75
reads

Pei-Sheng Lin

2011-11-11
12:50:00 - 16:00:00

400-1 , Mathematics Research Center Building (ori. New Math. Bldg.)

In this talk, we use a generalized linear mixed model, which simultaneously includes geographic clusters, covariates and spatial correlation in the model, for analysis of spatial clusters. We combine quasi-likelihood estimating equations and scan statistics to develop a method for multiple cluster detection. In the proposed method, parametric bootstrapping and quasi-deviance criteria are used to select hot/cool clusters for rejection. The proposed estimates of cluster coefficients are consistency even under misspecified cluster regions and correlation models. We conduct simulations to evaluate performance of the proposed method. The method is applied to an enterovirus data set in Taiwan for disease surveillance.