科研成果详情

Generalized estimating equations (GEE) (Liang & Zeger, 1986) are commonly used for longitudinal studies. In the literature little attention was paid to the choice of "working" covariance structures in GEE. Recently Wang & Carey (2003) showed that mis-specification of "working" covariance structures may lead to a great loss of estimate efficiency. In this paper we propose a data-driven approach for modelling covariance structures in GEE for longitudinal data. Our numerical analysis shows that the GEE estimate efficiency can be improved in terms of modelling of covariance structures.