It’s Hard to Be Normal: The Impact of Noise on Structure-agnostic Estimation
Jikai Jin, Lester Mackey, Vasilis Syrgkanis · 2025
Read It’s Hard to Be Normal: The Impact of Noise on Structure-agnostic Estimation by Jikai Jin, Lester Mackey, Vasilis Syrgkanis (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-02275.txt.
Introduction
Jikai Jin, Lester Mackey, and Vasilis Syrgkanis show a surprising fact about estimating treatment effects, in a paper published in two thousand twenty-five. The best achievable accuracy depends on the shape of the treatment's noise. Their setting is structure-agnostic causal inference, where machine learning black boxes estimate the nuisance parts and the analyst still wants an honest effect estimate.
The workhorse estimator is double machine learning. For Gaussian treatment noise, the authors prove it is minimax optimal, settling an open problem. For independent but non-Gaussian noise, the same estimator is always beatable. They build new procedures, called ACE, that exploit higher-order cumulants of the treatment. Whenever a certain cumulant is nonzero, the new estimators gain extra insulation against errors in the nuisance estimates. The title's moral: Gaussianity, being normal, is exactly the hard case for doing better.
Catalog record
| CREATOR | Jikai Jin, Lester Mackey, Vasilis Syrgkanis |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 229 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-02275