Design-based theory for causal inference from adaptive experiments
Xinran Li · 2026
Read Design-based theory for causal inference from adaptive experiments by Xinran Li (2026) 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-2602-21998.txt.
Description
In adaptive experiments, the chance a subject gets a treatment shifts as results roll in, which breaks the neat assumptions behind standard causal analysis. Xinran Li and Anqi Zhao build a theory for such experiments that drops the usual demand for identical, interchangeable subjects. Their finite-population framework allows units to differ and evolve over time, and covers treatment chances that stall, vanish, or never settle.
The authors justify causal estimates built from inverse-probability weighting and its augmented form, showing these stay valid even when outcome models come from black-box machine learning. They also add a covariance estimator that softens the overly cautious error bars typical of finite-population inference. The work gives experimenters firmer ground for trials that learn and adapt as they run.
Catalog record
| CREATOR | Xinran Li |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 171 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-21998