Statalist


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

RE : Heteroskedasticity and fixed effects (was: st: RE: Re: Weakinstruments)


From   GaulÚ Patrick <patrick.gaule@epfl.ch>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   RE : Heteroskedasticity and fixed effects (was: st: RE: Re: Weakinstruments)
Date   Thu, 17 Jul 2008 16:34:45 +0200

David Freedman's points are certainly interesting but I don't see how they contradict the suggestion that in practice it makes more sense to use robust standard errors rather than homoscedastic errors.

If I get his and your points correctly:

1)  If the model is seriously in error, robustifiying will not help getting better estimates of the coefficients. Getting standard errors right is irrelevant.
2) If the model is nearly correct, robustifying makes virtually no difference

In both cases where is the harm in using robust standard errors and what's the point to test for heteroskedasticity?

Best regards,

Patrick GaulÚ
________________________________________
De : owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] de la part de Maarten buis [maartenbuis@yahoo.co.uk]
Date d'envoi : jeudi 17 juillet 2008 14:22
└ : statalist@hsphsun2.harvard.edu
Objet : Re: RE : Heteroskedasticity and fixed effects (was: st: RE: Re: Weak instruments)

--- GaulÚ Patrick <patrick.gaule@epfl.ch> wrote:
> I would not worry about testing for heteroskedasticity. In practice,
> it just makes more sense to always use robust standard errors.

David Freedman (2006) has exactly the opposite view. He basically
distinguishes two scenarios: 1) the model is very wrong in which case
robustifying the standard errors makes a difference, but it also means
that all the coefficients are also wrong. So in this case you are
correctly testing meaningless hypotheses. 2) The model is almost right,
in which case robustifying makes virtually no difference.

In other words -robust- either makes no difference or when -robust-
does make a difference, the model is so much beyond repair that you'll
do the wrong thing anyhow. So, all -robust- does is add a false sense
of security.

A final note, though you will probably already know it as it is (or
should be) in any intro stats book: Do not test for heteroskedasticity
before you have looked at the functional form of the effects of your
predictors.


-- Maarten

David A. Freedman (2006) "On the So-Called `Huber-Sandwich Estimator'
and `Robust Standard Errors'". The American Statistician
60(4):299--302.


-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

http://home.fsw.vu.nl/m.buis/
-----------------------------------------

*
*   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