By Peter Gething, Andy Tatem, Tom Bird, and Clara R Burgert.
Abstract
Improved understanding of sub-national geographic variation and inequity in demographic and health indicators is increasingly recognized as central to meeting development goals. Data from DHS surveys are critical to monitoring progress in these indicators but are generally not used to support sub-national evaluation below the first-level administrative unit. This study explored the potential of geostatistical approaches for the production of interpolated surfaces from GPS cluster located survey data, and for the prediction of gridded surfaces at 5×5km resolution. The impact of DHS cluster displacement on these interpolated surfaces and the particular challenges of mapping in highly heterogeneous urban areas were also investigated. A flexible and robust geostatistical framework was proposed for generating and validating interpolated surfaces with georeferenced DHS survey data. The framework was tested by using four indicators from three different surveys. The accuracy of the interpolated surfaces varied between settings, and was driven by the spatial structure of each indicator and its relationship to available covariate data. The random displacement of DHS cluster geopositioning information reduced the precision of predicted maps, although the impact varied between settings and was generally modest. Over shorter distances, the greater degree of geographical heterogeneity that was associated with urban areas meant they were more sensitive to the impact of cluster displacement. High-resolution covariates and novel statistical approaches showed potential for improving mapping in these areas. This study demonstrated that, with appropriate modeling and validation, such data have broad utility for creating maps of a wide range of indicators that support improved geographically stratified decision-making.