| Summary: | 碩士 === 國立中正大學 === 數學系統計科學研究所 === 107 === Spatial data analyses are often used to study the relationship between region and natural phenomenon or the incidence of diseases. In most cases, spatial observations are not independent and it is necessary to model this dependence and allow for it in data analysis. However, the exact form of the spatial correlation is often complex and unknown. Adegboye et al. (2018) assume the spatial correlation may be induced by multiple latent sources and embed a series of generalized estimating equations, each reflecting a particular source of spatial correlation, into a larger system of estimating equations. However, it is still unkown how to perform model selection in this general type of analysis. To tackle this problem, we propose a new model selection criterion, the Generalized Spatial Information Criterion (GSIC), which is based on the expected quadratic error measuring the discrepancy between the true and the considered model for the marginal mean. The simulation results reveal that the proposed method performs quite well, regardless of whether the spatial correlation is anisotropy or not. Furthermore, the utility of proposed method is further illustrated by a real application.
|
|---|
