Bayesian reconstruction of multi-state mortality

Andrea Tamburini, Vienna Institute of Demography
Dilek Yildiz, Wittgenstein Centre for Demography and Global Human Capital

The connection between education and mortality has already been analysed. At the same time, the availability of data concerning the stratification of mortality according to level of education is rare and limited to developed countries and recent years. The focus up to this point has mainly been on how education, seen as a proxy for socio-economic status, affects life choices and habits and thus, indirectly, mortality. When mortality has been considered directly, the focus has been more on life expectancy in specific countries and for specific ages, without engaging in a systematic study with a broad geographical and temporal spectrum. Employing Bayesian models for estimating mortality rates, the focal point has been primarily on predicting future mortality rates and not on reconstructing past ones. In this paper we propose a mortality model that is able to combine patchy data in order to produce estimates of past mortality curves according to the level of education. We developed a hierarchical Bayesian model that uses mortality age schedules data by education attainment and combines them with data from DHS to produce age, sex and education-specific mortality curves. The model is applied in a case study for the female population of Turkey.

Keywords: Bayesian methods / estimation, Mortality

See extended abstract.

  Presented in Session 183. Measurement of Mortality and Causes of Death