Perceptual and Decisional Processes in Categorization

University Of Texas Austin, Austin TX

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): The long-term objective of the proposed research is to identify and quantify the perceptual and cognitive processes that are involved when an observer is presented with a categorization problem in which the prior probabilities (or base-rates) of the categories, and the costs and benefits (or payoffs) associated with categorization decisions are manipulated. With funding from NIH Research Grant # S R01 MH59196 my students and I made significant progress toward understanding the processes involved in decision criterion learning when base-rates and payoffs are manipulated, and toward understanding the complex interplay between several factors that influence base-rate/payoff learning. This work answered many questions, but also suggested many new lines of research. The purpose of this proposal is to expand our previous work in several new directions. The approach taken in the proposed research is to isolate and quantify the influence of several variables on decision criterion learning by comparing human performance with that of the optimal classifier--a hypothetical device that maximizes long-run reward. The aim is to test quantitative models of trial-by-trial and asymptotic performance by developing an "optimal" and several "sub-optimal" models, which instantiate important theoretical constraint. Four lines of research are proposed. Project 1 examines the effects of category distribution manipulations on base-rate and payoff learning. Theoretical work suggests that category discriminability, d', and category variance manipulations have a large effect on the rate of change in reward (or steepness) of the objective reward function which relates objective reward to the location of the decision criterion. If observers are sensitive to differences in steepness (called the flat-maxima hypothesis) then this should affect the speed and asymptote of learning. Project 2 examines the effects of payoff matrix manipulations on decision criterion learning. Theoretical work from our lab suggests that payoff matrix multiplication affect steepness, whereas matrix addition does not. Project 3 examines different types of feedback that might improve decision criterion learning. Especially promising is feedback based on the optimal classifier. Project 4 extends the studies in Project I - 3 to an explicit decision criterion task where observers adjust an observable decision criterion on each trial. These data are useful for testing learning models.