The Contingent Effects of Candidate Sex on Voter Choice Part 2

Effects of Candidate Sex and Gender-Office Congruence We begin with Figure 2, which shows the relative importance of candidate attributes on electoral support for the full sample of respondents and candidate pairings. The dots denote point estimates for the AMCEs, which indicate the average effect of each attribute on the probability that the candidate will be chosen. The horizontal bars show 95% confidence intervals. Our main interest here is the importance of candidate sex on voter decisions. Because each attribute is dichotomous, the estimated effects can be directly compared to one another. Note that within each category of attributes, one treatment is arbitrarily chosen as the omitted reference category, just as in a regression framework where one category serves as the baseline. For candidate sex, we set male as the baseline. [Figure 2 about here] The results of our experiment show that candidate sex is a significant voting heuristic, even under the presence of many other cues about candidates. On average, respondents are 1.3 percentage points less likely to vote for a female candidate. Importantly, because of complete randomization of all attributes, this effect cannot be attributed to other factors such as age, experience, issue priorities, or even personality traits that might differ (in reality or perception) between male and female candidates in real elections. In other words, it seems that voters are biased against female candidates as a baseline preference and not just because of traits inferred when evaluating a female candidate. We show in the appendix that the effect is larger – 2.5 percentage points – when the analysis is limited to opposite-sex pairings that include a male candidate and a female candidate. Relative to some other variables such as political experience and policy positions, the effect of candidate sex on voter decisions appears relatively small in magnitude. However, the magnitude of the gender effect is similar to the penalty faced by minority candidates and candidates with the lowest approval ratings and thus is likely to matter in a tight race where the winner is determined by a narrow margin. As explained above, scholars have suggested that while women might be discriminated against in presidential elections, female candidates might face less bias or even be favored in congressional elections. Although the logic of how voters might connect perceived traits of male and female candidates with specific offices is intuitive, the gender-office congruency theory has not been fully tested. To provide some resolution to this question, our experiment presented half of the candidate pairs as seeking the presidency and half as seeking seats in the House of Representatives. Analyzing these two groups separately provides a clean test of this theory. Incidentally, it also provides a check on the verisimilitude of our experiment by revealing whether respondents are taking the task seriously enough to differentiate between offices. Figure 3 shows the results of our analysis, which indicates the estimated marginal effect of candidate sex by the type of political office being sought. We find a clear and statistically significant difference between the two types of political office (p < .05). The overall effect of 1.3 points disadvantage presented earlier was an averaged that masked this difference by office. Figure 3 shows there is no female disadvantage in congressional elections, but voters punish female candidates running for president by 2.4 percentage points (p < .05). [Figure 3 about here] This finding suggests that female candidates face a greater challenge when they run for executive office than when they run for legislative office, but this is true even after controlling for stereotypes that voters associate with male and female politicians. As a result, the standard gender-office congruency theory does not fully explain why voters have a bias against women being elected as president.7 For example, because we randomly varied personal characteristics and experience in office, it is not the case that respondents preferred male candidates in the hypothetical presidential election because they inferred that the men were stronger leaders or more experienced. An additional explanation may be that the public simply has more experience with women in Congress, including even the former Speaker of the House Nancy Pelosi. In contrast, the public must use imagination to anticipate what a female president would do in office (Burden, Ono, and Yamada 2017). Research on the first election of black candidates finds that white voters are initally resistant becaue of the uncertainty and lack of experience with black elected officials (Hajnal 2003). It is plausible that the difference between presidential and congressional elections may be because having no experience of a female president brought some fears and uncertainties to voters about choosing a female candidate in presidential elections. We further show in the appendix that this result is not necessarily driven by people’s attitudes toward Hillary Clinton, who was running for the Democratic presidential nomination at the time when our survey was conducted. It is difficult to test for whether attitudes toward Clinton underlie views about a female president, because those who have bias against a female president are prone to dislike Clinton as well. However, as we show in the appendix, there is little evidence for a “Hillary effect” in our data. Even among those who do not have a negative 7 It is possible that voters are biased against female candidates only when it comes to the presidency and not for other executive offices such as governor and mayor. Additional experiments will be necessary to pursue these nuances of the gender-office congruency theory. favarability rating toward Clinton, we still found a tendency that people have a greater bias against female candidates in presidential elections than they do in congressional elections. Effects of Candidate Sex by Subgroups of Voters We have so far analyzed the overall effect of candidate sex on voter decisions. While we found a modest negative bias against female candidates, not all voters are likely to punish female candidates in the same manner. To examine the heterogeneity of treatment effects, Figure 4 compares the estimated marginal effects of candidate sex on voter decisions for several subgroups of voters. This figure shows how the disadvantage of a female candidate varies across the respondent’s characteristics such as sex, education level, age, social class, region of residence, race/ethnicity, and partisanship. [Figure 4 about here] Although treatment effects do not actually differ much across respondent attributes, we do uncover some differences across levels of self-identified social class. Whereas lower class voters tend to punish female candidates, upper class voters do not have any bias against female candidates. Differences by race and ethnicity, region, and age are modest to nonexistent. Our results further indicate that female voters do not vote disproportionately for female candidates at higher rates when other candidate attributes are varied. Instead, the results show that male voters prefer male candidates over female candidates. In short, we find no evidence for the gender affinity effect in the form that is often believed to exist among female voters, but we do find evidence for an opposite affinity among male voters.8 8 Note that we cannot rule out the possibility that the effects for male and female respondents are themselves not statistically significant from one another (p > .10). A more dramatic pattern emerges across voters depending on party identification.9 While Republicans have a negative bias against female candidates about two percentage points, Democrats do not punish female candidates. This is not especially surprising given the preponderance of women who identify as Democrats and serve as Democrats in elective office, although we find that it holds even when other characteristics of the candidate such as policy positions are held constant via randomization. What may be more surprising is that independents exhibit a larger negative bias against female candidates than do Republicans.10 The probability that independents support a candidate is 3.2 percentage points lower if that candidate is woman. This result is consistent with our hypothesis. We conjecture that among voters with partisan affiliations, partisanship acts as a strong force that leaves relatively little room for candidate sex to affect the voting decision. Partisanship is in fact a better diagnostic for how election officials will behave in office than perhaps any other characteristic, so partisan voters rationally focus on that dimension over candidate sex. Independents, in contrast, lack a 9 We measure party identification using the first of the standard branching questions. That is, “leaners” are not distinguished from other independents. However, we also have a measure of attitude toward each party. Our data suggest that 41.2% of those who identified themselves as independents are truly neutral to both parties; the rest of the self-identified independents (59.8%) have an attitude leaning toward either party, but the tilt is often slight and they are almost equally split between the two parties (28.9% for Republicans and 29.9% for Democrats). We discuss more details about the distribution of leaners among independents in the appendix. 10 The estimated effect for independents is greater than the one for Republicans by 1.2 percentage points, but the difference between them is not statistically significant (p > .10). predisposition to favor one of the candidates based on party. Without a clear indicator that one of the candidates will act as a faithful agent of their independent voter’s interests in office, their bias against female candidates is more readily apparent. In short, partisan labels provide a kind of “insurance” for partisans that guards against the uncertainty that a different kind of officeholder might bring. A candidate’s party brings a high level of predictability that allows partisan voters to discount other information. Independents, who lack the security provided by a party affiliation, end up relying more on other factors including candidate sex. Interestingly, this result conflicts with what respondents say when asked about candidate sex in isolation. For example, a survey conducted in November 2014 asked respondents whether they hoped to see a female elected president in their lifetimes or whether it did not matter to them. A majority of Democrats said they hoped this would happen, and more than one-third of independents did, but fewer than one in five Republicans did.11 This suggests that Republicans would be most likely to harbor bias against female candidates, or Democrats to harbor bias in favor of them, but this only appears to be true in the abstract when other factors about candidates are absent. Our conjoint experiment shows that in the context of making a decision between two candidates with realistic profiles, candidate sex actually matters most to independents because partisanship does not provide them with a reason in itself to override it. For a voter who identifies with a party, choosing a candidate of the same party provides some “insurance” about what the official would do in office regardless of their sex. 11 Pew Research Center, “American’s Views of Women as Political Leaders Differ by Gender,” http://www.pewresearch.org/fact-tank/2016/05/19/americans-views-of-women-as-politicalleaders-differ-by-gender/ To explore partisan differences further, Figure 5 presents the estimated marginal effects of candidate sex separately for the two competition contexts—competitions between differentparty candidate pairings (Democrat vs. Republican candidates) and competitions between sameparty candidate pairings (Democrat vs. Democrat candidates or Republican vs. Republican candidates). These two scenarios mimic the options that voters actually face in general elections and primary elections, respectively. We argue that independents have the greatest bias against female candidates because a candidate’s party label cannot work for them to ensure that their views are reflected in politics. If this is the case, the effect of candidate sex on voter decisions should be more determinative among partisan voters (especially among Republicans) in the context of electoral competition where the party label is not a point of differentiation. The results shown in Figure 5 are consistent with our expectations. When electoral competition is between two candidates from different parties, both Democrats and Republicans show no hostility toward female candidates (i.e., the estimated marginal effect of candidate sex for each group is not statistically discernible from zero). In contrast, when electoral competition is between the two candidates running from the same party, Republicans show biases against female candidates of 2.8 percentage points, while Democrats remain to have no bias against female candidates.12 These findings suggest that candidate sex becomes more important for Republicans when candidates cannot be differentiated by a party label; candidate sex does not matter among Democrats regardless of the electoral context. In other words, partisanship helps Republicans override the bias based on candidate sex. This implies that candidate sex is likely to 12 Democrats actually favor female candidates by 1.2 percentage points, but this estimated effect is not statistically significant at the .10 level. play a more important role for voters (especially Republicans) in evaluating candidates in primary elections or in nonpartisan elections where the party label—so consequential in general elections—is not a point of differentiation. [Figure 5 about here] In summary, the results of our conjoint experiment demonstrate that voters use candidate sex as a voting heuristic even when it is embedded among various other cues about candidates. At the same time, the overall effect of candidate sex on voter decisions is relatively small compared to other candidate attributes such as issue positions and public office experience, and its magnitude is almost the same as the effect of candidate race/ethnicity. In addition, the punishment to female candidates is limited to presidential rather than congressional elections, and appears only among male voters. More importantly, the results further show that the effect of candidate sex varies significantly among voters across party lines, and in particular, that independent rather than Republican voters showing the greatest negative bias against female candidates. We attribute this to the “insurance” that a party label provides to partisan voters and argue that the lack of such relevant information for independents allows their bias against female candidates to emerge. Republicans display bias in nonpartisan elections but appear to overcome the bias when they can differentiate candidates based on the party label. Conclusion Even though a majority of the population and of voters in the United States is female, women are sorely underrepresented in Congress and a woman has yet to be elected president. A complete explanation for the underrepresentation of women in elective office is necessarily complex and multifaceted, yet its elements are gradually coming to fruition. In particular, recent research has shown that part of the explanation is that potential female candidates are less likely to view themselves as qualified, are less willing to endure the demands of campaigning, and are less likely to be recruited than are men in similar circumstances (Kanthak and Woon 2015; Lawless 2012; Lawless and Fox 2005; Sanbonmatsu 2006). Previous studies have demonstrated that voters view candidates from gendered perspectives. This in itself could lead to the gender inequality in elected office. However, the scholarly literature has not made clear whether voters actually use candidate sex as a voting heuristic. This is partly because various other cues about candidates—such as candidate’s party label and even policy positions—are also available to voters. These other cues may overwhelm the effect of candidate sex on voter decisions, particularly in the contemporary era as public resistance to women in public life has declined and party cues have become so powerful. The design of our study allows for new insight on this aspect of women’s representation. It is difficult to evaluate the effect of candidate sex on vote decisions by using actual election results due to the presence of endogeneity problems such as entrance barriers for female candidates. Hence, multiple prior studies have conducted experiments to determine whether a candidate’s sex has a direct effect on voting decisions among people by manipulating candidate profiles in campaign advertisements and newspaper articles to make firmer causal inferences. However, those studies test the effects of multiple candidate attributes in separate experiments rather than simultaneously, and the number of varying candidate attributes in each experiment is quite limited. Omitting potentially relevant characteristics is a concern because voters have been shown to infer things about candidates based on their sex. Because the conventional experimental designs cannot fully decompose multiple treatment effects and rule out these inferential mechanisms, we instead employed a conjoint survey experiment. This design enables us to examine the relative importance of candidate sex on vote decisions by randomly varying a large number of candidate attributes in profiles to rule out endogenous effects, spurious effects, and mechanisms that are distinct from a baseline gender preference. The results of our study demonstrate that a candidate’s sex affects voter decisions even under the presence of many other information cues about candidates. Voters pay attention not only to a candidate’s party label and issue positions, but also to the candidate’s sex. However, the overall magnitude of the bias against female candidates is relatively small compared to other candidate attributes such as issue positions and public office experience, and its magnitude is almost the same as the effect of candidate race/ethnicity. Moreover, the bias does not occur because women disproportionately support female candidates; rather, the gender-based affinity effect is found only among male voters who prefer male candidates to female candidates. The effect of candidate sex also differs between presidential and congressional elections. However, in an apparently challenge to the gender-office congruency theory, we found that female candidates running for presidential elections face a greater challenge than those running for congressional elections even after isolating the effect of candidate sex as distinct from stereotypes that voters associate with male and female politicians. Thus, the role of gender stereotypes among voters cannot fully explain the difference between the levels of office that candidates are competing for. Most importantly, our results further showed that the effect of a candidate’s sex varies significantly among voters across party lines, and in particular, that independent rather than Republican voters show the greatest negative bias against female candidates. We attribute this to the “insurance” that a party label provides to partisan voters, who can be sure that the party affiliation of the candidate will represent their views in office regardless of the sex of the candidate, and argue that the lack of such relevant information for independents allows their bias against female candidates to emerge. Republicans override their bias against female candidates when they can differentiate candidates based on the party label. In contrast, independents have no such insurance, and always react more immediately to candidate sex. Our findings point to at least two intriguing paths for further research about conditions where candidate sex might be more consequential. The first path is to explore election environments where candidate information is less rich. The importance of candidate sex as an informational cue may be different depending on the types of other candidate attributes shown to voters. Indeed, the information cues available to voters differ depending on the context of electoral campaign. The sex of a candidate is usually easy to discern from advertising or from the names listed on the ballot, but policy information might be more elusive, particularly in lower level elections where knowledge about candidates is thinner. A study by Kirkland and Coppock (forthcoming) also uses conjoint experiments and finds some evidence for voter bias against male candidates in an environment where voters are given less information, notably the absence of any partisan information. Further research is needed to understand how the office in question and richness of candidate information might shift gender bias back and forth from one setting to another.13 In addition, there is also a possibility that exposure to candidate attributes related to 13 The Kirkland and Coppock (forthcoming) study differs from ours in two important ways. First, the authors compare elections with and without party labels, but do not examine elections where both candidates are from the same party. Second, their experiment does not tell respondents what office is being sought. Instead, the experiment varies the previous office held by the candidate, ranging from local offices such as city council to representative in Congress. It is possible that respondents infer what office is being sought by the backgrounds that are presented to them in each candidate pairing. gender stereotypes activates voters showing a bias against female candidates (see Bauer 2015). As a future extension, we need to systematically examine how the effect of candidate sex changes as voters across variation in the information environment. The second path is about how voter experiences with female elected officials might themselves change attitudes over time. The number of female representatives is increasing in Congress, especially on the Democratic side of the aisle. Voters often rely on stereotypes in evaluating unfamiliar candidates due to fear and uncertainty associated with them, but those uncertainties tend to dissipate as voters experience more members of underrepresented groups in elective office. Thus, the experience of having female representatives in Congress may have reduced the bias against female candidates among Democrats. Moreover, the presence of female president may change how voters evaluate female candidates in presidential elections afterwards. Researchers might test this possibility by examining whether female candidates are evaluated differently between voters who have any incumbent female representatives before and those who do not have such an experience.