Article Figures & Data
Figures
- Figure 1.
Experimental paradigm. Examples of experimental trials with either partial (top rows) or complete (bottom rows) feedback. Outcome types (factual vs counterfactual) are also specified. From these two factors (i.e., feedback and outcome type), three different experimental conditions originate: partial factual, complete factual, and complete counterfactual outcomes.
- Figure 2.
Behavioral results. The graphs on the top row display the correct choice rate in the partial (left) and complete (right) information conditions for the two contexts. In the bottom row, mean reaction times in partial (left) and complete (right) information trials are depicted.
- Figure 3.
Univariate and multivariate effects in the selected ROIs. A, The selected ROIs are depicted. fMRI analyses were conducted on nine preselected anatomic ROIs along the dorsal/ventral axis of mPFC. The set of ROIs comprises one dorsal, one middle, and one ventral area for each of three target regions: SFG, orbitofrontal/frontopolar cortex, and cingulate cortex. pOFC, posterior OFC; FPC, frontopolar cortex; vACC, ventral ACC; dACC, dorsal ACC; MCC, middle cingulate cortex; vSFG, ventral SFG. B, Results of the univariate analyses on outcome valence are shown. ROIs on the x-axis are arranged from the most ventral (left) to the most dorsal (right), according to their mean z-coordinate. The dashed line indicates a null effect. ROIs where there was a significant difference between counterfactual outcomes and both factual outcomes are highlighted. C, Univariate effects are shown only for outcomes in trials with complete feedback. The order of the ROIs on the x-axis is as in B. The interaction effect outcome type × ROI was significant, reflecting positive encoding of value for factual outcomes and negative encoding for counterfactual outcomes in more anterior mPFC ROIs but the opposite pattern in the dorsal regions. D, The graph displays the results of the multivariate analyses on outcome valence. The dashed line marks chance level (rounded-up to 50, actual chance level calculated with shuffle tests is 49.997). ROIs on the x-axis are arranged as in B and C. No ROI is highlighted on this graph as neither the main effect of ROI nor the interaction effect outcome type × ROI was significant in this analysis. Error bars display SEM. ***p < 0.001, **p < 0.01, *p < 0.05.
- Figure 4.
Comparison of different value coding models. A, Three different models for value coding are shown. Under absolute value coding (left), values are represented independent of the context, thus neural activity patterns encoding the same value in reward and punishment contexts (e.g., +0.0€ and −0.0€) are equivalent. With partially-adaptive coding (middle), values are rescaled so that positive outcomes in punishment contexts (e.g., −0.0€) have a higher value than negative ones in reward contexts (e.g., +0.0€). Finally, if the code is fully adaptive (right) neural patterns encoding a positive outcome in different contexts are equivalent, but different from those encoding negative outcomes, which are identical as well. B, Results of the analysis comparing different value coding models. The absolute coding model is the best at predicting neural activity patterns representing factual outcome when partial feedback is provided (left) in all ROIs. Instead, for counterfactual outcome value encoding (right) the fully-adaptive coding model is the one with higher prediction accuracy (see Materials and Methods, Comparing different valuecoding models). For factual outcome encoding when complete feedback is given (middle), the model with higher prediction accuracy is either the fully-adaptive or the partially-adaptive one, depending on the ROI. C, Interaction effect of coding model × type of outcome. The absolute model had higher prediction accuracy for factual outcomes with partial information than both the partially-adaptive and the fully-adaptive models. Instead, the fully-adaptive model had higher prediction accuracy than the absolute model for factual and counterfactual outcomes with complete feedback. pOFC, posterior OFC; FPC, frontopolar cortex; vACC, ventral ACC; dACC, dorsal ACC; MCC, middle cingulate cortex; vSFG, ventral SFG. Error bars represent SEM. ***p < 0.001, **p < 0.01.
