Comparing Theory-Driven and Data-Driven Attractiveness Models Using Images of Real Women's Faces

Facial attractiveness plays a critical role in social interaction, influencing many different social outcomes. However, the factors that influence facial attractiveness judgments remain relatively poorly understood. Here, we used a sample of 594 young adult female face images to compare the performance of existing theory-driven models of facial attractiveness and a data-driven (i.e., theory-neutral) model. Our data-driven model and a theory-driven model including various traits commonly studied in facial attractiveness research (asymmetry, averageness, sexual dimorphism, body mass index, and representational sparseness) performed similarly well. By contrast, univariate theory-driven models performed relatively poorly. These results (a) highlight the utility of data driven models of facial attractiveness and (b) suggest that theory-driven research on facial attractiveness would benefit from greater adoption of multivariate approaches, rather than the univariate approaches that they currently almost exclusively employ. Public Significance Statement What information is used to assess facial attractiveness? Most theories of female facial attractiveness have concentrated on the possible roles of symmetry, prototypicality, and femininity. Using a large sample of 594 faces, we show that these facial characteristics individually are relatively poor predictors of women's attractiveness. By contrast, an alternative model derived from basic shape and color information of images predicted women's facial attractiveness relatively well. Our study demonstrates the importance of multivariate approaches to facial attractiveness, as well as the utility of data-driven methods in detecting previously overlooked factors that influence perceptions of facial attractiveness.