[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st: RE: Durbin-Watson in Stata for non-time series
Thanks to everyone who responded. What I am trying to do is perform a
specification test to verify that I have correctly modeled a binary
choice probit. The probit is modeling a cost overrun as a function of
the number of times a contract budget has changed, contract length,
industry concentration, change in R&D budget, change in procurement
budget, inflation, and the technology readiness of the commodity.
My thesis advisor has asked me to calculate the D-W statistic for my
model, but, after everything I've read and the responses I received
here, it does not seem like an appropriate test for my case.
Thanks again for everyone's help.
[mailto:email@example.com] On Behalf Of Nick Cox
Sent: Thursday, November 16, 2006 12:49 PM
Subject: RE: st: RE: Durbin-Watson in Stata for non-time series
There is less contradiction here than might be inferred.
The module -spautoc- was not intended to be used post-modelling as it
does not specifically apply to residuals.
It is difficult to say more without clarification of what Major Ruter is
> Philip and Nick--
> What Nick says is quite (American sense) true, but see -ssc install
> spautoc- (by Nick Cox) or -findit spatcorr- (by Pisati) and page 20 of
> which notes of Moran's I :
> The statistic shows a striking similarity to the familiar
> Durbin-Watson test.
> (and refers to "an extensive discussion" in Anselin, L. and A. Bera
> (1998). Spatial dependence in linear regression models with an
> introduction to spatial econometrics. In: A. Ullah and D. E. A. Giles,
> Eds., Handbook of Applied Economic Statistics, pp. 237-289).
> But you may be interested in the cluster option for various regression
> commands, which estimates SEs robust to arbitrary intra-cluster
> correlations, assuming you have enough clusters (say 50 or more), e.g.
> 51 states (improperly counting the disenfranchised DC) as clusters in
> an analysis of US data.
> On 11/16/06, Nick Cox <firstname.lastname@example.org> wrote:
> > Durbin-Watson tests are for serial autocorrelation.
> > Serial autocorrelation is defined only for a time series, or at the
> > broadest for a one-dimensional spatial series in which influences
> > are propagated in one direction only (even for rivers or streams
> > this is difficult to believe). If your data are spatial you need
> > something else. If you want something else completely, please tell
> > us what you have in mind.
> > Nick
> > email@example.com
> > Ruter Philip E Maj AFIT/ENV
> > > I can find several places in Stata to calculate a Durbin-Watson
> > > statistic for time series and panel models. Is it possible to
> > > calculate a D-W statistic for cross-sectional data without a
> > > time-series
> > > component? Any Help would be greatly appreciated.
* For searches and help try:
* For searches and help try: