Generalized estimating equations for zero-inflated spatial count data

We consolidate the zero-inflated Poisson model for count data with excess zeros (Lambert, 1992) and the two-component model approach for serial correlation among repeated observations (Dobbie and Welsh, 2001) for spatial count data. This concurrently addresses the problem of overdispersion and distinguishes zeros that arise due to random sampling from those that arise due to inherent characteristics of the data. We give a general quasi-likelihood and derive corresponding score equations for the zero-inflated Poisson generalized linear model. To introduce dependence, a spatial- temporal correlation structure comprising forms for fixed time, fixed location, and neighbor interactions is required; construction using techniques from the theory of Markov point processes is investigated.