall into this latter category (Lanzetta, Sullivan, Masters, and McHugo, 1985; Sullivan and Masters, 1988.) What if voters have to choose between a candidate who is physically attractive and generally likeable but on the wrong side of important issues, and one who takes issue positions more favorable to the voter but who is noticeably less attractive and sporting quite unlikable traits? What then? Do the non-verbal (affective) cues inherent in physical image override the (cognitive) information provided by issue stands? And what if personality traits reinforce the physical image? Candidate attractiveness goes beyond the physical, including not just images, but personality traits that people find more or less endearing. In a series of studies, Rosenberg and McCafferty, (1987; Rosenberg, Bohan, McCafferty, and Harris, 1986) report experimental evidence that a photo showing a person’s physical appearance “communicates a clear and politically relevant image of that person’s character.” (1987, p. 114). Further, they find that controlling for candidate party and issue positions, physical appearance independently influences voter decisions, with attractive candidates more likely to garner support than unattractive ones. These studies focused only on physical appearance, and provided subjects with relatively little information about the candidates’ issues positions, giving only three stances for each candidate. As with most experiments that provide campaign information to subjects, the information was easily processed, and limited; for example, no information was provided about candidate backgrounds or personality traits. Budesheim and DePaola (1994) did provide some personality information as they examined the individual and joint effects of image (defined as both visual image and a limited set of personality traits) and issues on the evaluation of political candidates. Their results are suggestive of the importance of image broadly defined, as they found that issue positions have substantially less impact on evaluations than do the combination of a candidate’s picture and personality. Their focus is clearly on evaluation rather than the vote choice, but still it seems clear that image and attractiveness (defined broadly) are more important than issues when it comes to thinking about candidates. It is important to recognize that while candidate-centered information such as pictures or personality traits may trigger an emotional or affective reaction this does not mean they carry no content for cognitive processing. It is certainly reasonable to think that a candidate’s personality is relevant information for what it signals about the potential behavior of the candidate if elected; a politician’s personality is highly likely to have some influence on how he or she chooses from the myriad of actions available in every situation (Greenstein, 1969). Likewise, while we tend to think of issue positions in terms of how they are processed cognitively by voters, recent research suggests that one cannot simply separate affective reactio ns to candidate positions from the cognitive processing of those issues (Marcus, Neuman, and MacKuen. 2000, Lodge & Taber, 2000; Redlawsk, 2002.) Nonetheless, both conventional wisdom in the media and scientific research into the question has generally suggested that “image” in some way trumps issue agreement, and thus that affective, emotional reactions to candidates are perhaps more significant that cognitive ones. We are not so sure – or at least we see the question as far from settled because the research to date has not really explored what happens to candidate evaluation and choice in the type of environment that voters experience during a typical election campaign. Previous experimental studies have provided only very limited information about cand idates, and in a dry, static format such that subjects have little choice but to process it. They have failed to consider the role a competitive campaign plays over time, when voters are choosing between two or more alternative candidates. And they have certainly given no role to the voter is shaping his or her own information environment. 3

What is needed is an examination of the roles candidate image and issue stands play in both evaluation and choice when voters encounter a reasonable analogue of a political campaign, where a wide range of information can be either processed or ignored about each or several candidates over a period of time. Using a unique computer-based dynamic process tracing methodology (Lau, 1995; Lau and Redlawsk, 1997; 2001; Redlawsk, 2002) that allows us to track information search during a mock election campaign, we can directly test how candidate-centered information such as pictures and personality compares with issue-oriented information in influencing voters’ evaluations and decisions. This paper reports on an experiment in which subjects have to choose between a visually appealing candidate with likeable personality traits but who was relatively distant on important issues, and a less appealing, less likeable candidate who was much closer to the subject on the key issues of the campaign. Moreover, because our procedure allows all subjects a great deal of discretion in the information they choose to learn about any candidate, we can consider whether voters have different proclivities towards seeking out and learning different types of information about candidates, and whether those proclivities help explain the effectiveness of our manipulation. An obvious extension of Lau’s (1986, 1989) research on “chronically accessible political constructs” (see also Bargh, 198x) would suggest that voters should have such differential proclivities, to the extent people actively construct their information environments -- one of the basic assumptions of our research program – although we know of no evidence on this point in the larger chronic accessibility literature. Process Tracing Our experiment begins with process tracing a technique for following decision making as it occurs. While we have described the methodology in detail elsewhere (Lau, 1995; Lau and Redlawsk, 1997, 2001; Redlawsk, 2001, 2002) we will provide an overview of the technique here. Process tracing broadly defined simply is a means to follow the decision making processes in which people engage as they learn about alternatives and choose between them. Where most attempts at process tracing in political science have used a relatively simple static display of information in an easy to use grid format, such a format fails to adequately mimic the chaotic nature of an election campaign, where information comes and goes over time, and may often be overwhelming, at least for those paying attention. 1 We have addressed this problem by designing a dynamic process tracing board that presents election campaigns to subjects who learn about candidates over time as the campaigns progress. There is far more information available than anyone could possibly process, and thus voters must be selective in what types of information they want to consider. The system allows us to set up campaigns with varying numbers of candidates, with varying amounts and types of information, and to manipulate any number of underlying conditions about the candidates or the campaigns themselves. The dynamic process tracing board radically revises the standard technique to better model the campaign environment. Where the traditional information board is an ideal world, the dynamic board creates a complicated environment where information flows over time, coming and going as the campaign progresses. In choosing to examine one piece of information a voter may be forgoing the opportunity to learn something else since new information appears and older information disappears 1

See, for example, Herstein, 1981, whose study of voter decision making using a static information board where subjects were tracked as they selected cards from a board came up with the interesting finding that the partisanship of candidates was not an important factor in an election. Riggle and Johnson (1996) and Huang (2000; Huang and Price, 2001) have used computerized versions of the basic static information board to examine how voters process different kinds of information.

4

from the environment all the time. The information attribute labels "scroll" down the computer screen, rather than remaining fixed in place, and only a limited number are visible at any one time. As with the static board, most attribute labels include a candidate's name and the particular information that would be revealed if the label were accessed (e.g. "Martin's stand on Welfare"). As information behind the label is accessed by clicking on the label with the mouse, the scrolling continues in the background, creating a "cost" in terms of missed information, and thus mimicking the dynamic nature of election information flow. Only a small subset of a very large database of information is available at any one time, and the task of processing campaign information is much less manageable. In addition, the relative likelihood of any particular piece of information becoming available is controlled, so that some information (like party identification) is much easier to get (i.e., appears much more often) than other types of information (such as an obscure policy position). Our mock election campaigns also include 20 second political advertisements, which periodically take over the computer screen (as they do our television screens at home). With this design, the dynamic information board provides a much closer real world analog of a political campaign compared to the static information board. We can use the dynamic system to follow the search process as voters wade through the wide range of issue, candidate personality, group endorsement, and polling information that makes up the typical campaign. While the dynamic board is certainly not a “real” campaign, it mimics well the chaotic information environment that prevails in competitive elections as all candidates do their best to make themselves known and voters try to learn the information they need to make a choice. Procedure Subjects were recruited from the Central New Jersey area in the summer and fall of 1995. A nonprobability sample of 196 subjects participated in the study. All subjects were required to be eligible U.S. voters, although they did not have to actually be registered to vote. Subjects could not currently be attending college. Just over half (51%) of subjects were male, 79.7% were white, and 13.5% Black, with an average age of 40.5 (ranging from 18 to 82.) Subjects were well educated, with 37.0% holding at least a BA degree, though 26.2% had not attended college at all. Finally, subjects were evenly distributed through the range of family incomes from under $15,000 to over $75,000. The study began with the completion of a political attitudes questionnaire about political knowledge, interests, and preferences. After practicing on the computer, subjects participated in a simulated presidential primary election with two Democrats and two Republicans competing for their party’s nomination, followed by a general election campaign with one of the Democrats facing one of the Republicans. 2 The candidates in the elections, while fictitious, were designed to represent a realistic spectrum of ideologies across both major political parties. Before the primary election, subjects “registered” with a political party, and were subsequently constrained to vote only for candidates from that party, although information was available about candidates from both parties during the campaign. After the completion of the primary campaign (which lasted about 22 minutes) subjects voted and then rated all of the candidates on a 101-point feeling thermometer. Next they answered questions about the difficulty of their decision, learned which candidates were running in the general election during a short break, and the began that election (which lasted about 12 minutes). Following the general election campaign, subjects again voted, evaluated candidates, and answered questions about the 2

See Lau and Redlawsk (1997, Fig. 2, 588) for a detailed summary of the typical procedure.

5

difficulty of their decision. An unexpected memory test was then given about the candidates in the general election. Following this, subjects were debriefed, paid, and dismissed. Study Design In this study we use the dynamic board to examine the roles played by candidate pictures, personality, and issue positions in determining the vote. We designed an experiment pitting a relatively unattractive candidate who the voter agreed with on the issues, against a more personally attractive candidate who held many positions the voter disagreed with. We created two such pairings, one in the Democratic primary and one in the Republican primary. Attractiveness was operationalized in terms of both physical appearance and personality traits. For each candidate two pictures were created; one that was rated as relatively attractive in pre-testing and another image of the same person rated as unattractive. 3 These pictures were captured from the campaign ads of actual – but little known – candidates for Congress. Thus both the attractive and unattractive pictures were of the same person. Thus the variation in attractiveness that we manipulate is subtle and easily within the realm of what actually happens in many real campaigns. Some of us are just a little more attractive and a little more charming than others – and we all have our good days and our bad days – but we did not contrast beauty with the beast. Think Richard Nixon with and without his 4 o’clock shadow, rather than Nixon (with the shadow) vs. John Kennedy. The pictures were then paired with a set of personality trait statements, also rated as either relatively positive or relatively negative. The unattractive picture was paired with the more negative trait descriptions, while the attractive picture was paired with more positive trait descriptions. 4 Figure 1 displays the attractive and unattractive pictures of our four candidates, along with the likable and less likable personality descriptions associated with the different pictures. [Insert Figure 1 about here] We then took each candidate and assigned issue positions that were opposite of the candidate’s attractiveness. That is, unattractive candidates were assigned issue positions that were close to those espoused by a subject in response to a pre-experiment political attitudes questionnaire, while attractive candidates were assigned issue positions distant from the subject. But again, the variation in issue agreement was within the plausible range for American politics. We prepared eight different policy positions on every issue, four covering the range of stands typically found within the Democratic party (from the extreme left to slightly right of center), four covering the range of stands typically found within the Republican party (from slightly left of center to the extreme right). Each of our candidates was affiliated with one of the two major parties, and in choosing issue positions that were as close or as far as possible from the voter’s own positions, we selected from among the four which were available for that party. To make the general election choice as difficult as possible, we “fixed” the results of the primary elections so that the candidate the subject voted for in the primary election last the nomination, and the candidate the subject had rejected in the primary was the party’s nominee. The outparty’s candidate in the general election, however, was the same type of candidate the voter had preferred in the primary – 3

In order to establish the attractiveness of both the pictures and the personality traits, we had a separate set of 63 pretest subjects (college students) respond to a series of questions about each picture and/or personality description as they were randomly presented on a computer. The questions included “How attractive is the person?” and “How much would you like this person for a friend?” 4 The trait descriptions included a range of statements about the candidate’s personality, including descriptions from a friend, a staff worker, a political opponent, and a telling anecdote from the candidate’s mother.

6

attractive but relatively distant on the issues, or relatively unattractive but as close as possible on the issues. Thus is voters were to choose the same type of candidate in the general election they had preferred in the primary, they would have to defect form their party affiliation to do so. The result of this process was a clear but difficult cho ice in the primary and again in the general election between two “flawed” candidates. It is important to note here that there was no such thing as “fixed” candidates in this experiment, and in fact it was probably the case that no two subjects ever faced a campaign involving exactly the same candidates – at least if by “candidate” we are referring to more than just their name and appearance. Each of our four candidates had a name and an attractive and unattractive picture and set of personality descriptions, although which candidate was assigned to the attractive and unattractive conditions was randomly assigned within each party. 5 It is also important to note that because our procedure for the most part allows voters to select the information they wished to learn about the candidates, subjects were differentially aware of the tradeoffs between choosing either candidate. Every subject was exposed to two ads from each candidate during the primary, and three ads from each candidate during the general election. The last five seconds of every ad was the candidate’s (attractive or unattractive) picture with a voiceover offering the candidates campaign theme (e.g., “Vote Pat Thomas for a new tomorrow.”) Thus all subjects saw each candidate’s picture at least twice, and this was the minimum level of “treatment” to which all subjects were exposed. However, all voters had many opportunities to choose to examine any candidate’s picture again, to choose to consider brief personality descriptions of the candidates from different people who know them, and to choose to learn about a variety of different issue stands from every candidate. But no voter had to look at any of these things, and very few subjects looked at every personality description of every candidate, or every issue stand that every candidate took. So the “strength” of our manipulation varies across voters. Results The experiment was designed to make subjects have to choose between a candidate who was close to them on the issues by with an unattractive personality and picture and a candidate who was distant on the issues but substantially more attractive. The analyses that follow all focus on this choice. Because of the significance of party in a general election, compared to a primary where party is not a deciding factor, we analyze our two elections independently. For each election we examine the actual vote, the evaluations of the candidates, the difficulty of the decision for the voter, who is likely to vote close versus attractive, and finally, the accuracy of the vote decision. Primary Election We begin by examining the results from our primary elections. Subjects participated in either a Republican or a Democratic primary with two candidates, as described earlier. Because voters are constrained to vote only within their party (though they could see the campaigns for both parties) 5

Or at least that was our intention. However, a computer programming error that we only caught at the very end of our data gathering led to the same Republican candidates being assigned to the attractive and unattractive conditions in virtually every primary election. That the Republican candidates’ names are confounded with condition is probably irrelevant, but our efforts to obtain attractive and unattractive images of the same person went to naught in the Republican primary. One person’s attractive picture was almost always used for the attractive but distant candidate in the Republican primary, while a different person’s unattractive picture was almost always used in the unattractive but close on the issues condition. Fortunately, this bug had no effect on assignment to the various candidate combinations in the general election.

7

partisanship does not complicate these analyses. Table 1 reports the actual vote choice during the primaries, for both Republicans and Democrats. 6 Slightly more subjects voted for the close and unattractive candidate in their primary than for the distant/attractive candidate. At first glance this would seem to suggest that issue positions were generally seen to be more important than “image” – i.e., physical attractiveness and a good personality. Still, a substantial set of subjects chose the more distant candidate. These voters apparently ignored the fact that there was a candidate closer to them on the issues, or at least considered policy agreement to be less important than likableness. One possible explanation for this difference is expertise. Political experts may care more about issues that matter to them and may be more interested in the candidates’ positions on those issues. If so, experts might be more likely to vote for the close candidate, regardless of the personality/image of that candidate. During the pre-election questionnaire, subjects were asked a battery of questions about political interest, knowledge, and activity. These variables were combined into a single measure of political expertise, which was then divided at its median, with political experts coded high and novices coded low. Table 2 reports the vote in the primaries by political expertise, again examining Republicans and Democrats separately. The results are as expected. Experts are more likely to vote for the close/unattractive candidate than are novices (Democrats, X2 =2.723, p<.1; Republicans, X2 =7.996, p<.01.) It appears the political expertise is an important factor in determining whether voters will choose the candidate they are closest to on the issues or the one that has the nicest image. [Insert Tables 1 and 2 about here] In addition to voting for a preferred candidate, subjects in this study also evaluated all the candidates in the election on a 0-100 feeling thermometer scale. An interesting question is the extent to which these evaluations might vary for attractive compared to unattractive candidates. Obviously, voters are likely to rate their preferred candidate much more highly than one they rejected. Accordingly, we look at evaluations separately for chosen and rejected candidates in the primary. Table 3 reports the evaluations, first for all voters, then separately by political expertise. As expected, chosen candidates are liked much more than rejected candidates. But the most striking thing about the table is that there are almost no differences in evaluations between the close/unattractive and the distant/attractive candidates, whether preferred or rejected. Voters who choose the close/unattractive candidate generally rate him at the same level as those who choose the distant/attractive candidate. However, these overall results mask a small difference for political experts, who are somewhat more likely to rate the close candidate higher than the attractive one. There are no differences at all for rejected candidates, regardless of expertise. The pattern of these results suggests that there is no disconnect between the choice a voter makes and the evaluation given that choice. For those who prefer an attractive candidate, that preference – measured as the evaluation of the candidate – is just as strong as for those who prefer one closer on the issues. [Insert Table 3 about here]

6

While the computer bug described earlier meant that virtually all Republican voters faced the same candidate-condition pairing, because we are not concerned about partisanship we feel comfortable in combining both Republicans and Democrats in later primary analyses. Examination of the data for both primaries and the general election do not show any substantial differences between Republican and Democratic voters based solely on partisanship.

8

Despite the fact that candidate evaluations do not vary between the types of candidates, voters do report differences in the difficulty of their decision. As part of the post election procedure, after voting and then evaluating the candidates, subjects were asked how difficult their decision was to make, on a one to five scale with one labeled as extremely easy, and five as extremely difficult. We built an ANOVA model specifying factors that should influence how difficult subjects find their choice, including demographic factors such as age, education, partisanship, and gender, along with a control for the particular primary election (i.e. Democratic or Republican) in which the subject participated. We also included a measure of cognitive processing ability, based on the time subjects needed to read the instructions before the election (standardized), and a measure of how ideologically consistent subjects were in placing themselves on a range of issues (the Barton–Parsons index; see Barton and Parsons, 19xx). The two measures of interest here, political expertise and the chosen candidate (close or attractive) are entered both as main effects and with an interaction. The results of the analysis are presented graphically in Figure 2. No significant effects are found for expertise, or for the interaction of expertise and the type of candidate chosen. But the main effects for the type of candidate are substantial (F=6.552, p<.02.) Voters who choose the candidate closest on the issues rate their choice a significantly easier one than do those who choose the attractive candidate who is distant on issues. Thus while many voters were attracted to the candidate who looked better (and had a better personality) but disagreed with them on the issues, these voters found it much more difficult to actually choose that candidate, compared to voters who went for substance over image [that is, subjects who voted for the candidate they agreed with, even though he was not very likable]. This is an important finding, because it is clear evidence that issue positions in general are more important to voters than image, and that overriding them is difficult. [Insert Figure 2 about here] Perhaps we can get some purchase on just who it is who prefers an attractive candidate over a close candidate. As a first cut, we used a series of demographic variables to predict the likelihood of a vote for the attractive candidate. We built a simple logistic regression model that regresses the vote for an attractive candidate on the age, education, expertise, gender, ideology, and strength of partisanship or our subjects. The first data column of Table 4 presents this analysis. Two factors stand out, expertise and education. The univariate findings about expertise reported above continue to hold in this multivariate analysis. Experts are significantly less likely to vote for the attractive candidate in the primary election (b=-.894, p<.05). In other words, experts generally preferred that candidate they agreed with on the issues. However, controlling for political expertise, general education levels also predict the direction of the vote, with similar effects -- less educated subjects more likely to vote for the attractive candidate (b=-.187, p<.1.) While the overall model is not terribly strong, the results are suggestive that those who are more likely to be interested in and knowledgeable about political issues are also more likely to avoid voting for an attractive candidate with whom they disagree on the issues. [Insert Table 4 about here] A significant strength of our process tracing methodology is that we know what types of information voters examine during the election. No other research has been able to look at the relative importance of different types of information examined as a campaign progresses. This seems particularly important here since it is likely that a tendency to examine different kinds of information might lead to different vote decisions. The second data column of Table 4 reports an analysis in which we add measures of the types of information examined during the primary campaign to the basic 9

demographic analysis. We include measures of the share of pictures, personality items, issues, and ideology statements that our subjects accessed during the primary. 7 We expect that voters who spend more time on issues, and candidate ideology might be oriented towards more cognitive processing of the campaign, and perhaps more inclined to support a candidate based on the issue positions of that candidate. On the other hand, focusing on pictures and the personality of candidates might generate a somewhat more emotional response, and a tendency to be swayed by the attractiveness and personality of a candidate. We can test this hypothesis by inserting into the model our measures for how much of each type of information voters examined. Adding the information search variables does not change the significance of either education or expertise; both remain as before. Two additional demographic variables, conservative ideology and gender become marginally significant, with conservatives and wome n less likely to vote personality and image over issues. 8 More importantly, however, we get strong results from the information search variables, and in the expected directions. Examining a greater share of the available pictures and personality statements leads to a greater likelihood of a vote for the attractive candidate, ceterus paribus, while accessing a greater share of the issues substantially decreases the likelihood of such a vote. Candidate ideology, while not reaching statistical significance, is also in the expected direction. Clearly, the nature of the information that is examined, combined with key individual differences, has substantial impact on the vote. Voters who choose to focus on the more affective material are then more likely to vote for the attractive candidate with the good personality than are those who focus on the more cognitively oriented information. This analysis raises a subtle but important point influencing our interpretation of these results. We suggested above that voters have chronic proclivities to seek out different types of information about candidates, presumably because they find that type of information particularly useful in forming their impressions of candidates. We want to interpret the results of the second equation in Table 4 as illustrating that information search matters, because it determines, at least in part, the vote choice. But if the presumed proclivities to seek out different types of information are strong, then we may have illustrated nothing more than that there are different types of voters, and these voters are attracted to different types of candidates. This would be an important finding as well (see Lau, 1986, 1989), but in that case our information search variables are merely serving as ins truments to represent those different types of voters, rather than standing in for something that is important in its own right. To sort out these different possibilities, we first looked to see if any individual difference variables could predict differential search in our experiment. The most straightforward way to measure such proclivities is to simply ask people what types of information they generally use to make their vote decisions. 9 We asked subjects to think back to the last U.S. presidential election and to indicate how important each of eight different types of information was to making that particular decision. Three of 7

Subjects could also look at endorsements from various interest groups, poll results and other campaign hoopla, and background information specific to the different candidates, for example their age, education, religion, military and political experience, family, and so on. 8 Thus providing some small evidence that the often believed intent of George H.W. Bush in tapping Dan Quayle as his vice presidential running mate that Quayle would appeal to women was perhaps misguided. 9 Such a question, if asked in a pre-experimental questionnaire as ours was, could set up “demand” factors that would lead subjects, wanting to appear consistent, to seek out that same type of information during our mock election campaign, thus artificially inflating the correlation between responses to this question anda actually search. To try to reduce such demand characteristics, we asked subjects to think back to the last presidential election and to report the types of information they found particularly useful in making that particular vote decision, rather than vote decisions in general.

10

those questions (the candidate’s social, economic, and foreign policies) were related to issue stands and one to the general “likableness” of the candidates. We used the last item as an indication of a proclivity to seek out likeability/image information, and formed a scale from the first three items to measure a proclivity to seek out policy-based information. We then combined the first two measures of information search employed in the second equation in Table 4 – seeking out pictures and personality items about the candidates, and combined them into a single measure of candidate- image related search; and combined the last two measures of search employed in Table 4, issue search and candidate ideology, into a single measure of issue-related search. We regressed each of these two summary measures of information search on the rated importance of likeability and policy-based information, along with the same controls used in Table 4 (save for strength of party ID, which is a surrogate for political expertise), plus a count for the total number of items accessed during the election campaign. As seen in Table 5, the results are somewhat mixed, but generally as predicted. The rated importance of policy-based information is positively related to the amount of issue-based search, and negatively related to the amount of image-related search, in both the primary and general election campaigns. The rated importance of candidate likeableness is not as good a predictor, however. It has the wrong sign for the primary election, although not at all significant, but has the expected sign in the general election, just missing conventiona l significance levels (p < .07, onetailed). [Insert Table 5 about here] Thus we have reasonable individual-difference measures predicting differential information search. The question becomes, what happens when we replace the information search variables added to the second equation of Table 4, with our two indicators of a proclivity to seek out different types of information? The short answer is nothing. Whether these items replace the actual search variables, or are added to the equation along with actual search, they are never close to statistically significant, while the actual search measures remain significant. So it is actual information search – and not a proclivity to seek out different types of information, that is helping to determine the vote choice. Before moving on, let us attempt one further refinement to the vote choice model presented in Table 4. We have measures not only of the amount of issue- and image-related information each voter examined, but also the affective implications of that information. That is, while the individual personality items available for access for a candidate in the attractive condition were all positive, they were not all equally positive, nor were the personality items available about the unattractive candidate all equally negative. Similarly, no subject perfectly agreed with every policy stand the unattractive candidate took, nor totally disagreed with every stand the attractive candidate took. In both cases there were degrees of favorableness or unfavorableness that should have been associated with each individual item accessed, a degree that is only approximated by simple indicators of the amount of search itself. The first two columns of Table 6 report the results on an analysis where vote for the attractive candidate in the primary election is regressed on the six control variables from Table 4, two measures of actual information search (those used at dependent variables in Table 5, which each sum across two of the four indicators of search originally used in Table 4, and two indicators of the affective implications of that search. The way we have coded these variables, the two different measures of image information (i.e., total amount of search, and its affective implications) should have positive signs, while the two comparable measures if issue information should have negative signs. As can be seen in the first columns of the table, the affective implications of issue search has important predictive power, above 11

and beyond total amount of search (which also has a very significant effect). The affective implications of the image information does not influence the vote decision above the total amount of image related information examined, however. We have no ready explanation for this difference. [Insert Table 6 about here] Finally we turn to an important “so what” question. To what degree does the focus on candidate attractiveness versus issue congruence result in a high quality decision? That is, do those who vote for the closer candidate end up making a better decision than those who vote for the more attractive one? The initial answer seems obvious: Of course it is “better” to vote for the candidate who is closest on the issues. Presumably that is the rational thing to do, since an election is about determining who will get to carry out policy. By such a measure, we can simply declare that the 46.9% of our primary voters who chose the attractive but distant candidate failed to vote correctly. But this is too simple. After all, if as we suggested earlier, knowing a candidate’s personality traits means knowing something valuable about how that candidate will act in office, it is certainly possible that for some voters choosing the candidate who has the better personality over the one with better issue congruence is the right thing to do. Using our correct voting standard (Lau and Redlawsk, 1997) we assess the extent to which voters make what for them is the right choice. Our measurement of correct voting is guided by the subject’s own interests during the election. Rather than play the role of omnipotent seer determining a priori what is correct and what is not, we allow the nature of the information search to determine what would be correct for each subject. Our standard is actually quite simple. We assert that for each piece of information examined for one candidate during the election, the voter should have examined the same information for the other relevant candidate. Further, if a voter chooses not to examine certain information at all, we take that to mean that particular information was unimportant to the voter, and we do not include it in our assessment. Thus, our measure of a correct vote is a “normative- intuitive” measure, in that we normatively require information that is examined to be examined across the board, but only count that which the subject is actually interested in. 10 For each of our voters, then, we know how they voted and, by the above criteria of “correctness,” we know how they should have voted. The top half of Table 7 presents a simple analysis of correct voting in the primary election, examined separately for experts and novices. Of those who voted for the close unattractive candidate, 81.7% overall voted correctly. There is some difference between experts and novices, as might be expected, with 87.3% of experts making a correct choice, while only 73.2% of novices did so. But the comparison of interest here is between those who voted for the close candidate and those who voted for the attractive candidate. Overall, significantly fewer of those who chose the attractive candidate, by their own standards, voted correctly (67.4%, t=5.353, p<.05) with the difference especially clear for experts. Political experts who chose the attractive candidate were substantially led astray, with only about twothirds as many voting correctly as those who chose the close candidate (t=9.633, p<.01.) But turning to novices, the story is quite difference. Novices who chose the attractive candidate were no less likely to be correct than those who voted for the close candidate. In fact, political novices who voted for the attractive candidate were actually more likely to be making the correct choice than were experts who went for attractiveness over issues. [Insert Table 7 about here]

10

For complete details on how a correct vote is calculated, see Lau and Redlawsk, 1997, Appendix A.

12

General Election Where primary election analyses are relatively simple because there is no need to consider political party, voters in a general election not only had to decide between close/unattractive and distant/attractive candidates, but also whether to vote for the candidate within their own party (the “in party” candidate) or for the out party candidate. Thus our analysis of the general election, while proceeding on parallel with the primary analysis, has this additional factor to consider. Turning first to the general election vote (Table 8) we find very similar results to those of the primary. Overall just over 56% of voters chose the candidate closest on the issues, while 44% chose the more attractive candidate. We see little difference between Republicans and Democrats. But the more interesting question is what voters do when faced with an unattractive yet close in party candidate versus an attractive but distant out party candidate, or even more interestingly, the opposite – a distant/attractive in party candidate and a close/unattractive out party candidate. The bottom half of Table 8 examines these possibilities. When the in party candidate is close on issues, the decision is pretty clear: 76.1% vote for the candidate in their party, while 23.9% defect to the attractive candidate. On the other hand, when the in party candidate is distant on the issues but physically attractive, the decision is a bit tougher. Voters are still more likely to vote for the in party candidate, but a larger share cross party lines to vote for the closer candidate. What we find then, is that when the out party candidate is more attractive and has better personality traits, just under a quarter of all voters defect from their party. But more than a third (38.5%) defect when the out party candidate is closer on the issues than the in party candidate. This suggests that in fact physical attractiveness is less powerful than issue proximity in luring voters away from their own party. In this study Democrats were somewhat more likely to defect for a close out party candidate than were Republicans (46.3% vs. 30.0%) but partisans of both parties defected more for a close candidate than for an attractive one. [Insert Table 8 about here] As with the primary, expertise may play a role in determining who votes which way. Table 9 reports the general election vote by political expertise. Surprisingly, the results do not follow the primary. Instead, as the top part of Table 9 shows, political novices seem somewhat more likely to vote for the close/unattractive candidate, while political experts split their vote evenly between the two types (X2 =2.072, p<.15.) However, these overall results mask important differences when the in party candidate is close versus attractive. The bottom half of Table 9 reports a revised analysis that considers the status of the in party candidate. When the in party candidate is close on issues but unattractive, both experts and novices overwhelmingly vote for their party’s candidate. Likewise, then the in party candidate is distant, but attractive, most still stay within their party, though a larger number defect to the out party candidate who is closer on the issues. The small difference seen between experts and novices in the top of the Table are mostly eliminated in the revised analysis. Unlike the primary, where experts are more likely to vote for the closer candidate, in the general election, we see no significant differences between the two. Novices and experts alike are more likely to move towards a candidate close on issues but who is outside their party than towards an out party candidate who is attractive. [Insert Table 9 about here] The overall findings for the general election vote suggest two things. First, partisanship is stronger than either the closeness of the candidate on issues or the candidate’s attractiveness. Second, to the extent that physical attractiveness and positive personality traits draw voters away from their party, 13

they do so to a much lesser extent than issue agreement between the out party candidate and the voter. Despite the findings of earlier studies, candidate attributes such as appearance and personality do not trump issue agreement. Candidate evaluation in the general election looks very much like it does in the primary. Voters overall rate in party candidates who are close on issues slightly higher than those who are attractive, the opposite of what would be expected if physical attractiveness were more important than issue agreement (close=75.54, attractive=69.66) but the difference does not reach statistical significance. Likewise out party candidates who are close on issues are rated slightly more highly than attractive out party candidates, again the difference is not significant. There are also no significant differences in candidate evaluations based on political expertise. That partisanship is important can be seen in the difficulty of the vote choice. As with the primary, we built an ANOVA model with difficulty of the general election choice as the dependent variable. We included the same set of subject controls, as well as expertise, the type of candidate chosen, and an indicator of whether the vote cast was for an in party or out party candidate. The results are presented in Figure 3. The difference between difficulty for an in party choice (top chart) and an out party choice (bottom chart) is both significant and substantive (F=11.145, p<.001.) Choosing the out party candidate, whether close on issues or attractive, is seen by voters as more difficult, with choosing an attractive out party candidate not close on issues as especially difficult. But the pattern of differences between political experts and novices remains the same whether the vote is for an in party or out party candidate. Surprisingly, for novices a vote for a candidate close on issues is harder than for an attractive candidate more distant on the issues. For experts, the opposite is the case. The interaction effect is statistically significant (F=5.108, p<.05.) For some reason novices in a general election simply find it easier to vote for the attractive candidates, while experts show the pattern we would expect in finding a vote for a candidate more distant on issues to be more difficult. [Insert Figure 3 about here] Table 10 examines the demographic and information factors that might determine who is likely to vote for an attractive, yet distant candidate in a general election. The analysis parallels that of the primary election, with an important addition. To account for the effects of partisanship which are obviously important in the general election, an indicator of whether the in party candidate is attractive or distant is included in the analysis. 11 A positive coefficient on this variable would indicate a party vote, while a negative coefficient would indicate a defection from party to vote for the same “type” of candidate as in the primary. Our goal here is to understand the role played by the demographic and information factors apart from partisanship. The first data column of Table 10 examines just the demographics of the voters. As expected, the nature of the in party candidate is a significant positive predictor, simply indicating a strong tendency to vote for the party’s candidate in the general election, irrespective of whether that candidate was supported during the primary. More interestingly, the only significant demographic 11

A subject who has an attractive in-party candidate in the general election by necessity voted for the unattractive but close on issues candidate in her primary election, since whatever candidate was chosen in the primary “lost” that election and his opponent made it to the general election. This measure is used in the analysis instead of the simple partisanship of the voter because it captures both the effects of partisanship and the nature of the choice facing the voter. Since the dependent variable in the analysis is a vote for the attractive candidate (regardless of party) a positive coefficient on this measure would suggest a tendency to vote for the in party candidate, which is what is expected.

14

variable is the gender of the voter, with women less likely to vote for an attractive candidate, other things being equal, just as we found in the primary election. But unlike in the primary, neither education nor expertise predicts the direction of the vote in the general election. Turning to the second data column of Table 10, even after adding the information search measures, gender is the only significant predictor (other than the control for party.) None of the added information variables is significant, although the signs are in the expected direction. Clearly, the effects of party, which are not present in the primary, change the dynamic of the general election. [Insert Table 10 about here] The last two columns of Table 6 present another cut at this analysis, now considering the affective implications of the issue and image information considered about the general election candidates, along with the total amount of information sought. Gender and party remain significant, but now the affective implications of the image information, and the total amount of issue information, also have significant effects in the equation. Information now clearly matters, above and beyond party – but so does affect. Finally, we turn again to correct voting and back to Table 7, the bottom part of which reports our correct voting analysis for the general election. Overall correct voting in the general election is lower than in the primary, probably due to the more confusing environment once partisanship must be considered. But as with the primary, those voting for the close candidate were more likely to vote correctly than those voting for the attractive candidate, all other things equal. And again we see experts who vote for the attractive candidate doing significantly worse than those voting for the close candidate, while there is no significant difference for novices. Discussion If anything, the nature of our findings provides a more nuanced look at the question of whether image and personality trump issue positions. Earlier research, using limited pools of information in noncampaign experimental environments, showed a strong bias towards candidate physical attractiveness and personality over issues. While there is good psychological evidence that people do find attractive people more appealing, choosing political leaders is a much different task than simply evaluating someone as a friend or co-worker. While voters must undoubtedly evaluate political candidates, they must consider a wide range of factors, including the extent to which the candidate not only has the personality to be president, but also whether the candidate, if elected, will carry out one’s preferred policies. While some previous studies have attempted to provide policy information along with personality and pictures, none has been able to do it in as comprehensive a way as we can using dynamic process tracing. This new study presents voters with the full range of information that would be available in a typical presidential election. And, more importantly, our subjects generally choose the information they wish to learn about the various candidates. This is a crucial difference. Choosing information in an environment where time and other resources are limited means forgoing (other) information as well. Our voters could never learn everything about every candidate, and thus had to be guided by whatever considerations they thought most important. The result? Findings that are substantially different from those reported by others using much more limited studies. We do not find that the affect generated by image and/or personality automatically outweighs the more cognitive processing of issue and ideological information. When voters choose their preferred candidate, they rate their preference equally highly, whether that candidate is the attractive 15

one, who is distant on issues, or the unattractive one close on the issues. If pictures and personality were more important we would expect ratings for the attractive candidates to be substantially higher than for the unattractive ones. That this does not happen is an indication that for those who choose on the basis of issues, personality and image are less important, while for those who are oriented towards image, the issues do not override the more affective considerations. Our work provides important context for this question. We now know that attraction to image and personality versus issue positions varies by individual factors, at least in a primary election where party cannot be a deciding factor. Some voters certainly are attracted to image/personality. At least in the primary election, these tend to be less educated voters with less political expertise. Women, we find, are somewhat less likely to vote for the attractive candidate distant on the issues. And, importantly, the kind of information that voters actively seek out has substantial implications for their vote. Those voters who focus more on pictures and personality statements (and thus less on issues and ideology) are much more likely to vote for the attractive candidate even though he is distant on issues. Is this wrong? That is, should we establish some normative standard that says these particular voters fail to choose the right candidate? We have mixed feelings about this. At one level, the argument that voters should choose candidates more likely to implement the policies they desire seems to comport with basic democratic theory, where representatives should be he ld accountable to citizens’ desires. On the other hand, for political novices in our study, the empirical evidence is that those who vote for the attractive candidate are just as likely to vote correctly as those who vote for the closer candidate. Why? Because the vote is guided by the information that voters examine, and our novices seem to process what they learn in ways that make sense to them. Having said that, it might be argued that they look at the “wrong” information, that these political novices should not place so much importance on image and personality. Yet whether a candidate has the “right” personality to be president may be nearly as important as whether he is likely to pursue certain policies. In fact, one might argue it is more important, since policy success may be based at least in part on a president’s personality traits (Neustadt, 1960; Greenstein, 1969.) It is worth noting, however, that for political experts who tend to focus a greater part of their search on issue oriented information, choosing an attractive candidate over one close on the issues generally means lowering the chances of voting correctly, at least as we have defined it. SO, those experts who were more attracted to the candidate image and personality were perversely more likely to have made the wrong choice based on the information they themselves chose to examine about the candidates. Of course, we provided our subjects with a very clear but difficult choice that probably does not occur very often in “real” presidential politics. Real candidates are often more contradictory than that, so that the choice between image/personality and issue agreement is not usually so stark. In posing the puzzle we do for our voters, we put the affective and the cognitive in direct opposition and find that both play an important role in voter decision making, but that the cognitive generally outweighs the affective. As it turns out, then, and despite the old saying to the contrary, image really isn’t everything.

16

References

BartonA.H., and R.W. Parsons. 1977. “Measuring Belief System Structure.” Public Opinion Quarterly, 41(2): 159-180. Budesheim, T. L. and DePaola, S. J. (1994.) Beauty of the beast? The effects of appearance, personality, and issue information on evaluations of political candidates. Personality and Social Psychology Bulletin, 20(4): 339-348. Dion, K. K., Berscheid, E., and Walster, E. H. (1972.) What is beautiful is good. Journal of Personality and Social Psychology, 24: 285-290. Eagly, A. H., Ashmore, R. D., Makhijani, M. G., and Longo, L. C. (1991.) What is beautiful is good, but…: A meta-analytic review of research on the physical attractiveness stereotype. Psychological Bulletin, 110: 109-128. Glaser, J. and Salovey, P. (1998.) Affect in electoral politics. Personality and Social Psychology Review, 2(3): 156-172. Granberg, D.and Brown, T. A. (1989.) On affect and cognition in politics. Social Psychology Quarterly, 52(3): 171-182. Greenstein, Fred. (1969.) Personality and Politics: Problems of Evidence, Inference, and Conceptualization. Chicago: Markham Publishing Company. Herstein, J.A. (1981.) Keeping the voter's limits in mind: A cognitive process analysis of decision making in voting. Journal of Personality and Social Psychology 40:843 861. Huang, Li-Ning. (2000.) Examining Candidate Information Search Processes: The Impact of Processing Goals and Sophistication. Journal of Communication 59(1): 93-114. Huang, L-Ning and Vincent Price. (2001.) Motivations, Goals, Information Search, and Memory about Political Candidates. Political Psychology 22(4): 665-692. Lanzetta, J. T., Sullivan, D. G., Masters, R. D., and McHugo, G. J. (1985.) Viewers’ emotional and cognitive responses to televised images of political leaders. In S. Kraus and R. Perloff (Eds.), Mass media and political thought. Beverly Hills, CA: Sage. Pp. 85-116. Lau, R. R. and Redlawsk, D. P. (1997.) Voting correctly. American Political Science Review, 91(3): 585-598. Lau, R. R. and Redlawsk, D. P. (2001.) Advantages and disadvantages of cognitive heuristics in political decision making." American Journal of Political Science 45(October): 951-971.

17

Lodge, Milton and Charles Taber. (2000.) Three Steps Toward a Theory of Motivated Political Reasoning. In Arthur Lupia, Matthew McCubbins, and Samuel Popkin (Eds.) Elements of Reason: Cognition, Choice, and the Bounds of Rationality. London: Cambridge University Press. Marcus, George E., and Michael MacKuen. (1993.) Anxiety, Enthusiasm, and the Vote: The Emotional Underpinnings of Learning and Involvement during Presidential Campaigns. American Political Science Review 87(September): 672-685. Marcus, George E., W. Russell Neuman, and Michael MacKuen. (2000.) Affective Intelligence and Political Judgement. Chicago: University of Chicago Press. McGinne ss, Joe. (1988.) The selling of the president. New York: Penguin (Reprint Edition). Neustadt, (1991 [1960].) Presidential Power and the Modern Presidents (Rev. Edition), New York: Free Press. Ottati, V. C. (1990.) Determinants of political judgments: The joint influence of native and heuristic rules of inference. Political Behavior, 12: 159-179. Redlawsk, D. P. (2002.) Hot Cognition or cool consideration? Testing the effects of motivated reasoning on political decision making. Journal of Politics, 64(4): 1021-1044. Riggle, Ellen D.B., and Mitzi M.S. Johnson. (1996.) Age Differences in Political Decision Making: Strategies for Evaluating Political Candidates. Political Behavior 18(June): 99-118. Riggle, E. D., Ottati, V. C., Wyer, R. S., Kuklinski, J., and Schwartz, N. (1992.) Bases of political judgments: The role of stereotypic and nonstereotypic information. Political Behavior, 14: 6787. Rosenberg, S. W., Bohan, L., McCafferty, P., and Harris, K. (1986.) The image and the vote: The effect of candidate presentation on voter preference. American Journal of Political Science, 30(February): 108-127. Rosenberg, S. W. and McCafferty, P. (1987.) The image and the vote: Manipulating voters’ preferences. Public Opinion Quarterly, 51(Spring): 31-47. Sullivan, D. G. and Masters, R. D. (1988.) “Happy Warriors”: Leader’s facial displays, viewers’ emotions, and political support. American Journal of Political Science, 32, 345-368. Wattenberg, M. P. (1992.) The Rise of Candidate-Centered Politics: Presidential Elections of the 1980s. Cambridge, MA: Harvard University Press (Reprint Edition.)

18

Table 1 Voting in a Primary Election Choosing between “Close” and “Attractive” Candidates Democrats

Republicans Overall

Close/Unattractive Candidate

51.9% (54)

54.3% (50)

53.1% (104)

Distant/Attractive Candidate

48.1% (50)

45.7% (42)

46.9% (92)

100.0% (104)

100.0% (92)

100.0% (196)

Voted for:

Total

X2 =.115, 1df, n.s.

19

Table 2 Voting by Political Expertise in Primary Election Choosing between “Close” and “Attractive” Candidates Democrats Experts Novices

Republicans Experts Novices

Close/Unattractive

61.4% (27)

45.0% (27)

66.7% (36)

36.8% (14)

Distant/Attractive

38.6% (17)

55.0% (33)

33.3% (18)

62.2% (24)

100.0% (44)

100.0% (60)

100.0% (54)

100.0% (38)

Voted for:

Total

X2 =2.723, 1df, p<.1

X2 =7.996, 1df, p<.01

20

Table 3 Evaluations of Candidates when Close vs. Attractive Primary Election Close/Unattractive

Distant/Attractive

All Voters Chosen Candidate

82.84 (13.07) n=104

80.22 (15.07) n=91

t=1.29, n.s.

Rejected Candidate

56.91 (19.07) n=102

56.42 (21.96) n=89

t=.17, n.s.

Chosen Candidate

81.83 (15.72) n=41

81.61 (13.24) n=56

t=.08, n. s.

Rejected Candidate

56.67 (20.30) n=54

57.50 (21.27) n=40

t=-.19, n.s.

Chosen Candidate

83.49 (11.09) n=63

78.00 (17.58) n=35

t=1.90, p<.1

Rejected Candidate

57.28 (17.29) n=35

55.73 (22.54) n=62

t=.35, n.s.

Political Novices Only

Political Experts Only

21

Table 4 Vote for Attractive Candidate in the Primary Election B Age in Years

.001 (.009) -.187* (.107) -.894** (.351) -.361 (.325) -.152 (.117) -.011 (.155)

Education Political Expert Female Conservative Strength of Partisanship

Exp(B)

B

1.001 .829 .409 .697 .859 .989

Exp(B)

-.005 (.011) -.204* (.113) -.836** (.365) -.567* (.342) -.222* (.124) -.020 (.162)

.995

.952* (.525) 1.723** (.865) -2.901** (1.276) -.990 (.643)

2.591

.815 .434 .567 .801 .980

Information Examined Pictures Personality Issues Candidate Ideology

Constant X2 -2LL Pseudo R2 Classified

1.761** (.766)

5.817

5.600 .055 .372

3.027*** 20.631 (1.103)

14.870, 6df, p<.05 249.278 .075 60.7%

29.008, 10df, p<.01 235.141 .141 63.4%

*p<.1, **p<.05, ***p<.01 Note: Table entries are logistic regression coefficients, standard errors in parentheses. Dependent variable is a vote for the attractive but distant candidate. n=191.

22

Table 5 Predicting Information Search

Age Education Income Female Conservative Total Amount of Search Rated Importance of Candidate Likableness Issue Positions Constant (1.932)

Primary Election Campaign Image-Related Issue-Related Search Search -.013 (.018) -.046** (.021) -.116 (.107) -.012 (.126) .008 (.009) -.003 (.011) 1.164** (.550) -.578 (.648) .297 (.204) -.327 (.240) .176*** (.012) .254*** (.014)

General Election Campaign Image-Related Issue-Related Search Search .014 (.012) -.064***(.023) -.081 (.080) .118 (.147) -.008 (.007) .011 (.013) .264 (.413) 1.014 (.754) .300** (.153) -.538* (.280) .142***(.012) .327***(.022)

-.185 (.377) -.610*** (.156) -.439 (1.450)

.447 (.282) -.350***(.117) -.542 (1.057)

-.579 (.515) .588***(.214) 2.407

.478

.662

Adjusted R2

-.130 (.445) .393** (.184) .655 (1.710)

.674

.752 *p < .10

Note: N = 179 for all analyses.

23

*p < .05

***p < .01

Table 6 Effect of the Affective Implications of Search on Vote for Attractive Ca ndidate

Primary Election B S.E. -.44 (.34) -.01 (.01) -.20* (.11) -.23* (.13) -.04 (.16) -.71** (.37)

Gender Age Education Conservative Strength of PID Political Expertise Amount of Information Search Image Information 2.24* (1.24) Issue Information -3.55*** (1.26) Affective Implications of Search Image Information .08 (.31) Issue Information -.24** (.12) Attractive/Distant In-Party Candidate Constant 3.23*** (1.09) Model χ2 30.94*** .20 Nagelkerke Pseudo R2 % Correctly Classified 59.7

General Election B S.E. -.70** (.36) -.00 (.01) .05 (.12) .03 (.13) .05 (.17) -.17 (.40) -.17 (1.07) -2.52* (1.48) .58** (.25) -.12 (.11) 1.82*** (.38) -.04 (1.08) 43.76*** .21 69.6

p < .10 p < .05 p < .01 Note: N = 191 for both analyses.

24

Table 7 Correct Voting and Candidate Attractiveness % Voting Correctly Overall Experts

Novices

Primary Election Voted Close

81.7% (85)

87.3% (55)

73.2% (30)

Voted Attractive

67.4% (62)

60.0% (21)

71.9% (41)

t=5.353 p<.05

t=9.633 p<.01

t=.018 n.s.

General Election Voted Close

68.2% (75)

72.0% (36)

65.0% (39)

Voted Attractive

55.8% (48)

52.1% (25)

60.5% (23)

t=3.159 p<.1

t=4.134 p<.05

25

t=.200 n.s.

Table 8 Voting in a General Election Choosing between “Close” and “Attractive” Candidates Democrats

Republicans Overall

Close/Unattractive Candidate

59.6% (62)

52.2% (48)

56.1% (110)

Distant/Attractive Candidate

40.4% (42)

47.8% (44)

43.9% (86)

100.0% (104)

100.0% (92)

100.0% (196)

Voted for:

Total

X2 =1.098, 1df, n.s. ----------------------------------------------------------------------------------------------When In Party Candidate is Close/Unattractive: Voted for Close In Party Candidate

74.0% (37)

78.6% (33)

76.1% (70)

Voted for Attractive Out Party Candidate

26.0% (13)

21.4% (9)

23.9% (22)

When In Party Candidate is Distant/Attractive: Voted for Close Out Party Candidate

46.3% (29)

30.0% (15)

38.5% (54)

Voted for Attractive In Party Candidate

53.7% (25)

70.0% (35)

61.5% (60)

For Democrats, X2 =8.276, 1df, p<.01; for Republicans X2 =21.581, 1df, p<.001

26

Table 9 Voting by Political Expertise in a General Election Choosing between “Close” and “Attractive” Candidates Experts

Novices

Close/Unattractive Candidate

51.0% (50)

61.2% (60)

Distant/Attractive Candidate

49.0% (48)

38.8% (38)

100.0% (98)

100.0% (98)

Voted for:

Total

X2 =2.072, 1df, p<.15 --------------------------------------------------------------------------------------------------------In Party Candidate Close/Unattractive

In Party Candidate Distant/Attractive

Experts

Novices

Experts

Novices

71.4%

78.9%

39.7%

36.6%

Voted for: Close/Unattractive (25) Distant/Attractive

Total

(45)

(25)

(15)

28.6% (10)

21.1% (12)

60.3% (38)

63.4% (26)

100.0% (35)

100.0% (57)

100.0% (63)

100.0% (41)

X2 =.674, 1df, n.s

X2 =.101, 1df, n.s.

27

Table 10 Vote for Attractive Candidate in the General Election B Age in Years

.001 (.009) .008 (.116) -.216 (.381) -.721** (.346) .135 (.127) -.010 (.165)

Education Political Expert Female Conservative Strength of Partisanship

Attractive/Distant In Party Candidate

Exp(B)

B

1.001

-.003 (.011) .031 (.120) -.178 (.388) -.769** (.353) .108 (.129) .024 (.170)

1.008 .805 .486 1.145 .990

1.730*** 5.638 (.342)

Exp(B) .997 1.032 .837 .464 1.114 1.024

1.784*** 5.951 (.351)

Information Examined Pictures Personality Issues Candidate Ideology

Constant X2 -2LL Pseudo R2 Classified

-1.326 (.819)

.266

35.827, 7df, p<.001 225.460 .172 71.6%

.600 (.554) .025 (.966) -1.569 (1.480) -.141 (.481)

1.822

-1.028 (1.109)

.358

1.025 .208 .869

38.835, 11df, p<.001 222.452 .185 71.1%

*p<.1, **p<.05, ***p<.01 Note: Table entries are logistic regression coefficients, standard errors in parentheses. Dependent variable is a vote for the attractive but distant candidate. n=191.

28

Figure 1 Variations of Candidate Likeableness

29

Figure 1 Continued Variations of Candidate Likeableness

30

Figure 2 Difficulty of Primary Election Decision by Nature of Chosen Candidate

2.8

2.7

Difficulty of Decision

2.6

2.5

2.4

2.3

2.2 Novice 2.1

Expert Close/Unattractive

Distant/Attractive

Main effect of type of candidate voted for is significant at F=6.552, p<.02. Interaction effect between expertise and type of candidate is not significant. 31

Figure 3 Difficulty of General Election Decision by Nature of Chosen Candidate When Voting for In Party Candidate 3.8 3.6

Difficulty of Decision

3.4 3.2 3.0 2.8 2.6 2.4 2.2 Novice

2.0 1.8

Expert Close/Unattractive

Distant/Attractive

When Voting for Out Party Candidate 3.8 3.6

Difficulty of Decision

3.4 3.2 3.0 2.8 2.6 2.4 2.2 Novice

2.0 1.8

Expert Close/Unattractive

Distant/Attractive

Interaction effect between expertise and type of candidate voted for is significant at F=5.108, p<.05.

32