Heterogeneous Treatment Effect Event Study Design
Also known as: HTE event study, heterogeneous effects event study, group-time ATT event study, dynamic HTE design
Heterogeneous Treatment Effect Event Study Design is a causal-inference framework that uses event study regression to estimate how treatment effects vary across groups, cohorts, or time relative to a treatment event. Unlike classical two-way fixed-effects event studies — which assume a homogeneous effect — this approach explicitly models and recovers group-time average treatment effects (ATTs), addressing the contamination bias that arises when effects differ across treated units.
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 this design when you have panel data with a policy, treatment, or event affecting some units but not others, and you suspect that the treatment effect varies across cohorts, groups, or time since treatment — which is the realistic default in applied work. It is especially well-suited when treatment adoption is staggered across periods. It is not appropriate when you have only a single treated cohort and no reason to expect heterogeneity (a simple two-period DiD suffices), or when the panel is so short that multiple pre-treatment periods for testing parallel trends are unavailable.
Strengths & limitations
- Recovers honest, cohort-specific treatment effects instead of potentially sign-reversed TWFE averages when effects are heterogeneous.
- Provides a full dynamic event-study profile — how the effect evolves before and after treatment — while maintaining credibility under heterogeneous timing.
- Pre-trend tests are conducted for each cohort separately, offering sharper identification checks than a pooled TWFE test.
- Aggregation is flexible: the researcher can target average effects, cohort effects, or calendar-time effects depending on the policy question.
- Compatible with covariates and doubly robust extensions (Callaway & Sant'Anna 2021), adding robustness against misspecification of the parallel-trends assumption.
- Requires enough observations per cohort to estimate group-time ATTs precisely; thin cohorts increase variance substantially.
- The parallel-trends assumption must hold for each treated cohort against the comparison group, which is a stronger requirement than in a single-cohort design.
- Aggregated estimates depend on the chosen weighting scheme; different aggregations can yield different substantive conclusions, requiring transparent reporting.
- Implementation is more complex than a standard TWFE regression and requires specialised software (e.g., did, csdid, eventstudyinteract packages).
- With many cohorts and many periods, the number of ATT(g,t) cells grows rapidly, potentially leading to imprecise estimates if the panel is not large enough.
Frequently asked
Why can TWFE event-study coefficients be misleading under treatment effect heterogeneity?
In a staggered adoption setting, TWFE uses already-treated units as implicit controls for other treated units. When treatment effects differ across cohorts or grow over time, this comparison is invalid — earlier-treated units have non-zero 'treatment effect' status that contaminates the coefficient estimates, sometimes reversing their sign.
What is a group-time ATT and how is it identified?
ATT(g,t) is the average treatment effect for the cohort first treated in period g, measured at time t. It is identified by comparing the outcome path of cohort g to that of a clean comparison group — never-treated or not-yet-treated units — under the assumption that both groups would have followed parallel trends absent treatment.
How do I choose between the Sun-Abraham estimator and the Callaway-Sant'Anna estimator?
Both target the same group-time ATTs. Sun-Abraham uses a regression-based interaction-weighted approach that is easy to implement in standard regression software. Callaway-Sant'Anna offers doubly robust extensions and more flexible aggregation. In practice, results are often similar; the choice may depend on software availability and whether covariate adjustment is needed.
Can I still use a pre-trend plot to assess identification?
Yes, but pre-trend plots should be computed using the heterogeneity-robust estimator — not from TWFE — so that each relative-period coefficient reflects actual cohort-specific pre-trends rather than a contaminated TWFE average.
What sample size is needed?
There is no universal threshold, but each cohort (group g) needs enough units to estimate ATT(g,t) with reasonable precision. Thin cohorts — fewer than 20-30 units — will produce wide confidence intervals. The overall number of periods and cohorts also matters: more periods and cohorts improve the richness of the event-study profile but require larger panels.
Sources
- 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 ↗
- 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 ↗
How to cite this page
ScholarGate. (2026, June 3). Heterogeneous Treatment Effect Event Study Design. ScholarGate. https://scholargate.app/en/causal-inference/heterogeneous-treatment-effect-event-study-design
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 Event StudyCausal inference↔ compare