A cross-validation estimator that splits each A/B test's units eliminates the winner's-curse bias in evaluating decision rules across many weak experiments, with theory, simulations, and a Netflix case study reporting a 33% estimated gain.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ME 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Evaluating Decision Rules Across Many Weak Experiments
A cross-validation estimator that splits each A/B test's units eliminates the winner's-curse bias in evaluating decision rules across many weak experiments, with theory, simulations, and a Netflix case study reporting a 33% estimated gain.