Small Area Estimation of Poverty and Wealth Using Geospatial Data: What have We Learned So Far?

Abstract

This article offers a non-technical review of selected applications that combine survey and geospatial data to generate small area estimates of wealth or poverty. Publicly available data from satellites and phones predict poverty and wealth accurately across space, when evaluated against census data, and their use in model-based estimates improves the accuracy and efficiency of direct survey estimates. Models based on interpretable features appear to predict more accurately than estimates derived from convolutional neural networks. Estimates for sampled areas are significantly more accurate than those for non-sampled areas due to informative sampling. In general, estimates benefit from using geospatial data at the most disaggregated level possible. Tree-based machine learning methods appear to generate more accurate estimates than linear mixed models in common settings. Small area estimates using geospatial data can improve the design of social assistance programmes, particularly when the existing targeting system is poorly designed. AMS subject classification: 62P20