A latent process regression model for spatially correlated count data

This paper proposes a regression model for spatially correlated count data that generalizes the work of Zeger (1988, Biometrika 75, 621-629) developed in a time-series setting. In this approach, spatial correlation is introduced through a latent process, and the marginal mean function may contain spatial trends and covariates. Generalized estimating equations are used to estimate and perform marginal inference on the spatial trend and covariate effects. The feasibility of this approach is demonstrated using an example of the distribution of neuronal cell counts in a laboratory culture dish.