Interpretable Machine Learning for Causal Inference
Abstract
We develop a framework in which the predictive power of Machine Learning (ML) methods can be leveraged toestimate causal effects in experiments, both randomized and natural. Our approach allows one to use any black box ML method, without making the mistake of confusing prediction with causal inference. We also develop a set of new ML methods for estimating individual treatment effects. We pay particular attention to issues of interpretability and causal heterogeneity. We develop new theory, methods, and software.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 04, 2017
- Source ID
- N000141712176
Entities
People
- Jasjeet Sekhon
Organizations
- Office of Naval Research
- United States Navy
- University of California Regents