Fabian Soto¹, Emily Martin¹; ¹Florida International University, Miami, FL

Race influences both recognition and categorization of faces, yet most prior work relies on stimulus sets in which race and identity covary, preventing clear causal inferences about which facial properties drive these effects. We developed a controlled stimulus framework enabling independent manipulation of race and identity, allowing direct tests of how race-specific morphology shapes perception. Using anthropometrically validated 3D models of Caucasian, Asian, and African facial structure—together with an “alien” race matched in variability but unfamiliar to observers—we generated stimuli in which identity information was held constant across races while only race-specific shape varied. In Study 1, participants completed both a recognition-memory task assessing the other-race effect (ORE) and a race-categorization task assessing the other-race categorization advantage (ORCA). Because identity was matched across races, any differences reflect race morphology alone. Under these controlled conditions, we observed no ORE, but the ORCA remained robust. In Study 2, reverse correlation was used to test whether the features supporting identification of a familiar face differed across races. Despite clear visual differences among races, observers relied on the same identity-diagnostic information across race versions of the face, consistent with the absence of an ORE in Study 1. In Study 3, discrimination thresholds were collected across key regions of face space to compare population-encoding models. Encoding of familiar identities remained stable across races, but own-race faces structurally similar to the other race were easier to distinguish from the other race and harder to distinguish from nearby own-race identities. A model incorporating tuning shifts toward the other-race manifold best explained these effects. Together, these studies show that when race is isolated from identity, the ORCA persists while the ORE disappears, and representational changes are localized near the other race. These results suggest that identification and categorization biases arise from distinct underlying sources of information.

Acknowledgements: This research was funded by NSF grant 2319234 awarded to Fabian Soto.