科研成果详情

Generalized linear mixed models (GLMMs)have been developed by extending the standard generalizedlinear models (GLMs) and linear mixed models (LMMs) to accommodate correlated and overdisperseddiscrete data. Their broad applications are useful in various displaces, including medical clinical trail;epidemiologic studies; economics and social science studies. However, parameter estimation is highlychallenging due to the problem of analytically intractable integrated likelihood. Methods developedhave tended to be either computationally efficient but with very biased estimates, e.g., the PQL method(Breslow and Clayton, 1993), or very computationally intensive, e.g., Gibbs sampling (MCMC) (Zegerand Karim, 1991) approach. In this work, we propose to use the so-called Randomized (Scrambled)Quasi-Monte Carlo (RQMC) to solve the integration problem in the GLMMs. We investigate the per-formance of the proposed method by modellin the well-known salamander mating data that involvessix 20-dimensional analytically intractable integrals. We also conduct simulation studies to evaluatethe efficiency of our proposed method. A comparison has also been made with some commonly usedmethods discussed in the literature.