Average proportion of “stay” trials as a function of reward on previous trial (rewarded versus unrewarded) and transition type on previous trial (common versus rare) across the three trial types of interest. Lag-0, Lag-1, and Lag-2 WM-load refers to trials in which concurrent WM load occurred with the present trial, the previous trial, or the trial preceding the previous trial, respectively. Behavior on Lag-2 trials, in which no WM load was imposed on the current or previous trial, reflects the contribution of a model-based strategy. Conversely, behavior on Lag-0 and Lag-1 trials, in which concurrent WM load was respectively imposed on the current and previous trial, reflects only the contribution of a model-free strategy, suggesting that cognitive demand reverted subjects to model-free choice. Error bars depict standard error of the mean.