Cognitive control, attention, and the other race effect in memory

Abstract

People are better at remembering faces from their own race than other races-a phenomenon with significant societal implications. This Other Race Effect (ORE) in memory could arise from different attentional allocation to, and cognitive control over, same- and other-race faces during encoding. Deeper or more differentiated processing of same-race faces could yield more robust representations of same- vs. other-race faces that could support better recognition memory. Conversely, to the extent that other-race faces may be characterized by lower perceptual expertise, attention and cognitive control may be more important for successful encoding of robust, distinct representations of these stimuli. We tested a mechanistic model in which successful encoding of same- and other-race faces, indexed by subsequent memory performance, is differentially predicted by (a) engagement of frontoparietal networks subserving top-down attention and cognitive control, and (b) interactions between frontoparietal networks and fusiform cortex face processing. European American (EA) and African American (AA) participants underwent fMRI while intentionally encoding EA and AA faces, and ~24 hrs later performed an "old/new" recognition memory task. Univariate analyses revealed greater engagement of frontoparietal top-down attention and cognitive control networks during encoding for same- vs. other-race faces, stemming particularly from a failure to engage the cognitive control network during processing of other-race faces that were subsequently forgotten. Psychophysiological interaction (PPI) analyses further revealed that OREs were characterized by greater functional interaction between medial intraparietal sulcus, a component of the top-down attention network, and fusiform cortex during same- than other-race face encoding. Together, these results suggest that group-based face memory biases at least partially stem from differential allocation of cognitive control and top-down attention during encoding, such that same-race memory benefits from elevated top-down attentional engagement with face processing regions; conversely, reduced recruitment of cognitive control circuitry appears more predictive of memory failure when encoding out-group faces.

Fig 1. Task design and behavioral performance.

a) Representative stimuli and trial structure for encoding and subsequent memory (retrieval) task phases, separated by ~24 hr. b) Top graph: subsequent memory performance for same-race and other-race faces, by participant race. Bottom graph: distribution of same-race and other-race subsequent memory performance (d’), by participant race (purple and green points). Same-race and other-race d’ were strongly correlated (r = 0.72), and participants with greater same-race performance tended to demonstrate larger Other Race Effects (same-race–other-race d’; i.e., slope of the regression line (red line) < 1). AA = African American; EA = European American; Error bars reflect group SEM.

Fig 2. Subsequent memory effects.

Activity during the first encoding presentation of each face was predictive of subsequent memory success in regions previously implicated in successful encoding. Bar graphs: mean parameter estimates extracted from significant clusters, separately for same- and other-race face encoding trials (Error bars reflect group SEM). Activity is rendered on the 3D Caret inflated cortical surface or the 2D mean across-subject anatomical image, both in standardized MNI space; height and extent thresholds: p < 0.005, k = 33, FWE p < 0.05. r = remembered; f = forgotten; Hipp = hippocampus; VLPFC = ventrolateral prefrontal cortex.

Fig 3. Subsequent Memory Effects as a function of the Other Race Effect.

a) Left panels: 3D renderings of dorsal attention network (DAN, green) and cognitive control network (CCN, orange) ROIs. Right panel: Group-level SME “parent” seeds (purple) for PPI analysis (Fig 3c). b) Univariate activity as a function of subsequent memory and same-/other-race face status. There was a tendency towards greater SMEs for other- than same-race faces in left frontoparietal components of the CCN; OREs on the SME were qualitatively similar, but quantitatively reduced, in frontoparietal components of the dorsal attention network. Specifically, the ORE on the SME reached significance in lateral IPS (CCN) and was at trend level in IFS (CCN). c) PPI SME analyses as a function of the ORE. Right fusiform SME connectivity during encoding was greater for same- than other-race faces in the bilateral parietal DAN nodes (right hemisphere not shown); qualitatively similar patterns were observed in the CCN. r = remembered; f = forgotten; SFS = superior frontal sulcus; IFS = inferior frontal sulcus; mIPS-SPL = medial IPS and superior parietal lobule; lIPS = lateral IPS. Error bars reflect group SEM. * = ORE × SME interaction significant at p<0.05, ~ = p<0.10.

Fig 4. Other Race Effects on encoding activity, collapsing across memory success.

Voxel-level comparison of overall activity for same- vs. other-race encoding events revealed a large cluster spanning parietal components of the DAN and CCN. Height and extent thresholds: p < 0.005, k = 33, FWE p < 0.05. DAN (green) and CCN (orange) ROIs overlaid for reference.

Fig 5. Voxel-level subsequent-memory predictive fusiform functional connectivity.

a) Same-race face encoding success was predicted by widespread functional interactions between fusiform cortex and cortical networks associated with top-down attention and cognitive control. Other-race face encoding did not show similar memory-related functional interactions with right fusiform cortex. b) Consistent with the ROI-based analysis, left IPS showed significantly greater subsequent-memory predictive fusiform connectivity for same- than other-race faces. DAN (green) and CCN (orange) ROIs overlaid for reference. PPI effects rendered as described in Fig 2; height and extent thresholds: p < 0.005, k = 33, FWE p < 0.05.