Do nonpartisan ballots racialize candidate evaluations? Evidence from “Who Said What?” experiments

Abstract

At last count, U.S. voters were responsible for directly electing more than 510,000 public officials. Few of these contests feature lively campaigns or attract substantial media attention, often leaving the average voter to make decisions with limited information. We argue that the cognitive strategies voters use to make decisions in these low-information contests depend in part on the informational cues printed on their ballot—in particular, the presence or absence of partisan labels. Using two “Who Said What?” experiments, we show that voters engage in social categorization—and do so on the basis of race and ethnicity when candidates differ in their demographic background. We also find, however, that the availability of party labels shapes the degree to which voters categorize candidates based on their race and ethnicity. A central implication of our results is that efforts to increase minority representation should look beyond electoral institutions—such as district versus at-large elections—to the information printed on the ballot itself.

Keywords

nonpartisan elections party labels urban politics

The unusual election of Steve Rocco to the board of the Orange Unified School District in Southern California attracted national attention in the fall of 2004. Aside from his initial candidacy form, Rocco filed no campaign paperwork and appeared at no candidate forum or public event.¹ If voters hoped to find additional information about the mysterious candidate, they had only two options: The first was Rocco’s conspiracy-theory-laden website, https://andykaufmanlives.com . The second was a 1992 self-published autobiography, R.O.C.C.O. Behind the Orange Curtain, described as the “secret chronicles and public-record accounts of corruption, murder and scandal of corporate and political California, written by America’s premier legal technician.”²

While such oddities would have likely doomed most political candidates, Rocco won the contest by a comfortable eight-point margin on Election Day. His victory prompted the local PBS station to produce a documentary titled Rocco the Vote, “a cautionary tale about the need for informed participation in local elections.” Local observers pointed to two plausible explanations for Rocco’s improbable victory. In his candidacy form and on the ballot, Rocco described himself as a “teacher/writer” while his opponent’s listed occupation was “park ranger,” perhaps making Rocco the more appealing choice to fill an office dealing with education policy (see McDermott, 2005). The second, and perhaps most troubling, explanation was that his opponent’s Hispanic name—Phil Martinez—may have proved to be a major political liability in conservative Orange County.

Rocco’s election to the Orange Unified School District’s Board of Trustees is certainly not the only contest in which voter beliefs and assumptions about the candidates’ race and ethnicity may have played an important—and perhaps decisive—role. Indeed, questions of racial prejudice and discrimination remain at the heart of the study of American politics generally and voter behavior in particular. Existing research, however, has produced mixed findings about the degree to which minority candidates face systematic disadvantages at the ballot box.

Our analysis makes two contributions to this important line of research. First, we document the psychological processes through which racialized considerations enter the political calculus of voters. Our analysis examines how voters come to form their initial impressions of candidates. As we argue below, race may play a critical role during this impression-formation stage, even if racial effects cannot be readily detected in subsequent vote choices. The reason is that individual voters are likely to draw different inferences from a candidate’s demographic background and respond in varied ways to the same racial cues. Such heterogeneous treatment effects may offset one another and make it difficult to detect the net impacts of candidates’ race and ethnicity on aggregate vote choices, but do not mean that such factors play no role in individual candidate evaluations. By adapting an experimental procedure developed for the study of social cognition, we can examine the impression-formation process directly and measure the importance of racial and ethnic considerations at this initial stage. Unlike existing experiments that randomly manipulate hypothetical candidates’ physical descriptions or attributes and then examine the impact of these manipulations on subjects’ reported vote choices, our procedure is unobtrusive and minimizes the risk of social desirability response bias, a serious threat in the existing experimental work on the subject of candidate race and ethnicity.

Second, we examine how the informational cues printed on the ballot affect the extent to which voters rely on racial and ethnic cues when evaluating candidates running for low-profile offices such as a local school board. Building on Kam (2007), we show that voters are likely to place greater weight on racial and ethnic stereotypes in nonpartisan compared to partisan elections and our experiment allows us to document the underlying psychological processes responsible.

Although existing research is inconclusive about the extent to which minorities face systematic disadvantages in American elections, it does provide consistent evidence that voters often use candidate’s demographic background to make inferences about their competence, personalities, and issue positions (see, e.g. Huddy and Terkildsen, 1993; Jacobsmeier, 2015; McDermott, 1997; Moskowitz and Stroh, 1994; Sanbonmatsu, 2002; Sigelman et al., 1995)—all factors that can influence voting behavior indirectly.³ These studies do not, however, tell us whether such judgments reflect affective attitudes—with voters rating members of certain groups as less competent or more ideologically distant simply because they dislike them (Brady and Sniderman, 1985)—or an attempt to use “demographic cues” to infer, albeit imperfectly, candidates’ relevant but unknown political attributes.

Empirically disentangling these mechanisms is important because, although voter attitudes and prejudices toward social groups may be deep-rooted and fixed in the short term, relatively simple-to-implement changes in the information context can dramatically alter the kinds of cues voters consider when weighing competing candidates. After providing evidence that race-based evaluations reflect, at least in part, demographic cue-taking, we show that printing alternative cues on the ballot can moderate the extent to which voters rely on candidates’ race and ethnicity to form their impressions of candidates in low-profile elections. Our experimental findings show that the presence of partisan labels substantially reduces the importance of race and ethnicity in how voters initially categorize political candidates. These results suggest that candidates’ personal demographics loom larger in the minds of voters in nonpartisan elections compared with contests that utilize partisan ballots and thus have important implications for ongoing efforts to reach parity in representation for historically disadvantaged groups. Such efforts that have hitherto focused primarily on electoral institutions such as district versus at-large elections and the drawing of majority-minority districts but not on the design of the ballot itself.

Category-based impression-formation: An overview

While most studies of American political behavior focus on presidential and congressional elections, these contests represent just a small fraction of what a typical voter sees on the ballot. In 1992, the last time an exhaustive tally was completed, the Census Bureau counted more than 510,000 elected officials in America—or nearly 1 for every 500 people living in the country at the time (Census of Governments, 1995). Although some contests—including many statewide offices and mayoral elections in major cities—feature lively campaigns, result in substantial candidate and outside spending, and attract major media attention, the vast majority do not. This is one reason why some voters fail to submit complete ballots, abstaining from down-ballot races after casting their vote in the most salient contests (e.g., Bullock and Dunn, 1996). Many voters who show up on Election Day do, however, cast nearly complete ballots in typical elections, even when the ballots include contests about which they have only passing familiarity.

How do voters choose among competing candidates in these low-profile elections? We argue that they do so by relying on basic rules of thumb, the same rules of thumb individuals use to form impressions when they meet strangers. Social psychologists have studied this process extensively, and we build directly on the dual-process model of impression-formation developed in this literature (see, e.g. Fiske and Neuberg, 1990). The key lesson from this research is that people choose from a small menu of clearly defined strategies to decipher useful information or make inferences about new associates. At one end of the spectrum are strategies, often described as “System 1” thinking (for an overview, see Kahneman, 2011), that require minimal effort but provide only imprecise information. Social categorization, in which people sort new individuals into the most accessible social category cued by their appearance, speech, or other attributes, represents the primary strategy at this end of the spectrum. Having sorted strangers into categories, individuals draw on their knowledge about and attitudes toward the representative member of these groups and apply them to the newcomers. According to Fiske and Neuberg (1990), “Perceivers initially categorize others immediately upon encountering information for cuing a meaningful social category.…Once the category is cued, category-relevant cognitions, affect, and behavioral tendencies become accessible, although the perceiver will not necessarily act upon them. This initial categorization stage occurs immediately upon encountering any target individual, is extremely rapid, and is essentially perceptual” (p. 4).

While social categorization provides the initial basis for impression-formation, it is not always the endpoint. People may choose to seek out additional information to ascertain individual-specific attributes, updating their beliefs (System 2 thinking). But research in psychology clearly shows that how far individuals go beyond social categorization depends on their motivation and their willingness to engage in this costlier, but also more individuating, type of reasoning. Whether people move beyond social categorization therefore depends in large part on what decisions they must make and the perceived importance of the outcomes that depend on these decisions.

Particularly in low-profile elections for which the stakes are often uncertain and impacts remote, our expectation is that voters rarely proceed beyond social categorization as the basis for candidate evaluations for three reasons. First, in the absence of extensive campaigns and substantial media coverage, the cost of learning about candidate-specific attributes in local contests is relatively high. As Downs (1957) notes, when finding information is both costly and the perceived the impact of the decision low, voters will often refrain from carrying out extensive research. Indeed, one of the most consistent findings in public opinion research is that most voters do not know or cannot recall even basic facts about national institutions—such as the name of the chief justice or which party has the majority in the House of Representatives (e.g., Delli Carpini and Keeter, 1996). Accordingly, we do not expect voters to invest substantial effort researching the candidates for the local animal control commission.

Second, a voter who engages in this kind of research is unlikely to affect relevant outcomes, since a single voter has only a very small probability of being decisive in any given contest. Third, despite the fact that local quality-of-life issues often have a big impact on individual voters, the difference in turnout between national and local elections (e.g. Hajnal and Lewis, 2003) suggests that voters find local elections less interesting and likely use their already limited political attention to focus on higher profile contests.

For these reasons, we expect that social categorization represents a dominant mode of impression-formation in many low-salience, low-information elections. This does not mean, however, that all voters sort candidates into the same categories. Individuals make use of categories that are most accessible in their minds, which in turn depends on the political context. One important dimension of that context is what is printed on ballot that voters see. In many contests, political parties represent the most salient and accessible categories (Conover and Feldman, 1989; Rahn, 1993), especially when candidates’ partisan affiliations appear next to their names right on the ballot or voting machine screen. Because 80% of local elections in the U.S. are nonpartisan (Schaffner et al., 2001), however, voters in these contests may turn to other candidate attributes, including their gender (McDermott, 1997; Sanbonmatsu, 2002), occupation in states where this information is printed on the ballot (McDermott, 2005), and race and ethnicity (e.g., McDermott, 1998). In the case of the 2004 Orange Unified election, it is possible that the district’s use of nonpartisan ballots may have contributed to Steve Rocco’s victory by causing voters to focus on the candidates’ ethnic surnames and rely on ethnicity as the basis for their categorization.

Of course, social categorization may not always disadvantage minority candidates. How initial impressions map onto actual voter choices depends on several factors, including the racial stereotypes voters hold and which personal attributes voters believe are most relevant to the office at stake. For example, many people view Asian-Americans as a “model minority,” and rate this group above whites on some dimensions. Such stereotypes may result in an electoral boost for Asian candidates when voters categorize candidates based on their race. Similarly, Sigelman et al. (1995) show that voters view African-American candidates as simultaneously less competent and more compassionate. The net electoral impact of these stereotypes depends on precise weights voters assign to these attributes. Further, the electoral consequences of social categorization likely vary not only across elections, but also between voters, since the nature of racial attitudes and prejudices varies significantly within the population (e.g. Elmendorf and Spencer, 2014). Focusing empirical investigations exclusively on vote choice thus has the risk of missing many consequential intermediate but difficult to observe impacts of candidate race on voter evaluations. Accordingly, our two studies unobtrusively examine these consequences, and the psychological process that produces them.

Research design

To preview, our experiments produces three sets of findings. First, we show that voters engage in social categorization when they form their initial impressions of unfamiliar political candidates. Second, we document that voters engage in racial or ethnic categorization when they can easily discern the race or ethnicity of the candidates and when this attribute varies among the candidates. Third, and most importantly, we show that the degree to which voters categorize candidates on the basis of their ascriptive group membership depends on the information context—in particular, the presence of information about a candidate’s partisan affiliation. To explain how our approach speaks to each of these points, we now turn to describing our research design, a “Who Said What?” (WSW) experiment embedded in two separate surveys of voters.

In WSW experiments, subjects engage in a series of tasks. First, they simply listen to or read a series of quotations. Each statement appears one at a time accompanied by a picture of the individuals to whom each statement is attributed. Next, subjects complete a short distractor activity—in our case, a survey on their views of local public services in the first study and a political knowledge battery in the second. In the final phase, subjects are again shown the same statements and asked to identify the individual to whom the quotation was originally attributed. The recall task is meant to be unexpected—subjects are not initially told that they will be asked to match the statements to the pictures later in the experiment.⁴

Since recall may be imperfect, respondents often make a number of errors and misattribute statements to the wrong individuals. The difference in frequency of various misattributions produces the key quantity of interest in a WSW experiment. In the study that first developed the basic design of the experiment, Taylor et al. (1978) examined whether respondents were more likely to misattribute statements to individuals within the same gender group. When subjects matched a statement to the wrong individual, the researchers hypothesized that they would be more likely to misattribute a quote to an individual of the same gender as the original source rather than an individual of the opposite sex. This is precisely what they found: conditional on misattribution, subjects were much less likely to attribute the statement to people who had the opposite gender.

In the many years since the initial study, researchers have applied the WSW design to examine a number of cognitive phenomena. An extensive body of research in psychology has shown that the procedure is very effective at identifying the types of categories that are primed during the course of the experiment (Klauer and Wegener, 1998). Peteresen (2012) introduced the WSW procedure to political science in his examination of how the “deservingness” heuristic shapes voter attitudes toward welfare recipients. We follow Peteresen in embedding the experiment in two online surveys.⁵

A WSW experiment has a key advantage over traditional experimental designs typically used to study the effects of a candidate’s race on voter perceptions: Because the outcome of interest is misattribution errors, it is difficult for subjects to decipher the actual purpose of the experiment or change their response to mimic socially desirable behavior. The design also allows us to measure both explicit and implicit categorization (Klauer and Wegener, 1998).

Both of our experiments were embedded in surveys of voters. The first was fielded in the midst of a competitive local election in a major city in the Southwest in the fall of 2012. We surveyed voters in late September and early October 2012, between the June primary and November runoff elections. The subject pool in the experiment consisted of registered voters whose e-mail addresses were included in the official voter file and, although not a random sample of the local electorate, matched the composition of the overall electorate in the city fairly closely. Appendix A describes our recruitment method and provides some descriptive statistics that compare the subjects to the broader population.

After answering several questions about their preferred choices in the upcoming mayoral runoff election and two school board races, subjects were presented with pictures and accompanying statements of four individuals. The instructions displayed on the screen stated:

Between 2012 and 2014, [city name] voters will have a chance to elect nine new members to the City Council. In this part of the survey, four potential candidates will be introduced to you. These individuals will be presented one at a time. We will ask you to read a statement each one has made about their priorities for the city. Try to gain an impression of each person.

Later, you will be asked to answer a few brief questions about them.

Each portrait will be shown for 20 seconds, and the next portrait will come up automatically. So you do not have to press any keys during the introduction of these people. Just let yourself form an impression of them (emphasis in original).

For this part of the survey, the navigation controls were disabled so that subjects could not skip the pages featuring the candidate statements or advance to the next page if less than 20 seconds had elapsed.

Two of the pictures featured middle-aged white males while the remaining two pictures depicted Latino males. In the partisan condition, one white candidate and one Latino were described as Democrats and one candidate from each ethnic group were described as a Republican.⁶ In addition to the pictures, we also provided candidate names, drawing on the Census surname database to ensure that the names chosen unambiguously signaled the candidates’ ethnic background.

Below each picture, subjects saw a statement attributed to each candidate (see Figure 1).⁷ We composed the statements to be as realistic as possible, focusing on salient local issues such as declining service levels, and ensured that they did not contain or signal any partisan or ideological information about the candidates.⁸ A full listing of the names and the statements are included in Appendix B.

Figure 1.

Screen shots of nonpartisan and partisan stimuli.

Each picture and statement appeared in succession at 20 second intervals; the order in which these pairings appeared was randomized. The respondents were instructed only to examine the pictures and the statements of the candidates. They were unaware that they would later be asked to recall which candidate made each statement. After seeing each of the four picture and statement combinations, subjects engaged in a short distractor activity consisting of a survey about their attitudes toward local city services. The survey included two multiple choice questions and asked subjects to rate a total of 17 distinct city services in terms of how “essential” they were to the quality of life in the city.

Once the subjects had completed the distractor task, they were again presented with each statement (in random order). At this point of the experiment, subjects saw only the quotation and were asked to identify the candidate to whom it was originally attributed.⁹ Figure 2 provides a screenshot from the recall phase. After completing the recall task, subjects answered a number of other questions about local politics.

Figure 2.

Screen shot of recall task.

Our analysis focuses on the frequency of different types of misattributions. Specifically, we compare how often subjects misattributed statements to a different candidate within the same ethnic group (within errors) compared to misattribution to candidates from the other ethnic group (between errors).¹⁰ As we note above, social psychology experiments have established that the difference—within errors minus between errors—measures the extent to which subjects engage in social categorization on a given dimension. If subjects make more within-group errors, this would indicate that ethnicity has been cued in their mind and that they have mentally categorized the candidates at least in part along this dimension. We expect all subjects to engage in such categorization, but to do so at a higher rate when party labels are absent.

We highlight two additional features of the experiment and the subsequent analysis that may not be initially obvious to the reader. First, given the distribution of candidates we present, subjects who guess completely at random should make twice as many between errors as within errors. Our analysis accounts for this fact by standardizing the error rates prior to analysis—by adjusting the raw counts to account for the baseline difference in probability of guessing correctly at random¹¹ (see also Peteresen, 2012; Kurzban et al., 2001; Taylor et al., 1978). Second, although the frequency of both types of errors certainly depends on individual respondents’ memory, accuracy of recall is not a confound in the experiment. While the frequency of correct recalls clearly depends on each subject’s cognitive abilities, a respondent’s memory should not affect the relative number of within vs. between errors. In other words, our quantity of interest accounts for and is not sensitive to differences in each subjects’ memory abilities.¹²

Our second study largely mirrored the first, with several exceptions that we briefly discuss here. First, the follow-up experiment was fielded in May 2015 using a national sample drawn from an online panel of registered voters from Qualtrics. Although the sample was not designed to be nationally representative, the overall demographics were on par with the profile of registered voters in November 2014 as measured in the Current Population Survey (see Appendix C). Second, the Qualtrics survey also included two attention-screener questions recommended by Berinsky et al. (2014), and the results we present below include only subjects who passed both.¹³ The two screener questions are described in Appendix D. We included them to ensure that our findings were robust to excluding individuals who likely paid minimum attention to the survey and just clicked through to the next question as quickly as possible.

Finally, the second experiment featured six (instead of four) hypothetical candidates, who were described as “six potential congressional candidates from a nearby state.” Three of these candidates were white and three were African-American, with corresponding names chosen following a similar Census surname analysis as in the first study. Appendix E contains the pictures, names, and statements used in the second study. Increasing the number of candidates allowed us to vary the extent to which the partisan affiliations of candidates in the partisan condition matched existing partisan stereotypes. In each case, the experiment featured three Democratic and three Republican candidates, but how this mapped on to candidate race varied randomly among the respondents in the partisan condition. A third of the respondents in the partisan condition had all three black candidates affiliated with the Democratic Party (and all white candidates affiliated with the Republican Party); a third saw two of three black candidates described as Democrats (and one white Democrat); and the final third had only one black Democrat (and two white Democrats), with the remaining two black candidates being Republicans.¹⁴ Note that in the first case, the partisan and racial attributes of the candidates are perfectly aligned and reinforcing, so if we see social categorization in this condition, it is not possible to disentangle how much of it is being driven by categorization along partisan lines and how much is driven by racial categorization. Nevertheless, as we discuss below, this provides a useful comparison.

Results

Study 1

Our analysis of the first study focuses primarily on the behavior of white voters, who represented the overwhelming majority of the participants in our experiment (n = 1,247). Where relevant, however, we also contrast the behavior of white voters to the smaller subset of Latino respondents (n = 181). The key results for white respondents appear in Figure 3 (a).

Figure 3.

(a) and (b) Misattribution errors for white and Latino respondents (Study 1).

Overall, we find that respondents were significantly more likely to misattribute statements to candidates from the same ethnic group compared to errors across ethnic groups. White respondents in the nonpartisan condition made 0.73 within-group errors (out of four possible statements) on average, compared with .44 adjusted between-group errors (a difference of 0.29, p < 0.001).¹⁵

We also find, however, that the degree of cognitive racial categorization is moderated by the presence of party labels. Although subjects in the partisan group also turned to racial categorization (p < 0.001), the magnitude of the racial categorization effect was 44% smaller (0.17 vs. 0.29) than among subjects in the nonpartisan condition. The difference in the extent of racial categorization between the partisan and nonpartisan conditions was thus substantively large and statistically significant (p < 0.01).

This pattern is exactly in line with the predictions of the dual-process model. Voters appear to engage in some type of social categorization, but the salience of alternative categories and the dimension along which categorization takes place is shaped by the information context in which they make those decisions. When partisan information is absent, voters rely more on racial and ethnic cues when evaluating competing candidates for office.

Of course, an alternative explanation is that the pattern of misattributions we observe in the experiment may simply reflect racism or prejudice on the part of the subjects. Perhaps white subjects believe that all Latinos “look alike” and thus confuse them for one another during the experiment. This cannot explain why the rate of ethnic categorization declines when party labels are available, but does provide a competing explanation for our first finding. We offer two pieces of evidence suggesting that this was not the case in our study.

First, Figure 3 (b) repeats the analysis for our smaller sample of Latino respondents. Although far fewer Latinos participated in the experiment, the results among this group look strikingly similar to white respondents: Latinos also categorized the candidates on the basis of their ethnic group membership (p < 0.001). In fact, the extent of social categorization seems to be somewhat greater among the Latino subjects, although the small sample size does not provide enough statistical power to make comparisons between white and Latino respondents with a great deal of precision. As was the case among white subjects, the degree of social categorization among Latino respondents was also much greater in the nonpartisan condition compared to when candidates’ partisan affiliations are available, although small sample sizes mean that the difference in categorization between treatment conditions among Latinos was not statistically significant.

Second, if our initial results simply reflect the fact that white respondents are unable to discern clear differences between minorities, the results would be driven overwhelmingly by misattribution of Latino candidates’ statements in the experiment. Table 1, which summarizes respondents’ overall performance on the recall task, shows that this is not the case. In fact, there are no significant differences in the accuracy with which respondents could correctly recall the statements made by Latinos as compared to white candidates. Likewise, the rate of within-group misattribution by white respondents was not higher for Latino candidates’ statements.

Table 1.

Summary of attribution task performance, by subgroup (% of respondents).

Attribution Task	Correctly Attributed	Within (Co-ethnic) Misattribution	Between-Group Misattribution (Unstandardized)
White Respondents
Latino Candidate #1	59%	17%	24%
Latino Candidate #2	62%	16%	22%
White Candidate #1	64%	17%	20%
White Candidate #2	60%	18%	22%
Latino Respondents
Latino Candidate #1	58%	24%	18%
Latino Candidate #2	59%	16%	25%
White Candidate #1	61%	20%	19%
White Candidate #2	59%	18%	23%

Note: There are twice as many opportunities to make between-group errors as within-group errors when guessing at random. The rate of correct recall is not significantly different between Latino and white candidates for either white (p = 0.47) or Latino (p = 0.76) respondents. White respondents N = 1,247; Latino respondents N = 181.

Study 2

The primary purpose of the second study was to replicate our initial findings using a national sample and a different minority out-group (black instead of Latino candidates). Because we have a national sample, our survey questions focus on hypothetical candidates running for Congress, instead of city council to make the experimental stimuli relevant to all respondents.¹⁶ Of course, since all congressional races are partisan, our manipulation of partisan information is not particularly relevant for understanding the dynamics of congressional elections. The results may, however, shed light on differences in voter behavior between partisan general elections and congressional primaries, in which all candidates are from the same party and for party labels do not appear on the ballot.

Another motivation for the second study was to examine the robustness of our results by varying the degree to which partisanship and race/ethnicity represented cross-cutting candidate attributes. Since our first experiment included only four candidates, it was necessary to make these two attributes completely orthogonal—with one Democrat and one Republican within both pairs of co-ethnic candidates. As a result, this design created a potential limitation to the generalizability of our findings: Was the reduction in ethnic categorization we observed in the partisan condition driven by the fact that the candidates’ assigned partisan affiliations did not fit voters’ preexisting stereotypes? In other words, was the presence of a Republican Latino sufficiently surprising that it caused respondents to shift from System 1 to System 2 thinking, explaining why we observe a decline in ethnic categorization? By increasing the number of candidates to six, the second study allows us to test this possibility directly. As we noted previously, the second experiment included three different partisan treatments with varying degrees of correlation between candidates’ race and partisanship.

As in the earlier experiment, the second study produced evidence of significant racial categorization for respondents in the baseline nonpartisan condition.¹⁷ Overall, white respondents made 0.82 more within-group errors than adjusted between-group errors (p < 0.001) out of six total attributions. Figure 4 below shows how the partisan treatments affected this gap, in line with our theoretical predictions. The top panel of the figure reports the results after pooling the respondents from the two treatment groups where partisanship did not perfectly overlap with race, combining the respondents who saw one black Democrat and those who saw two black Democrats. As a reminder, each condition included three Democrats total, so the number of white Democrats presented in each case was equal to three minus the number of black Democrats. In the pooled specification, the introduction of partisan information reduced the rate of racial categorization by 26% (p = 0.03), somewhat smaller than the magnitude we found in the first study.

Figure 4.

Impact of partisan cues on within- vs. between-group misattribution errors. *The pooled specification combines respondents who saw one black Democrat and those who saw two black Democrats. Note that the total number of Democrats in each condition was three, so the number of white Democrats is equal to three minus the number of black Democrats.

The bottom panel of Figure 4 disaggregates this effect by partisan treatment and includes the third partisan treatment in which all three black candidates were Democrats and all three white candidates were Republicans. We observe the largest reduction in racial categorization when two (rather than just one) of the three black candidates were depicted as Democrats, the opposite of what we would expect if the results in the first study were simply driven by the lack of congruence between the candidate attributes and respondents’ existing partisan stereotypes of different racial and ethnic groups.¹⁸

Figure 4 also confirms that, when the partisan and racial attributes of the candidates were in perfect alignment, the sign of the effect flipped and we observe an increase in categorization along these two shared dimensions. As we note above, it is not possible to parcel out this combined effect into the portion driven by categorization along partisan lines and the remaining portion driven by racial categorization, however. Overall, the increase in categorization in the final partisan condition is 27% relative to the nonpartisan baseline, although it falls just short of statistical significance (p = 0.056).

Impact on candidate evaluations

Together, our two experimental surveys provide clear evidence that voters engage in racial and ethnic categorization when they form their impressions of unfamiliar candidates and that the magnitude of this effect depends on the availability of the candidates’ partisan affiliations. We now examine the downstream consequences of these psychological processes on how voters evaluate the candidates.

To do so, we build on existing studies showing that voters make inferences about candidates’ ideological positions at least in part based on their race. Previous experimental studies (e.g., McDermott, 1998; Sigelman et al., 1995) report that randomly manipulating the race of candidates changes how voters perceive their ideology. These studies show that respondents rate black candidates as more liberal when compared to identical white candidates. This relationship also appears in observational analyses of actual congressional elections. Using data from the American National Election Study and the Cooperative Congressional Election Study, Jacobsmeier (2015) similarly finds that voters perceive black congressional incumbents to be significantly more liberal than white incumbents with identical voting records, as measured by their DW-Nominate scores. Using simulation methods, Jacobsmeier (2015) calculates that this misperception costs black incumbents roughly 6.6% of the vote on average, a substantial electoral disadvantage.

Our second survey allows us to examine whether the psychological processes we document—social categorization along racial and ethnic lines—serves as the cognitive mechanism that accounts for these misperceptions. After completing the recall task, we asked respondents in the second survey to place one white and one black candidate on a 100-point liberal-conservative scale (see Appendix F). Because the partisan labels themselves likely affected the perceived ideology of the candidates, the analysis below focuses only on the subjects in the nonpartisan condition who did not receive any ideological cues. Since candidate statements were randomly assigned across the candidates and respondents, the only basis for any differences in the perceived ideologies of the black and white candidates are ideological and partisan stereotypes cued by their photographs and names. Consistent with Jacobsmeier’s (2015) findings and earlier experimental work, the white respondents assigned to the nonpartisan condition in our study ranked the black candidate as being approximately 13 points more liberal on the 100-point scale (white candidate mean: 59.9).

The specific hypothesis we test in this section is whether individual voters’ proclivity to categorize candidates along racial lines, as measured by the relative number of within- versus between-group misattribution errors they make, affects their ideological (mis)perceptions of the two candidates. As a reminder, the analysis is limited only to subjects in the nonpartisan condition. This means that any differences between subjects that we measure through their patterns of misattribution are purely observational—rather than being experimentally manipulated, as in the results above—and are explained only by underlying individual-level differences in the extent to which respondents engage in racial categorization. Accounting for the origins of these differences is beyond the scope of our study, but a worthy topic of future research. As we show below, our results are robust to controlling for a variety of individual-level demographic and political variables that may predict both candidate perception and respondents’ performance on the recall task.

In Table 2, we regress the perceived difference in ideology between the black and white candidates on our aggregate measure of racial categorization, the difference between the number of within- and between-group misattributions summed across all six candidate recall tasks. Model 1 reports only the bivariate relationship, while Model 2 also controls for voter partisanship, ideology, racial prejudice, income, education, and political knowledge.¹⁹ Regardless of whether controls are included, the patterns are strikingly similar: Respondents who engage in more racial categorization in the recall task perceive the black candidate to be significantly more liberal than their white counterpart, although there is no information—aside from the candidates’ names and photographs—to give them reason to do so. Indeed, increasing the difference between within-group and adjusted between-between errors from zero (roughly the 40th percentile in our data) to two (the 75th percentile) increases the perceived difference in liberal ideology by roughly 4 points using the estimate from Model 2. This is almost a third of the total difference in perceived ideology between the black and white candidate in our overall sample. Thus, the results document how cue-taking triggered by racial categorization serves as an important cause of ideological misperception of black candidates that, as Jacobsmeier (2015) documents, puts them at a considerable electoral disadvantage compared to ideologically identical white opponents.²⁰

Table 2.

Predicting the gap in perceived liberal ideology between white and black candidates.

	Model 1	Model 2
Within–Between Errors	1.75*	2.06*
	(0.93)	(0.94)
Controls	No	Yes
N	506	505

* p < 0.05 in one-tailed test.

Discussion

The results from our experiments speak to why and under which conditions voters rely on simple demographic cues to form impressions of candidates running for low-profile offices. Our findings are consistent with other research that documents how simple experimental manipulations altering a hypothetical candidate’s race, without changing any other aspect of their description, shifts voters’ perception of the candidate’s competence and issue positions. Our main contribution is to show that these changes in perceptions are not simply consequences of racial prejudice or stereotyping but emerge from natural underlying psychological processes that all humans have evolved to rely on. Although human brains appear to be hard-wired to rely on classification and cues to make inferences about other people, we additionally show that providing voters with information about candidate partisanship (e.g., by printing them on the ballot) can influence the salience of candidate demographics.

These insights offer both an overarching theoretical explanation and document the psychological mechanisms that help account for several findings in the existing literature that have largely been developed in isolation. First, field-experimental studies dating back to the 1950s (Kamin, 1958; Lorinskas et al., 1969) report much higher rates of racially polarized voting in hypothetical nonpartisan elections compared to voting behavior in contests featuring exactly the same candidates but with party labels present. By providing direct evidence of social categorization at the individual level and showing that such categorization increases in the absence of partisan information, we identify the psychological processes that can explain why nonpartisan elections racialize voting behavior. Second, Hajnal and Trounstine (2014) examine voting behavior in more than 100 local elections across 25 cities and find that voters’ race provides the most powerful predictor of vote choice—larger, in fact, than the explanatory power of partisanship or any other variable in their model. Since most of the cities in their sample use nonpartisan elections, race-based cue-taking triggered by social categorization likely represents one important cause of this pattern of behavior and help explain why candidate race and ethnicity appears to trump partisanship in local elections. One important lesson from our results is that the introduction of partisan elections would dramatically alter these dynamics.

Third, our results provide a plausible psychological mechanism for Meier and Rutherford’s (2014) finding that black school board candidates are more likely to be elected in heavily Democratic areas. In the original study, the authors speculate that this result emerged because Democratic elites are more likely to include black candidates in their slates. Since 87% of the elections in their sample use nonpartisan elections, however, an alternative explanation is that voters simply use candidates’ race as a cue about their ideological views or partisan affiliations in the absence of partisan labels. The stereotype that blacks are very likely to be Democrats—a stereotype accurately reflecting voter behavior in the U.S.—would thus aid black candidates running in heavily Democratic school districts but hurt their electoral prospects in Republican areas, exactly as the authors find.

Our results also help explain why candidates’ race and ethnicity affect the criteria voters bring to bear when making their decisions, and point to conditions under which this is most likely to occur. McConnaughy et al. (2010), for example, find that white voters’ sense of nativism affects their choice of candidates—but only when the menu of options includes a Latino candidate. Our experiments suggest that the presence of Latinos on the ballot encourages voters to categorize candidates based on their ethnicity, thus triggering the use of associated category-relevant attitudes and affects, such as nativism. While the authors conclude that nativism plays a role in voter decision-making when “A Latino [is] on the Ballot,” to borrow the title of the authors’ original study, our results suggest that nativism may be important especially when candidate partisan labels do not appear alongside the candidates’ name (as was the design in the McConnaughy et al. experiment).²¹

The central lesson from our findings is that voters’ inclination to engage in racial and ethnic categorization depends crucially on the context in which they make decisions—in particular, whether partisan information about the candidates is easily accessible. This is surely not the outcome that Progressive reformers intended when they pushed for nonpartisan elections at the turn of the 20th century. They hoped that the removal of partisan labels would purge local elections of partisan considerations, discourage participation among poorly informed voters, and help ensure that “the interested and involved individual…with other well-meaning and public regarding citizens [will] use the electoral process to select the most competent leaders who will then work for the common good” (Schaffner et al., 2001: 7). Although nonpartisan elections may encourage a certain number of otherwise uninterested voters to abstain from voting in down-ballot race, they also lead a substantial number of equally uninformed voters to turn to other types of information shortcuts as the basis for their decision. As the election of Steve Rocco showed in Orange County, nonpartisan contests may not always lead to the election of the most competent leaders if voters substitute racial and ethnic stereotypes in place of partisan considerations. Although partisan elections come with their own sets of disadvantages—for example, the activation of hostile affective attitudes toward members of the opposing party (e.g., Iyengar and Westwood, 2015)—it is hard to argue that nonpartisan elections necessarily improve the quality of voter decision-making, as the Progressive reformers hoped they would.

Finally, our findings should encourage both social scientists and political activists to rethink reform efforts designed to increase the political incorporation of minority groups. As we stress earlier, social categorization does not necessarily imply a systematic disadvantage for minority candidates. But our strong suspicion is that nonpartisan ballots do result in negative electoral consequences for minority groups, particularly in areas where racial prejudice and stereotyping are most prevalent (e.g. Elmendorf and Spencer, 2014). The political geography of negative stereotyping makes this particularly likely, as minority populations tend to be concentrated in regions where negative racial attitudes among whites are more prevalent. The widespread use of nonpartisan elections at the local level in these contexts is particularly troubling because local offices, including school boards and city councils, represent a vital stepping-stone for candidates interested in higher office, where prior political experience is often crucial for electoral success (Jacobson, 1989). If nonpartisan elections at the local level reduce the electoral prospects for minority candidates, this surely produces downstream (or, more appropriately, upstream) effects on representation at other levels of government. To date, advocates of minority incorporation have focused primarily on replacing at-large with district elections, drawing on strong empirical evidence indicating that the at-large elections creates substantial barriers to the success of minority candidates (but see Meier and Rutherford, 2014). Our results suggest that reformers would be well advised to look beyond local electoral institutions to the information printed on the ballot itself.

Supplemental material

Supplemental Material, sj-docx-1-ppq-10.1177_1354068821990625 - Do nonpartisan ballots racialize candidate evaluations? Evidence from “Who Said What?” experiments

Supplemental Material, sj-docx-1-ppq-10.1177_1354068821990625 for Do nonpartisan ballots racialize candidate evaluations? Evidence from “Who Said What?” experiments by Craig M Burnett and Vladimir Kogan in Party Politics

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The Ohio State University’s Institute for Democratic Engagement and Accountability and the Appalachian State University’s Department of Government and Justice Studies provided funding for this research.

ORCID iD

Craig M Burnett

Supplemental material

Supplemental material for this article is available online.

Notes

References

Berinsky

Margolis

Sances

(2014) Separating the shirkers from the workers? Making sure respondents pay attention on self-administered surveys. American Journal of Political Science 58(3): 739–753.

Brady

Sniderman

(1985) Attitude attribution: a group basis for political reasoning. American Political Science Review 79(4): 1061–1078.

Bullock

Dunn

(1996) Election roll-off: a test of three explanations. Urban Affairs Review 32(1): 71–86.

Census of Governments (1995) Volume 2: Popularly Elected Officials. Washington, DC: U.S. Government Printing Office.

Conover

Feldman

(1989) Candidate perception in an ambiguous world: campaigns, cues, and inference processes. American Journal of Political Science 33(4): 912–940.

Delli Carpini

Keeter

(1996) What Americans Know about Politics and Why It Matters. New Haven: Yale University Press.

Downs

(1957) An Economic Theory of Democracy. New York: Harper and Row.

Elmendorf

Spencer

(2014) The geography of racial stereotyping: evidence and implications for VRA ‘preclearance’ after Shelby county. California Law Review 102(4): 1123–1180.

Fiske

Neuberg

(1990) A continuum of impression formation, from category-based to individuating processes: influences of information and motivation on attention and interpretation. In: Zanna

(ed) Advances in Experimental Social Psychology, Vol. 23. Amsterdam: Elsevier Science, pp. 1–74.

10.

Hajnal

Trounstine

(2014) What underlies urban politics? Race, class, ideolgoy, partisanship, and the urban vote. Urban Affairs Review 50(1): 63–99.

11.

Hajnal

Lewis

(2003) Municipal institutions and voter turnout in local elections. Urban Affairs Review 38(5): 645–668.

12.

Huddy

Terkildsen

(1993) Gender stereotypes and the perception of male and female candidates. American Journal of Political Science 37(1): 119–147.

13.

Iyengar

Westwood

(2015) Fear and loathing across party lines: new evidence on group polarization. American Journal of Political Science 59(3): 690–707.

14.

Jacobsmeier

(2015) From black and white to left and right: race, perceptions of candidates’ ideologies, and voting behavior in U.S. house elections. Political Behavior 37(3): 595–621.

15.

Jacobson

(1989) Strategic politicians and dynamics of U.S. house elections, 1946-1986. American Political Science Review 83: 773–793.

16.

Kahneman

(2011) Thinking Fast and Slow. New York, NY: Farrar, Straus and Giroux.

17.

Kam

(2007) Implicit attitudes, explicit choices: when subliminal priming predicts candidate preference. Political Behavior 29(3): 343–367.

18.

Kamin

(1958) Ethnic and party affiliations of candidates as determinants of voting. Canadian Journal of Psychology 12(4): 205–212.

19.

Klauer

Wegener

(1998) Unraveling social categorization in the ‘who said what?’ paradigm. Journal of Personality and Social Psychology 75(5): 1155–1178.

20.

Kurzban

Tooby

Cosmides

(2001) Can race be erased? Coalitional computation and social categorization. Proceedings of the National Academy of Sciences 98(26): 15387–15392.

21.

Lorinskas

Hawkins

Edwards

(1969) The persistence of ethnic voting in urban and rural areas: results from a controlled election method. Social Science Quarterly 49(3): 891–899.

22.

McConnaughy

White

Leal

, et al. (2010) A Latino on the ballot: explaining coethnic voting among Latinos and the response of white Americans. Journal of Politics 72(4): 1–13.

23.

McDermott

(1997) Voting cues in low-information elections: candidate gender as a social information variable in contemporary United States elections. American Journal of Political Science 41(1): 270–283.

24.

McDermott

(1998) Race and gender cues in low-information elections. Political Research Quarterly 51(3): 895–918.

25.

McDermott

(2005) Candidate occupations and voter information shortcuts. Journal of Politics 67(1): 201–219.

26.

Meier

Rutherford

(2014) Partisanship, structure, and representation: the puzzle of African American education politics. American Political Science Review 108(2): 265–280.

27.

Moskowitz

Stroh

(1994) Psychological sources of electoral racism. Political Psychology 15(2): 307–329.

28.

Peteresen

(2012) Social welfare as a small-scale help: evolutionary psychology and the deservingess heuristic. American Journal of Political Science 56(1): 1–16.

29.

Rahn

(1993) The role of partisan stereotypes in information processing about political candidates. American Journal of Political Science 37(2): 472–496.

30.

Sanbonmatsu

(2002) Gender stereotypes and vote choice. American Journal of Political Science 46(1): 20–34.

31.

Schaffner

Streb

Wright

(2001) Teams without uniforms: the nonpartisan ballot in state and local elections. Political Research Quarterly 54(1): 7–30.

32.

Sigelman

Walkosz

, et al. (1995) Black candidates, white voters: understanding racial bias in political perceptions. American Journal of Political Science 39(1): 243–265.

33.

Taylor

Fiske

Etcoff

, et al. (1978) Categorical and contextual bases of person memory and stereotyping. Journal of Personality and Social Psychology 36(7): 778–793.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.67 MB