repeated measures anova for 3 groups

[Stuart Cathcart]

Hi, If i have 3 groups (g1 g2 g3) with repeated measures data (t1 t2 t3
t4), is it appropriate to split the file and conduct 3 two-group RMANOVA'S
(i.e. seperate RMANOVA'S comparing g1 vs g2, g1 vs g3, g2 vs g3). I would
do this after a 3 group RMANOVA finding sig G, T and GxT effects. I have
seen this done in a few publications, and was advised this was OK in an off-
site reply to a posting i made a while ago, but i wanted to check what
others thought as i dont see it done a lot. It seems a sensible way to
compare 3 groups on treatment effects, which is what i want to do. the
usual way of doing a 3 group anova then post-hoc t tests on group
differences at each time point doesn't seem appropriate, as the 3 groups
differ on the DV at all time points anyway. what im really interested in is
if the treatment effects are different for each group i.e. is there a
difference in the RATE OF CHANGE between each pair of groups across time.
Cheers.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Zdaniuk, Bozena-2]

"what im really interested in is
if the treatment effects are different for each group i.e. is there a
difference in the RATE OF CHANGE between each pair of groups across time."

Isn't it exactly the question that is answered by the interaction effect being (or not) significant in the mixed RM Anova analysis?
bozena


________________________________________
From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Stuart Cathcart [[hidden email]]
Sent: Sunday, September 07, 2008 12:47 AM
To: [hidden email]
Subject: repeated measures anova for 3 groups

Hi, If i have 3 groups (g1 g2 g3) with repeated measures data (t1 t2 t3
t4), is it appropriate to split the file and conduct 3 two-group RMANOVA'S
(i.e. seperate RMANOVA'S comparing g1 vs g2, g1 vs g3, g2 vs g3). I would
do this after a 3 group RMANOVA finding sig G, T and GxT effects. I have
seen this done in a few publications, and was advised this was OK in an off-
site reply to a posting i made a while ago, but i wanted to check what
others thought as i dont see it done a lot. It seems a sensible way to
compare 3 groups on treatment effects, which is what i want to do. the
usual way of doing a 3 group anova then post-hoc t tests on group
differences at each time point doesn't seem appropriate, as the 3 groups
differ on the DV at all time points anyway. what im really interested in is
if the treatment effects are different for each group i.e. is there a
difference in the RATE OF CHANGE between each pair of groups across time.
Cheers.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Stuart Cathcart]

Hi,

I have 3 groups. The GxT effect in the RMANOVA tells me there's a
difference in treatment effect between groups, but it doesn't tell me where
amongst the 3 groups this difference is. post hoc tests seem to compare
each group at each time, or compare each time within groups, but, what i
need is to see which groups differ on treatment effects. for example, is it
that g1 is decreasing more than g2 and g3, or maybe g2 decreasing more than
g1 and g3, etc. Doing 3 two-group RMANOVA's (g1 vs g2, g2 vs g3, g1 vs g3)
tests this (e.g. i can get GxT effects for each comparison of groups). As i
said i have seen this done in published studies, and been advised it's
appropriate, but i dont see it done much so was wondering what people at
large think.

cheers

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Maguin, Eugene]

Stuart,

I don't claim to know how to do this in GLM but it sounds to me like planned
contrasts might be helpful. They would allow you to compare G1+G2 vs G3. I
think you would get an F test for the main effects and interaction. I think
you could also set up planned contrasts to compare G1+G2 vs G3 at T3. The
key to doing all these sorts of custom contrasts is being able to manipulate
the K and M matrices. I've never succeeded but others on the list are
skillful with setting up these matrices. And, you may find the examples in
the documentation enough to get what you need. Also, search the archives.

Something else you might consider is to fit a polynomial to time by means of
a contrast statement because it sounds like you might be expecting slope
differences across the three groups.

Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD