9.6 Chapter Summary

Recap of DID Logic, Testing, and Practical Essentials

I. Key Takeaways

Identification core: Parallel trends assumption (pre-policy trends should be similar)
Empirical framework: Entity/time fixed effects + clustered standard errors
Dynamic characterization: Event study (check pre-trends + show dynamic effects)
Robustness: Placebo tests (fake time points, fake groups, permutation, leave-one-out, placebo outcomes)

II. Common Pitfalls

Concluding parallel trends hold based solely on "pre-trends not significant" (limited statistical power)
Inappropriate clustering methods (panel data should cluster at entity level, or two-way clustering)
Using "two-period two-group" DID formula for multi-period/staggered treatment
Results driven by a single treatment unit (without leave-one-out testing)

III. Practice Checklist

[ ] Convert data to long format (id, time, treated, post, y)
[ ] Event study: pre-policy coefficients ≈ 0, post-policy coefficients evolve over time
[ ] Main regression: includes C(id)+C(time), report clustered/two-way clustered standard errors
[ ] Placebo tests: fake time points/fake groups/permutation/leave-one-out/placebo outcomes
[ ] Conclusion interpretation: Return to mechanisms and identification, discuss limitations and external validity

Next Steps

Recommended further study:
- New multi-period DID methods (Sun & Abraham, Callaway & Sant'Anna, etc.)
- Instrumental variables and two-stage least squares (IV & 2SLS)
Related chapters:
- Next module (IV & 2SLS) entry → 10.1 Chapter Introduction

Previous: 9.5 Classic Cases and Python Implementation | Next: 10.1 Chapter Introduction