Jose Peña: Causal Machine Learning Using Normalizing Flows, with Applications in Sociology

Abstract

In this talk, I will present our work on deep learning for causal inference. Specifically, I will show how we use normalizing flows for computing causal effects at population, subpopulation and even individual level. Our end goal is to use these methods to assist the production of personalized public policies. To illustrate this, I will present our results on the impact of the International Monetary Fund (IMF) program on child poverty using real-world observational data of about 2 million children living in 67 countries from the Global-South. While the primary objective of the IMF is to support governments in achieving economic stability, our results indicate that the IMF program is beneficial for reducing child poverty, and the more personalized the program the better. I will also touch upon our approach to deal with unmeasured confounding via copulas, as well as on the combination of observational and experimental data.