Income Distribution and Economic Development: Insights From Machine Learning

The authors draw upon recent advances that combine causal inferences with machine learning, to show that poverty is the key income distribution measure that matters for development outcomes. In a predictive framework, the authors first show that LASSO chooses only the headcount measure of poverty from 37 income distribution measures in predicting schooling, institutional quality, and per capita income. Next, causal inferences with post-LASSO models indicate that poverty matters more strongly for development outcomes than does the Gini coefficient. Finally, instrumental variable estimates in conjunction with post-LASSO models show that compared to Gini, poverty is more strongly causally associated with schooling and per capita income, but not institutional quality. The authors' results question the literature's overwhelming focus on the Gini coefficient. At the least, their results imply that the causal link from inequality (as measured by Gini) to development outcomes is tenuous.