Bayesian Probabilistic Projections of Mortality

Jennifer L. Chunn, University of Washington
Adrian Raftery, University of Washington

Every two years, the United Nations Population Division (UN) publishes projections for populations of over 200 countries through the next 50 years. The UN accounts for uncertainty in population projections by projecting population size with total fertility rates that are higher and lower than those assumed in the main projection. However, UN projections do not include any uncertainty in mortality projections. Our paper presents a probabilistic projection model of life expectancy at birth, a summary measure of mortality levels at a given time. This is done by a random walk with drift model where the drift is determined by a Bayesian hierarchical model. We assess the performance of our model by fitting it to UN estimates from 1950 through 1995 and projecting life expectancy for males from 1995 to 2005 for 195 countries. Our 80% prediction intervals captured the 1995-2005 UN estimates 81% of the time.