Cluster and vce(robust) together

[Todd Motiwalla]

I understand this question is partially methodological and partially Stata-specific. With a model such as -xtreg, fe- I think I understand that vce(cluster) will provide both clustered SE as well as robust SE for heteroskedasticity.

With a model like -regress-, however, Stata does not do this, and indeed if you do -reg y x, vce(robust) cluster(county)- Stata will provide an error and tell you "only one of cluster() or vce(robust) is allowed".

So my question is can I do both with -regress- in some other way? Is that possible in Stata and does it make sense empirically?

[Clyde Schechter]

There is no need to separately specify the clustering and heteroskedasticity robustness with -regress-. If you just use -vce(cluster county)-, you get both types of robustness at once. (If you use -vce(robust)- you get only robustness to heteroskedasticity.)