Researchers Combine Satellite Data and Representation Learning to Assess Anti-Poverty Programs

Survey data alone do not provide the full picture for assessing who benefits from anti-poverty programs. In a collaboration with @UNICEFInnocenti and @UCLA led by @riccardocadeii, we show how to include satellite data using causal inference and representation learning. [1/3]

Our approach treats the CATE as a causal functional that interventions can manipulate (e.g., with new policies). We then turn pre-treatment unstructured data into interpretable representations and identify which factors causally drive heterogeneous treatment effects. [2/3]