Main results remain consistent across different exclusion criteria.

<p>In the main text, we use two main exclusion criteria: 1) participants who chose the option with the greater high payoff value less frequently as the probability of winning increased, suggesting a misunderstanding or lack of attention to the task, and 2) participants who always chose either the safe or the risky option in the Solo trials, and thus for whom bi-directional influence is not possible. To check the robustness of our main findings with respect to exclusions, we conducted the main analyses without exclusions; these analyses indicate that our main results remain largely consistent even when individuals excluded from our analyses in the main text are reinserted. <b>(a-c)</b> When all participants were included (control N = 46, insula N = 11, and dACC N = 11), we observed a trending result of the worse model fit (<i>P</i> = 0.12) and the increased follow weight (<i>P</i> = 0.11) for lesion participants. In addition, the correlation between worse model fit and the follow weight was significant (r = 0.45, <i>P</i> = 1.2e−04). <b>(d-f)</b> When participants who did not show reasonable choices in trivial cases remained excluded, but the ones who always chose either the safe or risky option were reincluded (control N = 25, insula N = 10, and dACC N = 8), we observed a trending result of the worse model fit (<i>P</i> = 0.092) and the increased follow weight (<i>P</i> = 0.027) for lesion participants. Furthermore, the correlation between worse model fit and the follow weight was significant (r = 0.46, <i>P</i> = 4.9e−04). Overall, although some of the results are statistically marginal, these data show that the main results remain consistent even when we are more lenient with our exclusion criteria.</p> <p>(PDF)</p>