Independence of residuals to use bootstrap with rdrobust

[Lia Moura]

Dear all,

How do I guarantee that the residuals from a discontinuous regression (sharp) are independent to use bootstrap? Is there any test to proving this? Or can I use bootstrap without problems in the RD case because the vce structures of the Stata already guarantee the independence of the residues?

Thanks

[Jeff Wooldridge]

Residuals are never independent of each other because they are estimated from the data. Do you mean errors? What is your data structure? If it's cross-sectional data and it was obtained via random sampling, just use the nonparametric bootstrap. Is there a reason you don't want to use the inference built into rdrobust?

[Lia Moura]

Thanks.


My data is a cross-section of a universe of insured of an insurance company. The idea is to check if the individual leaves the insurance when he turns 18 years old due to the increase in the value of the premium. I run the model for the total and for a sample that has income and education information. But this sample has selection bias. I use optimal bandwidth from Stata and manual bandwidth to check too, so I thought of using the bootstrap structure.

One of the codes I thought of using is:

bootstrap (_b[Conventional]) (_b[Bias-corrected]) ( _b[Robust]): rdrobust exit age,  c(18) h(2) all