Change point Detection in Mortality Trend in IRAN

In the analysis of mortality trend in the Iran that is basically a sequences of observations sorted by time, there is a point where the statistical properties of the mortality trend change so that the first k0 observations have a distribution of F0 and other n-k0 observations have a distribution of F1. The point k0 is unknown and called the change point. The aim of this paper was to detect the location of the change point and estimate it in the real mortality data of the country. In this study, Xt indicated the number of mortality in time t (year) and because of the numeric nature of the variable, we considered the Poisson model for the variable Xt. We assumed in early years (t0), Xt had a Poisson distribution with a mean of ʎ0 and for later years (t>k0), Xt had a Poisson distribution with a mean of ʎ1. In theory, we used the MIC method, a modification of the SIC method. For detecting more than one change point, we used the binary segmentation process in the mortality trend. The results showed that the change point occurred in 1993 and 1997. The finding of this study showed three periods with different rates in the mortality trend of Iran.