Voter bias against women cannot explain female underrepresentation in American politics; if anything, voters prefer women over men

Poutvaara, Panu; Graefe, Andreas (2022) : Do Americans Favor Female or Male Politicians? Evidence from Experimental Elections, Beiträge zur Jahrestagung des Vereins für Socialpolitik 2022: Big Data in Economics, ZBW - Leibniz Information Centre for Economics, Kiel, Hamburg. https://www.econstor.eu/bitstream/10419/264117/1/vfs-2022-pid-70505.pdf

Abstract: Women are severely underrepresented in American politics, especially among Republicans. This under-representation can arise from women being less willing to run for office, from voter bias against women, or from political structures that make it more difficult for women to compete. Here we show to what extent support for female candidates varies by voters’ party affiliation and gender. We carried out hypothetical elections in which participants made vote choices solely based on politicians’ faces. When deciding between candidates of different genders, Democrats, and particularly Democratic women, preferred female candidates, while Republicans chose female and male candidates equally often. These patterns remained when controlling for respondents’ education, age, and political knowledge and for candidates’ age, attractiveness, and perceived conservativeness. Our results suggest that voter bias against women cannot explain female under-representation. On the contrary, American voters appear ready to further narrow the gender gap in politics.

Keywords: Gender; Elections; Gender discrimination; Political candidates
JEL: D72; J16

5 Conclusion

Major gender gaps have opened in American politics in recent decades. Women are more likely than men to support Democrats (Gillion et al., 2020), Democratic voters are more likely than Republicans to support female candidates (Schwarz & Coppock, 2021), and the female share of congressional Democrats is almost three times that of congressional Republicans (Fig. 1B). We carried out hypothetical elections in 2016 and 2020 to disentangle how voter gender and partisanship interact in support for female candidates. Our results show that Democrats generally favored female candidates, and that preference for female candidates was particularly strong among Democratic women. In our 2020 survey, Democratic women chose the female candidate three times as often as the male candidate. Republican respondents, instead, chose female and male candidates about equally often. Our findings suggest that voter bias against women cannot explain female under-representation in American politics, even among Republicans. If anything, voters, on average, prefer women over men.

Our approach to study gender discrimination in voting complements vignette and conjoint survey experiments, which have become an established practice in political science research (Hainmueller et al., 2015; Hainmueller et al., 2014). In these studies, respondents state their preferences based on short, standardized descriptions of hypothetical candidates. Vignette and conjoint survey experiments allow studying simultaneously the effects of different cues, like gender, age, and reported experience. However, this comes at the cost that researchers define the characteristics that are presented to respondents, and how these are presented. Our approach of asking respondents to make vote choices based on candidate photographs does not require researchers to specify what textual cues are provided to respondents and in which order. Instead, we collected vote choices for hypothetical elections among all 736 Members of the European Parliament. One advantage of using MEPs was that they are real and elected politicians. Hence, the photographs likely incorporate cues that are relevant in politics, which may not be the case when using stock photographs. Another advantage of using MEPs was that American respondents are unlikely to recognize the candidates, which could have introduced bias. Finally, previous research has shown that evaluations of politicians’ photographs help to predict election outcomes around the world, providing external validity for using photographs (Antonakis & Dalgas, 2009; Ballew & Todorov, 2007; Berggren et al., 2010; Lawson et al., 2010; Todorov et al., 2005).

A major concern in all surveys is that subjects might change their behavior due to cues about what constitutes appropriate behavior (Zizzo, 2010). In our setting, the concern is that respondents would find supporting female candidates in hypothetical elections the appropriate choice, even if they would not vote for the female candidate in a real election. Our study design alleviates these concerns by randomizing gender combinations in hypothetical elections. We also did not refer to gender – but only to voting under very little information – in our task description. Furthermore, recent research has found that experimenter demand effects are small in online surveys even when respondents are provided a hint on the hypothesis that researchers are testing (de Quidt et al., 2018; Mummolo & Peterson, 2019). Comparing conjoint and vignette experiments with real referendums in Switzerland also suggests that estimates from survey experiments perform remarkably well in predicting actual voting outcomes (Hainmueller et al., 2014).

Our results emphasize the critical role of supply side factors as remaining barriers to closing the gender gap in political representation, such as women’s reluctance to enter politics and discrimination by party elites and donors, as well as the weight of historical f emale under-representation through incumbency advantage. Given that voters with prior exposure to female leaders are more likely to vote for women (Baskaran & Hessami, 2018; Beaman et al., 2009; Bhavnani, 2009), recent increases in the share of elected female politicians, and the election of Kamala Harris as the first female Vice President of the United States, could foreshadow a narrowing gender gap in years to come.