I have a question regarding some potentially existing empirical tests in the difference-in-difference context.

Assume that I have the usual setting of observing some hypothetical outcome for a treatment and a non-treatment group both before and after the treatment. As commonly known, the difference between:

$Outcome_{treated, after} - Outcome_{treated, before}$


$Outcome_{non-treated,after} - Outcome_{non-treated, before}$

is known as the difference-in-difference estimator, which I call $\delta_1$.

Now assume that for some reasons, I can consider a part of the non-treated group as also being treated. Hence I simply assign some of the non-treated entities the treatment status. All entities that already had a treatment status remain in the treatment group. Compared to the previous setting, the treatment group hence increases and the non-treatment group decreases. We can moreover consider the 'old' treatment group as a subset of the 'new' treatment group. I estimate the model again and obtain a new difference-in-difference estimator $\delta_2$.

Now my question: I'm very much interested in knowing whether the treatment effect is now significantly stronger, i.e. whether $\delta_1$ - $\delta_2 \neq 0$ at some common significance level.

Does anyone of you knows how I could test for this? I know that I can put everything in a regression setting with two binary variables (for the treatment vs. non treatment and the after vs. before) plus an interaction term. The regression coefficient of this term is then $\delta$. But how can I compare these two coefficients in the two models with each other?

I'd appreciate any feedback. Many thanks in advance


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.