CAREER: Relational Causal Inference

University Of Illinois At Chicago, Chicago IL

Investigators

Abstract

Technologies powered by big data and artificial intelligence are ubiquitous in our daily lives. However, their sole reliance on data correlations implies an inability to reason about cause and effect, and the consequences of their own behavior. As human interactions and decision making are increasingly supported by automated digital platforms, there is an urgent need for causal inference techniques to adapt to the inherently relational data that these platforms produce and to support responsible data-driven technologies. This project introduces relational causal inference as the theory and algorithms that address this need. Through interdisciplinary collaborations, the project will demonstrate the applicability of relational causal inference to answering scientific questions from multiple disciplines. Through curriculum development and mentorship, it will equip high school, undergraduate, and graduate students with the skills necessary to develop modern-age data literacy and research skills. This project will investigate relational causal inference as the theory and algorithms for causal inference from observational data with underlying interference between units. The goal of this project is to identify and resolve barriers to relational causal inference from noisy, heterogeneous, relational data and to make causal inference practical and accessible for real-world applications. To achieve this goal, the project will integrate data-driven causal methods with relational learning and structural causal modeling and it will encompass the full data science lifecycle, from processing raw relational data to creating robust causal models and extracting actionable knowledge. The project has four main thrusts: 1) developing robust and scalable tests for relational dependence, 2) enabling relational causal discovery with latent variables and selection bias, 3) developing algorithms for heterogeneous peer effect estimation, and 4) enabling domain-specific discovery through relational causal inference. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.