Predicted poverty score, fifths of each city's cells
The score is the proxy means test (PMT) score that PASP uses to decide who is eligible. A higher score means a household is better off. Scores are much higher in Maputo than in the other cities, so each city is coloured on its own scale. The bairro map uses the same classes as the grid.
How these maps were made
The poverty score comes from the 2014/15 household budget survey (IOF) and the first round of PASP enrolment interviews: about 15,000 urban households with GPS locations. We used a convolutional neural network to measure the density and roof quality of buildings in Google satellite images of each cell. A random forest model then combined these measurements with distance to the city centre and OpenStreetMap roads to predict the average score in each cell.
A bairro's score is the average of the cells whose centre falls inside it. Ten percent of the surveyed cells were held back to test the model.
0.66R² at bairro level (out-of-sample)
0.83Rank correlation at bairro level (out-of-sample)
57,535115 m grid cells
186Bairros
Client: Government of Mozambique & the World Bank. Method and evaluation: Sohnesen, Fisker & Malmgren-Hansen, Using Satellite Data to Guide Urban Poverty Reduction, Review of Income and Wealth.