Cultural effects on the association between election outcomes and face-based trait inferences

Abstract

How competent a politician looks, as assessed in the laboratory, is correlated with whether the politician wins in real elections. This finding has led many to investigate whether the association between candidate appearances and election outcomes transcends cultures. However, these studies have largely focused on European countries and Caucasian candidates. To the best of our knowledge, there are only four cross-cultural studies that have directly investigated how face-based trait inferences correlate with election outcomes across Caucasian and Asian cultures. These prior studies have provided some initial evidence regarding cultural differences, but methodological problems and inconsistent findings have complicated our understanding of how culture mediates the effects of candidate appearances on election outcomes. Additionally, these four past studies have focused on positive traits, with a relative neglect of negative traits, resulting in an incomplete picture of how culture may impact a broader range of trait inferences. To study Caucasian-Asian cultural effects with a more balanced experimental design, and to explore a more complete profile of traits, here we compared how Caucasian and Korean participants' inferences of positive and negative traits correlated with U.S. and Korean election outcomes. Contrary to previous reports, we found that inferences of competence (made by participants from both cultures) correlated with both U.S. and Korean election outcomes. Inferences of open-mindedness and threat, two traits neglected in previous cross-cultural studies, were correlated with Korean but not U.S. election outcomes. This differential effect was found in trait judgments made by both Caucasian and Korean participants. Interestingly, the faster the participants made face-based trait inferences, the more strongly those inferences were correlated with real election outcomes. These findings provide new insights into cultural effects and the difficult question of causality underlying the association between facial inferences and election outcomes. We also discuss the implications for political science and cognitive psychology.

Fig 1. Experiment procedure and image display screens.

(a) The schematic diagram of the full experiment. (b) Example of image display screens in the competence evaluation block. At the beginning of each block, there were instructions on the screen indicating which trait the participant was asked to evaluate. Then for each pair of candidates, participants first focused on the cross of the fixation screen which lasted for 1–2 seconds; then the images of the pair of candidates were up for 1 second; participants could make a decision as soon as the images appeared; after the images disappeared, participants had a maximum of 3 seconds to enter their choice. As soon as a valid key was pressed (i.e., press “A” if their choice was the candidate on the left and press “L” if their choice was the candidate on the right), a grey screen was up for a 1 second inter-stimulus interval.

Fig 2. The number of participants whose trait inferences agreed with the outcomes of more electoral races in one country than the other.

The blue histogram represents the numbers of participants whose trait inferences agreed with the outcomes of more U.S. than Korean elections. The orange histogram represents the numbers of participants whose trait inferences agreed with the outcomes of more Korean than U.S. elections. For brevity, the category U.S. = Korean was omitted from the graph. All participants (N = 80).

Fig 3. Distribution of response times and average agreement.

The histogram represents the distribution of response times over all trials (n = 28540) across all participants, candidate pairs, and traits, excluding missing data, data for recognized candidates, and seven trials with response times less than 100 milliseconds. The line represents the average agreement over trials with response times within the given interval, omitting those for response-time intervals with less than 200 trials.