Questions: Parallel Trends Assumption: Validity and Testing

Question 1

A researcher runs a DiD study and finds that treated and control groups have statistically identical pre-treatment trends. A critic argues the study could still be invalid because the groups might have diverged after the treatment date even without the policy. Is the critic right?

• A. No — statistically parallel pre-trends fully validates the parallel trends assumption; the DiD estimator is identified
• B. Yes — parallel pre-trends provide indirect supporting evidence but cannot prove counterfactual post-treatment behavior
• C. No — if a formal pre-trends test passes at the 5% significance level, the assumption is verified by definition
• D. Yes — parallel trends can only be validated using synthetic control methods, not pre-trend inspection

Answer: (B)

Parallel trends is a counterfactual claim about what would have happened post-treatment. Pre-treatment trend similarity is encouraging evidence — it suggests the groups were on comparable trajectories — but it doesn't rule out divergence caused by anticipation effects, differential seasonality, or structural changes coinciding with the treatment window. Option A is the most common mistake: researchers who equate parallel pre-trends with a verified assumption skip the further robustness checks (placebo tests, alternative control groups, sensitivity analysis) that build real credibility.

Question 2

What is the purpose of a placebo test in a difference-in-differences study?

• A. To confirm that the treatment coefficient is statistically significant at conventional levels
• B. To assign fake treatment to an untreated group (or a false treatment date) and check whether a spurious 'effect' appears, which would undermine the parallel trends assumption
• C. To test whether the control group's pre-treatment trend is stationary over time
• D. To verify that treatment assignment was random across the sample

Answer: (B)

A placebo test checks whether the DiD estimator finds an 'effect' where no real treatment occurred. If you assign treatment to a group that was never treated, or shift the treatment date to a period before the actual policy, and still detect a large estimated effect, something other than the treatment is driving the pattern — a confounding trend, a contemporaneous event, or a violation of parallel trends. Option D describes a randomization check appropriate for RCTs; DiD is used precisely when treatment is not random, so placebo tests provide a different kind of credibility evidence.

Question 3

The parallel trends assumption is fundamentally a counterfactual claim: it asserts what would have happened to the treated group in the absence of treatment, which can never be directly observed.

• T. True
• F. False

Answer: True

This is the core epistemological point about DiD identification. You observe the treated group's post-treatment outcome and the control group's post-treatment outcome, but you never observe the treated group's counterfactual path without treatment. Parallel trends is the bridge between what you observe and what you need to infer causation — and because the counterfactual is unobserved, the assumption is untestable directly. All DiD credibility analysis is indirect: building circumstantial evidence that the assumption is plausible, not proving it.

Question 4

If a researcher finds no statistically significant pre-treatment trends in an event study regression, the parallel trends assumption is proven and no further robustness checks are needed before publishing causal estimates.

• T. True
• F. False

Answer: False

Passing a pre-trends test is a necessary but not sufficient condition for credibility. Pre-trends can be absent for several reasons other than genuine parallel counterfactual paths: low statistical power, a short pre-period, or the treated group being selected precisely because it was trending similarly until the moment of treatment. Moreover, pre-trends tests only cover the observed pre-treatment period; they say nothing about post-treatment behavior. A credible DiD paper combines pre-trends evidence with placebo tests, sensitivity to alternative control groups, and (for staggered designs) robust estimators like Callaway-Sant'Anna.

Question 5

Why can't the parallel trends assumption be directly tested using post-treatment data, and what can researchers do to build credibility for the assumption before drawing causal conclusions?

Model answer: Post-treatment data confounds the treatment effect with any counterfactual trend — you can't separate 'what the treatment caused' from 'what would have happened anyway' because you only observe one path. To build credibility, researchers: (1) plot and formally test pre-treatment trend parallelism across multiple periods; (2) run placebo tests assigning treatment to untreated groups or fake dates; (3) try alternative control groups and check whether estimates are stable; (4) use event study specifications with leads and lags to look for anticipation and to visualize the parallel-trends evidence. None of these prove the assumption, but together they build a case for its plausibility.

The fundamental problem is the missing counterfactual — you need to know what the treated group would have done without treatment to test whether the assumption holds post-treatment, but that's exactly what you're trying to estimate. This circularity means DiD credibility must come from indirect evidence gathered before and around the treatment window, not from the post-treatment data itself.