but the \(f\) for the differenceindifferences estimator doing better is far more modest. This is because the assumptions for the differenceindifferences estimator are weaker and hence it can deal with a wider range of settings than the standard fixed effects estimator.
3. Which one of the following is NOT an assumption of the differenceindifferences estimator?
a) Treatment and control groups are parallel in the absence of treatment
b) There are no unobserved confounders
c) The treatment effect is constant over time
d) The treatment effect is the same for all units
Answer: d)
The differenceindifferences estimator does not require the treatment effect to be the same for all units. This is because the estimator compares the change in the outcome for the treatment group to the change in the outcome for the control group. As long as the change in the outcome is the same for both groups, the estimator will be unbiased even if the treatment effect differs across units.
Figure 3.12: Visualization of a differenceindifferences estimator. This figure shows the average outcomes for the treatment and control groups before and after the treatment. The treatment effect is estimated by comparing the change in the outcome for the treatment group to the change in the outcome for the control group.