A universal efficient robust likelihood method for general correlated data

Abstract

An efficient robust likelihood method is proposed for regression analysis of general correlated data without knowing the true underlying joint distributions. This robust likelihood is able to provide valid likelihood ratio tests, score tests and other inferential tools whose creation generally requires a valid distributional assumption. Simulation study also shows that our parametric robust approach is more efficient than several competitors such as the popular semi-parametric generalized estimating equations methodology, especially when the within-cluster correlation is less structured. A number of examples are also used to demonstrate the efficacy of the proposed method. (This is a joint work with Professor Tsung-Shan Tsou.)?