 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: re: tsset with non-integers
<>
Thank you all for your thoughts and comments, which have shed light on
inaccurate suggestions from a book on statistics using SPSS.
Best,
Frank
On May 7, 2009, at 1:32 PM, John Antonakis wrote:
Oh....forgot to say; welcome to Stata. You won't regret it. SPSS has
some very serious limitations.
Also, I just check and saw what SPSSs command does: it seems to check
for serial correlation in adjacent observations. As Kit said, unless
the data are specifically ordered for that purpose, checking for
seriel correlation makes no sense.
As for the clustering, I used the term in a generic sense to talk of
nested observations. If observations are not independent then you need
to model this non-independence using the -cluster- option for the vce.
Also, intercepts might vary between clusters. If so use -xttest0-
after running -xtreg- to see if you need to model the random intercept.
HTH,
John.
____________________________________________________
Prof. John Antonakis
Associate Dean Faculty of Business and Economics
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
Faculty page:
http://www.hec.unil.ch/people/jantonakis&cl=en
Personal page:
http://www.hec.unil.ch/jantonakis
____________________________________________________
On 07.05.2009 19:02, Frank Gallo wrote:
<>
SPSS offers users the post regression estimation option to test the
independence of residuals terms using the Durbin-Watson test so that
users can check the independence assumption. My data are not of the
time series type. As a stata beginner, I am trying to learn a stata
equivalent approach. Thank you.
Best,
Frank
On Thursday, May 07, 2009, at 09:47AM, "Nick Cox" <[email protected]
> wrote:
Why can't you -tsset- the data?
If your data are time series, you should know the times (dates).
Specify the time variable to -tsset-.
If they aren't, then serial correlation is presumably not defined
or applicable anyway.
Nick [email protected]
Frank Gallo
Hi Eva,
The way I understood Kit's response was that both suggested
approaches (estat bgodfrey and wntestq) require the user to tsset
the data, which I cannot do. I am looking to test the
autocorrelation of residuals terms so that I can check the
regression assumption of indepence of errors. Thank you.
On Thursday, May 07, 2009, at 09:31AM, "Eva Poen"
<[email protected]> wrote:
Frank,
please tell us what you don't understand about the replies you have
got so far. Kit gave recommendations for testing for autocorrelation
in residuals. If it is serial correlation that you are after, you
have
time series (or panel) data, and should tell Stata about it, using
the
-tsset- command. See -help tsset-.
If it is not serial correlation but some other kind of dependence,
you
need to be more specific about the type of data you have.
Kit discouraged the use of -durbinh-, but if you insist on using it
anyway, you need to -tsset- your data first. In any case, residuals
are very unlikely to be integers. Your _time_ variable (e.g. years
or
quarters or months) could be integer. But that has nothing to do
with
your residuals.
Eva
2009/5/7 Frank Gallo <[email protected]>:
<>
Hi Nick & Kit,
Thank you for your responses. My goal is only to check the
independence of residual terms following a regression run. My
reading of the durbinh command lead me to believe that it would
help me achieve my goal. Can you suggest an alternative option to
check the independence of residual terms that are non-integers?
Thank you.
Best,
Frank
On Thursday, May 07, 2009, at 06:32AM, "Kit Baum" <[email protected]>
wrote:
<>
Frank said
What I would like to do, which I cannot find exactly in the
archives,
is to check the independence of the residual terms (e) from a
regression. I would like to run the -durbinh- command...
Nick Cox answered the technical question re -tsset-. I do not
recommend you rely on the -durbinh- command. It is a special
case of
the Breusch-Godfrey test in which you only consider AR(1) vs
i.i.d.
The -estat bgodfrey- postestimation command allows you to test for
higher-order autocorrelation as well (which might well be
present even
if an AR(1) coefficient is insignificant). Also consider using -
wntestq-, which is an unconditional test of the residuals'
autocorrelation function. (B-G is a conditional test in that it
uses
the X matrix from the regression, whereas the Lung-Box-Pierce
"Q" test
may be applied to any time series). All will require that the
data are
properly -tsset-.
*
* 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/