I am trying to study the effects of a program, where everyone gets treated. I have 50% of the population in t and the rest of the 50% in t+1 period. I understand this gives me a slight opportunity of creating a control group for just one time period and run something like DID. Or I can run a Segmented TS analysis.

I was wondering if there is any better way?

  • $\begingroup$ Are the subjects in t the exact same subjects in t+1, with the only differences being the introduction of an intervention and the passing of time? Also, given no loss to follow-up. Do you have any other time measures (e.g., observations) beyond these two points? Since this would dictate if you should think about using an interrupted time series. Most likely you are looking at a before-and-after study design. $\endgroup$ – hlsmith Nov 3 '17 at 16:33
  • $\begingroup$ Does this mean that half of the population gets treated at time 0 and the other half gets treated at 1, but you have all the data in both periods? Do you expect the treatment to work instantly and persists? Are there reasons to expect general equilibrium effects/SUTVA violations? $\endgroup$ – Dimitriy V. Masterov Nov 3 '17 at 16:33
  • $\begingroup$ @hlsmith Yes they are the same. The dataset has two randomly assigned rotating panel groups. I have about 10 waves of pre-treatment analysis and 3 post treatment for one group and 2 for the other. $\endgroup$ – FightMilk Nov 6 '17 at 14:30
  • $\begingroup$ @DimitriyV.Masterov I have no reason why SUTVA may be violated. As for general equilibrium effects, it is something I might need to control for, and this is where it is getting tricky. $\endgroup$ – FightMilk Nov 6 '17 at 14:33

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.