Abstract: Understanding patterns in mortality across subpopulations is essential for local health policy decision making. One of the key challenges of subnational mortality rate estimation is the presence of small populations and zero or near zero death counts. When studying differences between subpopulations, this challenge is compounded as the small populations are further divided along socioeconomic or demographic lines. In this paper, we build on principal component-based Bayesian hierarchical approaches for subnational mortality rate estimation to model correlations across subpopulations. The principal components identify structural differences between subpopulations, and coefficient and error models track the correlations between subpopulations over time. We illustrate the use of the model in a simulation study as well as on county-level sex-specific US mortality data. We find that results from the model are reasonable and that it successfully extracts meaningful patterns in US sex-specific mortality. Additionally, we show that ancillary correlation parameters are a useful tool for studying the convergence and divergence of mortality patterns over time.
Note: This paper is the final version of the work presented at these conferences.
The preprint can be downloaded here.