Estimation of death underreporting at small areas using empiric Bayesian and EM algorithm methods

Abstract

Level and standard of mortality are major demographic estimation problems in Brazil. Demographists dealing with mortality in Brazil still do not fell assured of the real behavior of this population dynamics component. On the other hand, there is a need for mortality indicators available for more disaggregated geographic levels, mostly municipalities. The difficulty is that the more disaggregated, the more complex is the task for estimating any social or demographic indicator. In this study, we aimed to estimate and to propose the correction of death underreporting at the municipal level, according to age, using two methods: the empiric Bayesian estimator (BE) and the EM (Expectation-Maximization) algorithm. For the two methods to be operational within comparable municipalities, two steps were performed: we grouped the municipalities according to a mesoregion; and we grouped them into two homogeneous groups, created from a cluster analysis using the variables level of urbanization, proportion of death from external causes and the population of each municipality. We used data collected in 2000 from the State of Rio Grande do Norte. For the entire State, we estimated underreporting to be 11% using the BE estimator, and 12.9 % using the EM algorithm. Another important finding was the capability to assess the level of death coverage by age groups in the municipalities and, at any level of aggregation.