areg with cluster

[Marry Lee]

Hello!

In Stata book, I found that

areg with vce(robust) or vce(cluster clustvar) works similarly, calling robust after regress to produce the Huber/White/sandwich estimator of the variance or its clustered version.

Why does it say "works similarly"?
While it should not be the case, as cluster accounts for serial correlation while robust accounts for heteroscedasticity. I tried this with an unbalanced panel and I got the exact same results using both options.


Is the cluster option with areg providing the cluster-robust estimator allowing for both serial correlation and heteroscedasticity?
Thank you.

[Andrew Musau]

While it should not be the case, as cluster accounts for serial correlation while robust accounts for heteroscedasticity.

This is true for panel data estimators in Stata where the option -robust- is equivalent to -cluster(clustervar)-. Panel data commands have the prefix -xt- in their names, e.g., xtreg and xtpoisson. For regress and areg, the option -robust- calculates White standard errors. If you need cluster robust standard errors, you use the option -cluster(clustervar)-.

[Marry Lee]

Thank you Andrew Musau for your answer.

You say:

If you need cluster robust standard errors, you use the option -cluster(clustervar)-.

So using areg with the option -cluster(clustervar)- accounts for both serial correlation and heteroscedasticity?

[Andrew Musau]

So using areg with the option -cluster(clustervar)- accounts for both serial correlation and heteroscedasticity?

Yes, assuming that you have a wide panel \(N>>T\).