Candidate Gender Composition and 2018 Midterm House Race Outcomes
By Alexander Agadjanian (@A_agadjanian)
Much has been made about the record number of women who ran for Congress in 2018. But how did this greater gender diversity--and different candidate gender scenarios--play out in eventual election outcomes? Below is a quick, initial illustration of this relationship.
For the 387 House races in my dataset in which candidates from both major parties ran (i.e. received votes), I regress Democratic two-party vote percentage (= 100 * Dem% / (Dem% + Rep%)) on a primary four-category predictor that captures candidate genders and two controls. The predictor has four categories to represent the four combinations of two-party candidates' genders (1. male Republican vs. male Democrat—baseline reference category in the regression—2. female Republican vs. female Democrat, 3. male Republican vs. female Democrat, and 4. female Republican vs. male Democrat). The set of controls include a dummy variable for whether or not the House race is for an open seat and a dummy for whether the seat was previously controlled by Republicans. The coefficients from this model with 95% confidence intervals are shown in the below graph.
As expected, if a Republican had previously won the district, the Democratic vote share in 2018 was much lower (27.36 points less) than if a Democrat had previously won it (where Democratic vote share was 66.93 percent, all else equal in the model). The main focus, though, are the middle coefficients—those capturing Democratic vote share associated with different gender candidate pairings. The comparison group is a race where a male Republican and male Democrat ran against each other. Democratic vote share was lowest in this group, given the positive signs on the three coefficient terms you can see in the graph. For opposite gender pairings, the vote share "effects" are pretty small and do not attain statistical significance at the 0.05 level. However, for a race that involved a female Republican running against a female Democrat, Democratic two-party vote share was 7.48 points higher (a difference that is statistically significant). In other words, controlling for the district's prior winner and whether it was an open seat, Democrats did better when two female candidates ran against each other than when two male candidates ran against each other.
The evidence presented here should just be treated descriptively, as there are many other factors left unaccounted for that can only be addressed with a rigorous research design (like this one). However, to speculate, past work might suggest that Republican female candidates—having won their party’s nomination—perform worse in general elections because they receive less support from their party’s elite and donor ranks. On the other side of the aisle, perhaps nominating a female candidate motivated greater turnout and support from Democrats in 2018. Those two dynamics would work in tandem to produce the result seen above. But again, that’s only speculative, and more research needs to be done to better understand what’s at work here.
As suggested by a reader, a more appropriate way to model the relationship in this piece is it use a continuous measure of 2016 vote choice as a control variable, instead of the dummy variable indicating the party that previously controlled the district. When controlling for a measure like this--specifically, 2016 presidential vote share at the district level--the significant effect noted in the piece goes away (2016 presidential vote was related to both 2018 house vote and 2018 gender candidate composition).
Alexander Agadjanian (@A_agadjanian) is a research associate at the MIT Election and Data Science Lab.