Conditional Empirical Likelihood Approach to Unbalanced Longitudinal Data Analysis

• Activity Number: 301
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Peisong Han*+ and Peter Song and Lu Wang
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: GEE ; marginal models ; robustness

Abstract

We propose a conditional empirical likelihood approach to analyzing unbalanced longitudinal data, in which stratification is invoked to deal with unbalanced visit patterns. Unlike the currently popular marginal modeling approaches (e.g. GEE), the proposed approach does not require any explicit specification of variance-covariance matrix, but only correct specification of the marginal mean model. As a result, our approach is robust against model misspecifications in the aspects of marginal variances and/or within-subject correlations. We show that our estimator is consistent and asymptotically normally distributed under certain regularity conditions. In addition, utilizing the objective function in the proposed approach, we establish a likelihood-ratio type of test, which resembles, in both forms and asymptotic properties, to the classical likelihood ratio test. We conduct simulation studies to compare the proposed method with some popular marginal modeling approaches. We also illustrate this method through real data analysis.