Exclusion Restriction
The core identifying assumption of instrumental variables: the instrument affects the outcome only through its effect on the treatment, and through no other channel. Combined with instrument independence (Randomization/as-good-as-random assignment), Monotonicity, and a first stage, it delivers the LATE. Unlike independence — which random assignment can guarantee — the exclusion restriction is substantive and untestable, so it must be defended with a clear, consistent causal story.
Relied on by
Instrumental variables (IV), fuzzy RD, and fuzzy DiD.
Referenced by
- Angrist2022-EmpiricalStrategiesIlluminatingPath
- AngristKrueger1999-EmpiricalStrategiesLaborEconomics
- AngristPischke2010-CredibilityRevolution
- Method-buildout pass (2026-06-17): AngristKrueger2001-SearchForIdentification, Imbens2014-IVEconometriciansPerspective, Murray2006-WeakAndInvalidInstruments, RosenzweigWolpin2000-NaturalNaturalExperiments (its violation is the central critique); fuzzy RD in LeeLemieux2010-RDDInEconomics, ImbensLemieux2008-RDDGuideToPractice, CattaneoEtAl2020-RDDHandbook; noncompliance in AtheyImbens2017-EconometricsOfRandomizedExperiments; surveys AbadieCattaneo2018-EconometricMethodsProgramEvaluation, ImbensWooldridge2009-ProgramEvaluation.
New-papers pass (2026-07-04): FreyaldenhovenEtAl2019-PreEventTrendsPanelEventStudy (the unaffected covariate must respond to the confound but not the policy; policy leads serve as excluded instruments).
New-papers pass (2026-07-20) — shift-share/Bartik designs, where exclusion is argued at the level (share or shock) taken as exogenous: BlanchardKatz1992-RegionalEvolutions and Card2001-ImmigrantInflows and AutorDornHanson2013-ChinaSyndrome (applied); GoldsmithPinkhamEtAl2020-BartikInstruments (exclusion at the share level) versus BorusyakEtAl2022-QuasiExperimentalShiftShare (exclusion at the shock level).