Models and Anti-Models: The Structure of Payoff-Dependent Social Learning

We conducted an experiment to describe how social learners use information about the relation between payoffs and behavior. Players chose between two technologies repeatedly. Payoffs were random, but one technology was better because its expected payoff was higher. Players were divided into two groups: 1) individual learners who knew their realized payoffs after each choice and 2) social learners, who had no private feedback about their own payoffs, but in each period could choose to learn which behavior had produced the lowest payoff among the individual learners or which behavior had produced the highest payoff. When social learners chose to know the behavior producing the highest payoff, a model of imitating this successful behavior matches the data very closely. When social learners chose to know the behavior producing the lowest payoff, they tended to choose the opposite behavior in early periods, while increasingly choosing the same behavior in late periods. This kind of rapid temporal heterogeneity in the use of social information has received little or no attention in the theoretical study of social learning.