In many countries, particularly in sub-Saharan Africa, Demographic and Health Surveys (DHSs) are the main way of estimating HIV prevalence nationally in the general population. Some DHSs record the longitude and latitude of the survey clusters.
We present three methodological approaches for mapping spatial variations in HIV prevalence using the DHSs. These approaches are applied to simulated DHS samplings from a model country. The estimated surfaces are then compared with the model’s initial surface.
We demonstrate that a method using kernel estimators with adaptive bandwidths size of equal number of persons observed can be used to estimate the main regional trends in epidemics. Application to Burkina Faso’s 2003 DHS data provides a plausible image of that country’s epidemiological situation.