ABSTRACT
Performance in a wide variety of cognitive tasks – classification, identification, psychophysical judgment, preference – takes the form of choice from a set of alternatives. Prior to the appearance of Luce’s choice model in 1959, it was standard for choice behavior to be interpreted by means of domain-specific, data-based models. Luce’s approach, in contrast, was to start with a very general, rationally based axiom and derive theorems and functions that constrain the forms of performance measures and relationships among them in the various task domains. Further, an extension of the 1959 formulation termed the similarity choice model has become the most influential complement to signal detection theory as a basis for distinguishing the effects of stimulus discriminability and response bias in psychophysical and preferential choice data. Following a review of the role of the similarity choice model in cognitive research, I address the central theoretical question of whether choice behavior is basically probabilistic or deterministic – concretely, do predictions of choice probability derived from the model apply to the choices made by individuals on particular experimental trials or only to data obtained by averaging over groups of individuals or blocks of trials? The theoretical work of several investigators taken together with some relevant empirical results of my own research lead to a provisional resolution of the issue.
