No Anticipation
The assumption that units do not respond to a treatment before it is implemented: potential outcomes in pre-treatment periods are unaffected by future treatment status. It is what licenses using pre-treatment periods as a clean baseline and interpreting event-study “lead” coefficients as a test of Parallel-Trends rather than as real effects. Often paired with parallel trends as the second pillar of modern DiD identification.
Relied on by
DiD and event-study designs (especially staggered/dynamic estimators); also a synthetic-control requirement (no pre-treatment response to the intervention).
Referenced by
- AtheyImbens2022-DesignBasedDiD
- CallawaySantAnna2021-DiDMultiplePeriods
- CallawayGoodmanBaconSantAnna-ContinuousTreatment
- SunAbraham2021-EventStudies
- Wooldridge2021-TWFEMundlakDiD
- DeChaisemartinDHaultfoeuille2020-IntertemporalTE
- DeChaisemartinDHaultfoeuille2023-TWFESurvey
- RambachanRoth2023-MoreCredibleParallelTrends
- RothPretrends2022-PretestWithCaution
- RothSantAnna2023-WhatsTrendingInDiD
- BakerEtAl2025-DiDPractitionerGuide
- Method-buildout pass (2026-06-17), SCM: AbadieEtAl2010-SyntheticControlMethods, AbadieEtAl2015-ComparativePoliticsSCM, Abadie2021-UsingSyntheticControls
New-papers pass (2026-07-04): KahnLangLang2020-PromiseAndPitfallsOfDiD (implicit in the pre/post framing of the counterfactual).
New-papers pass (2026-07-13): BorusyakEtAl2024-RevisitingEventStudyDesigns (no-anticipation stated explicitly and tested; fully dynamic specs without never-treated units are under-identified when anticipation is not ruled out).