SDC was calculated from the models’ softmax likelihood by equalizing selleck compound Pc/a for choosing and avoiding using the following equation: SDC = abs(Pc/a − 0.5) × 2. The result ranged from 0 (maximal insecurity) to 1 (absolute preference of one option). Feedback-locked data were analyzed separately for the categorical conditions fictive and real. Predictors included

the PE (δt), variable learning rate (αt), and a dichotomous regressor indicating a switch of response (coded as 1) or a stay (coded as 0) on the next trial that the same stimulus was shown again. Standardized b values can be assumed to be Gaussian due to the central limit theorem and thus could be tested via two-tailed CT99021 purchase one-sample t tests, which were done separately at each data point in a whole-brain approach across subjects. Resulting p values were corrected for multiple comparisons using false discovery rate (FDR) following the method suggested by Benjamini and Yekutieli (2001), which has been shown to provide solid control of the family-wise error rate (FWER) in EEG data ( Groppe et al., 2011). However, as FDR in itself does not provide strong (local) control of the FWER, it was applied to all concatenated b value data sets per model. This ensured that

all corrections were done with the same threshold value for each regressor in the models. H0 was rejected for all p < 0.00070 in the feedback Farnesyltransferase and p < 0.00045 in the stimulus-locked model. Nonsignificant data points are masked in white in the topography plots and Movie S1. Both conditions in the feedback-locked epochs were contrasted via paired two-tailed t tests thresholded at the same level as noted above. We compared both real and fictive feedback processing directly via paired two-sided t tests of the regression b values,

thresholded at the same level determined by FDR. This revealed that feedback processing indeed differed significantly for all PE effects. The late parietal effect did not differ significantly when it was inverted for fictive feedbacks, assuming that counterfactual thinking was employed (by multiplication with −1) before contrasting. Contrasts for alpha and switch regressors did not reveal significant differences between both conditions. Artifact-free raw EEG was averaged from 370 to 430 ms at electrode (Pz) that showed the biggest overlap between effects of the switch, PE, and learning rate predictors in the regression analysis (Figures 4C, 4D, and S4) and SDC effects locked to stimulus onset. As we observed a positive covariation in the regression analysis for switching behavior, we hypothesized that higher EEG amplitudes should be associated with a higher likelihood to switch. Additionally, because the absolute EEG amplitudes differed between both conditions (Figure 3), the analyses for real and fictive feedback were performed separately.