Multi-period Event Study Design
Multi-period Event Study Design for Dynamic Treatment Effects · Also known as: multi-period event study, dynamic event study, relative-time event study, leads-and-lags design
The multi-period event study design estimates causal treatment effects at each point in time relative to the treatment onset, using panel data with multiple pre- and post-treatment periods. By plotting the full path of treatment coefficients rather than a single average, it reveals how effects build up, fade, or remain stable over time — and allows formal tests of pre-treatment parallel trends across many periods simultaneously.
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 the multi-period event study when you have panel data spanning multiple periods before and after a treatment, you want to see how the effect evolves over time rather than just an average, and you need to formally test the parallel-trends assumption across pre-treatment periods. It is the right choice for studying policies, economic shocks, or interventions whose effects are expected to grow, dissipate, or be heterogeneous across time horizons. Do not use it when you have only one pre-treatment period (a two-period DiD is sufficient), when the panel is too short to support estimation of multiple lags and leads, when staggered adoption is present but you rely on the naive TWFE estimator without correction, or when the treatment date varies widely across units with very sparse per-period cells.
Strengths & limitations
- Provides a full dynamic picture of treatment effects at each relative time period rather than a single aggregate number.
- Enables formal pre-trend tests across multiple pre-treatment periods, giving stronger evidence for or against the parallel-trends assumption.
- Reveals heterogeneity in timing: effects may build up, peak, or dissipate — information that a pooled DiD hides.
- Graphical output (event-study plot) is highly transparent and immediately interpretable for non-specialist audiences.
- Compatible with heterogeneity-robust estimators (e.g., Callaway-Sant'Anna, Sun-Abraham) when treatment timing is staggered.
- Requires a sufficiently long panel; short panels cannot support estimation of many relative-time coefficients without severe imprecision.
- Coefficient estimates become noisy at the tails of the relative-time window when few units contribute to those periods.
- In staggered adoption designs, the standard two-way fixed-effects (TWFE) estimator can produce biased or sign-reversed estimates if treatment effect heterogeneity is strong — requiring more complex heterogeneity-robust methods.
- Parallel-trends assumption is still untestable in the post-treatment period; passing pre-trend tests does not guarantee validity after treatment.
Frequently asked
How is this different from a standard two-period difference-in-differences?
A two-period DiD produces one estimate: the average treatment effect. The multi-period event study produces a separate estimate for each relative time period (e.g., k = -3, -2, -1, 0, 1, 2, 3), revealing the full dynamic path and allowing pre-trend tests across multiple pre-treatment periods.
What is the reference period and why does it matter?
One relative-time indicator must be omitted to avoid perfect multicollinearity; its coefficient is set to zero by definition. The convention is to omit k = -1 (one period before treatment). All estimated coefficients represent deviations from that baseline period, so the choice of reference period affects the visual level of the event-study plot but not the differences between adjacent estimates.
When should I use a heterogeneity-robust estimator instead of TWFE?
In staggered adoption designs — where different units are treated at different calendar times — the standard TWFE estimator implicitly uses already-treated units as controls for later-treated units, which can produce biased or sign-reversed coefficients when treatment effects vary across cohorts. Sun and Abraham (2021) and Callaway and Sant'Anna (2021) provide estimators that avoid this problem.
How many pre-treatment periods do I need?
At least two pre-treatment periods are needed to plot a pre-trend and run even a basic test. Three or more periods provide a meaningful joint test of parallel trends. Many empirical studies use four to six pre-treatment leads and a similar number of post-treatment lags, balancing informativeness against cell sparsity at long horizons.
What should I do if the pre-trend test fails?
A failed pre-trend test — non-zero, statistically significant pre-treatment coefficients — means the parallel-trends assumption is implausible for this design. Consider whether a different comparison group, a propensity-score-matched control, or a regression discontinuity design better handles the pre-existing differences between treated and control units.
Sources
- Jacobson, L. S., LaLonde, R. J., & Sullivan, D. G. (1993). Earnings losses of displaced workers. American Economic Review, 83(4), 888-909. link ↗
- Freyaldenhoven, S., Hansen, C., Perez-Skiba, A., & Shapiro, J. M. (2021). Visualization, identification, and estimation in the linear panel event-study design. NBER Working Paper 29170. link ↗
How to cite this page
ScholarGate. (2026, June 3). Multi-period Event Study Design for Dynamic Treatment Effects. ScholarGate. https://scholargate.app/en/causal-inference/multi-period-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