Causal Estimand
The specific causal quantity a study aims to learn, defined in terms of Potential-Outcomes. Common estimands:
- ATE — Average Treatment Effect, over the whole population.
- ATT — Average Treatment effect on the Treated, .
Holland’s “statistical solution” to the Fundamental-Problem-of-Causal-Inference replaces the unobservable unit-level effect with an average causal effect; that average is the estimand the Randomization-based or observational design then tries to identify and estimate.
Relied on by
All methods — each targets a specific estimand and should state which.
Referenced by
Rubin1974-EstimatingCausalEffects, Rubin1977-AssignmentOnCovariate, Rubin1978-BayesianCausalEffects, Rubin1980-RandomizationAnalysis, Holland1986-StatisticsAndCausalInference.
DiD/IV pass — every paper note in those folders names an estimand, e.g. AngristPischke2010-CredibilityRevolution, Angrist2022-EmpiricalStrategiesIlluminatingPath (the LATE), CallawaySantAnna2021-DiDMultiplePeriods (group-time ATT), GoodmanBacon2021-DiDVariationInTiming, RothSantAnna2023-WhatsTrendingInDiD.
Method-buildout pass (2026-06-17): ATT/ATE targets in AbadieEtAl2010-SyntheticControlMethods, Imbens2004-NonparametricATEReview, Imbens2015-MatchingMethodsInPractice, CaliendoKopeinig2008-PSMImplementationGuidance, SmithTodd2005-ReconcilingPSMEvidence, LeeLemieux2010-RDDInEconomics (effect at the cutoff), AtheyImbens2017-EconometricsOfRandomizedExperiments, BruhnMcKenzie2009-InPursuitOfBalance, RosenzweigWolpin2000-NaturalNaturalExperiments, AbadieCattaneo2018-EconometricMethodsProgramEvaluation, ImbensWooldridge2009-ProgramEvaluation.
New-papers pass (2026-07-04): ArkhangelskyImbens2024-FixedEffectsGeneralizedMundlak (FE estimands under heterogeneity are sampling-scheme-dependent weighted averages); FreyaldenhovenEtAl2019-PreEventTrendsPanelEventStudy (the policy effect in a linear panel event-study); KahnLangLang2020-PromiseAndPitfallsOfDiD (what the DiD interaction identifies depends on the assumed scale/counterfactual).
New-papers pass (2026-07-13): BorusyakEtAl2024-RevisitingEventStudyDesigns (target is a researcher-chosen weighted sum of heterogeneous effects, specified explicitly and separately from the identifying assumptions).