A High Bar or a Double Standard? Gender, Competence, and Information in Political Campaigns

With funding provided by the National Science Foundation.

See www.​processtracing.​org. A screenshot of a DPTE campaign and an open information box are provided in the Supporting Materials.

Very few differences were found between the two sub-samples or in their performance in the study. Details can be provided upon request.

For a discussion of the Primary Election, see the Online Appendix A.

Subjects were asked to choose which party’s primary election they’d like to vote in and their “in-party candidate” is considered to be the candidate that ran in the party whose primary they chose. 4 Democrats in the study chose to vote in the Republican primary, while all self-identified Republicans voted in the Republican primary. Among independents, 3 Republican leaners voted in the Democratic primary and 6 Democratic leaners voted in the Republican primary. Among pure independents, 21 voted in the Democratic primary and 18 voted in the Republican primary. I control for this in my analyses by including a “primary partisan match” variable, which is explained in detail below.

This study also included a race manipulation in which in-party candidates could also either be black or white. Because this manipulation was randomly assigned and did not affect the nature of the results for candidate gender, I leave it out of the analyses that follow.

Pictures were taken from the websites of various state legislatures and presented headshots of the candidates against a blank, single-colored background. See Online Appendix A to see images. Importantly, all in-party candidates in the general election were also “incompetent-looking.” That is, the images associated with the in-party candidate scored low in “competence” ratings in a pretest of candidate images. Pretest details are also available in Online Appendix A. The out-party candidate was always a competent-looking, white man from the out-party’s primary. This is the result of a manipulation in the primary election portion of the study in which both a competent-looking and an incompetent-looking candidate ran for the party’s nomination. In order to answer unrelated questions, the incompetent-looking in-party candidate always “won” and advanced to the general election. While this is a confound, because all in-party candidate images (both men and women) in the general election were incompetent-looking, an incompetent appearance is constant across all groups. I can therefore still compare female candidates to male candidates.

See Clifford et al. (2015), Berinsky et al. (2012), Weinberg et al, 2014, Buhrmester et al. (2011), Paolacci and Chandler (2014) and Crump et al (2013) for analyses of how MTurk samples compare to other types of internet and in-person samples. Evidence suggests that findings from MTurk studies do not differ in important ways from those conducted on other kinds of samples. Concerns about MTurk sample demographics center around the fact that MTurkers tend to be more liberal than nationally representative samples (Berinsky et al. 2012; Huff and Tingley 2015), which can pose a problem for certain kinds of studies. In the case of gender stereotypes, a more liberal sample presents a tougher case than one that is more conservative, as conservatives are more likely to hold traditional views on gender, and Republican women tend to fare worse than Democrats (King and Matland 2003; Dolan 2010).

While subjects in Experiment 1 were asked to choose which primary election they wished to vote in, which allowed true independents to choose which candidate would become their “in-party,” true independents in Experiment 2 (those who do not lean toward Democrats or Republicans) were randomly assigned to either the “Democrat” condition or the “Republican” condition. There were 55 independents to whom this applied, half of whom experienced manipulations to the Democratic candidate, half of whom experienced manipulations to the Republican candidate. I include a measure of the strength of a subject’s party ID in order to control for this.

Data and syntax used for these analyses are available on the Dataverse site for Political Behavior (https://​dataverse.​harvard.​edu/​dataverse/​polbehavior).

I also constructed models incorporating a number of covariates typical to vote choice models, including party ID, ideology, demographics and political sophistication, as well as controls for the amount of information accessed (both total information accessed and the number of competence-related items accessed) by each subject, and found that none of these influenced the pattern of results. I present these simplified models for parsimony. Full models can be provided upon request.

The likelihood of an in-party vote is lower than might be expected in a general election. This is likely due to the fact that independents are included in these analyses, as well as the fact that my manipulations were designed specifically to discourage in-party votes.