Abstract
Visual processing of single faces is associated with reduced attentional demands relative to non-face objects, but evidence that this applies to multiple faces is limited. Although face ensemble studies show that multi-face information can be extracted in parallel as a summary statistic, whether we can divide attention across multiple faces to support single-face tasks is not known. We used a divided attention paradigm in which participants made three different kinds of face judgments: (1) gender (male/female), (2) orientation (upright/inverted), and (3) color (red/gray). Judgments were made for either one (single-task) or two (dual-task) faces in a simultaneously-viewed face pair. Our paradigm enabled us to measure dual-task deficits for each type of judgment and to assess the magnitude of deficits with respect to three benchmark models. The unlimited capacity parallel model predicts little if any deficit. The all-or-none serial model predicts a deficit magnitude consistent with an ability to process only one face at a time. Finally, the fixed-capacity model predicts an intermediate result. In addition to measuring dual-task deficits, we also examined the relationship between the two responses in the dual-task condition. Specifically, we used conditional accuracy to test for evidence of response trade-offs, which we again assessed with respect to the three benchmark models. Consistent with previously published results, the unlimited capacity parallel model was the best predictor of the color judgment results. Dual-task deficits for both the gender and orientation judgments were surprisingly small and consistent with the predictions of a fixed-capacity model. The fixed-capacity model also best predicted the trade-off results for gender judgments, but the unlimited-capacity model best predicted the orientation judgment results. Our overall findings show no evidence of all-or-none serial processing of face pairs, unlike results reported in similar divided attention studies of semantic categorization for non-face objects.