Event Study Design in Education Research
Event Study Design for Causal Inference in Education Research · Also known as: event study, education event study, policy event study, dynamic difference-in-differences
An event study design tracks how educational outcomes evolve before and after a clearly defined event — such as a school finance reform, accountability policy, or curriculum change — for affected and unaffected units. By estimating period-by-period treatment effects relative to a baseline period, it delivers both a causal estimate of the policy's impact and a transparent test of the parallel-trends assumption underpinning difference-in-differences.
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 an event study design when a policy, programme, or institutional change affects some educational units (districts, schools, cohorts) at a discrete point in time, and you have panel data covering multiple periods before and after. It is especially valuable when the timing of the event varies across units (staggered rollout), because the dynamic estimates reveal heterogeneous timing effects. It is not appropriate when the event date is endogenous or self-selected by units in response to outcome trends (selection on trends), when pre-event data are unavailable, or when only a single cross-section exists. Avoid applying it when fewer than three pre-period observations are available per unit, since you cannot reliably assess pre-trends.
Strengths & limitations
- Provides a built-in falsification test: pre-event coefficients directly reveal whether the parallel-trends assumption holds.
- Captures the full dynamic trajectory of a treatment effect — immediate impact, ramp-up, and long-run persistence — rather than a single average.
- Handles staggered policy adoption across units using modern heterogeneity-robust estimators.
- Naturally accommodates rich administrative education datasets (district-year panels, cohort-level test score files) without randomisation.
- Transparent and visually compelling: the event-time coefficient plot communicates both identification credibility and substantive findings in one figure.
- Requires sufficient pre-event observations per unit to assess parallel trends; thin pre-periods yield imprecise or untestable pre-trend tests.
- Staggered adoption with many event cohorts demands careful implementation; the classic two-way fixed effects estimator is biased when treatment effects are heterogeneous across cohorts or time.
- Poorly suited to events that unfold gradually or ambiguously, where the event date cannot be pinned to a single period.
- Statistical power may be limited in small education samples (few districts or schools), and clustering inflates standard errors further.
Frequently asked
How is an event study different from a standard difference-in-differences?
Standard DiD collapses the pre/post comparison into a single treatment-effect coefficient. An event study retains all period-by-period comparisons, producing a coefficient for each lead and lag relative to the event. This adds a pre-trend test (checking k < 0 coefficients) and reveals dynamic effects — information lost in a single DiD estimate.
Why must I omit the period k = −1?
The period immediately before the event serves as the normalisation baseline. Omitting it avoids perfect multicollinearity among the lead/lag indicators. All coefficients are then interpreted as the deviation from the pre-event baseline, making zero the natural reference point for assessing pre-trends and treatment effects.
What goes wrong with two-way fixed effects in staggered designs?
When districts adopt a policy in different years, already-treated districts can enter the comparison group for later-treated ones. If treatment effects vary by cohort or grow over time, this contamination introduces negative weights on some groups, potentially reversing the sign of the aggregate estimate. Callaway-Sant'Anna or Sun-Abraham estimators solve this by constructing clean comparison groups.
How many pre-event periods do I need?
At least three pre-event periods are recommended to run a meaningful parallel-trends test. More periods improve power and credibility. With only one or two pre-event observations, you can plot the pre-event coefficient but cannot reliably reject a diverging trend.
Can I apply this when not all districts are ever treated?
Yes — never-treated districts are an ideal comparison group because they are free from contamination. When never-treated units exist, use them as the control group exclusively to ensure clean identification.
Sources
- Jacobson, L. S., LaLonde, R. J., & Sullivan, D. G. (1993). Earnings Losses of Displaced Workers. American Economic Review, 83(4), 685-709. link ↗
- Lafortune, J., Rothstein, J., & Schanzenbach, D. W. (2018). School Finance Reform and the Distribution of Student Achievement. American Economic Journal: Applied Economics, 10(2), 1-26. DOI: 10.1257/app.20160567 ↗
How to cite this page
ScholarGate. (2026, June 3). Event Study Design for Causal Inference in Education Research. ScholarGate. https://scholargate.app/en/causal-inference/event-study-design-in-education-research
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
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Interrupted Time SeriesCausal inference↔ compare
- Panel Fixed EffectsEconometrics↔ compare
- Synthetic ControlCausal inference↔ compare