Right now, I am designing an experiment, a Randomized Controlled Trial, with two treatment groups (T1: Access to Subsidy and No access to borrowing money; T2: No Access to subsidy but access to borrowing money) and one control group (T0: no access to subsidy neither to borrowing money). So I will have three dummies as independent variables (mutually exclusive) and I will assess their impact on one outcome continuous variable (Y).

I will collect monthly data throughout 5 years! Panel Data structure!

My goal is to see which one of my treatment groups achieves higher outcome and I would like to check if that difference is statistically significant!

What test should I perform? I would like to use a dif in dif analysis but I am not sure how to "write down" an regression equation with three groups at the same time I control for fixed effects!

Any suggestions?

Thanks for the help!


If you want to do a Diff-In-Diff, you need to estimate a panel fixed effects regression with a post dummy, and post-T1 and post-T2 interactions. The last two are the parameters you care about. The T1 and T2 dummies will be absorbed by the fixed effects, so there is no need to include them.

However, it is not clear why this would be necessary with an experiment. DID is necessary when you have a time-invariant difference between treatment and control, but randomization typically ensures that is not the case.

  • $\begingroup$ Thank you for the answer! But I did not quite get if I am allowed to compare the coefficients from postT1 and postT2 and see if their difference is stat. sig. different from zero and what is the sign of it! Would the regression look like this: Yit = C + B1*(posttT1i) + B2*(posttT2i) + eit And how would I test if B1 = B2 ? $\endgroup$ – Ruc Oct 28 '16 at 18:23
  • $\begingroup$ Also, for Stata commands would it be something like: xtset groupid month xtreg Y post-T1 post-T2 ? $\endgroup$ – Ruc Oct 28 '16 at 18:31
  • $\begingroup$ You can test each Tk-post coefficient is equal to zero to see if treatment is effective. You can also test that their difference is zero to see if one is more effective than the other. These would be simple linear Wald tests in Stata that can be done using the test command and there are plenty examples of how to do such tests on this forum. Your spec seems correct for the regression part, but you need the fe option. $\endgroup$ – Dimitriy V. Masterov Oct 28 '16 at 18:34
  • $\begingroup$ @RicardoGabriel Did this clear things up? $\endgroup$ – Dimitriy V. Masterov Oct 31 '16 at 23:55
  • $\begingroup$ Sorry, yes solved my concern $\endgroup$ – Ruc Feb 19 '17 at 19:59

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.