Megan Lall¹ (), Eric Mah¹, Brett Roads², Jim Tanaka¹; ¹University of Victoria, ²University College London

Most people are experts at recognizing faces, however, research has shown that we are less expert at recognizing faces from other races - the Other-Race Effect (ORE). An influential framework to account for the ORE is the face space model where faces are represented as points in a multi-dimensional similarity space with each dimension signifying a specific facial feature or attribute. According to this approach, own-race faces are perceived as more distinct and therefore, their representational points are more spatially separated in face space. In contrast, other-race faces are perceived as more similar and therefore, their points are more densely clustered in face space. To examine the face space representations of participants for own- and other-race faces, we employed PsiZ, a novel method for obtaining psychological embeddings–rich multi-dimensional representations of psychological similarity spaces that are inferred from behavioural similarity judgements. We predicted that the psychological embeddings will be more differentiated for own-race faces and more densely packed arrangements of other-race faces. For this study, we recruited 60 African, 60 Caucasian and 60 Chinese online participants who made similarity judgments to blocks of 20 African, 20 Caucsian and 20 Chinese faces. Our main results indicated that there was limited evidence to support the face space account of the ORE. Inspection of the psychological embeddings by race of the participant and face showed that own-race faces were not more differentiated in face space than other-race faces. However, upon examining faces by different racial groups, distinct patterns emerged. African participants exhibited an ORE for Caucasian faces, while Caucasian participants demonstrated an ORE for Chinese faces. Notably, Chinese participants did not display a discernible ORE, indicating variability in cross-race recognition effects among the studied groups. Participant similarity judgements were moderately correlated with simulated judgements based on VGG-16 perceptual features, with some differences by race.