Figure 3.

Most inverted pairs do not replicate. How prevalent are true inversions of category preference in FFA and PPA? We investigated this by computing the PRIP. A, Computation of the PRIP. The bar graphs display single-image activation of right FFA (defined at 128 voxels in one subject) for each session. Colored circles connected by black arrows highlight four inverted pairs (i.e., nonpreferred image ranked before preferred image) in session 1. Only 1 of the 4 inverted pairs replicates in session 2, the other 3 revert to category-preferential order (i.e., preferred image ranked before nonpreferred image). If these 4 example pairs were the only inverted pairs in session 1, the PRIP would be 0.25. Color coding is the same as in Figure 1. B, Results of statistical group analysis of the PRIP for category-selective regions FFA and PPA and control regions hIT and EVC. The PRIP was averaged across subjects, allowing for different particular image pairs to be inverted in each subject. Since inverted pairs are defined based on the notion of category preference, the analysis was based on nonface–face pairs for FFA and nonplace–place pairs for PPA. hIT and EVC do not have a strong category preference and were tested for both types of pairs, serving as a control for category-selective regions. We used a two-sided label-randomization test to determine whether the PRIP differed significantly from 0.5; the level we expect under the null hypothesis that the apparently inverted pairs actually have equal activation. A PRIP significantly >0.5 indicates that most inverted pairs replicate. A PRIP significantly <0.5 indicates that most inverted pairs revert to category-preferential order. Results show that most inverted pairs revert to category-preferential order for FFA and PPA for most ROI sizes. The p values were corrected for multiple comparisons as described in Figure 1. Black boxes highlight the ROI sizes used in Figures 1 (FFA and PPA) and 2 (hIT and EVC).