Generalized outcome-based strategy classification: Comparing deterministic and probabilistic choice models

Abstract

Model comparisons are a vital tool for disentangling which of several strategies a decision maker may have used—that is, which cognitive processes may have governed observable choice behavior. However, previous methodological approaches have been limited to models (i.e., decision strategies) with deterministic choice rules. As such, psychologically plausible choice models—such as evidence-accumulation and connectionist models—that entail probabilistic choice predictions could not be considered appropriately. To overcome this limitation, we propose a generalization of Bröder and Schiffer’s (Journal of Behavioral Decision Making, 19, 361–380, 2003) choice-based classification method, relying on (1) parametric order constraints in the multinomial processing tree framework to implement probabilistic models and (2) minimum description length for model comparison. The advantages of the generalized approach are demonstrated through recovery simulations and an experiment. In explaining previous methods and our generalization, we maintain a nontechnical focus—so as to provide a practical guide for comparing both deterministic and probabilistic choice models.