A0766
Title: Bayesian optimal designs for generalized linear mixed models based on the penalized quasi-likelihood method
Authors: Yao Shi - Qingdao University (China) [presenting]
Abstract: Generalized linear mixed models are widely used in data analysis, while the complexity of the information matrices for such models makes optimal design questions challenging. Moreover, based on some previous research, locally optimal designs for such models can be sensitive to the local parameters. The focus is on the Bayesian optimality criterion to get a robust optimal design. The evaluation of the Bayesian criterion is based on the penalized quasi-likelihood method and on a non-informative prior, to get rid of the influence from the prior choices. Bayesian designs found by a particle swarm optimization algorithm are presented and discussed. The robustness of such a Bayesian design is studied by comparing it with locally optimal designs. Finally, as an illustration, Bayesian optimal designs are derived for a real study.
