Statalist The Stata Listserver

[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: clustering necessary for xtreg, fe i(id)?

From   "Schaffer, Mark E" <>
To   <>
Subject   st: RE: clustering necessary for xtreg, fe i(id)?
Date   Wed, 3 May 2006 17:44:50 +0100


> -----Original Message-----
> From: 
> [] On Behalf Of 
> Scott Cunningham
> Sent: 03 May 2006 15:54
> To:
> Subject: st: clustering necessary for xtreg, fe i(id)?
> I have an individual-level panel dataset and am estimating a 
> fixed effects model using -xtreg-.  I had been operating off 
> an assumption
> - which I thought was correct but now think is incorrect - 
> that - xtreg, fe i(id)- would automatically adjust for 
> correlation between the individual errors.

Have a look at any standard exposition of fixed/random effects.  When
you write the fixed/random effects equation, there is a time-invariant
panel-specific term which is the fixed or random effect.  Thus the
correlation between individual observations is assumed to have a very
specific form: since the term is time-invariant, the correlation between
period s and period t is the same no matter how close or far apart s and
t are.

> But this is 
> incorrect, no?  I should include the cluster(id) syntax if I 
> want to correct for this problem, right?

Using the cluster option allows for the possibility that the specific
form of correlation takes a form other than the fixed/random effects
form.  If so, then the coefficient estimates will be consistent, but the
usual standard errors will be wrong.  The cluster option will give you
standard errors that are consistent if the intertemporal correlation
isn't of the fixed/random effects form.

Hope this helps.


> sc
> *
> *   For searches and help try:
> *
> *
> *

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index