Figure 4.

Analysis Testing Whether the OFC Signal in Experiment 2 Fulfils the Criteria of a Signal Encoding Associations between Outcomes and Their Causal Choices

Each panel shows the observed temporal evolution of a GLM contrast over intratrial time (contrast parameter estimates ± SEM). Data are averaged across all OFC voxels that survived the (orthogonal) contingency contrast on AA? triplets. Outcomes (reward/nonreward) refer to the outcome of the middle trial in each triplet. Vertical bars separate decision, delay, outcome, and interval phases.

(A) Consistent with a signal encoding contingent associations between choices and outcomes, only in the outcome phase of DIRECT trials do contingent associations elicit an increased lOFC signal.

(B) In NBACK blocks, it is the noncontingent trials that yield lOFC activity.

(C) Taken together, contingent and noncontingent trials lead to exactly opposite signals in DIRECT and NBACK blocks (addition of the contrasts [BA+B - BA−B] + [BA−B - BA+B]).

(D) In AA? triplets, contingent choices are identical in DIRECT and NBACK blocks; accordingly, lOFC shows the same effect in both conditions (contrasting [BA+B - BA−B] - [BA−B - BA+B] triplets).

(E) BA? triplets show a highly significant contingency effect in the lOFC.

(F) Thus, AA? triplets show an equally strong contingency effect as BA? triplets. Overall, the figure shows that lOFC activity is incompatible with predictions made by the reward and reward prediction error hypotheses but corresponds precisely to the predictions made by the contingency hypothesis. Note that all plots were produced by right-aligning data from the decision phases so as to line up with the decisions themselves. The jittered duration of the delay phase thus causes a discontinuity between the delay and the monitor phases in this visualization. See also Figure S6.