Problems with rdrobust

[Liz Kelley]

Hi there,

I'm using rdrobust in a project to find the LATE of getting admitted to a better school as admission is based on cutoff scores and I'm running into the following issue:

When I use the full sample data, rdrobust works fine, but when I limit the data to 10 points on each side of the cutoff, I get the following error:

rdrobust outcome margin if margin >=-10 & margin<=10
rdrobust() not able to compute the loc. poly. bandwidth (h) above the threshold. Please run rdbwselect() for more information}

and when I run rdbwselect(), I see this message

Invertibility problem in the computation of preliminary bandwidth above the thresholdInvertibility p
> roblem in the computation of bias bandwidth (b) below the thresholdInvertibility problem in the co
> mputation of bias bandwidth (b) above the thresholdInvertibility problem in the computation of loc
> . poly. bandwidth (h) below the thresholdInvertibility problem in the computation of loc. poly. ba
> ndwidth (h) above the threshold

I'm not really sure how to deal with this. Can somebody help me understand what's going on? I would really appreciate any advice or suggestions!

[Phil Bromiley]

You didn't get a quick answer. This is a user written program so it is quite possible that the active folks on the list don't use it. I don't use this statistical technique.

One thing to try is to drop the observations you don't want to use before the rdrobust. In most estimators, this wouldn't matter, but it might be a problem in this specific user written estimator. Beyond that, you'll need to delve more deeply into the theory cited for this estimator. As a last resort, contact the authors.

[Sammara Cavalcanti]

Liz Kelley, have you managed to solve this problem? I'm having exactly the same errors!

[Jose Quiroga]

Hi! I have the same issue as Sammara, did you manage to solve the problem Liz Kelley?

[Charles Calmeyn]

Hello Liz Kelley! I'm confronted with the same problem as you. I am researching the influence of grade retention on grades in their future academic career, however, the rdrobust doesn't work as well. Did you manage to solve your problem?

Kind regards,
Charles Calmeyn

[Panos Tzouvanas]

Hi,

I have exactly the same problem.
The p() option specifies the order of the local polynomial to be used to construct the point estimator.
When I change the p and q options, this seems to be working:

.rdrobust depvar runningvar , c(0) q(1) p(0) all

But, I am not sure if this is right.

Kind regards,
Panos