Date:         Fri, 4 Sep 1998 10:31:27 -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: Repeated measures or independant samples

I don't know of any way to get an independent samples mean comparison out of
SPSS with the data set up in two columns (aside from programming your own
computations).
 
The assumption that doing an independent samples test rather than a repeated
or paired samples test will result only in a power loss under an alternative
rather than increasing Type I error under the null is predicated on the
assumption of a certain level of positive correlation among the time points.
That's usually the case with repeated measures data, but it's not the only
thing that can logically or mathematically occur.
 
This is actually a pretty complicated situation, with many things to
consider, particularly with regard to missing data mechanisms. One approach
would be to just apply mixed models methods to the problem, which would
generally require rearranging the data so that there are multiple cases per
subject. But the best thing I could advise would be to have a good
statistician familiar with your substantive area look at the problem
carefully and make appropriate recommendations.
 
David Nichols
Principal Support Statistician and
Manager of Statistical Support
SPSS Inc.
 
        -----Original Message-----
        From:   Damon A. Muller [SMTP:fvp@RAV.ORG.AU]
        Sent:   Wednesday, September 02, 1998 12:48 AM
        To:     SPSSX-L@UGA.CC.UGA.EDU
        Subject:        Repeated measures or independant samples
 
        Hi Folks,
 
        I'm afraid this might be a bit of a basic question, but my stats is
a
        little rusty, and in my present job I don't have a local stats
mentor I can
        go to for help!
 
        We are conducting research on the efficacy of several domestic
violence
        interventions, such as perpetrator groups and couples counselling,
and we
        are administering pre- and post-intervention questionnaires, as well
as
        another questionnaire at a 6-month follow-up period, to clients.
 
 
        Ideally, this would call for a repeated measures design.
Unfortunately, we
        are dealing with a very difficult sample, and getting all of the
data is
        proving to be extremely difficult. As such, we have lots of pre and
post
        data, and as much 6-month data as we can wring out of them. But,
there are
        lots of gaps, and there are very few people for whom we have all
three sets
        of data :(
 
        If I had better data, I could use, for example, a paired-sample
t-test to
        compare pre and 6-month scores on tests such as the ABI (a
pencil-and-paper
        test for measuring abusive behaviour). This would be the 'propper'
way to
        do it, IIRC.
 
        Would it be a Bad Thing to use an independant-samples t-test with
the above
        data, above and beyond the loss of statistical power? Are there
perhaps
        better ways for me to approach the analysis?
 
        Now for the ObSPSS: Iff I can use an indepentand sample t-test (or
        similar), is there any easy way to convice SPSS (version 8.0, Win95)
to use
        data that is laid out accross the page, with no grouping variable,
rather
        than down it?
 
        Thanks in advance for your help/flames/whatever,
 
        damon (who wishes he hadn't lost his undergrad stats books!)
 
 
        -----
        Damon Muller (fvp@rav.org.au)
        * Research Assistant (Family Violence Intervention & Research
Project)
        * Webhacker (http://www.rav.org.au/fvirp)
        * Occasional student