Shangcheng Zhao

Presentation Title

Social impressions and election: facial appearance and beyond

Abstract

Facial impressions of competence have been shown to predict political election outcomes above chance. However, real-world voting decisions integrate information beyond facial appearance. In this study, 300 U.S. citizens were recruited prior to the 2024 election. Participants briefly viewed images of 140 House of Representatives candidates and reported recognition, perceived job capability, and voting intentions. Consistent with prior research, we replicated a 64% prediction accuracy of election outcomes based solely on facial appearance. Extending beyond facial cues, we examined how visual and semantic information contribute to subjective evaluations and actual vote share. Using deep neural networks and large language models, we extracted visual features from candidate images and semantic content from 2024 web news. Results show that while visual and semantic features similarly predict competence judgments in zero-acquaintance contexts, semantic information plays a larger role when candidates are recognized. Additionally, semantic information was a stronger predictor of actual vote share. These findings underscore the crucial role of semantic information in both impression formation and real-world electoral decisions.