Itsa commande

[Peter Egas]

Hello everyone,

I'm facing a problem regarding intervention analysis.

I'm studying the effect of the increase of two new cost-sharing policies on the number of outpatient visits in the last 20 years.
So, I used an intervention analysis with itsa on Stata: itsa outvisit, single trperiod(199. 201.) lag (1) fig posttrend.
My model was: Yt=β0 + β1 * timet + β2 * intervention1t + β3 * time after intervention1t + β4 * intervention2t + β5 * time after intervention2t + et with t=199. to 201. Until now, no problem.

But next, I would like to create a model which calculate a global average increase rate for all policies, which that sets the beta β2 and β4 as an unified coefficient β2. i.e, this model: Yt=β0 + β1 * timet + β2 * (intervention 1 and intervention 2)t + β3 * time after (intervention1 + intervention 2)t + et with t=199. to 201.

Is it possible to do that with itsa? Does anyone have an idea?

Thank you very much

Peter

[Ariel Linden]

Hi Peter,

First off, it is a recommended practice when posting a query to this forum to indicate the provenance of the command you are using. In this particular case, you are using -itsa- which is a user-written program that is available from both the SJ as well as downloadable from SSC.

As for your question: I am not sure I understand what you mean by "global average increase"? You indicated that you have interventions that were initiated sequentially, so there is no overlap in the effect during the first intervention, whereas there may be some overlapping effect (perhaps additive) during the second intervention period. In any case, there cannot be a "global effect" where you combine the two intervention effects, if they do not overlap in time.

That said, I would compare the treatment effect at a particular point in time (let's say the last period in your sample), to the counterfactual, that is, what the predicted value would be at that point in time absent either of the interventions.

For example, using the data in the -itsa- help file:

Set up
. sysuse cigsales
. tsset state year

Run a single-group ITSA with two sequential interventions
. itsa cigsale, single treat(3) trperiod(1982 1989) lag(1) fig posttr

Calculate value at period 30 absent an intervention
. lincom _b[_cons] + _b[_t] * 30

Calculate value for period 30 adjusted for the interventions (treatment effect at time 30)
. lincom _b[_cons] + (_b[_t] * 30) + (_b[_x1982] * 1) + (_b[_x_t1982] * 18) + (_b[_x1989] * 1) + (_b[_x_t1989] * 11)

Calculate the difference between treatment and counterfactual
. lincom (_b[_x1982] * 1) + (_b[_x_t1982] * 18) + (_b[_x1989] * 1) + (_b[_x_t1989] * 11)


I currently have a paper submitted for review in the Stata Journal that provides many different measures of interest following the estimation of an ITSA model. I will post a notification if it gets accepted.

Ariel