Do you estimate an RDD & want to do subgroup analyses while holding other covariates constant? Then you might be interested in our new approach using reweighting.
Includes Stata command rddsga (RDD SubGroupAnalysis). Comments welcome!
With @alvcarril
dropbox.com/s/m0d48plkt8il…
Includes Stata command rddsga (RDD SubGroupAnalysis). Comments welcome!
With @alvcarril
dropbox.com/s/m0d48plkt8il…
Before I started using regression discontinuity design (RDD) in practice, I imagined that doing subgroup analysis would be as simple as in an IV or RCT setting.
It turns out it's not. One cannot simply interact the treatment dummy with a characteristic of interest.
It turns out it's not. One cannot simply interact the treatment dummy with a characteristic of interest.
In the "standard" case that I was used to before, where one can interact the characteristic of interest with the treatment dummy, holding other observable characteristics constant is simple: just interact them with the treatment dummy too.
Since that doesn't work in the RDD case, we struggled a while to find a solution.
We eventually created the following approach:
We eventually created the following approach:
If there is a binary characteristic of interest, say A and B: Reweight the samples of A and B such that they are similar along other observables. Then run the RDD separately within the two reweighted samples.
Let us know if you find the Stata commend useful or if it creates any problems! We are still working on continuously improving it.
