JRSSB Nov 17, 2026

A decorrelation method for general regression adjustment in randomized experiments

Authors
Fangzhou Su Wenlong Mou Peng Ding Martin J Wainwright
Research Topics
Machine Learning
Paper Information
  • Journal:
    Journal of the Royal Statistical Society Series B
  • DOI:
    10.1093/jrsssb/qkag145
  • Published:
    November 17, 2026
  • Added to Tracker:
    Nov 20, 2026
Abstract

Abstract We study regression adjustment with general function class approximations for estimating the average treatment effect in the design-based setting. Standard regression adjustment involves bias due to sample re-use, and controlling this bias either leads to suboptimal dependence on the sample size or requires restrictive assumptions. Our main contribution is to introduce a novel decorrelation approach that circumvents these issues. We prove guarantees, both asymptotic and nonasymptotic, relative to the oracle functions that are targeted by a given regression adjustment procedure. The same independence structure also yields conditional normal approximation bounds and finite-sample coverage guarantees that do not require consistency of the regression step. We illustrate our method by applying it to various high-dimensional and nonparametric problems, exhibiting improved sample complexity and weakened assumptions relative to known approaches.

Author Details
Fangzhou Su
Author
Wenlong Mou
Author
Peng Ding
Author
Martin J Wainwright
Author
Research Topics & Keywords
Machine Learning
Research Area
Citation Information
APA Format
Fangzhou Su , Wenlong Mou , Peng Ding & Martin J Wainwright (2026) . A decorrelation method for general regression adjustment in randomized experiments. Journal of the Royal Statistical Society Series B , 10.1093/jrsssb/qkag145.
BibTeX Format
@article{paper1931,
  title = { A decorrelation method for general regression adjustment in randomized experiments },
  author = { Fangzhou Su and Wenlong Mou and Peng Ding and Martin J Wainwright },
  journal = { Journal of the Royal Statistical Society Series B },
  year = { 2026 },
  doi = { 10.1093/jrsssb/qkag145 },
  url = { https://doi.org/10.1093/jrsssb/qkag145 }
}