The impact of rewarding medium effort and the role of sample size

We take the point-of-view of designers of incentive systems who cannot reward the behavior they most desire, and must decide whether to reward a less desired behavior. For example, when it is difficult to distinguish between the desired high-effort strategy from a low-effort “mimicry” strategy, policy makers may choose instead to reward medium levels of effort. Experiments 1 and 2 examine situations in which participants (the subjects of the policy) base their decisions on personal experience obtained in 100 choices with immediate feedback. The results show that rewarding medium levels of effort is effective in "reckless shortcut" settings (where a low-effort strategy impairs participants' expected return but provides the best payoff in most cases), and is counterproductive in "treasure-hunt shortcut" settings (where a low-effort strategy maximizes the expected return, but provides the worst payoff in most cases). Experiment 3 suggests that this pattern is intensified when participants base their decisions on the experience of other agents. The aggregate choice rates can be captured with a two-stage naïve sampler model that assumes reliance on small samples of past experiences, and a decrease in the sample size when learning from the experience of others. Importantly, the descriptive value of the reliance on small samples hypothesis is not a result of a tendency to rely on the most recent experiences. These findings can be explained by assuming that people rely on those experiences, which seem most similar to the current task, and that the experiences of others seem less similar.