Spatial correlation affects responses collected from adjacent clusters, and it should take into consideration in modeling responses. In this study, a new approach was introduced to estimate both covariates effects and variance of random effect in models for count responses by incorporating clusters dependence. Generalized linear Poisson mixed model was used to model the count responses under three different designs of covariates. The performance of suggested approach was checked by a simulation study. The marginal generalized quasi-likelihood approach was used for estimation of parameters based on Gauss-Newton iterative procedure. For the weak and moderate spatial correlation (up to 0.5), proposed approach gave close simulated means to true values and very low simulated standard errors for all estimates. For the values of spatial correlation higher than 0.5, the suggested procedure was not able to give results. The proposed method is suitable for estimation of parameters when there is a weak or moderate spatial correlation among adjacent clusters.
Keywords: Spatial analysis, Poisson mixed model, Cluster counts, Intercorrelations