We 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, we first show that LASSO chooses only the headcount measure of poverty from 37 income...