Statalist The Stata Listserver


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

Re: stepwise like SPSS? [was: RE: st: Means test]


From   "Zurab Sajaia" <zsajaia@hotmail.com>
To   <statalist@hsphsun2.harvard.edu>
Subject   Re: stepwise like SPSS? [was: RE: st: Means test]
Date   Wed, 14 Feb 2007 19:17:55 -0500

I've not posted it on SSC yet since I'm still developing it, any user is most welcome to try my -bestreg- command, that implements Furnival's 'Leaps and
Bounds' algorithm for all subset comparisons.
Standard advantages over stepwise procedures should apply, and also I have included grouping, when groups of dummy variables are considered as one factor and added/dropped together.

Try typing;
.net describe bestreg, from(http://siteresources.worldbank.org/INTPOVRES/Resources)

(note, since Stata is case-sensitive, the url above should copied asis)

Any comments/suggestions would be greatly appreciated.

Zurab


----- Original Message ----- From: "Nick Cox" <n.j.cox@durham.ac.uk>
To: <statalist@hsphsun2.harvard.edu>
Sent: Wednesday, February 14, 2007 5:32 PM
Subject: stepwise like SPSS? [was: RE: st: Means test]



Replying to a different thread and not changing the title
are not recommended.

You want what I see as full-blown intravenous stepwise.

StataCorp's support for any flavour of
stepwise -- by comparison with SPSS -- is lukewarm at best.

You could build on some user-written tools to write
your own, but I guess that's not in prospect if you
are just beginning.

Throw "stepwise" at a random bunch of users at a Stata users'
meeting, and I think the reactions would be some mix of

snake oil!!
wishful thinking!
dodgy, but useful with care

and somebody would suggest that you go and read Frank Harrell's
"Regression modeling strategies" (Springer 2001) on why
you shouldn't.

Nick
n.j.cox@durham.ac.uk

David Zatz

I am moving to Stata from SPSS. We have a somewhat special
situation, where
we frequently work on exploratory data, and have in the past
used SPSS'
stepwise regression to isolate key variables from a list of
about 60-90
variables as a way to guide our path analysis. (We do not
solely rely on
regression but do use it to help lead us to key "unique variance"
questions.)

Stata's stepwise regression does not give us any indication
of increase in
R2 for each model, adding variables in, though we could
manually do this at
a great increase in time. Indeed, I've now spent several
hours looking at
various solutions, including stepwise regress, rsquare,
pathreg, etc. and
haven't found anything that goes through all the various potential
combinations and spits out a number of models, with total R2, R2
contributed by each new entry, and the various coefficients and betas
involved with each. Is this possible in Stata?
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   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