Abstract
Individuals vary widely both in their ability to recognize faces and in how they rate faces for dimensions such as attractiveness. We examined the pattern of differences for typical categorical judgments made about a face, such as the sex or race of the person. Face images were from the Chicago Face Database for individuals labeled as male or female and Asian or White, with 10 faces selected from each of the 4 categories. These were cropped to remove external features, and then paired and morphed to form 40 stimulus sets. The morphed faces for a set were shown simultaneously as a graded series arranged in a circle on a display, and spanned the two original identities in steps of 0.05. Participants selected the face closest to the category boundary for each set, with all stimuli for a given category (sex or race) randomly intermixed and with the morph configuration rotated randomly each trial. Reliability for repeated measurements was high (r~0.9), with average between-subjects variance 3x the within-subjects level. For each of the categories, the boundaries selected depended significantly on both the observer and the stimulus pair, and interobserver correlations were generally weak across stimuli showing that the observers did not merely differ in a general bias but rather in face-specific biases, similar to a pattern that has been found for trustworthiness (Sutherland et al., 2020). These results suggest that observers may partly use distinct and idiosyncratic strategies for making basic categorical distinctions between faces.