Parameter estimation for the averaging model of information integration theory

• Main Author: Zalinski, James S
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-1987
• Subjects: Psychological tests; Quantitative psychology
• ISBN: 9798206853087

The broad applicability of the averaging model of information integration theory for modelling cognitive judgments has been established in twenty years of research. Until now no studies had been completed for the purpose of investigating different procedures for estimating averaging model parameters, or for documenting the numerical and statistical properties of the estimated parameters. The present research attempted to fill this void. Three sets of analyses were completed. The first set of analyses used Monte Carlo techniques to estimate averaging model parameters using simulated data from standard factorial designs. These simulations demonstrated that reasonably accurate estimates of the averaging model parameters could be consistently obtained from realistically simulated experimental situations. The numerical and statistical characteristics of the estimated parameters should be sufficient to provide for their quantitative comparison. The most significant finding of the present research, within the area of specific estimation techniques, was the enhanced results obtained by placing upper and lower bounds on the estimated scale parameters. The present research has established that bounding the scale estimates is an effective procedure for reducing parameter estimate instabilities, variances and biases. In the second set of analyses, a computer program was developed using the Fisher information technique (Fisher, 1925) to analytically estimate equal weight averaging model parameter variance-covariance matrices for general two- and three-dimension factorial designs. When given as input the specifications of an experimental factorial design, the true averaging model parameters values, and a value for the response error variance, the program provides as output an estimated variance-covariance for the weight and scale parameters. The accuracy of the variance-covariance matrices estimated using the Fisher technique was validated by comparing them to Monte Carlo simulation results generated with identical input specifications. In general, a high degree of agreement was found between the two sets of matrices. In the third set of analyses, a second Monte Carlo simulation study was completed to evaluate alternative sets of equations for estimating averaging model parameters. In the "regression approach," algebraic manipulation produces sets of estimation equations for the weight parameters that remove the need to estimate the scale parameters. Unfortunately, the overall results obtained using the regression approach were not positive. While weight estimates could be consistently obtained, their properties did not fare well when compared to previously mentioned results.