Fig. 4: Results of PCA on CSVA model feature weights, across OTC voxels.

A A group-level principal components analysis (PCA) was conducted on CSVA model feature weights across all OTC voxels where model fit was significant and better than that of the Semantic Only model. The scree plot shows the amount of variance explained by each of the top ten PCs (in red). PCs from a PCA analysis conducted on stimulus features (using the combined design matrix from all 6 subjects) are shown in black. Asterisks indicate group PCs that explain significantly more variance than the stimulus PCs (one-tailed jackknife test, *p = 0.03, this is the smallest possible p value given the jackknife test used), see Methods for details. B Results of a leave-one-out cross validation analysis of the similarity in feature loadings between individual subject PCs and group PCs. The correlation matrix presented gives the correlation of feature loadings for the top three PCs extracted from PCA conducted on each individual subject’s data and the top 3 PCs from the group-level PCA conducted on the data from all remaining subjects. All correlations shown are significant at p < 1E-8 (assessed by one-tailed permutation tests) except for subject 6 PC 1 where p = 0.0065, see Table S2). This indicates a shared representational structure across subjects.