In the face of diversity: Face ethnicity influences the use of face features for social trait perception

Psychological science is constrained by a lack of diversity (e.g. Cook & Over, 2021). One notable example is social trait face perception research (e.g., trustworthiness, dominance; e.g., Oosterhof & Todorov, 2008) which is based primarily on White faces (but see e.g., Sutherland et al., 2018). Recent work suggests that face ethnicity influences these social judgements (e.g., Freeman & Johnson, 2016; Xie et al., 2021) though it remains unknown how this affects the causal features that subtend social perception. To examine this, we modelled the 3D face features (shape/complexion) that drive perceptions of trustworthiness and dominance in Black, East Asian, and White faces using reverse-correlation (Zhan et al., 2019). In a between-subjects design, we generated 2400 face identities per face ethnicity by adding randomly sampled principal components representing individual identity variance to an average face (Black, East Asian, or White; gender-balanced). Participants (N=60, White Western, gender-balanced) rated each face on trustworthiness and dominance (e.g., very submissive-very dominant) in separate tasks. To model the 3D face features that drive these perceptions, we linearly regressed the stimulus features presented on each trial with each participant’s responses. We then compared the resulting 3D face models across face ethnicity (N=20 per ethnicity, social trait, sex of face) using a combination of data-reduction and machine learning techniques. Results revealed that social trait perception is driven by a core set of facial features (e.g., affect-related cues: frowning/smiling) plus ethnicity-specific variations, including the mouth in Black faces and the eyes in East Asian faces. Our results provide new insights into how demographic facial cues influence social trait perception with direct implications for current theoretical accounts, highlighting the importance of diversity in psychological science (Jack et al., 2018).