Valid Post-Contextual Bandit Inference

Ramon van den Akker, Bas J.M. Werker, Bo Zhou · 2025

Read Valid Post-Contextual Bandit Inference by Ramon van den Akker, Bas J.M. Werker, Bo Zhou (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-2505-13897.txt.

Description

Ramon van den Akker, Bas Werker and Bo Zhou study valid inference after contextual bandit experiments, in work from 2025. Bandits sample treatments adaptively, after which researchers still want to test effects. Standard tests can fail in that setting, so the authors derive the limit experiment for multi-armed contextual bandits under weak-signal conditions.

Their limit takes a form they can express with stochastic differential equations. From it they get null distributions and local power for common test statistics. Two nuisance threats to size control are addressed, one by translation-invariant sampling and one by suitable pivots. New regularized versions of several sampling rules are proposed, and simulations support the theory.

Catalog record

CREATORRamon van den Akker, Bas J.M. Werker, Bo Zhou
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE147 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-13897

Browse the full library · About the Tower of Babel