Résumé en anglais

I here use GIS and HIV data from five African countries to estimate the effect of road proximity on HIV infection. I find a negative effect of the distance to the nearest paved road on the probability of being infected with HIV: a one standard-deviation fall in this distance (approximately 2.4 km) increases the probability of infection by 0.6–2.0 percentage points. Using slope as an instrument for road distance continues to produce a negative and significant estimated coefficient. However this relationship may also reflect selection and reverse causality in individual choice of location, and I extensively discuss the role of migration. While the number of lifetime sexual partners is significantly influenced by the presence of roads in some recent years, the effect of road distance on access to protection has disappeared.