Accuracy of small area mortality prediction methods: evidence from Poland

Abstract

We investigate the forecasting accuracy of several simple methods for predicting mortality in small regional areas in Poland. We focus on methods that scale country-level forecasts appropriately and, therefore, can be used by official statistical agencies to improve population projections. We examine data from 379 sub-NUTS-3 districts in Poland for the period 2006–2019, divided into three subperiods. The first period is treated as the training sample and the latter two the testing subperiods. The mortality surface method delivers the most accurate forecasts of the mortality profiles whereas using the district-level standardized mortality rates (SMR) calculated for several broad age groups to scale the country-level mortality forecasts gives the best life expectancy at birth predictions. The latter approach is far better than using the NUTS-2-based standardized mortality rate (SMR), as practiced by the Polish statistical agency. For single age-groups predictions, the SMR-based methods deliver relatively accurate forecasts for young cohorts, but their forecasting accuracy deteriorates significantly with age.