Solving Generalised Estimating Equations With Missing Data Using Pseudo Maximum Likelihood Estimation Is Equivalent to Complete Case Analysis

Arminger and Sobel (1990) proposed an approach to estimate meanand covariance structures in the presence of missing data. These authors claimed that their method based on Pseudo Maximum Likelihood (PML) estimation may be applied if the data are missing at random (MAR) in the sense of Little and Rubin (1987). Rotnitzky and Robins (1995), however, stated that the PML approach may yield inconsistent estimates if the data are (MAR). We show that the adoption of the PML approach for mean- and covariance structures to mean structures in the presence of missing data as proposed by Ziegler (1994) is identical to the complete case (CC) estimator. Nevertheless, the PML approach has the computational advantage in that the association structure remains the same. Keywords: Correlated Data Analysis, Generalised Estimating Equations, Marginal Models, Missing Data, Pseudo Maximum Likelihood 1 Introduction Arminger and Sobel (1990) introduced an approach to estimate mean- and covariance structures in ..