Regression discontinuity plot with clustered standard errors

[Sipke Bontekoe]

Dear Stata Listers,

I'm using clustered standard errors in my RD design. For my linear split model I cannot figure out how to plot my RD with clustered standard errors displayed to read the discontinuity from the graph.

*linear split regression
reg y treatment_dummy running_variable treatmentdum_runningvar_interaction, cluster(running_variable)

This is the twoway scatter I can draw with robust standard errors, but I want to display clustered standard errors.

twoway (scatter y running_variable)                                     ///
    (lfitci y running_variable if treatment_dummy==0, ciplot(rline))    ///
    (lfitci y running_variable if treatment_dummy==1, ciplot(rline)),   ///
    xline(cutoff)

Also, cmogram and rdplot seem not to offer this option.

Would you know a way I can plot my regression discontinuity with clustered standard errors?


Thanks so much for your time and help!
Sipke Bontekoe

[Bruce Weaver]

Hello Sipke Bontekoe. I think you mean you want to display the fitted values and CIs from the regression model that uses clustered SEs, and the problem is that -lfitci- does not use those clustered SEs to generate the CIs. I wonder if you can achieve what you want via -margins- and -marginsplot-, with recast(line), recastci(rarea) and addplot() options for the latter. I don't have time to try a toy example right now, but here is a basic example from the documentation that might help steer you in the right direction.

* Example from:
* https://www.stata.com/manuals/rmarginsplot.pdf

clear
use https://www.stata-press.com/data/r16/auto
generate tons = weight/2000
format tons %6.2f
separate mpg, by(foreign)
regress mpg foreign##c.tons##c.tons
margins, at(tons=(.8(.05)2.4)) over(foreign)
marginsplot, addplot(scatter mpg0 tons || scatter mpg1 tons) recast(line) noci

HTH.

[Sipke Bontekoe]

I found my answer!

help twoway_lfitci
And then: estopts() for estopts(cluster(xvar))

You can add this option into your lfitci specification

[Bruce Weaver]

Thanks for sharing what you found.