How to obtain heteroscedasticity robust standard errors?

[Wes Beliski]

Hello everyone,

Since my error terms are heteroscedastic, I am trying to figure out how I can obtain heteroskedasticity-robust standard errors in Stata. I found some examples on this forum, but these don't exactly match my situation.

I have longitudinal data; my dependent variable was measured at 3 different time points. Thereby, I have a continuous outcome variable. I am analyzing my data with a linear mixed model analysis. Therefore, my command is:

mixed [dependent variable] [independent variables] || id:

I can obtain robust standard errors by adding ", vce(robust)" to this command or ", vce(cluster id)" (both options give the exact same output). However, I understand that these are the HC1 type of robust standard errors, which are not the heteroskedasticity-robust standard errors. Unfortunately, if I add ", vce(hc2)" or ", vce (h3)" to my command, I receive an error.

How do I obtain heteroskedasticity-robust standard errors in my situation?

Thanks for any help.

[Carlo Lazzaro]

Wes:
the standard errors produced by the -hc- available from -regress- (but not for -mixed-) are not that different.
In addition,

vce(hc2) and vce(hc3) specify alternative bias corrections for the robust variance calculation.

(see -regress- helpfile), that does not imply that the default robust option is flawed.

[Jeff Wooldridge]

You should use vce(robust) so the standard errors are robust to the assumed error correlation structure, which is a restrictive exchangeable structure. You get automatic robustness to heteroskedasticity, too.