Using cluster analysis to explore mortality patterns associated with tropical cyclones.

Understanding the circumstances and conditions surrounding disaster-attributed deaths may contribute to designing and implementing emergency preparedness and response programmes. This paper introduces a three-step cluster analysis of multiple binary variables to investigate mortality patterns related to tropical cyclones. It is designed to overcome the difficulties of performing cluster analysis in a disaster database that is composed in part of nominal variables and is unavoidably incomplete owing to missing information. The first step in the process codes all variables as binary data in order to accommodate the nominal variables. The second step calculates Spearman's rank correlation coefficients for pairs of variables. And the third step subjects the correlation coefficients to cluster analysis. Data related to 1,575 deaths attributed to tropical cyclones (also known as typhoons) that struck Taiwan between 2000 and 2015 are used to illustrate the method. The results yield two distinct groups of variables that are worthy of further exploration.