A Regression Analysis of Problem Solving in a Binary Choice Task

Abstract

This study was designed to investigate three problems concerning human behavior in a two-choice learning situation. Rather than attempt a summary of research on two-choice learning we refer the reader to Bush and Mosteller (1955) and Feldman (1959). The first concern of this study was to develop and test certain trial-to-trial measures of the decision process. The basic trialto-trial measure is s's choice, left or right, on each experimental trial. Previous studies have utilized various transformations of this basic choice measure, such as the learning parameters of the stochastic learning models (Bush & Mosteller, 1955; Estes, 1950; Luce, 1959), run statistics (Bush & Mosteller, 1955; Goodnow, et at., 1959;3 Luce, 1959), conditional probabilities (Hake & Hyman, 1953), and stay-shift strategies (Goodnow & Pettigrew, 1955). In an attempt to tap more of rhe problem solving behavior in the situation we defined some new transformations of the choice data, including a modification of the Goodnow and Petcigrew (1955) strategy measure. Additional raw data were collected in the form of subjective probability estimates. Our aim was to define a criterion of concept attainment in the binary choice situation and to evaluate the power of our measures in predicting this criterion. The statistical technique for evaluating these measures was a regression model. We split the data at the mid-point of the trial series and ran regressions of the attainment criterion for the last half on the values of the selected measures for the first half as a means of testing the predictive power of the measures. The second concern of the study was to investigate the effect of subjective probability estimation on choice behavior. Originally there was some concern that this additional measure would bias S's conception of the choice situation and modify his pattern of choices. If this were the case the additional measure would change the situation rather than provide supplementary information about the original choice task. A control group, which made no probability estimates, was included to evaluate the extent of the bias. 'This research was conducted under the auspices of the Laboratory of Social Relations of Harvard University and was submitted as a Seniot Honors thesis toward the degree of Bachelor of Arts. Work on the project was facilitated by a grant from the Ford Foundation and an allotment of time on the IBM 704 by .the-~assachusen s Institute of Technology Computation Center. The authors wish to express their appreciation to Dr. Frederick Mosteller for his advice and encouragement. -Currently at the University of California at Berkeley. 'Goodnow, J. J., Rubinstein, I., & Lubin, A. Response to changing patterns of events. Unpublished memorandum. Problem Solving Projects, Walter Reed Army Instirute of Research, 1959; Goodnow, J. J., Shanks, B., Rubinstein, I., & Lubin, A. What is the human subject responding to in a two-choice task? Unpublished memorandum. Problem Solving Projects, Walter Reed Army Institute of Research, 1959.