Linearization: Turning Ratio Metrics into Per-User Metrics
A simple algebraic transformation that converts ratio metrics into independent per-user values, making T-tests and sensitivity techniques directly applicable
Variance-Reduction
Delta Method for Ratio Metrics in AB Testing
Correct variance estimation for ratio metrics using the Delta Method — and why naive approaches quietly break your confidence intervals
CUPED: Reducing Metric Variance with Pre-Experiment Data
How CUPED uses pre-experiment covariates to cut metric variance and boost AB test sensitivity — with a rigorous proof of what works and what silently breaks
Bootstrap Methods for AB Testing
Classical and Poisson bootstrap as non-parametric tools for normalizing skewed metrics and enabling valid T-tests on small samples
Bucketing: Variance Reduction and Faster AB Tests
How aggregating observations into buckets normalises skewed metrics, reduces variance, and cuts computation — with a probability proof for choosing the right bucket count