A Bayesian approach to estimate the completeness of death registration

TL;DR

This paper introduces a hierarchical Bayesian model with Global-Local priors to accurately estimate death registration completeness across different geographic levels, improving mortality data reliability.

Contribution

It presents a novel Bayesian modeling approach that accounts for demographic covariates and variability within and between countries to estimate death registration completeness.

Findings

Model improves estimates of death registration completeness.

Application to Colombia shows practical utility.

Method outperforms existing approaches.

Abstract

Civil registration and vital statistics (CRVS) systems should be the primary source of mortality data for governments. Accurate and timely measurement of the completeness of death registration helps inform interventions to improve CRVS systems and to generate reliable mortality indicators. In this work we propose the use of hierarchical Bayesian linear mixed models with Global-Local (GL) priors to estimate the completeness of death registration at global, national and subnational levels. The use of GL priors in this paper is motivated for situations where demographic covariates can explain much of the observed completeness but where unexplained within-country (i.e. by year) and between-country variability also play an important role. The use of our approach can allow institutions improve model parameter estimates and more accurately predict completeness of death registration. Our models are based on a dataset which uses Global Burden of Disease (GBD) death estimates based on the GBD 2019 and comprises 120 countries and 2,748 country-years from 1970-2019 (Demographics Collaborators, 2020). To illustrate the effectiveness of our proposal we consider the completeness of death registration in the departments of Colombia in 2017.