Is vce(cluster) option robust to cross-sectional dependence?

[YYating LLL]

Hi All, I recently heard that -vce(cluster)- option in -xtreg- robust to cross-sectional dependence. I see thread here: https://www.statalist.org/forums/for...d-effect-panel
and a relevant question here: https://www.stata.com/support/faqs/s...cs/references/

I researched Stata manual a lot, but can't find solid reason why -vce(cluster)- option produce SE robust to cross-sectional dependence. Can anyone drop me a reference to support this? and if this is true, what makes it different from using -xtscc-? (https://www.stata-journal.com/sjpdf....iclenum=st0128)

Thank you!

[Sebastian Kripfganz]

When using xtreg, the vce(robust) option is equivalent to the vce(cluster id) option, where id is the cross-section identifier. This computes standard errors that are robust to correlation within cross sections but not across. In some situations, in particular when the time dimension is relatively large, you could use vce(cluster time), where time is the time identifier, to account for cross-sectional dependence.