Abstract
Risk as a factor in decision-making has been well-studied in both economics and psychological literatures. However, research on risk, especially in the psychological literature, has focused more on avoidance of negative outcomes and less on the adaptive nature of risk-taking. Thus, this study sought to explore how the choice of more high vs low risk options influenced learning the association between stimulus features and good outcomes. Our task was a multiarmed bandit with six visually distinct slot machines that produced points as outcomes in a 2x3 design with two levels of expected value (EV = 12 or 20) and three levels of risk (low, medium, high). The number of points gained/lost was displayed as feedback on every trial, however, on trials with extreme outcomes (which were more common with high-risk machines), extra feedback was given. The primary dependent measure was the proportion of choices for each machine. Performance was defined as learning to choose the machines that produced higher gains (EV-20). Subjects who did not learn were analyzed separately and the others were divided into high and low performers. In the positive feedback condition, both groups showed a preference for high-risk over medium-risk machines, suggesting that positive feedback increased risk-taking. High performers showed a preference for low-risk over medium-risk machines, suggesting that making more low-risk choices enhanced learning. An analysis of sequences of choices suggested that low performers were more likely to explore across all six machines, while high performers tended to exploit their preferred machines.