Patch leaving in humans: can a generalist adapt its rules to dispersal of items across patches?

Abstract

We used a computer game to examine three aspects of patch-leaving decisions in humans: how well do humans perform compared to the optimal policy, can they adjust their behaviour adaptively in response to different distributions of prey across patches and on what cues are their decisions based? Subjects earned money by catching fish when they briefly appeared within a pond; the timing of appearances was stochastic but at a rate proportional to how many fish remained. Caught fish were not replaced and ponds varied in how many fish they initially contained (according to three different distributions). At any point subjects could move to a new pond, but travel took some time. They delayed this switch much too long. Furthermore, regardless of the distribution of prey, subjects spent longer at ponds where they had found more items (contrary to optimality predictions in two of the environments). However, they apparently responded not to the number of captures directly (despite this appearing on screen) but to the current interval without a capture, to the interval preceding the last capture, and to the time spent at the current pond. Self-reports supported this order of cue importance. Subjects often left directly after a capture, perhaps an example of the Concorde fallacy. High success rate in the preceding patch decreased residence time and subjects appeared to be learning to leave earlier over the latter two thirds of the experiment. Minimization of delay to the next capture alone might explain some of the suboptimal behaviour observed.