The effects of facial appearance on measures of generalized trust

Abstract

Abstract Research suggests various associations between generalized trust and a wide range of economic, political, and social dimensions. Despite its importance, there is considerable debate about how best to measure generalized trust. One recent solution operationalizes generalized trust as the average of trust ratings across a small set of trust domains and human faces. Here, we investigate whether heterogeneity in facial appearance affects the psychometric properties of these new instruments. In a survey experiment conducted with a sample of U.S. adults (n = 5001), we randomly assigned respondents to one of five conditions that varied the features of human and AI-synthesized faces. Irrespective of the condition, respondents rated each face along four trust domains. We find that facial heterogeneity has negligible effects on the measurement validity and measurement equivalence of these new instruments. Small differences are observed on a subset of faces for some psychometric tests. These findings contribute to a growing body of work using faces to measure generalized trust, and demonstrate the utility of using AI-synthesized faces in social science research more broadly.