Statistics, experimentation, causal reasoning · insight-0025
With staggered rollouts, do not rely on a single two-way fixed-effects DiD coefficient
- Evidence type
- documented fact
- Difficulty
- hard
- Roles
- Data scientist
- Industries
- aviation, retail, public-sector
Summary: situation, action, reason, boundary
Situation
A self-service kiosk was rolled out to airports at different months, and the team estimates one coefficient with airport and month fixed effects.
Action
Estimate effects by adoption cohort and time since adoption using an estimator designed for staggered timing, with not-yet-treated or never-treated units as controls, and aggregate the cohort effects with explicit weights.
Reason
Already-treated units are poor controls when their own treatment effect is still changing; modern estimators avoid those comparisons.
Boundary
Phased launches across stores, airports, regions or customer cohorts.
Symptom or risk
When effects differ by airport or grow over time, the pooled coefficient compares newly treated airports with already treated ones and can mislead, even taking the wrong sign.
Exceptions
A single adoption date for all treated units reduces to the classic two-group design.
Trade-offs
Cohort-level estimates are noisier and need more explanation than one headline number.
Pitfalls
Dropping never-treated units because "they are different" without checking whether not-yet-treated units suffice.
Lessons and concepts
Claims and their sources
With staggered treatment timing and heterogeneous effects, TWFE regressions can rely on comparisons with already-treated units and can produce coefficients with the opposite sign of all unit-level effects.