Design- and Model-Based Approaches to Small-Area Estimation in a Low- and Middle-Income Country Context: Comparisons and Recommendations

Abstract

Abstract The need for rigorous and timely health and demographic summaries has provided the impetus for an explosion in geographic studies in low- and middle-income countries. Many of these studies present fine-scale pixel-level maps in an attempt to answer the needs of the current era of precision public health. However, even though household surveys with a two-stage cluster design stratified by region and urbanicity are a major source of data, cavalier approaches are taken to acknowledging the survey design. We investigate the extent to which accounting for the sample design affects the predictive performance at the aggregate level of interest for health policy decisions. We consider various commonly used models and introduce a new Bayesian cluster-level model with a discrete spatial smoothing prior. The investigation is performed through a simulation study in which realistic sampling frames are created for Kenya, based on the population and demographic information, with a survey design that mimics a Demographic Health Survey (DHS). We find that including stratification and cluster-level random effects can improve predictive performance. Spatially smoothed direct (weighted) estimates and area-level models accounting for stratification were robust to the underlying population and survey design. Continuous spatial models showed some promise in the presence of fine-scale variation; however, these models require the most “hand holding.” Subsequently, we examine how the models perform on real data, estimating the prevalence of secondary education for women aged 20–29 and neonatal mortality rates, using data from the 2014 Kenya DHS.