Experimental design.

<p> <b>(A)</b> Subjects were engaged in two decision-making tasks in which they were instructed to choose from one of two actions (A or B) with the goal of obtaining and maintaining maximum earned reward as indicated by a centrally located reward bar. <b>(B)</b> Example of reward structure: Central bar height (reward) depends on two variables: 1) current decision (A or B, Red and Blue trace correspondingly) and 2) the percentage of choice A (%A) made over the past 20 trials (x-axis). The initial %A is set to be 50%. In cases in which participants chose choice A in <i>more</i> than 33% of the previous 20 trials, a B choice results in greater reward than an A choice. In cases in which participants chose choice A in <i>less</i> than 33% of the previous 20 trials, an A choice results in greater reward than a B choice. Thus, to the right of the crossing-point of the two reward functions, ‘B’ choices both increase reward and move subjects to left on the x-axis. However, as subjects move left past the crossing-point, A choices begin to yield greater reward than B choices and move subjects right on the x-axis. In this example, the crossing-point represents the optimal allocation to A, as the average return at all other allocations is less than at the crossing-point of the two reward structure (indicated by the dashed line). <b>(C)</b> In the first task, participants made 125 decisions with reward determined by the matching shoulders (MS) structure (left panel), followed by 125 decisions with reward given by the rising optimum (RO) structure (right panel). <b>(D)</b> In a second task, participants began with the flat returns (FR) structure (left panel) and switched to a pseudorandom (PR) task (right panel). In the FR task, all choice strategies yield the same average return (dashed line). In the PR condition, reward was randomly determined independent of choice but was set to give the same mean and variance of rewards as was earned in the FR structure.