An adaptive approach to resource allocation

Resource allocation tasks require the decision maker to allocate a limited resource (e.g., budget or staff) to competing activities (e.g., commodities or jobs) in order to maximize an explicit objective (e.g., utility or productivity). In contrast to past research that has compared optimal to observed performance, the present study investigated how individuals learn to improve their resource allocation policies from outcome feedback. A total of 64 subjects received 50 training trials under one of four conditions produced by a 2 × 2 factorial design. One factor was the form of the unknown objective function—it had either a single global maximum, or it had a local maximum below the global maximum. The second factor was a manipulation of prior information —subjects were either told the maximum possible objective function value (but not its location), or this information was withheld. Several measures of learning were analyzed including (a) the cumulative proportion of subjects reaching a criterion as a function of training, (b) the path of the resource allocation policies selected across trials, (c) the magnitude of change in allocations as a function of training, and (d) the probability distribution of search directions selected following success and failure. A learning principle called hill-climbing was useful for interpreting the results.