Skip to content

Methods

The results contained here are based on Demographic Health Surveys (DHS). DHS data are collected using a stratified two-stage cluster sampling design and this sampling must be taken into account when we produce estimates. This can be achieved by using weighted estimation. This generally works well for the first geographical administrative level (so-called Admin-1), since DHS surveys are powered to produce reliable estimates at Admin-1. For the second administrative level (Admin-2), data sparsity becomes an issue and so we use the most popular small area estimation (SAE) approach, which is the Fay-Herriot model. Under this model, the weighted estimates are smoothed over space, which increases the precision of estimation (reduces uncertainty) in each constituent area, because the data from neighboring areas are aiding in the estimation.

We include maps of point estimates, and of the uncertainty, measured through the width of a 90% uncertainty interval. Ridge plots show the complete uncertainty distribution associated with prevalence estimation. On the point estimation maps, areas with high uncertainty have hatching. On the uncertainty width maps, areas with sparse data may yield ill-defined uncertainty, and these are colored gray.

Developers

This website was developed by researchers at the University of Washington and the University of California Santa Cruz, with support from the Gates Foundation.