Tight resource-rational analysis

Abstract

Resource-rational analysis is used to develop models that assume that people behave optimally given the structure of the task environment and the cost of cognitive operations. We argue in favor of a tight resource-rational analysis, an extension in which model parameters are independently constrained. As a case in point, we demonstrate how to develop a tight resource-rational model of the video game Space Track. Our approach consists of four steps. First, we measure performance-critical parameters in independent micro-tasks, which we input into mathematical models of cognitive processes. Second, we validate these models in other process-specific micro-tasks. Third, we rely on a theory of the cognitive architecture (i.e., ACT-R) to derive estimates of the time costs of these processes. Finally, we generate predictions for the main task, Space Track, by assuming that subjects are doing their best given their abilities. The generated individualized predictions were close to observed subject asymptotic performance, which demonstrated the viability of our approach, even in tasks of similar complexity to that of Space Track.