Staggered Difference-in-Differences
Staggered Difference-in-Differences (Callaway-Sant'Anna / Sun-Abraham Estimators) · Also known as: staggered DID, staggered adoption DID, heterogeneous treatment DID, Callaway-Sant'Anna estimator, Sun-Abraham estimator, Kademeli DID (Staggered Difference-in-Differences)
Staggered Difference-in-Differences is a generalisation of DID for panel designs in which treatment is rolled out to different groups at different times. Introduced in the modern form by Callaway and Sant'Anna (2021) and Sun and Abraham (2021), it corrects the bias that classical two-way fixed-effects (TWFE) estimators suffer when treatment effects are heterogeneous across cohorts and over time.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use staggered DID when you have panel or repeated time-series data and treatment is adopted at different times by different groups, with each unit's first treatment date known. It requires a parallel-trends assumption that must hold separately for each cohort, treatment that is absorbing (once treated, stays treated) or whose reversibility is modelled explicitly, and the availability of never-treated or not-yet-treated units to serve as controls. A reasonable sample is needed (at least about 100 observations) so each cohort has enough events. Prefer it over classical TWFE whenever treatment timing varies and effects may differ across cohorts.
Strengths & limitations
- Corrects the negative-weighting bias that contaminates classical two-way fixed-effects estimates under heterogeneous, staggered treatment.
- Estimates a clean effect for every cohort and period, which can be aggregated into an overall effect or a dynamic event-study path.
- Uses only valid comparison groups (never-treated or not-yet-treated), so already-treated units are never misused as controls.
- Requires the parallel-trends assumption to hold separately for each cohort, which is harder to satisfy than in a single two-group design.
- Needs a sizeable sample (about 100 or more observations) so that each cohort has enough events; otherwise the heterogeneity correction is unreliable.
- Depends on the availability of clean control units; without never-treated or not-yet-treated groups, identification breaks down.
Frequently asked
Why not just use a two-way fixed-effects regression?
Under staggered adoption with heterogeneous effects, TWFE implicitly uses already-treated units as controls for later-treated ones and can assign negative weights to valid effects, biasing or even reversing the estimate. Staggered DID estimators avoid this by only comparing against clean (never-treated or not-yet-treated) controls.
What is a treatment cohort?
A cohort is the set of units that first receive treatment at the same time. Defining cohorts by their first treatment date is what lets the method estimate a separate, clean effect for each group and period.
What does the parallel-trends assumption require here?
It requires that, absent treatment, each treated cohort's outcome would have moved in parallel with the comparison group's. Because there are multiple cohorts, this must hold for each cohort separately, not just on average.
What if there are no never-treated units?
The method can use not-yet-treated units as the comparison group. If even that is unavailable, or if parallel trends cannot be defended per cohort, an event-study design or a synthetic-control approach may be more appropriate.
Sources
- Callaway, B. & Sant'Anna, P. H. C. (2021). Difference-in-Differences with Multiple Time Periods. Journal of Econometrics, 225(2), 200-230. DOI: 10.1016/j.jeconom.2020.12.001 ↗
- Sun, L. & Abraham, S. (2021). Estimating Dynamic Treatment Effects in Event Studies with Heterogeneous Treatment Effects. Journal of Econometrics, 225(2), 175-199. DOI: 10.1016/j.jeconom.2020.09.006 ↗
How to cite this page
ScholarGate. (2026, June 1). Staggered Difference-in-Differences (Callaway-Sant'Anna / Sun-Abraham Estimators). ScholarGate. https://scholargate.app/en/causal-inference/did-staggered
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Event Study DesignCausal inference↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Regression DiscontinuityCausal inference↔ compare
- Synthetic ControlCausal inference↔ compare