Answers are grounded in this paper and cite their sources.
Spatial Poisson (SP) models are used to understand the geographic distribution of event rates. Bayesian and Maximum Likelihood Estimation (MLE) methods are well developed but computationally intensive and only implemented in software for specific spatial correlation structures. Researchers emphasized the urgency of advancing estimation techniques. We developed a method of moment estimation (MME) approach for a generic three-parameter family of SP models (mean event rate, overdispersion, and spatial correlation). These models can accommodate any spatial correlation structure, but we studied only four models. Two of them are the well-known Conditional Autoregressive and Leroux models, defined through the precision matrix, while two other models, called the Constant Neighbor Correlation (CNC) and Shortest Path models, are defined through the variance–covariance matrix. Our numerical implementation provided results identical to the closed-form solutions for the CNC model. Simulation was performed to investigate the performance (on bias, mean squared error, coverage probability, and computational burden) of the MME for these four SP models and to compare the MME with other estimation methods. Estimation of the four SP models on county-specific 5-year mortality rates (2018–2022) for heart, lung, and blood disease in Texas was illustrated. The MME is computationally more efficient and, in the worst case, non-inferior in performance to MLE and the Bayesian estimation methods. It is better at verifying the existence of overdispersion than Bayesian estimation and showed better estimates in the case study. The MME can be used more broadly because it is flexible in the choice of statistics and matrix inverting techniques.
No takes yet. Share an insight, caveat, or question.
Heuvel et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: