Glossary Entry

Conditional Average Treatment Effect

The average treatment effect among units sharing covariates x, written tau(x); the quantity uplift models estimate to decide whom to target.

Statistics Decision Making Models

Also called: CATE, heterogeneous treatment effects, uplift modeling, uplift model

Seed source: EconML documentation

The CATE function tau(x) = E[Y(1) - Y(0)

X = x] describes how a treatment’s effect varies across kinds of units, which is what a targeting decision actually needs: a program with a positive average effect can still be wasted on segments where its effect is zero or negative.

Industry calls estimating tau(x) uplift modeling, and sorts customers into persuadables (target them), sure things and lost causes (spending on them is waste), and do-not-disturbs (contact backfires). Meta-learners (S-, T-, X-learners) and causal forests are the standard estimators. The caveat: a CATE model inherits the identification of its training data, so heterogeneity modeling layers on top of a randomized experiment or a defensible unconfoundedness argument, never in place of one.