Date:         Fri, 11 Feb 2000 13:49:22 -0600
Reply-To:     "Nichols, David" <nichols@SPSS.COM>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "Nichols, David" <nichols@SPSS.COM>
Subject:      Re: three way interaction

To test specifically the three way interaction you need to see the
difference in -2 log likelihood values obtained when fitting a model with
that interaction included and with it excluded. If there are no other
variables involved, then the model with the three way is a saturated model,
so the -2 LL value is 0 (actually, it's -2 times the kernel of the log
likelihood), and you don't need to bother to fit that. Probably the easiest
way to do this is to choose the Logit option and fit a model with just the
two main effects of the predictors on the dependent, which is equivalent to
a loglinear model with all three two way interactions. The likelihood ratio
test for the fit of the data to the model that GENLOG will produce is then
the test you want. It's also pretty simple to get this using the Model
Selection approach, or the General Loglinear approach.

David Nichols
Principal Support Statistician and
Manager of Statistical Support
SPSS Inc.

> -----Original Message-----
> From: Andrea Varrone [mailto:avarrone@CALVIN.MED.YALE.EDU]
> Sent: Monday, February 07, 2000 12:28 PM
> To: SPSSX-L@LISTSERV.UGA.EDU
> Subject: [SPSSX-L] three way interaction
>
>
> Hi.
>
> I am analyzing data for which I would like to use tests for three
> way-interaction in SPSS.
> I have a set of clinical and imaging variables, which are my
> predictors,
> and two diagnosis groups (dependent or outcome variable,
> categorized as
> 0/1). I would like to test the interaction between clinical
> and imaging
> variables in the classification of either diagnosis group.
> My predictor variables are categorical.
> I have tried to use loglinear models. However, I don't know whether I
> should use the "General" or the "Model selection" option. In
> both cases
> should I use a saturated model or a custom "interaction" or
> "all-3 way" model?
>
> I thank you anybody who can provide insight on that.
>
> Thanks,
>
>
> Andrea Varrone, MD
>
> NeuroSPECT Center
> 40 Temple Street, Suite 5A
> New Haven, CT 06510
> Tel: 203-764-4315
> Fax: 203-764-4307
>