Computer Simulation of Multiple Strategies in Human Binary Prediction

Abstract

Most models of human binary-prediction behavior (two-choice probability learning) have been of the mathematical learning theory variety and have given little consideration to S's memory for the event sequence and to the kinds of strategic behavior he may be using. This study considers reasons for dissatisfaction with that approach and proposes a new type of model which is based on a memory structure like that described by Atkinson and Shiffrin. The model is of the computer-simulation type and proposes that S uses each of several prediction strategies in approximate proportion to its relative frequency of success. The theory is developed from a number of general assumptions and specific axioms and is tested on the data from a new binary-prediction experiment. The discussion concludes with comparisons to results from studies in mathematical learning theory and suggestions for further development of the new type of model.