Nonparametric Estimation of Average Treatment Effects under Exogeneity

Causal Question / Estimand

The ATE and ATT for a binary treatment under selection on observables — and a unified review of the estimators that target them.

Identification Strategy

Identification of treatment effects follows from two assumptions: unconfoundedness (Ignorability) — assignment is independent of potential outcomes given covariates — and Overlap — every covariate value has positive probability of both treatment states. Together these point-identify the ATE/ATT nonparametrically. The review then organizes the estimators that exploit this: regression on covariates, Propensity-Score methods (weighting, blocking/subclassification), matching, and doubly robust combinations, and discusses their large-sample efficiency (the efficiency bound) and variance estimation.

Key Assumptions

Ignorability (unconfoundedness / exogeneity), Overlap (positivity), SUTVA, and — for the propensity-score route — correct specification or nonparametric estimation of $e(X)$.

Threats to Validity

Failure of unconfoundedness (unobserved confounders); thin or violated overlap, where estimates are driven by extrapolation and variance explodes; misspecified propensity or regression models. Notes the role of trimming and overlap diagnostics.

Setting / Data

n/a — methodological review (econometric survey).

Key Claims

  • Under unconfoundedness + overlap, ATE/ATT are nonparametrically identified, and a large family of estimators (regression, matching, weighting, blocking) target the same parameter.
  • The propensity score reduces dimensionality but does not relax the identifying assumptions; combining outcome and propensity models yields efficiency/robustness.
  • Overlap is as consequential as unconfoundedness — limited overlap is a central practical obstacle.

Connections

Citation

Imbens, G. W. (2004). Nonparametric Estimation of Average Treatment Effects under Exogeneity: A Review. The Review of Economics and Statistics, 86(1), 4–29.