Reproduce “Density Ratio-Free Doubly Robust Proxy Causal Learning” 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
Bariscan Bozkurt, Houssam Zenati, Dimitri Meunier, Liyuan Xu, Arthur Gretton
The paper combines outcome-bridge and treatment-bridge estimates for hidden-confounding causal effects, using proxies and a correction for their overlap. DRKPV and DRPMMR use conditional mean embeddings without explicit density-ratio estimation or treatment smoothing. Both outperform baselines on dSprite; in Job Corps Setting 1, DRKPV’s distance to a Kernel-ATE oracle is 0.87 ± 0.30, a semi-synthetic result rather than ground-truth error.
Key insights
- A novel doubly robust identification formula ensures consistent causal effect estimation if either the outcome bridge or treatment bridge function is correctly specified, providing robustness against model misspecification.
- Avoiding explicit density ratio estimation and kernel smoothing is critical for scaling PCL effectively to continuous and high-dimensional treatment variables.
- The use of Conditional Mean Embeddings enables the derivation of simple, closed-form estimators, improving computational efficiency and theoretical tractability.
- Rigorous uniform consistency guarantees confirm the theoretical soundness of the proposed non-asymptotic kernel-based estimators.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.