Robust inferences in longitudinal models for binary and count panel data in the presence of outliers

Generalized quasi-likelihood (GQL) estimation approach is known to produce consistent and efficient estimates for the regression parameters involved in longitudinal models for binary and count data. It is also known that if the data contain one or more outliers, then the GQL approach may not even produce consistent estimates for the regression effects. As a remedy to this inference problem, recently Bari and Sutradhar (2010) developed a fully standardized Mallows’s type quasi-likelihood (FSMQL) approach in the independence set up. In this paper, as a generalization of the FSMQL approach for the independent data, we develop a robust GQL (RGQL) approach for consistent estimation of the regression parameters in the longitudinal models. It is demonstrated through a simulation study that the proposed RGQL approach produces almost unbiased and hence consistent estimates for the regression effects. The simulation study also exhibits the adverse effects of the outliers on the traditional GQL estimation approach, which is known to be constructed by ignoring the outliers. © 2010, Indian Statistical Institute.