A latent set of facial features unifies models of social trait perception

Humans make spontaneous judgments about who to trust, love, despise or avoid basedon facial appearance. Influential models of such social judgments posit that they are based ontwo main evaluations—the person’s intentions and their ability to carry them out. Two centralmodels characterize these dimensions as trustworthiness and dominance, and warmth andcompetence respectively. However, emerging evidence shows that the facial features that drivethese central perceptions diverge. Here, we show that disparate models of social trait perceptioncan be unified when considering social traits in terms of latent facial features. By modelling the3D facial features that drive central social trait judgments using a classic data-driven methodcombined with a modern generative model of the human face, we showed that two latent sets offacial features drive social trait perception. It is the unique combination of these sets of features that gives rise to individual trait perceptions. Additionally, we highlight differences in social traitperception not only based on face sex but also across individual observers. Together, our resultsunify existing models of social trait perception and highlight the importance of representingdiversity in modern models of human behavior.