Potential Outcomes

The framework in which each unit $u$ has a potential response under each treatment level — e.g. $Y_t(u)$ under treatment and $Y_c(u)$ under control — and the causal effect for that unit is the comparison $Y_t(u)-Y_c(u)$. Only one potential outcome is ever observed per unit (the other is counterfactual). Formalized for causal inference by Rubin1974-EstimatingCausalEffects and named the “Rubin Causal Model” by Holland1986-StatisticsAndCausalInference.

Relied on by

The entire potential-outcomes program: RCT, DiD, RDD, PSM, SCM — all are strategies for recovering an average of $Y_t-Y_c$ that the Fundamental-Problem-of-Causal-Inference makes unobservable per unit.

Referenced by

Rubin1974-EstimatingCausalEffects (origin), Rubin1977-AssignmentOnCovariate, Rubin1978-BayesianCausalEffects, Rubin1980-RandomizationAnalysis, Holland1986-StatisticsAndCausalInference (names it the “Rubin Causal Model”).

Credibility-revolution pass: AngristPischke2010-CredibilityRevolution, AngristKrueger1999-EmpiricalStrategiesLaborEconomics, Angrist2022-EmpiricalStrategiesIlluminatingPath.

Method-buildout pass (2026-06-17): AbadieEtAl2010-SyntheticControlMethods, Imbens2004-NonparametricATEReview, AtheyImbens2017-EconometricsOfRandomizedExperiments, Imbens2014-IVEconometriciansPerspective, AbadieCattaneo2018-EconometricMethodsProgramEvaluation, ImbensWooldridge2009-ProgramEvaluation, AbadieEtAl2020-SamplingVsDesignUncertainty, Keele2015-CausalMediationAnalysis, VanderWeele2016-MediationAnalysis.

New-papers pass (2026-07-04): KahnLangLang2020-PromiseAndPitfallsOfDiD (DiD identification stated in potential-outcomes terms against the experimental benchmark).