Perception Action Cognition Occasional Talks

Estimating the “average race” in a dynamic crowd of faces

Inspired by recent studies of “ensemble” face processing, we asked whether observers could accurately estimate the ethnic composition of an array of faces. We also examined whether such estimates were influenced by the race of the observer. We developed a new task in which a set of 16 faces was continuously shuffled within a moving 4 x 4 grid. This design prevents explicit estimation and blocks spatial sub-sampling. Across trials we systematically varied the proportion of Asian and Caucasian faces. Two groups of observers were assigned different target and distractor races. Their task was to indicate whether there were more targets or distractors present on each trial. Each group consisted of 8 Asian and 8 Caucasian observers. We fitted cumulative normal functions to the response distributions and extracted the PSE and JND as dependent measures. The same 2 (Observer Race) x 2 (Target Race) ANOVA was used for analysis. For PSE, there were main effects of Observer and Target that did not interact. Observers consistently weighted other-race faces more heavily than own-race faces and PSEs were generally shifted in the direction of the Target. For JND, there was a main effect of Observer and a Observer x Target interaction. While Asian observers were generally more sensitive, both groups had smaller JNDs when the target was the other-race. Overall, our findings indicate that observers can rapidly estimate the racial composition of a group of faces, but these estimates are influenced by their own race.