CUPED Variance Reduction Calculator
CUPED (Controlled-experiment Using Pre-Experiment Data) reduces the variance of an A/B test or clinical trial estimate by adjusting the outcome for a correlated pre-experiment covariate. Because sample size scales with variance, removing predictable variance lets you reach the same power with fewer participants.
How it works
The outcome is adjusted using a pre-experiment covariate that is correlated with it; the stronger that correlation, the larger the variance reduction and the smaller the required sample size. This calculator estimates the reduction and the resulting sample size, and can estimate the correlation directly from an uploaded dataset.
When to use it
- You have a stable pre-experiment metric (e.g., baseline revenue, prior activity, a pre-period lab value) correlated with your primary outcome.
- You want to reduce sample size or shorten trial duration while preserving power.
- The covariate is measured before treatment assignment, so adjustment does not depend on the treatment.
Assumptions & limitations
- Gains depend on the strength and stability of the pre/post correlation; a weak or unstable covariate yields little reduction.
- Measuring the covariate before treatment assignment helps preserve unbiased treatment-effect estimation under the planned adjustment model, but model misspecification, missing data, or post-hoc covariate selection can still introduce bias.
- Type I error control depends on prespecifying the adjustment model and implementing it correctly; the adjustment should be planned, not chosen after seeing outcomes.
For the full methodology, derivation, and worked examples, read the complete guide.