http://spssx-discussion.165.s1.nabble.com/REPEATED-MEASURES-MANOVA-question-tp5720874p5720878.html
"Any help," okay. Alarm bells are going off.
Change scores, for three groups that differ in brain function.
Is there any chance at all that the groups are equal at PRE?
The chance that the answer is "no" is the first cause of alarm.
- When PREs aren't equal, the comparison of Changes is
problematic even for one score. Do you really want to look at
simple change, or should you look at regressed change (ANCOVA)?
- That sometimes depend on how the samples are selected: You
do expect "regression toward the mean" if the severity was a guide
for selection (maybe, "mild" versus "severe"?).
The best guide for figuring out what (or whether)something is really
happening is to examine the scatter plots of the data points, in
addition to the plots of the means. Is there a basement/ceiling effect,
or do the measurements have no problems to worry about?
MANOVA is not a powerful way to look for specific patterns when
you do have specific patterns in mind, because it wastes its d.f.'s
in testing "all possible patterns." This is the second cause of alarm.
If you have one or two specific hypotheses in mind, you will get your
firm answer by constructing specific contrasts and testing those
separately. That gives you the ANSWER as well as giving you the test
with its best power. Use Bonferroni correction to preserve the overall
alpha.
Hope this does help.
--
Rich Ulrich
> Date: Mon, 24 Jun 2013 20:01:01 -0700
> From:
[hidden email]> Subject: REPEATED MEASURES MANOVA question
> To:
[hidden email]>
> Hi, i'm new to using repeated measures manova and though i have trawled the
> net a bit, i can't find the precise answer i need.
>
> I am analysing some data that has:
> 4 measures (4 different and related tests from each subject)
> 1 within-subject factor (time at T1 and T2)
> 1 between-subjects variable (brain function at 3 levels - either norm, mild,
> or severe)
>
> I've performed a repeated measures manova and get my multivar and univar
> results. However, i am interested in how each test score changes from T1 to
> T2, for each brain function group. I can get the marginal means for this
> interaction via the 'options' box and it gives me the means and CIs, but how
> can i get SPPS to provide some p-values?
>
> Any help grateful.
>