Facial Masculinity Measurement Protocol

The 20 male faces were part of a larger photoset of 62 male and female faces from the same population of adults. A geometric morphometric analysis of all of these faces was used to generate morphological masculinity scores for each face in a manner analogous to that use used for previously for bodies [47] . First, using criteria established by Stephan et al [48] , the x-y coordinates of 129 facial landmarks (Fig. S1 – supplementary material) were delineated for each face using Psychomorph [49] . Geometric morphometric techniques were then used to calculate a masculinity index for each face. Morphologika [50] (link) was used to carry out Procrustes registration of the landmark data - a best fit procedure that removes scale, rotational and translational differences between shapes [51] –[53] .
Next, to identify dimensions of variation in facial landmark configuration, Morphologika was used to conduct Principle Components Analysis (PCA) of the Procrustes-registered landmark data. A Kaiser-Guttman criterion was used to select Principle Components (PCs) for inclusion in subsequent analysis; i.e. those with eigenvalues greater than the average eigenvalue were retained. This led to the retention of the first 11 PCs which together accounted for 84.7% of the variance in facial landmark configuration (see Table S1, supplementary material for details).
Step-wise discriminant analysis (SPSS 13) was then used to establish which of the 11 PCs were best able to discriminate between the male and female faces. The resulting discriminant function incorporated eight of the PCs (Wilks' λ = 0.163; df = 8; χ2 = 101.6, p<0.00001), and yielded correct sex classifications for 96.8% of faces (see Table S1, and Fig. S2, supplementary material, for details). Discriminant function scores were therefore used as an index of morphological masculinity, with high scores indicating a more masculine facial structure (see Table S1, supplementary material for details).