WHOMP: Improving Upon Randomized Controlled Trials via Wasserstein Homogeneity
Published in Under review (Journal of the American Statistical Association), 2024
Summary
Comparative experiments often require splitting a population into subgroups (e.g., stratification, cohort construction, or controlled partitioning) while minimizing accidental bias and maximizing statistical efficiency.
This paper proposes WHOMP, a principled criterion built on Wasserstein homogeneity, and provides efficient algorithms to compute near-optimal subgroup splits at scale.
Key contributions
- Optimality criterion: A Wasserstein-based homogeneity objective that directly targets subgroup balance in a distributional sense.
- Algorithms: Practical estimators and solvers designed to be used in real experimental pipelines.
- Theory + analysis: Analytical comparisons clarifying when and why WHOMP improves upon classical baselines.
- Empirical validation: Numerical experiments demonstrating consistent gains across multiple realistic settings.
Resources
- Paper (PDF): https://arxiv.org/pdf/2409.18504
- Code: (add link when public)
- Slides: (add link if available)
Suggested BibTeX
```bibtex @misc{xu2024whomp, title = {WHOMP: Improving Upon Randomized Controlled Trials via Wasserstein Homogeneity}, author = {Xu, Shizhou and Strohmer, Thomas}, year = {2024}, note = {Under review at Journal of the American Statistical Association}, howpublished = {arXiv preprint}, }
Recommended citation: Shizhou Xu, Thomas Strohmer. (2024). “WHOMP: Improving Upon Randomized Controlled Trials via Wasserstein Homogeneity.” Under review at Journal of the American Statistical Association.
Download Paper | Download Slides