0
$\begingroup$

I am using a sample of roughly 325,000 immigrants from the American Community Survey to assess the impact of years in the U.S. on income levels. I ran the Breusch Pagan test on my regression model but the income levels are heteroscedastic because there is a strong left skew.

To correct for this, I ran my regression with the vce(robust) option in stata. I now want to test my regression for heteroscedasticty but when I attempt to run estat hettest I get an error:

hettest not appropriate after robust cluster() r(498);

my question

What are appropriate tests to run to test for heteroscedasticity of the error terms when using robust standard errors?

$\endgroup$
1
$\begingroup$

I cannot help you with the Stata code, but in general, it does not seem convincing to retest your regression for heteroskedasticity after some robustification of the variance covariance matrix of the slopes.

After all, you still use the same (OLS) point estimator, and will thus get the same residuals exhibiting the same heteroskedasticity - all the robustification does is estimate the variance covariance matrix differently to take this into account.

$\endgroup$
  • $\begingroup$ Is there a way to remove heteroscedasticity? I've tried basic things like log and power transformations, which didn't help. $\endgroup$ – Stan Shunpike Feb 13 '17 at 19:15
  • 2
    $\begingroup$ Why insist on removing heteroskedasticity if your inference is already made robust to it? $\endgroup$ – Christoph Hanck Feb 14 '17 at 15:02
  • $\begingroup$ @RichardHardy, done - I suppose the new tag description will clarify the differences between ARMA and ARIMA? $\endgroup$ – Christoph Hanck Mar 21 '17 at 13:05
  • $\begingroup$ @ChristophHanck, the current tag description already does that, but we can always try to improve it further. $\endgroup$ – Richard Hardy Mar 21 '17 at 13:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.