Optimal method for time series analysis of mortality rates and their relations with other factors

Abstract

Influence of factors which fluctuate in time is usually described in term of residuals present in general linear or nonlinear models. The aim of this paper has been to identify relationship between mortality rates and major social and economic factors and to compare three methods in detecting (hidden) relations between them. Data on frequencies for all causes of death, e.g. mortality rates have been collected for Central Serbia for 21 years, from 1973. until 1993. We have used two models based on the general linear model, variance-covariance regression and correlation and Box-Jenkins time series method and one nonlinear model based on neural networks. The relationship has been established, according to our results. Among presented methods we consider the analysis based on multilayer neural networks techniques has several advantages over classical approach. © The authors 1996.