The Statistical Computation of Heat Disorder Risk with HGLM

Abstract

This study provides the prediction result of heat disorder incidence risk using Hierarchical Generalized Linear Model (HGLM) on the basis of the relationship between climate variables (temperature and relative humidity) and heat disorder incidence. Basic descriptive statistics were calculated to track down to any change in climate variables over the past 43 years. The empirical (probability) density functions were simulated by four different times of 1970s, 1980s, 1990s and the recentest. Furthermore, we compared the statistics, regional ranges and regional standard deviations, of weather variables in 1973 and in 2015(t-test was applied). Understanding the variables with these statistics, two types of response variable (the monthly sum of heat disorder and the monthly sum of heat disorder per 1 million people) were modeled to predict the risk with explanatary climate variables. Especially, a spatial correation structure was included in the models as a random effect. This spatial correlation sturcture had the location information of each region in terms of adjacency. We found that this could decrease the significance of nominal region variable. With the estimates obtained from the chosen model, we compared the simulated heat disorder incidence risk during the unobserved period to the observed current data.