A profile likelihood approach for longitudinal data analysis

Inappropriate choice of working correlation structure in generalized estimating equations (GEE) could lead to inefficient parameter estimation while impractical normality assumption in likelihood approach would limit its applicability in longitudinal data analysis. In this article, we propose a profile likelihood method for estimating parameters in longitudinal data analysis via maximizing the estimated likelihood. The proposed method yields consistent and efficient estimates without specifications of the working correlation structure nor the underlying error distribution. Both theoretical and simulation results confirm the satisfactory performance of the proposed method. We illustrate our methodology with a diastolic blood pressure data set.