Abstract:

Based on a Gaussian pseudolikelihood method, we proposed a methd to estimate the consistency of the positive definite correlation matrix of generalized estimating equations. We proved that the Gaussian pseudolikelihood estimators of AR(1), exchangeable and MA(1) working correlation matrices
 were positive definite, even if the correlation matrix was misspecified. Simulation result shows that the regression parameter estimator of the generalized estimating equations based on the positive definite correlation matrix estimator is efficient.

Key words: Gaussian pseudolikelihood, generalized estimating equation, positive definite, working correlation matrices