Fig. 1.

Basic experiment and competing theories. (A) Experiment 1. Stimuli in the display were either “targets” or “distractors.” Subjects earned a reward for fixating a target for at least 100 ms during the display period but did not earn a reward for fixating a distractor. There were two types of targets, horizontal bars (H) and vertical bars (V), and one of each was present in each display. Distractors were diagonal bars whose orientation was varied across blocks to manipulate the “feature-contrast” of the targets (orientation difference between the target and distractors). Subjects expressed their choice by fixating on the chosen item and were asked to try to maximize their total earnings. The experiment consisted of several blocks of 50 trials. Across blocks, we varied the value and feature-contrast of the targets. At the beginning of each block, subjects were informed about the value of targets H and V (e.g., value of H is 20 points, value of V is 10 points), and they received training. (B) We compared the performance of three different computational models. In M1, fixations are deployed to the location that is most likely to contain the target with the maximum feature-contrast for the block. In M2, fixations are deployed to the location that is most likely to contain the target associated with the highest value for the block. In M3, fixations are deployed to the location associated with the maximum expected reward for the trial, as predicted by an ideal Bayesian observer model.