Fig. 3.

Results of experiment 1. Empirical data appear in blue with solid lines and the winning model in orange with dashed lines. Bars represent within-subject confidence intervals (28). In the figure and in the following description, “value” refers to the sum of the ratings along one path of the decision tree. (A) First-stage choice accuracy as a function of the difference between the maximum value and average of the other two values. A trial is considered correct if the first-stage choice does not rule out the optimal path. (B) Second-stage choice accuracy as a function of the absolute difference between the ratings of the items remaining at the second stage. A trial is considered correct if the higher-rated item is selected. Only trials where a second-stage choice had to be made are included. (C) First-stage reaction time for correct trials. A trial is considered correct if the best overall path was selected. (D) Second-stage reaction time for correct trials. A trial is considered correct if the best overall path was selected. (E) First-stage reaction time for correct trials, as defined in C and D, as a function of the paths that appear together in the tree. For example, “Max and second best” means that the two paths with the two largest values were grouped on one side (pressing left or pressing right at the first stage, depending on the paths’ location, would leave both of them in play), and the smaller-valued path was on the other side by itself. (F) Overall choice accuracy, taking both stages into account.