RD Regression with multiple thresholds

[Kathrin Me]

Hello all,
I currently try to set up an RD-regression with 10 thresholds. For a single threshold the regression should be as follows:

gen Ntilde1000 = wpop / 1000
gen logNtilde1000 = ln(Ntilde1000)
gen Gtilde1000 = exptot / 1000
gen logGtilde1000 = ln(Gtilde1000)
 
rdrobust logGtilde1000 logNtilde1000 if logNtilde1000 >= -0.15 & logNtilde1000 <= 0.15, c(0) p(3) kernel(epanechnikov) h() vce(nn)

I managed to set up the loops for the generation of the necessary variables:

foreach i in 1000 2000 3000 5000 10000 20000 30000 50000 100000 200000 {
     gen Ntilde`i' = wpop / `i'
     gen logNtilde`i' = ln(Ntilde`i')
}
foreach t in 1000 2000 3000 5000 10000 20000 30000 50000 100000 200000 {
     gen Gtilde`t' = exptot / `t'
     gen logGtilde`t' = ln(Gtilde`t')
}

As a next step I have to set up an RD-regression for all 10 thresholds. I tried to create a loop for this as well:

foreach i of logGtilde`i' {
     foreach t of logNtilde`t' {
           if `i' == `t'
     rdrobust i t if logNtilde`t' >= -0.15 & logNtilde`t' <= 0.15, c(0) p(3) kernel(epanechnikov) h() vce(nn)
}
}

Unfortunately, the loop is not working. Stata is still very new to me, so maybe someone knows how to fix it. My goal is to have something like this:

rdrobust logGtilde1000 logNtilde1000 if logNtilde1000 >= -0.15 & logNtilde1000 <= 0.15, c(0) p(3) kernel(epanechnikov) h() vce(nn)
 
rdrobust logGtilde2000 logNtilde2000 if logNtilde2000 >= -0.15 & logNtilde2000 <= 0.15, c(0) p(3) kernel(epanechnikov) h() vce(nn)
 
rdrobust logGtilde3000 logNtilde3000 if logNtilde3000 >= -0.15 & logNtilde3000 <= 0.15, c(0) p(3) kernel(epanechnikov) h() vce(nn)
…

Even if it would work, would I get a single value (including a p-value)?

I would appreciate your help a lot!! Best, Kathrin

[George Ford]

As long as the 1000/2000/etc is the the same for all variables, then use this:

foreach i in 1000 2000 3000 5000 10000 20000 30000 50000 100000 200000 {
    rdrobust logGtilde`i' logNtilde`i' if logNtilde`i' >= -0.15 & logNtilde`i' <= 0.15, c(0) p(3) kernel(epanechnikov) h() vce(nn)
}

Don't think you'll get a single p-value across 10 regressions. Possibly use test after each regression and accum, but not sure that's valid here.

[Kathrin Me]

Thank you George! I managed inbetween