Event Study Design

An event study design estimates treatment effects separately for each period before and after an event or policy adoption, using leads and lags relative to a reference period to show dynamics and test pre-trends.

사건연구 설계는 사건이나 정책 도입 전후의 각 시기별 처치효과를 기준 시기 대비 선행·후행 항으로 추정하여 효과의 동태를 보여주고 사전추세를 검정하는 방법입니다.

How do I show how a policy's effect evolves over time and check for pre-existing trends?

정책 효과가 시간에 따라 어떻게 변하는지 보여주고 사전추세를 확인하려면 어떻게 하나요?

An event study design replaces a single difference-in-differences estimate with a set of coefficients for each period relative to treatment, so pre-treatment coefficients test for differing trends and post-treatment coefficients trace how the effect develops. It fits panel data with treated and comparison units observed over time, and with staggered adoption it should use estimators that are robust to heterogeneous effects.

When to use Event Study Design

• Policy adopted by some units at known times, with outcomes observed before and after
• Need to show that treated and comparison units had similar trends before treatment
• Effects that may build up or fade over time after adoption

When to use something else

• Only a single average effect with one treatment date is needed: use difference-in-differences
• Only one treated unit: use the synthetic control method

Assumptions and requirements

• Parallel trends in the absence of treatment
• No anticipation of treatment before adoption
• An omitted reference period, usually the period just before treatment
• Estimators robust to heterogeneous effects when adoption is staggered

What the analysis reports

• Coefficients for each event time with confidence intervals
• Event study plot
• Joint test of pre-treatment coefficients
• Average post-treatment effect

How to report it

Pre-treatment coefficients were not jointly significant, F([df1], [df2]) = [value], p = [value], and the effect on [outcome] reached [value] (95% CI [[value], [value]]) [k] periods after adoption.

Related methods

• Difference-in-Differences
• Synthetic Control Method (SCM)
• Interrupted Time Series

Run Event Study Design on Quorant