Accurate and spatially-disaggregated estimates of economic well-being and poverty are fundamental to research and policymaking on economic development and poverty eradication in developing countries. A central challenge researchers and policymakers face, however, is that such estimates typically do not exist. Our goal is to fill this gap by providing spatially disaggregated estimates of consumption per capita and poverty for sub-Saharan Africa over time.
"Spatial economic development in Africa" provides spatial estimates of consumption per capita as well as the share and the total number of people living with less than $1.90 a day (2011 PPP dollars) in 10x10 km pixels in 42 sub-Saharan African countries from 2003-2018. A brief summary of our methods is found here
and a full description is found
This webpage makes it possible to explore the data using cutting-edge visualization tools. It also allows users to download the data. Finally, this page also provides the files necessary for adopting and implementing the methodology used here.
Traditional approaches to measuring economic well-being in the developing world usually rely either on household surveys or, more recently, on satellite images of nightlights. The utility of survey data is limited by the fact that they are conducted infrequently and cover only a fraction of a country’s territory. The use of nightlights partially overcomes these limitations, as global coverage exists over time. However, nightlight measures provide very little information for roughly half the population in Africa who live in areas that are dark at night. And in areas that are not dark, actual levels of material well-being are typically impossible to discern from measures of luminosity since nightlight intensity has no substantively interpretable metric.
To address these limitations, we divide Sub-Saharan Africa in 10x10 km2 cells and use machine learning techniques to predict consumption per capita and poverty levels. One challenge associated with this task is the lack of a good training variable, as geocoded surveys measuring consumption per capita in Africa are very scarce. We address this challenge by providing a framework for transforming asset-based measures of well-being from the Demographic and Health Surveys (DHS) into consumption per capita measures. DHS data from more than 900,000 households makes it possible to estimate the average level of consumption in over 34,000 geocoded DHS clusters. Random forest models that consider a wide range of predictors, including nightlights, are used to train prediction models on the cluster-level measure of consumption. We compute the estimates of consumption using the random forest model, and use these estimates, along with the household surveys from DHS, to compute poverty rates non-parametrically in all 10x10km2 cells in Africa over time.
is Professor of Political Science at Columbia University and
is a tenured researcher at the Institute for Economic Analysis (CSIC) and affiliated professor to the Barcelona School of Economics. Invaluable research assistance for the project has been provided by Ella Bayi, Martin Devaux, Dylan Groves, Salif Jaiteh and Adriana Mallol.
The project was made possible through financial support from “la Caixa” Foundation Research Grants on Socio-economic Well-being through the project “Inequality, Political Instability and Long-term Development," Columbia University, the Barcelona School of Economics, and from the project PID2021-124256OB-I00 funded by MCIN/AEI/10.13039/501100011033 and by ERDF A way of making Europe.