Figure 4.

Trials to extinction versus partial reinforcement schedule during training. These plots are plots of the trials at which the curves in the bottom panels of Figure 3 intersect the even odds criterion, that is, the trial during the extinction phase at which the odds shift in favor of the conclusion that there has been a decrease in the rate or probability of reinforcement. These plots approximately replicate the empirical results plotted in Figure 2B, although these are slightly concave upward, whereas the empirical plots are slightly concave downward. To a first approximation, a Bayesian change computation, based on the assumption that extinction is the consequence of an approximately optimal change-detecting computational mechanism, predicts quantitatively the relation between partial reinforcement and trials to extinction shown in Figure 2B.