Statalist


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

st: RE: Clustered standard errors in -xtreg- with dfadj


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: Clustered standard errors in -xtreg- with dfadj
Date   Sun, 8 Mar 2009 19:06:34 -0000

Lisa,

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> Lisa M. Powell
> Sent: 08 March 2009 14:34
> To: statalist@hsphsun2.harvard.edu
> Subject: st: Clustered standard errors in -xtreg- with dfadj 
> 
> Dear List members,
> 
> I would like to follow up on some of your email exchanges 
> (see email exchange at the
> bottom of this email) regarding the inclusion
> of the dfadj command when clustering standard errors in an FE 
> panel model.
> 
> I have an unbalanced sample of individuals over 4 waves of data. I
> include county level variables in the analyses which are my key
> variables of interest and I would like to cluster on the county id.
> Given that individuals can move over time, they are not fully nested
> within the clusters (the counties).
> 
> In STATA 10, if I try to run: xtreg outcome $econ5, i(caseid)
> vce(cluster fip) fe;
> then I get the message of: panels are not nested within 
> clusters r(498);
> which is indeed the case.
> 
> I gather that in earlier version of STATA such as 8.0 where one did
> not have to specify nonest using the cluster command even when the
> panels were not nested that STATA as the default implemented 
> dfadj ? Is that correct?

I can't answer that offhand but it should be clear in the Stata 8
manuals.  Or you can compare Stata 8 and Stata 10 (with -nonest-) output
and see if they're the same.

> Therefore, I have added the option "nonest".
> (xtreg outcome $econ5, i(caseid) vce(cluster fip) nonest fe)
> 
> If I additionally add the option "dfadj" then as expected my standard 
> errors increase.
> (xtreg outcome $econ5, i(caseid) vce(cluster fip) nonest dfadj fe)
> 
> In another post, a few years ago, Mark Schaffer responded as 
> follows which was very helpful:
> The problem arises when the panels cut across clusters. Some dof 
> adjustment is needed, but what should it be? I don't know, 
> and I haven't 
> seen any paper that describes exactly what it should be. When writing 
> -xtivreg2-, I opted simply for the program to exit with 
> error. Official 
> -xtivreg- now has the -nonest- and -dfadj- options; these 
> didn't exist 
> when I wrote -xtivreg2-. You'll see that they allow you to go between 
> the extremes of no dof adjustment (which will give you SEs 
> that are too 
> small) and a full dof adjustment (which will give you 
> "conservative" SEs 
> that are on the large side).
> 
> My question now:  Is there a "partial" adjustment that can be done?

When this first came up a few years ago, I had a brief correspondence
with, I think, David Drukker and Vince Wiggins from StataCorp (I think
it was "and" but it might have been "or").

As I recall, the problem is that your question hasn't been answered in
the theoretical literature, and deriving an answer would be hard.  Maybe
someone has tackled this problem by now, but I haven't seen anything.

The only thing I can suggest is to do a Monte Carlo simulation of your
problem and see what sort of adjustment is needed to make the
cluster-robust VCV estimator behave reasonably.  (Or possibly, as you
suggest, try bootstrapping.)

--Mark

> Alternatively, is there a way to bootstrap the standard 
> errors to account for the non-nested county clusters? 
> 
> Any further advice that you could offer would be much appreciated.
> 
> many thanks,
> Lisa
> 
> 
> 
> *
> *From*  "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk
> <mailto:M.E.Schaffer@hw.ac.uk>>
> *To*    <statalist@hsphsun2.harvard.edu
> <mailto:statalist@hsphsun2.harvard.edu>>
> *Subject*       RE: st: Clustered standard errors in -xtreg-
> *Date*  Thu, 28 Dec 2006 13:55:02 -0000
> 
> --------------------------------------------------------------
> ----------
> 
> Thomas,
> 
> > -----Original Message-----
> > From: owner-statalist@hsphsun2.harvard.edu
> > [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of
> > Thomas Corneli?en
> > Sent: 28 December 2006 12:29
> > To: statalist@hsphsun2.harvard.edu
> > Subject: Re: st: Clustered standard errors in -xtreg-
> >
> > > Thomas Cornelissen wrote:
> >
> >
> > > >> I am comparing two different ways of estimating a linear
> > > >> fixed-effects
> > > >> model:
> > > >>
> > > >> Method 1: Use -regress- and include dummy variables 
> for the panels.
> > > >> Method 2: Use -xtreg, fe-.
> > > >>
> > > >> These two deliver exactly the same estimates of 
> coefficients and
> > > >> their standard errors (if I do not cluster the 
> standard errors).
> > > >>
> > > >> However, if I use the option -cluster- in order to get 
> clustered
> > > >> standard errors (clustered on the panel ID), I get different
> > > >> results with the two ways of estimating the model.
> > > >>
> > > >> Why is this ?
> 
> <snip>
> 
> > Is there a rationale for not counting the absorbed regressors
> > when standard errors are clustered ?
> >
> > Haven't degrees of freedom been used for absorbing the
> > variables and therefore the absorbed regressors should always
> > be counted as well?
> 
> The short answer to your first question is "yes" - you don't have to 
> include the number of absorbed regressors in a degrees of freedom 
> adjustment for the cluster-robust covariance estimator.  The slightly 
> longer answer is to appeal to authority, e.g., Wooldridge's 2002 
> textbook.  The cluster-robust covariance estimator is given in eqn. 
> 10.59 on p. 275, and you will see there is no dof adjustment.  The 
> standard covariance estimator is discussed on pp. 271-2, and the dof 
> adjustment is given explicit attention.
> 
> This is why the more recent versions of Stata's official -xtreg- have 
> the -nonest- and -dfadj- options for fixed effects estimation.
> 
> -nonest- relates to nesting panels within clusters; the 
> cluster-robust 
> cov estimator doesn't require a dof adjustment but only if panels are 
> nested within clusters.  If panels are not nested within 
> clusters, then 
> some kind of dof adjustment is needed.
> 
> -dfadj- will impose the full dof adjustment on the cluster-robust cov 
> estimator.  If panels are nested within clusters, then you 
> would never 
> need to use this.  But since some kind of dof adjustment is needed if 
> panels are not nested within clusters, you can use this 
> option to go all 
> the way and impose the full dof adjustment.
> 
> Cheers,
> Mark
> 
> Prof. Mark Schaffer
> Director, CERT
> Department of Economics
> School of Management & Languages
> Heriot-Watt University, Edinburgh EH14 4AS
> tel +44-131-451-3494 / fax +44-131-451-3296
> email: m.e.schaffer@hw.ac.uk
> web: http://www.sml.hw.ac.uk/ecomes
> 
> 
> 
> 
> -- 
> Lisa M. Powell, PhD
> Research Associate Professor
> Senior Research Scientist
> Institute for Health Research and Policy
> University of Illinois at Chicago
> tel: 312-413-8468
> fax: 312-996-0065 
> 
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
> 


-- 
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   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