Constructing preference from sequential samples: The impact of evaluation format on risk attitudes.

The cognitive mechanism underlying decisions based on sequential samples has been found to be affected by whether multiple alternatives are evaluated together or whether each alternative is evaluated individually. In this experiment, we examined whether evaluation format can also lead to different preference orders among risky alternatives. We hypothesized that due to differences in computational demands posed by the two evaluation formats, there would be differences the type of the cognitive mechanism deployed: a statistical summary mechanism that trades off the mean reward and risk of an alternative, or a selective-accumulator mechanism that sums the rewards of each alternative, with a higher weight to more extreme payoffs. Each participant rated the same set of alternatives (sequences of payoffs from slot machines) in both a one-by-one and a grouped evaluation format. The mean and the variance of the payoff distributions of each alternative were varied orthogonally. As predicted, in the grouped (but not in the one-by-one) condition the impact of the variance on participants’ ratings interacted with the mean payoff. Specifically, participants were risk averse for alternatives with a low mean payoff and risk seeking for alternatives with a high mean payoff. Computational modeling showed that the majority of participants were best described by a risk‒return model in the one-by-one condition, but by a selective-accumulator model in the grouped condition. Our results underline the importance of studying the cognitive foundations of risk attitudes in order to understand how they are shaped by the structure of a given decision task.