Probabilities of Causation: From Foundations to Decision Making (Oct 2)

Probabilities of Causation: From Foundations to Decision Making

Ang Li · Department of Computer Science Seminar

Abstract

Decision makers in education, psychology, and health seek not only effective interventions, but also principled explanations of whether those interventions are necessary, sufficient, or both for achieving desired outcomes. These questions require counterfactual reasoning beyond what average causal effects alone can reveal. Probabilities of causation (PoCs) provide a framework for quantifying these counterfactual relationships at the population and subpopulation levels, while also supporting individual-level analysis and personalized decision making.

In this talk, I will present our research on PoCs within Structural Causal Models, connecting mathematical foundations with causal decision making at the population, subpopulation, and individual levels. I will begin with theoretical advances in defining, bounding, and identifying PoCs. I will then examine their quantification under realistic conditions, including additional knowledge of causal structure, limited observational and experimental data, and finite-sample estimation and sample size requirements. Building on these foundations, I will introduce decision-making frameworks based on PoCs, including unit selection for targeting individuals or subpopulations with desired counterfactual response patterns. Finally, I will discuss applications and ongoing research in education, psychology, and health. Together, these research directions advance the foundations of counterfactual reasoning and enable more trustworthy, explainable, and personalized causal decision making.