Implicit race bias decreases the similarity of neural representations of black and white faces

Abstract

Implicit race bias has been shown to affect decisions and behaviors. It may also change perceptual experience by increasing perceived differences between social groups. We investigated how this phenomenon may be expressed at the neural level by testing whether the distributed blood-oxygenation-level-dependent (BOLD) patterns representing Black and White faces are more dissimilar in participants with higher implicit race bias. We used multivoxel pattern analysis to predict the race of faces participants were viewing. We successfully predicted the race of the faces on the basis of BOLD activation patterns in early occipital visual cortex, occipital face area, and fusiform face area (FFA). Whereas BOLD activation patterns in early visual regions, likely reflecting different perceptual features, allowed successful prediction for all participants, successful prediction on the basis of BOLD activation patterns in FFA, a high-level face-processing region, was restricted to participants with high pro-White bias. These findings suggest that stronger implicit pro-White bias decreases the similarity of neural representations of Black and White faces.

Fig. 1

Classification performance for Black and White faces in the regions of interest. Error bars represent ±1 SE. The brain images illustrate the location of the three regions in which classification was significantly better than chance (p < .05). OP = occipital pole; OFA = occipital face area; FFA = fusiform face area; AMY = amygdala; STS = superior temporal sulcus; CAUD = caudate; IFG = inferior frontal gyrus; INS = insula; OFC = orbitofrontal cortex.

Fig. 2

Link between implicit pro-White bias and ability of the classifier to predict the race of faces participants were viewing. The scatter plot (a) presents implicit bias as a function of correct classification performance in fusiform face area (FFA); the best-fitting regression line is also shown, β = 0.61, p = .025. The bar graph (b) shows correct classification performance in occipital pole (OP), occipital face area (OFA), and FFA separately for participants with low and high scores on the Implicit Association Test (IAT), as determined by a median split. Error bars indicate ±1 SE. The asterisk indicates a significant difference between groups (p < .05).