We investigate whether faces are processed holistically in ensemble judgments. We first address holistic processing operationalized as a failure of selective attention to parts. In Experiment 1 (n=35), we compared judgments of the average identity for top-halves of faces in ensembles, when the most prevalent top identity is paired with a relatively homogenous vs. diverse set of bottom parts. Bayesian evidence supported the conclusion that bottom-halves could be selectively ignored. In Experiment 2 (n=45), we compared judgments of the average identity of top-halves of faces in ensembles with or without bottom-halves present. Bayesian evidence supported the conclusion that bottom-halves had no influence on ensemble judgments about the tops. We then addressed the role of configural information, another common meaning given to holistic perception. In Experiment 3 (n=42), we measured whether the configural arrangement of the same face parts influences judgments of the average identity of faces, and compared the configural effect to the addition of one more top and one more bottom halves, presented separately. We found influences of the parts on their own and of their configuration, with the configural effect relatively small compared to the part-based effect. Our results suggest both individual face parts and their configuration influence judgments of average face identity. However, ensemble perception of faces does not limit selective attention to faces parts, a holistic effect that is well-documented in single face judgments. This supports the idea that ensemble face perception differs in important ways from single face perception. These findings also provide an empirical dissociation of two effects, failures of selective attention to parts and sensitivity to configural information, that are sometimes attributed to the same construct, i.e., holistic perception. Finally, our results also support the learned attention account, but not the representational account, of holistic processing.
Acknowledgements: This work was supported by the David K. Wilson Chair Research Fund (Vanderbilt University).