Multiple Time Periods and Staggered Adoption
What You'll Learn
You'll learn how to extend Difference-in-Differences (DiD) beyond the simple two-period case when different units adopt treatment at different times.
The Staggered Adoption Challenge
In many real-world scenarios, treatment doesn't happen uniformly. Some states adopt a policy in 2015, others in 2018, and some never adopt it. This is called staggered adoption or staggered rollout.
The classic DiD framework (one treatment group, one control group, two time periods) breaks down here because:
- You have multiple "before" and "after" periods for different groups
- Some treated units can serve as controls for others (before they're treated)
- The parallel trends assumption must hold across all these comparisons
The Event Study Design Approach
An event study design is the solution. Instead of comparing just pre/post treatment, you estimate treatment effects for each time period relative to when each unit was treated.
How it works:
- Define "event time" = (current year - treatment adoption year)
- Event time -1 is one year before treatment; +1 is one year after
- Estimate separate effects for event times: ..., -2, -1, 0, +1, +2, ...
- Normalize one pre-treatment period (usually -1) as the reference
This lets you:
- Test parallel trends visually: Pre-treatment coefficients should be near zero
- See dynamic effects: Does the policy effect grow or fade over time?
- Handle staggered timing: Each unit's treatment effect is estimated relative to its own adoption date
Key Takeaway: When treatment timing varies across units, use event study designs to estimate time-varying treatment effects relative to each unit's adoption date, allowing you to test parallel trends and capture dynamic causal effects.