Cross-Domain Classification of Facial Appearance of Leaders

Abstract

People often rely on visual appearance of leaders when evaluating their traits and qualifications. Prior research has demonstrated various effects of thin-slicing inference based on facial appearance in specific events such as elections. By using a machine learning approach, we examine whether the pattern of face-based leadership inference differs in different domains or some facial features are universally preferred across domains. To test the hypothesis, we choose four different domains (business, military, politics, and sports) and analyze facial images of 272 CEOs, 144 4-star generals of U.S. army, 276 U.S. politicians, and 81 head coaches of professional sports teams. By extracting and analyzing facial features, we reveal that facial appearances of leaders are statistically different across the different leadership domains. Based on the identified facial attribute features, we develop a model that can classify the leadership domain, which achieves a high accuracy. The method and model in this paper provide useful resources toward scalable and computational analyses for the studies in social perception. © 2020, Springer Nature Switzerland AG.