Predictions in Generalized Linear Mixed Model

• Activity Number: 129
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Chenghsueh Yang*+
• Companies: University of California, Riverside
• Keywords: Generalized linear mixed model ; Pseudo-Likelihood ; Newton-Raphson ; Laplace approximation ; Empirical best linear predictor ; Gauss-Hermite quadrature

Abstract

A randomized clinical trial design is used as an example for comparing effects of drugs. Constructing Generalized linear mixed model (GLMM) can tell different effects between drugs. However, GLMM confronts a problem of maximizing the high-dimensional integration of the likelihood function. Wolfinger (1993) proposed both Pseudo-Likelihood and Restricted Pseudo-Likelihood to tackle this problem. Both methods are based on a pseudo variable followed by application of the Newton-Raphson algorithm. McCulloch (2001) suggested applying the Gauss-Hermite quadrature for approximation of the likelihood function. Laplace approximations have also been proposed for approximating the integrals in closed form. More recently, a modified Laplace approximation has been proposed. In this talk I will also discuss the mean squared error performance of the empirical best linear predictor.