A Different Kind of Disadvantage: Candidate Race, Cognitive Complexity, and Voter Choice

In our conception, prejudice may either be implicit (and thus, unknown to the respondent), or explicit but suppressed for some other reason (e.g. social desirability or conflicting egalitarian attitudes). In either case, we expect cognitive complexity to increase the role of prejudice in decisions.

Our own analysis of General Social Survey Data from 1998 to 2008 confirms that both white liberals and white Democrats hold such stereotypes and that the differences in their ratings of whites and African Americans on these dimensions are significantly different from zero. Results available from the authors.

While implicit attitudes are automatic and unconscious, explicit attitudes are deliberate and intentional. Research shows that both implicit and explicit attitudes can affect vote choice and policy preferences (Mo 2015; Perez 2010; Payne et al. 2010; Galdi et al. 2008; Lodge and Taber 2000, 2005; Burdein et al. 2006), but studies show that explicit attitudes tend to dominate in studies of vote choice (Ditonto et al. 2013; Kalmoe and Piston 2013; Kinder and Ryan 2017). Our theory focuses on how explicit attitudes—such as prejudicial negative stereotypes of blacks and beliefs about black candidates’ ideology—influence voter behavior.

Some egalitarians may be expressing what is known as social desirability bias—the belief that revealing one’s racism is inappropriate (Krumpal 2013), while others might feel firmly committed to egalitarian ideals or hold genuine racial sympathy toward non-whites (Devine and Monteith 1993; Tesler 2016).

Political liberals may also have a greater commitment to diversity/racial equality than political conservatives. Such a commitment will produce an observationally equivalent outcome in which liberals are more likely to select black candidates than conservatives under low cognitive load.

Some research (Oliver 2012; Anzia 2014; Berry 2009) suggests that because of the general low levels of interest in local politics, local electorates are often disproportionately comprised of knowledgeable, motivated residents like home owners and municipal employees. Thus, the level of knowledge among voters may interact with cognitive complexity to affect the extent to which voters use racial stereotypes when casting ballots in real elections. In the Online Appendix we take advantage of distinctions in voter knowledge across real election contexts to test this possibility. Our analyses in Fig. OA1. suggest that more cognitively complex elections (at large contests where voters must select multiple candidates versus district contests where they select only one) do yield fewer black officeholders particularly in elections that are more likely to have low-information voters (on-cycle elections with higher turnout versus off-cycle elections with lower turnout). These tests suggest support for our theory in real elections and indicate the effects of cognitive complexity in provoking use of racial stereotypes among voters should be more common among lower than higher information voters.

Replication materials are available in the Political Behavior Dataverse at https://​doi.​org/​10.​7910/​dvn/​mnao24.

In total, there were 46 survey questions in Experiment 1 and 41 questions in Experiment 2.

This paper is focused particularly on testing the role of cognitive complexity in shaping voter behavior—a test not present in existing research on candidate race and vote choice. Our experiments are designed to mimic one type of electoral context—low information elections, in which voters may see images of candidates on campaign mailers, etc. but have little other information about the candidates. Studies in Ireland, where ballots for local office include photographs, find that voters rely heavily on photographs as shortcuts for decision-making in low information elections (Buckley et al. 2007). To evaluate the effects of candidate race and ethnicity, we sought to remove as many other cues from our experimental ballots as possible—for example, we chose to cue candidate race using images of individuals in professional dress rather than names which could potentially also signal socioeconomic status or other cues.

We used a total of 80 pictures in our experiments. These include 40 pictures of governors and senators from prior studies (Olivola and Todorov 2010; Todorov et al. 2005) that were calculated to be within two competency score standard deviations from one another and an additional 40 pictures of state or local elected officials from Arizona, Florida, Georgia, Hawaii, North Carolina, and Texas gathered online. Using MTurk, a separate sample of 966 participants rated each photo on attractiveness, competency, and trustworthiness similar to Todorov et al. 2005. There is no significant difference between the two sets of pictures with regard to competency. Our photos were judged to be very slightly more attractive and trustworthy. Summary statistics on these measures are in the Online Appendix Table OA2.

Regarding racial and ethnic identification, respondents correctly identified the race or ethnicity of black and white candidates 89% of the time and Asian candidates 74% of the time. Latino candidates were only identified as Latino 39% of the time. When respondents made mistakes, Black and Asian candidates were nearly always categorized as some other racial minority (they were only perceived as white 2.6% of the time). But, Latino candidates were generally perceived as white (47% of the time). As many Latinos racially identify as white, this may simply reflect the complexity of Latino identification in the broader population and in actual elections with Latino candidates. Regarding ideology, when estimating the ideology of our candidate photos on a 7-point scale, with higher values meaning more conservative, on average respondents rated black candidates at 3.168 (i.e. solidly liberal) and white candidates at 4.701 (i.e. moderate leaning toward conservative)—a highly significant, one standard deviation difference (see appendix Table OA3).

Prior to the cognitive load manipulations, respondents completed a set of text-based voting tasks separate from the dependent variables in this analysis. These are discussed in (Crowder-Meyer et al. 2017) but do not test the cognitive difficulty mechanism and are not discussed here. After the text-based voting task, all respondents engaged in two simple cognitive processing tasks—completing an analogy and an anagram—prior to being randomly assigned to the eye blink manipulation.

As this was an online, anonymous survey we had no way to check the accuracy of the eye blink totals. On average our eye blink counters reported 22 blinks while answering five questions.

In cities where councilors are elected at-large, there are commonly twice as many candidates as seats available for election in at-large contests (Trounstine 2008).

We have exactly ten photos for each of eight categories: white men, white women, black men, black women, Latino men, Latina women, Asian men, and Asian women.

For details on the task and samples, see the Online Appendix.

We present the results for black candidates relative to white candidates. Estimated effects for Asian and Latino candidates are discussed in footnotes and are available from the authors by request.

No controls for respondent attributes are needed because candidate race was randomized across respondents. Including characteristics like age, gender, marital status, and income does not affect the conclusions and the coefficients on such variables are essentially zero.

Hainmueller et al. (2014) define the AMCE as “the increase in the population probability that a profile would be chosen if the value of its lth component was changed from t₀ to t₁, averaged over all the possible values of the other components given the joint distribution of the profile attributes p(t)” (p. 11).

Our results are more mixed for Asian and Latino candidates. In Experiment 2, Asian and Latino candidates were not significantly more or less likely to be selected than whites in at-large versus district elections. In Experiment 1, Asian candidates even appear to benefit from the imposition of a higher cognitive load. Our theory predicts that voter behavior under heavy cognitive load is driven by racial stereotypes, thus these results are consistent with the presence of more mixed positive and negative stereotypes about Asians and Latinos in the US (see e.g. Kao 1995). Additionally, our findings for Asian candidates may indicate that positive stereotypes about Asian-Americans’ work ethic and intellect lead voters to prefer Asian candidates when stereotypes play a larger role in their decision-making (Kinder and Kam 2010).

Respondents were asked their ideology on a 7-point scale. Liberals are those who said they are extremely liberal, liberal, or slightly liberal, while conservatives are extremely conservative, conservative, or slightly conservative. Respondents who insisted on choosing moderate even after a follow-up question prompting them to choose an ideological position or who said that they haven’t thought much about it are excluded from the following analysis.

One alternative explanation for our findings is that liberal voters are simply creating more diverse councils in the at-large elections when they have the opportunity to select more than one candidate—choosing Asian or Latino candidates rather than black candidates among some of the three candidates they choose. We test this possibility by focusing on elections in which only one black candidate appeared. If our liberals are actually making more diverse slates, and are not affected by cognitive processing, the one black candidate should do equally well in district and at-large elections. This is not what we find. Instead, liberals are significantly less likely to choose the one black candidate in at-large elections than in district elections – with an effect size even larger than in the full sample. The coefficient on black candidates interacted with at-large elections is − 0.19 (SE = 0.07) in this subset of the data compared to − 0.07 (SE = 0.039) in the full sample.

This finding that black candidates do better in district elections is consistent with existing research outside of the experimental lab. Racial and ethnic minorities are more likely to achieve descriptive representation in real district (as opposed to at-large) elections as well (Marschall et al. 2010). To date, scholars have argued this effect is a product of geographic segregation (e.g., Trounstine and Valdini 2008; Sass 2000; Vedlitz and Johnson 1982). However, our results indicate that geography may not be the sole reason for this pattern. Districts may enhance descriptive representation of racial and ethnic minorities for an additional reason: the ease of decision making in a race with fewer candidates.