Statalist The Stata Listserver


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

st: RE: xtivreg2 and clustering


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: xtivreg2 and clustering
Date   Thu, 1 Feb 2007 14:20:19 -0000

Alejandro,

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> Alejandro Molnar
> Sent: Thursday, February 01, 2007 12:23 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: st: xtivreg2 and clustering
> 
> Hi,
> 
> I'm attempting to cluster standard errors at the level of 
> panel units while including unit-time interactions among the 
> regressors (using xtivrge2), and have run into the "number of 
> clusters must be greater than instruments" problem. Using the 
> -fwl- option has been suggested in the past (i.  
> http://www.stata.com/statalist/archive/2006-03/msg00434.html 
> ; ii. http://www.stata.com/statalist/archive/2006-11/msg00254.html)
> 
> Both in my application and the example provided in the 
> command's help, I get the result that the standard errors 
> (for the coefficients I don't "partial out" through fwl) are 
> the same whether I use the fwl option or not.
> 
> I was expecting them to be different. I'm now thinking I 
> should have expected them to be the same: since the fwl 
> option estimates the same model (no new assumptions about 
> errors, just a different order to the mechanics), why should 
> the standard errors be different?

That's correct.  The model is the same.  The only reason the SEs might
differ is if you use a small sample correction (the -small- option) that
accounts for the number of regressors.  -xtivreg2- doesn't count the
partialled-out ones.

> So my question is: Is what I've said above wrong and 
> estimates should be different, or is the fwl option intended 
> as housekeeping, to make sure I don't go interpreting 
> standard errors I don't understand?

Housekeeping, but the main use is to help you get a VCV that is full
rank.

> Also, could someone please point me to articles about 
> clustering in this case (e.g. showing why the VC isn't of 
> full rank), since I'd like to gain some intuition about the problem.

Now *that's* a good question!  I'd like to see a proper reference
myself.  There has been some discussion on Statalist of this and related
topics, and Vince Wiggins gave a good explanation with respect to
singleton dummies a few years ago:

http://www.stata.com/statalist/archive/2003-06/msg00646.html

There is the very terse mention in Baum-Schaffer-Stillman (SJ 2003 3:1),
but we don't offer an intuitive explanation.

--Mark

> 
> Thanks in advance,
> Alejandro
> 
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
> 

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index