TEST contrasts in MIXED

[Kylie]

Hi all,

I have repeated measures data analysed as a MIXED model - time (7
levels), group (3 levels) and their interaction, all fixed:

MIXED
depvar BY group time
/FIXED = group time time*group | SSTYPE(3)
/METHOD = REML
/PRINT = SOLUTION
/REPEATED = time | SUBJECT(StudyID) COVTYPE(AR1)
/EMMEANS = TABLES(time*group) COMPARE(time).

We use the EMMEANS subcommand above to get tests of time1 vs time7 for
each group, which is one of the main hypotheses of interest. Another
set of specific hypotheses that we wish to test for is between group
pairwise tests of the change from time1 to time7. Ie:

(Group1Time1 - Group1Time7) vs (Group2Time1 - Group2Time7)
(Group1Time1 - Group1Time7) vs (Group3Time1 - Group3Time7)
(Group2Time1 - Group2Time7) vs (Group3Time1 - Group3Time7)

I believe this should be possible using the /TEST subcommand, but all
examples I have seen have been contrasts between specific cells of an
interaction, not the difference between two cells. Can anyone suggest
how to form these contrasts for this model?

Thanks,
Kylie.

[Kylie]

Just for completeness, here is the solution that I have got working:

/TEST 'Group[1]Change v Group[2]Change' group*time -1 0 0 0 0 0 1 1
0 0 0 0 0 -1 0 0 0 0 0 0 0

Kylie.

[Bruce Weaver]

Kylie, thanks for posting that. Too often people figure out the
answer to their own question, and never tell the rest of us.

--
Bruce Weaver
bwe...@lakeheadu.ca
www.angelfire.com/wv/bwhomedir
"When all else fails, RTFM."