Combining Survey and Census Data in Time and Space in a Developing World Context

Abstract

Obtaining reliable estimates of health indicators at a granular level in space and time is important for informing health intervention and public policy decisions. In low and middle income countries, the data available come from a multitude of sources including vital registration systems, complex surveys, and disease registries. These data sources are often of varying quality. In particular, the information on health outcomes may be aggregated over space and time. Overall, this poses a modeling challenge as there is a mismatch between the underlying process, the observed data, and the inferential resolution desired. This work tackles three main issues that are commonly faced in this setting by developing Bayesian models tailored to the specific problem at hand. In particular, this work considers incorporating data where the outcome has been aggregated over space (common in census data), the outcome is associated with a point in space but the exact location is unknown (common in household survey data), and the outcome has been aggregated over time (common in modeling child mortality using census data).