There has been attempts to document the effect of smiling by candidates in ballot pictures on votes for candidates. However, existing studies are plagued by confounding by comparing across candidates with differences in smiling and many other characteristics. We vastly improve over the previous studies by using an algorithm to code smiling (and beauty/feelings) in pictures of politicians in Colombian mayoral elections across several elections. This allows us to estimate the effect of smiling for the same candidate in different elections substantially removing confounding by using candidate fixed effects. Furthermore, we control for time varying confounders that could lead to candidates to change their smile across elections. After coding 8806 pictures, we find that if a candidate did not smile (smile index = 0) and later had a smile (smile index = 1), this can increase the vote share by 2.4 percentage points, which is a sizeable amount. Moreover, using measures of beauty, we explore how the effect of smiling can be boosted by candidate’s beauty. This aims to tie the separate literatures on smiling and voting, and beauty and voting, and shows how a simple smile can make a difference in who gets more votes.
Existing literature posits that, in low-information settings, voters base their choices on candidates’ looks. This often favours disproportionately right-wing politicians who happen to enjoy an appearance premium. We argue that this is endogenous to politicians self-selecting into parties. We show that the Republican party is a more homogenous looking party and individuals most similar to the 'ideal' Republican stereotype benefit from an advantage in open-seat elections, compared to Democrats. The electoral premium enjoyed by Republicans is due to a greater similarity between its members. As a result, Republican voters spot more easily 'in-group' members and use this similarity as a cue for their choices in low-information settings. Additionally, we predict that the winner in the Republican primaries is the one looking most like the `average’ Republican, whereas this is not the case for Democrats. Using machine learning analysis of portraits of winners and runner-ups for the US Congress and Senate we confirm these predictions.