A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Parallel trends requires that absent treatment, treated and control groups follow identical outcome trends. Untestable using post-treatment data alone, but examinable using pre-treatment periods: if trends diverge before treatment, the assumption is questionable. Placebo tests and sensitivity analysis are essential for credibility.
Plot pre-treatment trends for treated and control groups before running any regression. A visual inspection is often more informative than a formal pre-trends test. Then run an event study specification and check whether pre-treatment coefficients are near zero.
From difference-in-differences estimation, you know the DiD estimator identifies a causal effect by comparing changes over time in a treated group to changes in a control group. The whole logic rests on a single identifying assumption: the parallel trends assumption. It states that, had the treatment never happened, the treated and control groups would have moved in lockstep over time — their outcome trends would have been parallel. The DiD estimator attributes any deviation from that parallel path to the treatment.
The fundamental difficulty is that parallel trends is a counterfactual claim. You observe what the treated group actually did after treatment, but you never observe what it would have done without treatment. This makes the assumption strictly untestable in the post-treatment period. This is not a minor technical caveat — it is the central credibility challenge of every DiD study. No statistical test can directly verify it using post-treatment data.
What you *can* do is look at the pre-treatment record. If treated and control groups had parallel trends before the treatment began, that pattern gives indirect evidence that they would have continued in parallel. The standard diagnostic is to plot both groups' outcome means over multiple pre-treatment periods and visually inspect whether their trajectories run parallel. More formally, you can run an event study regression that includes leads and lags of treatment: the coefficients on pre-treatment leads should be near zero and statistically insignificant if the parallel trends assumption holds. Significant pre-treatment trends ("pre-trends") are a red flag — they suggest the groups were on diverging paths before treatment, which undermines the DiD identification.
Placebo tests offer another layer of scrutiny. If you assign treatment to a group that wasn't actually treated (or choose a fake treatment date for the real group) and the DiD estimator finds a large "effect," that is evidence against the parallel trends assumption — something other than the treatment is producing the pattern. Sensitivity analysis using different control groups, different time windows, or weighting schemes (like synthetic control or callaway-santanna estimators for staggered rollout designs) can further probe robustness. A compelling DiD paper does not merely apply the formula — it builds a case that the parallel trends assumption is plausible, using all of these tools together.