BROADENING THE TESTS OF LEARNING-MODELS

For many years psychological studies of the learning process have used a simulated medical diagnosis task in which symptom configurations ar e probabilistically related to diseases. Participants are given a set of symptoms and asked to indicate which disease is present, and feedba ck is given on each trial. We enrich this standard laboratory task in four different ways. First, the symptoms have four possible values (lo w, medium low, medium high, and high) rather than just two. Second, sy mptom configurations are generated from an expanded factorial design r ather than a simple factorial design. Third, subjects are asked to mak e a continuous judgment indicating their confidence in the diagnosis, rather than simply a binary judgment. Fourth, cumulated performance sc ores, payoffs, and the availability of a historical summary of the out comes are varied in order to assess how these treatments modulate perf ormance. These enrichments provide a broader data set and more challen ging tests of the models. Using 123 subjects each in 480 trials, we co mpare five existing learning models plus several variants, including t he well-known Bayesian, fuzzy logic, connectionist, exemplar, and ALCO VE models. We find that the subjects do learn to distinguish the sympt om configurations, that subjects are quite heterogeneous in their resp onse to the task, and that only a small part of the variation across s ubjects arises from the differences in treatments. The most striking f inding is that the model that best predicts subjects' behavior is a si mple Bayesian model with a single fitted parameter for prior precision to capture individual differences. We use rolling regression techniqu es to elucidate the behavior of this model over time and find some evi dence of over-response to current stimuli. (C) 1998 Academic Press.