Methods/Statistics
Simultaneous Ranking and Clustering of Small Areas based on Health Outcomes Ronald Gangnon* Ronald Gangnon Elizabeth Blomberg
A common task in public health communication is ranking and/or clustering different geographic units (small areas), e.g. counties in the United States, based on health (or socioeconomic) outcomes/determinants. We consider a nonparametric empirical Bayes finite mixture model for small area health outcomes (binary, count or quantitative) that is suitable for simultaneously ranking and clustering small areas. Small areas are ranked using optimal point estimates of the joint ranks, which minimize the expected integrated squared error loss on the health outcome (mean or proportion) scale. Small areas are simultaneously clustered by assigning small areas to the optimal (minimum mean square error loss) cluster (mixture component) for their estimated rank positions. We illustrate the utility of our approach with an analysis of percent low birth weight (<2,500g) births for Arizona counties, 2014-2020.
