Candidate Faces and Election Outcomes Prepared for Publication on Andrew Gelman’s Blog

Recent research finds that inferences from candidate faces predict aggregate vote margins. Many have concluded this to mean that voters choose the candidate with the better face. We implement a survey with participant evaluations of over 167,000 candidate face pairings. Through regression analysis using individualand district-level vote data we find that the face-vote correlation is explained by a relationship between candidate faces, incumbency, and district partisanship. We argue that the face-vote correlation is not just the product of simple voter reactions to faces, but also of party and candidate behavior that affects which candidates compete in which contests. Recent work in psychology demonstrates that the naive, rapid evaluations by survey participants of the facial competence of candidates linearly predict United States Congressional candidate vote share (Todorov, Mandisodza, Goren, & Hall 2005). Todorov et al. (2005) showed college students the faces of candidates contesting United States House and Senate elections in 2000, 2002, and 2004, and asked them to unreflectively choose the more competent looking candidate in each contest. They find that the candidate more frequently selected as appearing competent won the actual election in 66.8 percent of House and 71.6 percent of Senate contests. They also find that the proportion of paired evaluations in which one candidate’s face is judged more competent than the opponent’s correlates to the difference in vote share between the two candidates in both the House (r = .40) and in the Senate (r = .44). In Figure 1 we reproduce Todorov et al.’s (2005) main finding using participant evaluations to predict Democratic vote share. Similar findings have subsequently shown that candidate faces predict gubernatorial elections (Benjamin & Shapiro 2006, Ballew II & Todorov 2007) and that executive faces predict corporate profits (Rule & Ambady 2008). Although Todorov et al. (2005) were careful to discuss their results as predictive rather than causal, there has been a tendency among popular commentators and some scholars to interpret the results as an indictment of the ability of American voters to make reasoned decisions. An item in The New York Times Magazine, writing about Little, Burriss, Jones, & Roberts (2007), declared that “Faces Decide Elections” (Skloot 2007), and a National Public Radio segment suggested we “[f]orget political polls . . . voters prefer candidates who look competent, even if they are not” (Hamilton 2005). In fact, some have implied the need for electoral reform so that voters will not be duped by candidates’ faces (Zebrowitz & Montepare 2005). Todorov et al. (2005) measure relative candidate facial competence in a laboratory environment. This innovative approach to measuring a candidate characteristic may have led We thank Alex Todorov for generously sharing with us his data.