SUTVA
The Stable Unit Treatment Value Assumption: a unit’s potential outcomes depend only on its own treatment, not on the treatments assigned to others (no interference), and there is a single, well-defined version of each treatment (no hidden variation). SUTVA is what lets and be well-defined quantities in the Potential-Outcomes framework.
Named by Rubin (1980, 1986); the no-interference idea is already operating in the population-level “statistical solution” of Holland1986-StatisticsAndCausalInference and in Rubin1980-RandomizationAnalysis.
Relied on by
All potential-outcomes methods: RCT, DiD, RDD, PSM, SCM.
Referenced by
Rubin1980-RandomizationAnalysis (names it); Holland1986-StatisticsAndCausalInference (population-level no-interference in the statistical solution).
DiD/IV pass: Angrist2022-EmpiricalStrategiesIlluminatingPath, AtheyImbens2006-NonlinearDiD, AtheyImbens2022-DesignBasedDiD, CallawaySantAnna2021-DiDMultiplePeriods, SantAnnaZhao2020-DoublyRobustDiD, DeChaisemartinDHaultfoeuille2018-FuzzyDiD, and the broader DiD note set.
Method-buildout pass (2026-06-17) — the no-interference/no-hidden-versions base across SCM (AbadieEtAl2010-SyntheticControlMethods, AbadieEtAl2015-ComparativePoliticsSCM, Abadie2021-UsingSyntheticControls), PSM (Imbens2004-NonparametricATEReview, Imbens2015-MatchingMethodsInPractice, CaliendoKopeinig2008-PSMImplementationGuidance, SmithTodd2005-ReconcilingPSMEvidence), RDD (LeeLemieux2010-RDDInEconomics, ImbensLemieux2008-RDDGuideToPractice, CattaneoEtAl2020-RDDHandbook), RCT (AtheyImbens2017-EconometricsOfRandomizedExperiments, BruhnMcKenzie2009-InPursuitOfBalance), IV (AngristKrueger2001-SearchForIdentification, Imbens2014-IVEconometriciansPerspective, Murray2006-WeakAndInvalidInstruments, RosenzweigWolpin2000-NaturalNaturalExperiments), and surveys/inference (AbadieCattaneo2018-EconometricMethodsProgramEvaluation, ImbensWooldridge2009-ProgramEvaluation, AbadieEtAl2020-SamplingVsDesignUncertainty, Keele2015-CausalMediationAnalysis, VanderWeele2016-MediationAnalysis, LudwigEtAl2011-MechanismExperiments). Its no-interference clause is relaxed by AngelucciDiMaro2016-SpilloverEffects.
New-papers pass (2026-07-04): ArkhangelskyImbens2024-FixedEffectsGeneralizedMundlak (no cross-group spillovers; flagged as restrictive for blood-drive data).
New-papers pass (2026-07-13): BorusyakEtAl2024-RevisitingEventStudyDesigns (potential outcomes indexed by own treatment path only).