Reproduce “Optimal transport weights for causal inference” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Eric Dunipace
Causal Optimal Transport (COT) introduces a model-free weighting methodology for causal inference in observational studies, directly balancing covariate distributions through the minimization of Sinkhorn divergence. The method achieves semiparametric efficiency and double robustness, consistently showing the lowest RMSE and bias in simulations, and superior inference performance in a real-world case study compared to existing methods.
Key insights
- COT provides a unifying theoretical framework, demonstrating that methods such as Synthetic Control Method, Nearest Neighbor Matching, Optimal Matching, Energy Distance, and Mean Maximum Discrepancy can be viewed as special cases or closely related formulations of the COT framework.
- The COT estimator is proven to be semiparametrically efficient and exhibits a form of double robustness, ensuring consistency if either the optimal transport distance between weighted distributions converges sufficiently fast or if chosen balancing basis functions are correctly specified.
- COT offers a mechanism for nonparametric imputation of missing potential outcomes by estimating the conditional mean outcome function via barycentric projection, with a theoretical convergence rate of n^(-2/(2d'+9)).
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.