A test of a model for multiple-choice behavior.

Abstract

A mathematical model was outlined for expts. in which subjects are required to predict on each of t trials which event of n possible events will occur. Excellent agreement between data and model was found when subjects were informed, following each of their predictions, which one of the n possible events actually occurred. Here, the probability that the subject will predict a particular one of the events is a negatively accelerated growth or decay function of trials, approaching as a limit the relative frequency of the event being predicted. When subjects were informed only that their prediction was correct or incorrect, model-generated curves fitted the response data over the early trials of the experiment, but failed to do so in later trials when some subjects abruptly increased their frequency of prediction of one alternative. The experiment confirmed the prediction that rate of change of the probability function is negatively related to n, but failed to verify the prediction that asymptotic probability decreases with increasing n.