Fig. 5

PE encoding regions predictive of memory updating. a For the subset of questions answered incorrectly in Test1 and accurately in Test2 (i.e., successful learning from feedback), confidence ratings in Test2 answers were highly correlated with responses observed in the VS and DLPFC ROIs during feedback (Supplementary Table 1). On the right, average single-trial beta plots for these ROIs show that the greater the reduction in activity elicited by negative feedback in Test1, the stronger was the confidence in correct (blue), but not incorrect (red) Test2 answers. b For all questions answered erroneously in Test1, the response to negative-feedback information in cingulate, right IPL, and right putamen PE clusters, was predictive of subsequent accuracy in Test2 (Supplementary Table 1). Greater deactivation in these regions was associated with feedback that was subsequently incorporated into memory vs. feedback that was not (incorrect-to-correct vs. incorrect-to-incorrect trials). Subsequent memory correlative activity was not detected in unsigned PE ROIs