Seminars

View all Seminars  |  Download ICal for this event

What Makes Treatment Effects Identifiable? A Learning-Theoretic Approach

Series: Bangalore Theory Seminars

Speaker: Anay Mehrotra, Yale University

Date/Time: Aug 22 11:00:00

Location: CSA Auditorium, (Room No. 104, Ground Floor)

Abstract:
Sometimes, like when studying the effects of smoking, it is impossible to run a randomized control trial and we must rely on observational data; where we observe who chose to receive treatment but do not control the assignment. This leads to a fundamental challenge: we only see one outcome per individual (their health either as a smoker or non-smoker, never both), making it impossible to directly observe causal effects. Unlike typical machine learning tasks where given enough data and computational power learning is always possible, causal inference requires assumptions to enable learning. Unfortunately, these assumptions frequently fail ?? severely limiting when causal effects can be identified.

In this talk, I will present a learning-theoretic characterization of when treatment effects are identifiable from observational data, which unifies existing approaches and enables exact identification of treatment effects in scenarios where current methods only provide approximate bounds. These scenarios include classical models (e.g., sensitivity analysis and regression discontinuity designs). The characterization bridges causal inference with classical learning theory and opens exciting new avenues for causal estimation when standard assumptions fail.

This talk is based on joint work with Yang Cai, Alkis Kalavasis, Katerina Mamali, and Manolis Zampetakis.


Microsoft teams link:

Link


We are grateful to the Kirani family (Link and the Walmart Center for Tech Excellence (Link for generously supporting this seminar series


Hosts: Rameesh Paul, KVN Sreenivas, Rahul Madhavan, Debajyoti Kar, Nirjhar Das