Theoretical and methodological congruence with face perception research: an alternate paradigm for facial attractiveness

This dissertation critiques the prevalent and contemporary explanatory framework for facial attractiveness hypotheses that are based on certain assumptions of the relationships among hormones, facial growth, and immune system function. I propose an alternative explanatory framework based on face perception research. The facial attractiveness and facial perception literatures currently are not integrated; however, much of our knowledge about face recognition is relevant to understanding how people make judgments of facial attractiveness. In particular, computational methods used in face recognition are vital to testing competing facial attractiveness hypotheses. Three studies that test the two major hypotheses proposed to explain facial attractiveness, averageness and sexual dimorphism, are presented. Each study was designed to provide critical tests of these hypotheses as well as demonstrate how face representation models can be used for this purpose. Results show that both averageness and sexual dimorphism are correct, explaining different aspects of facial variation that covary with attractiveness judgments. Modeling results show that facial averageness should be construed as the degree of similarity between a face and a hypothetical gender-neutral prototype rather than a sex-specific prototype. Finally, this research demonstrates that unsupervised learning algorithms (principal components analysis and independent components analysis) can explain moderate amounts of variance in attractiveness. A supervised connectionist model, however, can explain all of the variance between faces in mean attractiveness ratings, generalizing almost perfectly to predict attractiveness judgments made to novel images of faces.