Figure 3.

Prediction Errors Are Coded Relative to SD

(A) Normalized RPE responses. Significant coding of normalized RPEs (I.e., RPE/SD) after removing all shared variance between normalized and non-normalized RPEs. To facilitate comparison of normalized and non-normalized regressors, parametric modulators were Z scored prior to model estimation. Z scores were calculated per subject, across all SD conditions. Whereas BOLD responses were significant on cluster level as well as in the a-priori-defined ROI in the midbrain (SN/VTA complex), activity in the ventral striatum only became significant when we decreased the search volume to a 9-mm sphere centered on the peak area showing significant adaptation in the previous analysis (see Figure 2).

(B) Increased average (± SEM) RPE coding slopes (parameter estimates) for normalized RPEs compared to (non-normalized) RPEs in the SN/VTA complex and ventral striatum. Coding slopes for the SN/VTA complex and ventral striatum were averaged over all voxels in the a-priori-defined ROI. Data were Z scored per participant across parameter estimates for normalized and non-normalized RPEs to control for potential outliers.

(C) Significant learning rate coding slopes in a cluster including the occipital cortex and cerebellum.

Vstr, ventral striatum; SN/VTA, substantia nigra/ventral tegmental area; param. est., parameter estimates. Ψ, neural adaptation.