Methods

Vectorial  and raster collected data sets have been gathered into one single shapefile using Quantum GIS (QGIS ,référence) software. Larger scaled data sets (domestic violence on women and neo-mortality rate) have been resampled to the Second-Order Administrative Division (SOAD) scale, or equivalently, the district scale. To do this, they were exported as a raster image with resolution of XX m2 with the Rasterize (référence?) module of QGIS, and finally polygonized at the SOAD scale using the Spatial Statistics plugin (référence?) of QGIS. They has been polygonized by taking the majority value contained inside each SOAD polygon, in order to avoid edge effects between two provinces. Neo-mortality rate have finally been multiplied by the population of women per district, which is contained in the census data set, and multiplied by 1000 to obtain an estimation of the number of new born mortality per district.

Travel time to closest facility has been estimated at the district scale by taking the mean of the values contained inside the district polygons.

Datasets have been explored using GeoDa software. Regressions were produced with the regression tool of GeoDa. The multivariate linear regression was done according to the following formula:

The dependent variable, neo_mort, is the mortality rate per 1000  during the first 28 days of life. The independent or explicative variables are: health_dist, which is proportional to the distance in minutes that it takes to reach the nearest health center; elec, that represents the percentage of households with access to electricity; household_size, that represents the average size in family members for each district; vio_tot, that represents the rate of domestic violence on women.