Figure 4.

Neuroimaging design and results. (A) Each fMRI trial contained one tone–face pair. To ensure attention, participants pressed a key on infrequent “oddball” trials where either the tone or face was replaced by two rapid tones or faces, respectively, and otherwise withheld their response. Oddball trials were discarded from analysis. (B) There were four cue conditions: only the face indicating sex (ΔFace), only the tone indicating sex (ΔTone), the face and tone indicating the same sex (ΔCongruent), and the face and tone indicating different sexes (ΔIncongruent). Within each condition, the sex could either be male or female; ΔIncongruent trials were labeled based on the sex of the face. We assessed the neural evidence for facial sex in each condition by attempting to discriminate voxel patterns for male and female trials using multivariate pattern classification. (C) ROIs were generated using automated meta-analyses of published neuroimaging data (Yarkoni, Poldrack, Nichols, Van Essen, & Wager, 2011). (1) left auditory cortex/STG, (2) right auditory cortex/STG, (3) left occipitotemporal cortex, (4) right occipitotemporal cortex, (5) left amygdala, (6) right amygdala, and (7) right inferior frontal gyrus. (D) Accuracy of the four classifiers (in units of d′) for each ROI. Dotted line indicates root quadratic sum of ΔFace and ΔTone d′, as in Figure 3. L/R indicate left/right hemisphere. (E) Mean percent signal change for each condition and ROI, averaged across trials, voxels, and participants. Error bars are the SEM across participants. Lines and asterisks reflect paired t tests. • p < .10, *p < .05, **p < .01, ***p < .001.