Date:         Mon, 11 Nov 2002 22:55:04 -0500
Reply-To:     Mike Benton <bikepunk005@yahoo.com>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Mike Benton <bikepunk005@yahoo.com>
Subject:      Re: logit/loglinear analysis...i think
Comments: To: "Parise, Carol A." <PariseC@sutterhealth.org>
Content-Type: text/plain; charset="iso-8859-1"

Hi,

You can use Log-Linear modeling in SPSS, but only through syntax.

For example (albeit brief):

You have a DV with two levels, an IV (IV1) with two levels, and two IVs (IV2
and IV3) with three levels.

For a Log-Linear model you will need to start off with a baseline, follow
your theoretical relationships, and go from there.

****BaseLine****.

   LOGLINEAR DV (0,1) IV1 (0,1) IV2 (0,2) IV3 (0,2)
   /PRINT NONE
   /DESIGN DV, IV1, IV2, IV3 IV1 by IV2 by IV3.

  ****IV1****.

    LOGLINEAR DV (0,1) IV1 (0,1) IV2 (0,2) IV3 (0,2)
    /PRINT NONE
    /DESIGN DV, IV1, IV2, IV3 IV1 by IV2 by IV3
     DV by IV1.

  ****IV2****.

    LOGLINEAR DV (0,1) IV1 (0,1) IV2 (0,2) IV3 (0,2)
    /PRINT NONE
    /DESIGN DV, IV1, IV2, IV3 IV1 by IV2 by IV3
     DV by IV2.


  ****IV3****.

    LOGLINEAR DV (0,1) IV1 (0,1) IV2 (0,2) IV3 (0,2)
    /PRINT NONE
    /DESIGN DV, IV1, IV2, IV3 IV1 by IV2 by IV3
   DV by IV3.

   ****IV2 and IV3****.

     LOGLINEAR DV (0,1) IV1 (0,1) IV2 (0,2) IV3 (0,2)
     /PRINT NONE
     /DESIGN DV, IV1, IV2, IV3 IV1 by IV2 by IV3
      DV by IV3 DV by IV2.

   ****IV1 and IV2****.

     LOGLINEAR DV (0,1) IV1 (0,1) IV2 (0,2) IV3 (0,2)
     /PRINT NONE
     /DESIGN DV, IV1, IV2, IV3 IV1 by IV2 by IV3
      DV by IV2 DV by IV1.

    ****IV1, IV2, and IV3****.

      LOGLINEAR DV (0,1) IV1 (0,1) IV2 (0,2) IV3 (0,2)
      /PRINT NONE
      /DESIGN DV, IV1, IV2, IV3 IV1 by IV2 by IV3
       DV by IV2 DV by IV1 DV by IV3.

As for interpretation, for logistic and Log-Linear I like to reference:
http://www2.chass.ncsu.edu/garson/pa765/logistic.htm
 and
http://www2.chass.ncsu.edu/garson/pa765/logit.htm

Also, Desktop Data Analysis with SYSTAT by Wilkinson, Blank, and Gruber has
and excellent chapter on Log-Linear regression.

Finally, you can also follow the publishing trend for your field.  Some may
hate Log-Linear and require logistic regression.

HTH.

Mike H. Benton

----- Original Message -----
From: "Parise, Carol A." <PariseC@sutterhealth.org>
Newsgroups: bit.listserv.spssx-l
To: <SPSSX-L@LISTSERV.UGA.EDU>
Sent: Monday, November 11, 2002 8:32 PM
Subject: Re: logit/loglinear analysis...i think


> If you have a set DV, then sounds like logistic regression may be the way
to go. Whether or not your DV is 2 categories or ordinal wil make a
difference on whether you use binary or multinomial logistic regression.
>
> > -----Original Message-----
> > From: Melanie Rogers [SMTP:melanie@gtw.net]
> > Sent: Monday, November 11, 2002 10:25 AM
> > To:   SPSSX-L@LISTSERV.UGA.EDU
> > Subject:      logit/loglinear analysis...i think
> >
> > I need to run several analyses.  The dependent variable is always
> > categorical and has either two or three levels.  For each analysis, I
have
> > three independent variables.  (I don> '> t think the number of levels
for each
> > independent variable makes a difference here, but just in case------one
IV
> > has two levels and the remaining two IVs each have three levels.)  My
> > question is: What type of analysis do I need to run?  It seems to me
that I
> > should do logit analysis.  I am using SPSS 9.0 and have attempted to
perform
> > logit analysis, but the output does not make any sense to me.  If logit
> > analysis is the appropriate test to run, could someone please walk me
> > through what I need to do?  Perhaps I am running the analysis
incorrectly.
> > If logit analysis is NOT the appropriate test to run, I> '> d appreciate
other
> > suggestions.  Thanks!
>