Uplift Modeling: From Average Effect to Policy
Uplift modeling estimates the effect per user and turns it into who to treat. Meta-learners, the gain curve and the validation that cannot exist.
Complete, honest guides on CRO, A/B testing and experimentation: statistics, sample size, test design and what to do with the results. No fluff. (page 5 of 24)
Uplift modeling estimates the effect per user and turns it into who to treat. Meta-learners, the gain curve and the validation that cannot exist.
Causal mediation splits an effect into direct and indirect. The arithmetic, the post-treatment segmentation trap, and the assumption nobody tests.
When the feature cannot be switched off, you randomize the invitation. What an encouragement design measures, what it costs in sample, and where it breaks.
The row that never arrived changes the result. How to classify missing data, why complete cases overstate the effect, and what imputation actually fixes.
Interrupted time series: segmented regression separates step, trend and season from the growth that was already there before the launch.
Propensity scores collapse dozens of covariates into one number so you can pair treated with control. What that fixes, what it cannot, where it backfires.
Unmeasured confounding turns every observational read into doubt. Sensitivity analysis makes it a number: how strong the hidden factor would have to be.
Without randomizing users, difference-in-differences compares the treated change with the control change. How to run it, and why standard errors lie.
Regression discontinuity compares users just above a threshold against users just below. When it holds, when it lies, and the one test that settles it.