Nonparametric Hierarchical Modeling for Detecting Boundaries in Areally Referenced Spatial Datasets

With increasing accessibility to Geographical Information Systems (GIS) software, researchers and administrators in public health are increasingly encountering spatially referenced datasets. Inferential interest of spatial data analysis often resides not in the statistically estimated maps themselves, but on the formal identification of “edges” or “boundaries” on the map. Boundaries can be thought of as a set of connected spatial locations that separate areas with different characteristics. A class of nonparametric bayesian models are proposed in this paper to account for uncertainty at various levels to elicit spatial zones of rapid change that suggest hidden risk factors driving these disparities. Simulation study are conducted to illustrate the new approaches and compare with existing methods. “Boundaries” on Pneumonia and Influenza hospitalization map from the SEER-Medicare program in Minnesota are detected using the proposed approaches.