Measuring sexual dimorphism in human faces.

Facial shape is one of the most widely studied sexually dimorphic traits in humans. Sexually dimorphic facial shape has been linked to processes in neurodevelopment, immunocompetence, social perception, and mate preference. However, research into these associations has produced conflicting results, owing in part to the diverse methods used to quantify sexual dimorphism of the face. Our study compares two commonly used methods for measuring morphological sexual dimorphism: regression scoring and Canonical Variates Analysis (CVA; or linear discriminant analysis). We test both methods on a large sample of adult males (n = 540) and females (n = 540) with three-dimensional (3D) descriptions of the whole face, both with and without prior decomposition of the allometric component. Our results show that CVA outperforms regression scoring, resulting in scores that are more accurate in classifying the sexes and recreating the male-female shape axis (i.e., the difference in shape means based on reported sex). We also find that height is positively associated with regression scores after controlling for sex (p < 0.01), but not with CVA scores. These results suggest the need for a possible reassessment of previous claims that taller males have more male-like facial shapes, as well as a broader re-evaluation of the literature that considers the significance of method selection in shaping research outcomes. We establish a foundation for more accurate comparisons of facial sexual dimorphism and its relationship to various domains of human health and biology.