Testing Parallel Trends
You'll learn how to empirically validate the crucial parallel trends assumption using pre-treatment data and placebo tests.
Why Testing Matters
Remember that Difference-in-Differences (DiD) relies on the parallel trends assumption: the treatment and control groups would have followed the same trend absent the intervention. But this is fundamentally untestable for the post-treatment period (we can't observe the counterfactual). Instead, we validate it indirectly using the pre-treatment data we can observe.
Pre-Treatment Period Analysis
The most common validation: examine whether the treatment and control groups exhibited parallel trends before the treatment began. If they diverged before treatment, the parallel trends assumption is questionable.
Visual check: Plot both groups' outcomes over multiple pre-treatment periods. The lines should move in sync.
Statistical test: Run a DiD-style regression using only pre-treatment data, with placebo "treatment" dates. If you detect a significant "effect" when no real treatment occurred, parallel trends is violated.
Placebo Tests
Create fake treatment dates in the pre-treatment window and estimate "effects." For example, if treatment actually occurred in period 10, pretend it happened in period 5 and run DiD. Any significant estimate suggests pre-existing divergence, not parallel trends.
Think of it like checking a thermometer's accuracy before using it: you test it in ice water (known condition) before trusting it to measure an unknown temperature.
What You're Really Doing
These tests don't prove parallel trends would have held post-treatment—but they build confidence. Consistent pre-trends suggest the assumption is plausible; divergent pre-trends mean you should reconsider your identification strategy or add controls.
Key Takeaway: Test parallel trends by verifying that treatment and control groups moved together pre-treatment through visual inspection, placebo DiD regressions, and statistical tests for pre-existing differences.