Multi-period Difference-in-Differences (Staggered DiD)
Multi-period Difference-in-Differences with Staggered Adoption · Also known as: staggered DiD, multi-period DiD, staggered difference-in-differences, heterogeneous timing DiD
Multi-period Difference-in-Differences extends the classic two-period DiD framework to settings where units adopt treatment at different points in time. Formalised by Callaway and Sant'Anna (2021) and Goodman-Bacon (2021), it decomposes the overall treatment effect into group-time average treatment effects and addresses the bias that arises when conventional two-way fixed-effects regressions are applied to staggered adoption designs.
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 multi-period DiD when you have panel data with staggered treatment adoption — that is, different units receive the intervention at different calendar dates — and you want unbiased causal estimates free from the contamination problem of two-way fixed effects. Ideal when you have at least a handful of cohorts and sufficient never-treated or not-yet-treated units to serve as a clean comparison. Avoid it when virtually all units are treated at the same time (use standard DiD instead), when the panel is very short with only one pre-period, or when treatment timing is determined by the outcome itself (reverse causality).
Strengths & limitations
- Corrects the well-documented negative-weighting bias of two-way fixed-effects regressions applied to staggered adoption designs.
- Produces group-time treatment effects that reveal treatment effect heterogeneity across cohorts and over time since adoption.
- Event-study plots of cohort-specific dynamics make anticipation effects and effect build-up directly visible.
- Uses never-treated and not-yet-treated units as clean controls, avoiding the invalid use of already-treated units as a comparison.
- Callaway-Sant'Anna estimator allows both unconditional and conditional (covariate-adjusted) parallel-trends assumptions.
- Requires a sufficient number of never-treated or not-yet-treated units; if nearly all units are eventually treated, the control group shrinks and precision falls.
- With many cohorts and many time periods, the number of ATT(g,t) cells grows rapidly, and some cells may have too few observations for reliable estimation.
- The parallel-trends assumption must hold separately within each cohort, which is a stronger requirement than in a single-cohort design.
- Aggregation of cohort-time effects into a summary ATT involves weighting choices that must be made explicit and defended.
- Software implementation (e.g., csdid in Stata, did package in R) is more complex than a standard two-way fixed-effects regression.
Frequently asked
Why does standard two-way fixed effects fail with staggered treatment?
TWFE uses all available 2x2 DiD comparisons, including ones where already-treated units serve as controls for later-treated cohorts. When treatment effects are heterogeneous across cohorts or evolve over time, these comparisons receive negative weights, and the TWFE coefficient can be far from any reasonable average treatment effect — and can even have the wrong sign.
What is a treatment cohort?
A cohort groups all units that first receive treatment in the same calendar period. Units treated in 2010 form one cohort, units treated in 2012 form another, and so on. Never-treated units form a separate group used as a clean comparison.
What comparison group should I use?
The preferred comparison is never-treated units, because they are clean controls for every cohort across every post-treatment period. If no never-treated units exist, not-yet-treated units (those not yet treated as of period t) can serve as a comparison, though this requires stronger assumptions.
How do I present results from a multi-period DiD?
Report both the aggregated overall ATT and an event-study plot showing ATT(g,t) by relative time to treatment adoption. The pre-period estimates serve as a parallel-trends check, and the post-period dynamics reveal whether effects grow, shrink, or persist.
Which software implements multi-period DiD?
The csdid command in Stata and the did package in R (by Callaway and Sant'Anna) implement the group-time ATT estimator directly. The sunab command in Stata implements the Sun-Abraham interaction-weighted estimator as an alternative.
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 ↗
- Goodman-Bacon, A. (2021). Difference-in-differences with variation in treatment timing. Journal of Econometrics, 225(2), 254-277. DOI: 10.1016/j.jeconom.2021.03.014 ↗
How to cite this page
ScholarGate. (2026, June 3). Multi-period Difference-in-Differences with Staggered Adoption. ScholarGate. https://scholargate.app/en/causal-inference/multi-period-difference-in-differences
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.
- Difference-in-DifferencesEconometrics↔ compare
- Dynamic Difference-in-DifferencesCausal inference↔ compare
- Panel Data Difference-in-DifferencesCausal inference↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Synthetic Control MethodCausal inference↔ compare