skip to main content skip to footer

Some Considerations in Probability Learning

Author(s):
Wainer, Howard
Publication Year:
1967
Report Number:
RB-67-19
Source:
ETS Research Bulletin
Document Type:
Report
Page Count:
50
Subject/Key Words:
Learning Strategies, Predictor Variables, Probability, Reinforcement

Abstract

A binary choice learning experiment with noncontingent reinforcement was run. It was found that if the subjects clearly understood that there was no pattern in their schedule of reinforcement, e.g., that it was completely random, they would very quickly choose the maximal gain strategy decreed by game theory. If, on the other hand, they believe that there is a pattern, and that if they "figure it out" they could be right on each trial, they would probability match in the early stages of the experiment as predicted by Estes' probability matching hypothesis. After about one thousand trials, however, the subjects' response levels come close to the extreme asymptotic generalization predicted by Edwards. An amendment to the Estes model is proposed which allows it to predict the observed behavior.

Read More