Test for Misspecification of Correlation Matrix in Dependent Binary Regression using Generalized Estimating Equations
by
Balakrishna Hosmane
Division of Statistics, Northern Illinois University, DeKalb, Illinois, USA
Coauthors: Ayrin Calachan Molefe, Northern Illinois University, DeKalb, IL, USA

The generalized estimating equations (GEE) introduced by Liang & Zeger(1986) and extended by Prentice(1988) has become very popular for analyzing longitudinal data where correlations exist among the repeated observations. One of the important applications is in modeling dependent binary data, e.g., in analyzing adverse reactions to a drug in clinical trials. Through the use of a "working" correlation structure to approximate the unknown dependence structure, the GEE yield consistent estimators of the regression parameters and of their variances. The misspecification of working correlation can lead to substantial decrease in efficiency of the estimates. In this work, we propose a goodness-of-fit test for misspecification of working correlation structure and establish its asymptotic properties. A Monte Carlo simulation is carried out to assess the small sample behavior of the proposed test and application to a real life example is presented.

Date received: August 8, 2001