Process models and stochastic theories of simple concept formation

Abstract

A class of information-processing models, stated in computer programming language, is constructed for a concept attainment experiment previously studied by Bower and Trabasso. The stochastic theory of Bower and Trabasso can be derived formally from the process models, but the process models, with fewer degrees of freedom, make more specific predictions over a wider range of experiments than the stochastic theory—hence are more universal, precise, and parsimonious in the sense of Popper. The formal process models are shown to be useful in discovering inconsistencies and unstated assumptions in informal descriptions of the psychological processes underlying the stochastic theory. The fit of the “fine-grain” statistics of the stochastic theory to the data is shown to be independent of the psychological content of the theory.