Glossary Entry

Average Treatment Effect

The mean difference between potential outcomes with and without treatment across a population, E[Y(1) - Y(0)]; the headline quantity most causal analyses target.

Statistics Metrics

Also called: ATE, average treatment effect on the treated, ATT

Seed source: Causal Inference - The Mixtape, ch. 4

The ATE averages individual treatment effects over everyone, sidestepping the impossibility of observing any single one. Restricting the average to those actually treated gives the ATT (average treatment effect on the treated), and to the untreated gives the ATU; the three differ whenever the people who take a treatment respond differently from those who do not.

Which average a method delivers is part of its fine print: adjustment and weighting methods usually target the ATE, difference-in-differences targets the ATT, and instrumental variables deliver a complier-specific average. Labeling an estimate with whose effect it measures is a core discipline of causal work.