Difference-in-differences
Estimate effects from comparative changes over time
Difference-in-differences compares outcome changes in treated and comparison groups to estimate a treatment effect under a parallel-trends assumption. With staggered treatment timing, conventional two-way fixed-effects averages can mix inappropriate comparisons when effects differ across cohorts or time. Modern group-time approaches make treatment timing, comparison groups and effect aggregation explicit.
Choose this for policy or organisational changes with credible untreated comparisons and observations before and after treatment. A defensible argument for untreated parallel trends matters more than adding many control variables.
Strengths
- Removes stable group differences and common time shocks
- Can examine event-time patterns and treatment heterogeneity
Limitations
- Parallel untreated trends cannot be proven from observed post-treatment data
- Anticipation, spillovers and changing composition threaten identification
Know the boundary
A non-significant pre-trend test does not establish parallel trends. Two-way fixed effects is not automatically a valid staggered DiD estimator.