Although women constitute a majority of the American electorate and vote at higher rates than men, women hold only about 20% of seats in Congress and no woman has been elected president. The reasons for the underrepresentation of women are many: differences between men and women in resources, political networks, tolerance for political campaigns, and self-confidence. A nagging question that has been obscured by these factors is whether voters are biased female candidates running for national office. Our study addresses this question directly using experiments to determine if voters prefer a male candidate over a female candidate when other “upstream” differences between the candidates are neutralized.
Real world data make it difficult to determine if voters are biased for or against candidates of different sexes. This is because voters often infer other characteristics of candidates from visible traits such as sex. For example, absent other information, a voter might assume that a candidate who is female is more liberal and shows weaker leadership skills compared to a male counterpart. This is a simple application of broad stereotypes about the differences between men and women. Relying on these inferences, such a voter might disfavor the female candidate not because of sex per se but because of a desire for a candidate who is less liberal or has more leadership skill. If these inferences could be disrupted with additional information, does voters still show biases based on sex?
To distinguish between these kind of stereotype-based decisions and more fundamental biases, we conducted “conjoint” experiments. These experimental designs are often using in marketing studies to determine how consumers respond to different combinations of product attributes. In a similar fashion, we present respondents with opposing candidates who possess different combinations of attributes. In a nationally representative survey fielded in early 2016, we showed respondents pairs of hypothetical candidates for Congress or President. Each candidate had 13 different attributes – things such as age, race, personal style, issue positions, of course sex – randomly selected. But running the experiment many times we could estimate the effect of a candidate’s sex on voting patterns.
We find that the effects of candidate sex are highly contingent in at least two important ways. First, respondents were slightly biased against female candidates running for President (about 2.4% on average) but not against women running for Congress. We theorize that this difference between offices could be due a lack of experience with women in the White House compared to the more regular exposure to female legislators (including a former Speaker of the House). At least some voters appear reluctant to imagine how a woman would behave in an office only held by men. The bias against female presidential candidate might also reflect some residual hesitation about women serving as Commander-in-Chief due to the national security responsibilities that have traditionally been associated with men.
Second, there is a difference in how women fare in primary elections versus general elections. Our experiment provides evidence on this because half of the candidate pairs are from the same party while the other half are from different parties. In the general election setting, self-described Democrats or Republicans show little bias. These individuals instead rely on partisanship as the main decision heuristic as a powerful guide to which candidate is best. In contrast, independents display the largest bias against female candidates; this reflects the absence of party as a form of “insurance” about what a politician would do in office. The situation is different in primary settings, where Republicans and independents both display some bias against female candidates. Our study thus provides some of the first experimental evidence that equivalent male and female candidates get different responses even from the same voters depending on whether other informational cues are available. Candidate sex does not have one effect but many depending on the type of election, characteristic of the voters, and real-world experiences with women in office.