Dr. Guohua Yan

Department of Mathematics and Statistics

University of New Brunswick

Title: " Modelling heterogeneity and serial correlation in longitudinal data using non-normal random effects "

In this talk I will introduce a class of generalized linear mixed models for longitudinal data based on the Tweedie distributions. The main results depend only on the first- and second-order moment assumptions of unobserved random effects, accommodating flexible temporal structures. An optimal estimating function for the regression parameters is developed using the best linear unbiased predictor of random effects, allowing an efficient common fitting algorithm for the whole class. In addition, the estimators for both regression and dispersion parameters are consistent. This approach unifies population-averaged and subject-specific inferences. I will illustrate the method by analysing the patient-controlled analgesia data and the Framingham cholesterol data.

This is joint work with Renjun Ma and Tariq Hasan.