[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: random effects with clustered s.e.

From   Stas Kolenikov <[email protected]>
To   [email protected]
Subject   Re: st: random effects with clustered s.e.
Date   Mon, 16 Nov 2009 22:18:39 -0600

On Mon, Nov 16, 2009 at 8:04 PM, Shuaizhang Feng <[email protected]> wrote:
> I find that stata 11 allows vce(cluster clustvar) option in xtreg, re models. Has anyone used this before? Does it make sense to cluster s.e. with random effects model? What is the difference with OLS with clustering?

-xtreg, re- is a GLS estimation method, which is different from OLS.
Each is optimal under its own assumptions, and if you need to
-cluster-, these assumptions are likely violated. An obvious case for
-xtreg, re vce(cluster PSU)- is a longitudinal study with a complex
survey design, in which the panels are nested in clusters. You run
-xtreg, re- to get a good account of within-panel correlations that
you know how to model (via a random effect), and you top it with
-cluster(PSU)- to account for the within-cluster correlations that you
don't know how or don't want to model. As Austin said, you need to
make sure you have approximately infinitely many clusters for this
estimator to work.

Stas Kolenikov, also found at
Small print: I use this email account for mailing lists only.
*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index