Statistics Glossary → Parallel trends assumption
Parallel trends assumptionconcept
The key assumption of difference-in-differences: without the intervention, the average outcomes of the treated and untreated groups would have changed by the same amount over time.
Difference-in-differences subtracts the before-after change in the control group from the before-after change in the treated group. For this difference to be read as a causal effect, the control group's change must correctly represent the treated group's 'counterfactual' change without the intervention. The groups may differ in level; what matters is that their trends are parallel.
The assumption cannot be tested directly for the post-intervention period. If several pre-intervention periods are available, an event study can examine whether pre-trends were parallel, but parallel trends before the intervention do not guarantee that they would have stayed parallel afterwards. The assumption depends on the scale: trends that are parallel in levels may not be parallel on a log scale. Other events that coincide with the intervention and affect only one group also break it.
An everyday example
One city raises its minimum wage and a neighbouring city does not. In the five years before the rise, restaurant employment in both cities grew by about 2% a year. If, after the rise, employment grows by 1% in the first city and 3% in the neighbour, difference-in-differences gives an effect of −2 points; but this rests on the assumption that the first city would also have grown by 3% without the rise.
In MerQur
In the Difference-in-Differences analysis it underpins the classical 2 × 2 design and the two-way fixed-effects model; the pre-trend test of the event-study option shows how well the assumption is supported in the pre-intervention periods.
Analyses where this concept appears
Difference-in-Differences (DiD)