FRIDAY, December 3, 2004
Time: 2:00 - 3:00 PM
Constant Hall Room 1037

Title: Statistical Analysis of Longitudinal and Multivariate Discrete Data

Deepak Mav
Department of Mathematics & Statistics, Old Dominion University

The generalized linear models developed in the seminal paper by Nelder and Wedderburn (1972, Journal of Royal Statistical Society A, 135, pp. 370-384) have been extensively used for statistical modeling of discrete outcomes. These models include univariate Poisson log-linear models and multinomial logistic regression models. In this talk we will discuss a general class of multivariate Poisson distributions obtained using binomial thinning operators. Unfortunately, these models have complicated multivariate probability mass functions, making it extremely difficult to implement maximum likelihood to estimate the parameters involved. However, the models can be simulated using simple algorithms. In this talk we will first discuss moment based methods of estimating the parameters, including the Gaussian and the quasi-least squares methods. Using simulations, we will study performance of these procedures via asymptotic relative efficiencies and coverage probabilities of simultaneous confidence regions, for the log-linear models described in the paper by Thall and Vail (1990, Biometrics, 46, pp. 657-671).