Fig. 4.

Comparison of choice data to different model predictions in the fMRI sample. a Posterior predictive checks show that—among the models with a single predictor—the model comprising the probability of foraging success captures choice data better than any of the other models. b Posterior predictive checks show that—among the models with two predictors—the model comprising the probability of foraging success and the optimal policy (at a horizon of 5 time steps) captures choice data better than any of the other models. Overall most models make quite similar predictions since they all include the probability of foraging success. c Posterior predictive checks show that the model comprising both the probability of foraging success and the optimal policy (horizon-5) provides a better fit to the data than the model that only comprises the probability of foraging success. Error bars are SEM. Per data bin, circles depict mean empirical data points and colored lines and crosses depict mean model predictions (averaged for simulated data according to each participant’s model fit). In several cases, error bars are smaller than the marker sizes, which scale with the average number of trials contributing to the respective data points. See Table 1 and Table 2 for lists of all variables. See Supplementary Fig. 4 for the behavioral sample. See Supplementary Fig. 5 and Supplementary Fig. 9 for posterior predictive checks of the winning model with choice data split according to the nine other heuristics and combinations thereof. See Supplementary Fig. 6 for parameter estimates of a full model including all candidate variables. See Supplementary Fig. 7 and Supplementary Fig. 8 for further posterior predictive checks of the winning model with choices split jointly according to the energy state and the probability of foraging success or the weather type. See Supplementary Fig. 10 and Supplementary Fig. 11 for comparisons of different time horizons. h-5: horizon of 5 days; h-1: horizon of 1 day; cont.: continuous; bin.: binary