Resemblance and Discrimination in Elections

Abstract

Discrimination affects hiring, mating and voting decisions. Whilst discrimination in elections mainly relates to gender or race, we introduce a novel source of discrimination: candidate resemblance. When candidates' partisanship is not known, voters select those that resemble most elected co-partisans. Using a machine learning algorithm for face comparison, we find a stronger resemblance effect for Republicans compared to Democrats in the US. This happens because Republicans have a higher within-party facial resemblance than Democrats, even when accounting for gender and race. We find a similar pattern in the UK, where Conservative MPs are more similar looking to each other than Labour. Using a survey experiment, we find that Tory voters reward resemblance, while there is no similar effect for Labour. We estimate that facial dissimilarity decreases the candidate's re-election probability by 5-14 percentage points. The results are consistent with an interpretation of this behaviour as a form of statistical discrimination.