Hey Good Lookin: Does facial competence actually matter for election outcomes?

Are politicians successful when they are better looking? Todorov and colleagues (here and here) created a stir by demonstrating that election outcomes can be predicted based on very quick evaluations of the “competency” of candidates' faces. Could it be that voters are so superficial that a great share of them are merely persuaded by the attractiveness of candidates?

A recent article in the Quarterly Journal of Political Science takes on this question with a clever study. Atkinson et al. note that previous studies have not accounted for the possibility that candidate selection depends on the competitiveness of a race. They in fact show that challenging parties tend to run better-looking candidates when they have a greater chance of winning.
They demonstrate this in two-steps. First, they develop a survey instrument to estimate the competency of candidates’ faces. The method is clever. They show subjects a randomly drawn pair of papers and ask them to indicate which candidate looks more competent. They then derive a competency score that assumes some transitivity on competence, assumes it is probabilistic and not determinative, and assumes some continuum of competence. They develop their own estimator to score candidates (though using the existing Bradley-Terry method would have been easier!).

Next, they marry these scores to exit poll surveys to estimate the effect of competence on different types of voters, i.e. independents and partisans. They show that there is no competence effect for House races, but small effects in Senate races. These effects are larger for independent voters than for partisans. Finally, they use these estimates to test whether facial competence could have been pivotal in any Senate races. They find it was not.

A few thoughts on this paper and the facial competence literature more generally. First, I quite like that Atkinson et al have noted the selection effects which undermine the bold claims made by earlier papers. There is a general lesson in this: we can elicit large effects in laboratories. We should search for these effects, but we should be sure to test their external validity as close to actual vote choice as possible. What is more, they develop a nice instrument for replicating these findings in other countries. I should hope we could do something similar in Canada and other democracies, though I think a Bradley-Terry estimation could just as easily be used. Second, I don’t quite understand why such studies don’t also test for the effects of whether voters find some faces more “attractive” than others. Or, why not ask voters “Which person looks like they would understand your needs better?” There are any number of qualities we can infer from faces and I see no reason why competence should be the only one, or even chief among them.