Yale researchers Yang Cai, Alkis Kalavasis (FDS Postdoc), Katerina Mamali, Anay Mehrotra, and Manolis Zampetakis have won the Best Paper Award at the 2025 Conference on Learning Theory (COLT) for their foundational work on causal inference.
The paper addresses a central challenge in data science: how to determine cause and effect from observational data. Imagine trying to figure out whether giving kids candy makes them happier. If you can’t run a controlled experiment — maybe you just observe who got candy and who didn’t — how can you be sure the candy caused the happiness, rather than something else like mood or personality?
To answer such questions, researchers often assume unconfoundedness, the idea that nothing hidden is influencing both who gets candy and how happy they are. But in many real-world settings, this assumption doesn’t hold.
This paper offers a unified condition that precisely characterizes when causal effects are still identifiable from observational data, even when common assumptions break down. The framework applies to a wide range of settings — from standard ones to more structured designs like regression discontinuity — and brings clarity to longstanding questions in the field.
The result is a rigorous foundation for causal reasoning, with wide potential implications for science, policy, and machine learning.
Read the paper: https://arxiv.org/abs/2506.04194
