Date: Mon, 13 Apr 1998 15:21:49 -0500
Reply-To: "Nichols, David" <nichols@SPSS.COM>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: "Nichols, David" <nichols@SPSS.COM>
Subject: Re: logit and tobit with spss 6.1.2
SPSS doesn't do Tobit models unless you can find something written by
someone outside of SPSS in the form of a macro or script, but whether a
procedure will easily do multinomial logit models depends on what type of
predictors you have. If they're all categorical, the GENLOG procedure will
do them (Statistics->Loglinear Models->Logit). For such models with
continuous predictors, the COXREG approach discussed by John Hendrickx is
the best thing I can currently recommend.
As an aside, though it is true that the SPSS MATRIX procedure doesn't
explicitly honor the WEIGHT command, you can make macros handle weighted
data if you're aware of the issue and take some care, generally by reading
the vector of weights in explicitly and folding it into computations as is
appropriate.
David Nichols
Principal Support Statistician and
Manager of Statistical Support
SPSS Inc.
nichols@spss.com
----------
From: Markus Quandt [SMTP:Markus.Quandt@UNI-KOELN.DE]
Sent: Monday, March 16, 1998 11:40 AM
To: SPSSX-L@UGA.CC.UGA.EDU
Subject: Re: logit and tobit with spss 6.1.2
In article <m0yEXlb-00031eC@baldur.fh-brandenburg.de>,
Michael Stobernack <stoberna@FH-BRANDENBURG.DE> wrote:
>I'm looking for a possibility to estimate a multinomial logit
>and a tobit model with spss 6.1.2.
>
>Does anyone know a source for a spss-macro to run these
>estimations?
J. Hendrickx has a webpage at
http://www.socsci.kun.nl/maw/sociologie/resources/mlogist/
where you will find macros for two different approaches to
multinomial logit estimation in SPSS. From my own experience I
know that the macro supplied by Steffen Kuehnel works very well
for most problems except those that require weighted data sets, as
the SPSS matrix language cannot deal with weights. The macro by J.
Hendrickx embeds the 'canned' SPSS Cox regressions procedure in
such a way as to have it estimate multinomial logits. I guess this
is going to run faster if you have very large datasets, but
cannot base it on experience.
Sorry, no ideas about tobit models. You might have to fall back on
GAUSS etc.
MQ
--
--------------------------------------------------------------------
Markus Quandt quandt@wiso.Uni-Koeln.DE
Universitaet zu Koeln
University of Cologne, Germany
