Categorization and Decision Rules: A Comparison of Likelihood Ratio and Prototype Models

Abstract

Two conceptually different approaches to classification were compared. One, based on statistical decision theory, postulates that decisions are made on the basis of a likelihood ratio, which implies the existence of an indifference region in the stimulus space. The other postulates that decisions are based on relative distances from abstracted prototypes. The models have different implications for the abstraction process involved in concept formation, and predict different outcomes in a classification task. In an experiment designed to compare the two models, subjects saw 9 examples from each of two categories and then sorted additional stimuli. The dependent variable was an average confidence rating. Neither the likelihood ratio model nor the prototype model was sufficient to account for the data. Apparently subjects used both the indifference region and the axis connecting the two prototypes to determine their responses.