Poverty, Inequality and Development Studies with Machine Learning

This chapter provides a hopefully complete ‘ecosystem’ of the literature on the use of machine learning (ML) methods for poverty, inequality, and development (PID) studies. It proposes a novel taxonomy to classify the contributions of ML methods and new data sources used in this field. Contributions lie in two main categories. The first is making available better measurements and forecasts of PID indicators in terms of frequency, granularity, and coverage. The availability of more granular measurements has been the most extensive contribution of ML to PID studies. The second type of contribution involves the use of ML methods as well as new data sources for causal inference. Promising ML methods for improving existent causal inference techniques have been the main contribution in the theoretical arena, whereas taking advantage of the increased availability of new data sources to build or improve the outcome variable has been the main contribution in the empirical front. These inputs would not have been possible without the improvement in computational power.