Detection of Hotspots on Spatial Data Using Principal Component Analysis

Echelon analysis(Myers et al., 1997) is a method to investigate the phase-structure of spatial data systematically and objectively. This method is also useful to prospect the areas of interest in regional monitoring of a surface variable. The spatial scan statistic(Kulldorff, 1997) is a method of detection and inference for the zones of significantly high and low rates based on the likelihood ratio. These zones are called hotspots. Kurihara and Hong(2002) detected the hotspots area for geospatial lattice data based on the echelon analysis. This method is valid for the detection of univariate lattice data. With our approach, we can detect hotspots area for multivariate lattice data. The present paper takes an approach to echelon based on the spatial scan statistics and data reduction method such as principal component analysis(PCA). We apply this method to epidemiological data concerning some causes of death. We will attempt to investigate whether there exist the regional differences in mortality ratio for each cause or not, and if there are regional difference in each cause, we find which area is high mortality ratio or low mortality ratio.