The structure and perceptual basis of social judgments from faces

Abstract Two questions have motivated much of our research. What is the structure of social judgments from faces and what is the perceptual basis of these judgments? Twelve years ago, we proposed a simple 2-dimensional model, according to which faces are evaluated on perceived valence and power, and introduced data-driven computational models, which reveal the physical variation of faces that drive specific judgments. First, using data from our lab collected more than 10 years later and from a large cross-cultural replication project, we show that the simplistic 2D structure model is remarkably stable and generalizes to all world regions, in which data have been collected. We also discuss misunderstandings and limitations of structure models. Second, we make the assumptions of the data-driven modeling approach transparent and discuss recent developments. The computational framework allows for precise parametric manipulation of the appearance of faces, for control of shared variance between judgments, and for relating models at different levels of face processing. We also explore analyses mapping multiple social judgments simultaneously to the physical face space. We conclude with two future directions: models of hyper-realistic faces and the underappreciated role of stable individual differences in judgments.