Small Area Estimation of Monetary Poverty in Mexico Using Satellite Imagery and Machine Learning

ABSTRACT Estimates of poverty are an important input into policy formulation in developing countries, making the accurate measurement of poverty rates a first‐order problem for development policy. This paper shows that combining satellite imagery with household surveys can improve the accuracy and precision of estimated poverty rates in Mexican municipalities, a level at which the survey is not considered representative. It also shows that empirical best prediction (EBP) based on a twofold household‐level model outperforms EBPs based on other common small area estimation models. These results indicate that the incorporation of household survey data and widely available satellite imagery can improve poverty estimates in developing countries, even for small subgroups.