Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: st: Heteroskedasticity Test in svy
From
Nick Cox <[email protected]>
To
"'[email protected]'" <[email protected]>
Subject
RE: st: Heteroskedasticity Test in svy
Date
Fri, 18 Mar 2011 13:17:32 +0000
Good outcome. Thanks for the closure.
Nick
[email protected]
R G
Steve et al
Sorry for the delay. Actually, I figured it out - I had a variable that had no
observations, and that was (obviously) messing with the XX matrix. Reading
Stata's explanation (which admittedly I read in a hurry) it is pretty clear:
"Is there a regressor that is nonzero for only 1 observation or for one cluster?
The VCE you have just estimated is not of sufficient rank to perform the model
test. This can happen if there is a
variable in your model that is nonzero for only 1 observation in the
estimation sample."
----- Original Message ----
From: Steven Samuels <[email protected]>
To: [email protected]
Sent: Thu, March 17, 2011 5:59:36 PM
Subject: Re: st: Heteroskedasticity Test in svy
Robin-
I apologize for the tone of my last post; you did nothing to deserve it. You can
ordinarily compute the F test from the e(b) and e(V) matrices that -svy: reg-
returns, but, as Nick says, Stata must have had a good reason for not doing so
itself. Show us the -svy: reg- command line and the output and perhaps we can
figure out how you can present the results.
Steve
--
On Mar 17, 2011, at 11:28 AM, Nick Cox wrote:
This is a kind of a curious argument here.
Stata is saying, as I understand it, "We are not going to show you this, because
we don't think it makes much sense." That wouldn't be the end of the story if
you had better technical reasons why StataCorp are wrong, which certainly does
happen.
But wanting a number just to placate the audience....
If I am ever in the audience, can I have an opt out?
(intended seriously, but not aggressively)
Nick
[email protected]
Robin Hertner
Does this mean then I don't need to test for heterskedasticity at all when
runing svy: reg? I'm comparing different models - my normal diagnostic run
through apart from t-tests of the coefficients, R2, LR tests/Wald tests, and
normality of residuals, is to look at remaining heteroskedasticity.
As an aside, in running svy: reg on my models, the model's F-test is blank.
Stata gives the explanation that this isn't a problem, it's just not computed.
Is there a way that I can compute this other than manually? I'm not worried
about it, but it's one of those stats that conferences/publications likes to
see.
From: Stas Kolenikov <[email protected]>
On Thu, Mar 17, 2011 at 8:18 AM, R G <[email protected]> wrote:
> I'm using Stata's svy commands, and am trying to figure out if there is an
> equivalent to imtest, White to do a heteroskedasticity test (besides plotting
> residuals and variables).
It does not make sense in the context of design-based inference
paradigm. There are no information matrices to talk about, and
variance estimators are generalizations of White's
heteroskedasticity-robust estimators, i.e., correct for
heteroskedasticity already. If you want to pursue small gains in
efficiency by modeling variance of the residuals, you can try to do
that, but you run into risk of misspecifying the variance model which
would only make things worse.
*
* 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/