Difficulty with contrasts in Repeated Measures Mixed ANOVA (2X3)

[cbr...@psych.udel.edu]

Hello,

I am attempting to run contrasts for a Mixed ANOVA design using the
Repeated Measures General Linear Model (GLM) function. My design
includes both within and between subjects factor. My within subjects
factor has two time points (measures pre- and post- test mood), and
the between subjects factor identifies the 3 different experimental
groups.

This analysis gives me a significant interaction between my within-
subjects factor and one of my between-subjects factors (mood*group
interaction). I believe I have a fairly good idea of what is going on
by looking at a graph, however, I need to run contrasts to obtain
statistics for the significant interaction. I want to look at whether
the change between time 1 and time 2 of my within subjects measure
(the change in mood from before to after the test) is different
between the 3 groups (one group shows a decrease in mood, and the
other 2 groups show an increase in mood, but post-test mood itself
does not differ significantly between the groups). However, the
contrasts available in SPSS appear to look at between group
differences for time 2 (post-test mood). I have looked at the syntax
reference guide with several people very familiar with SPSS, and they
still cannot figure out how to obtain contrasts that look at between-
subject difference in mood change across the groups. I was told that
older versions of SPSS software may have done this, and I would be
happy for any insight on how to get SPSS 18 to give me these
contrasts.

[Bruce Weaver]

First, it is more conventional to use ANCOVA for this design
(particularly if there is random assignment to groups), with the
followup score as the outcome variable, the baseline score as the
covariate, and Group as the fixed factor. E.g.,

http://www.bmj.com/content/323/7321/1123.full

But now addressing your model...

The F-test for the Group x Time interaction in your model is equivalent
to the F-test from a one-way ANOVA (with Group as the factor) on the
change scores (Followup - Baseline). So if you are determined to stick
with your current model (which essentially analyzes change scores), you
could compute the change scores, run an ANOVA via ONEWAY, and use
Fisher's LSD for pair-wise comparisons of the 3 groups. (Fisher's LSD
is out of favour with a lot of folks for not controlling familywise
error adequately; but it does control FW error just fine when you have 3
groups, and is more powerful than the other usual alternatives. E.g.,
see http://core.ecu.edu/psyc/wuenschk/stathelp/Pairwise.htm, and search
for "K=3". See also the chapter on multiple comparison procedures in
Dave Howell's book "Statistical Methods for Psychology").

HTH.

--
Bruce Weaver
bwe...@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/Home
"When all else fails, RTFM."