On Mon, May 7, 2012 at 3:19 PM, Bedda Rosario <[log in to unmask]> wrote:
SPM List,

We ran a repeated measures ANOVA (flexible factorial) in SPM to test for interactions between two experimental factors. 

We have 3 groups (n1=4, n2=16, n3=6)  and 2 conditions (pre- and post-).  Therefore, we have a total of 52 scans.

We have included three factors in the model: SUBJECT, GROUP and CONDITION.
SUBJECT: independence-yes, variance-equal
GROUP: independence-yes, variance-unequal
CONDITION: independence-no, variance-equal

We specified three main effects (one for each factor: SUBJECT, GROUP and CONDITION) and the interaction between GROUP and CONDITION.

 
We have the following questions

1.  Is it correct to test for the interaction in this model (that is, using the flexible factorial, mixed effects design)?

Yes. The only contrasts that can't be tested are between-subject effects.
 



2.  If we use this model to test for the interaction, we are not sure how to extend the contrast weights from Glascher and Gitelman for a 3 x 2 design. 
Is this correct?

Ho = G1C1 - G1C2 = G2C1 - G2C2
        G2C1 - G2C2 = G3C1 - G3C2

F Contrast
zeros(1,24) 0 0 0 0 0 1 -1 -1 1 0 0
zeros(1,24) 0 0 0 0 0 0 0 1 -1 -1 1


Yes. Assuming that the order of the factors is subject, group, condition, group*condition.
 


3.  If we test the interaction with the flexible factorial, and there is no interaction between the factors (GROUP x CONDITION),
is it correct to run a full factorial model to test for main effects for GROUP and CONDITION including the interaction in the model
or should we average the groups and conditions and run a one-way ANOVA and a paired t-test separately to find the main effects for GROUP and CONDITION?

If there is no interaction, you can either test the condition effect in this model or run a new model without the group*condition term, but keep in the group term.

For the GROUP effects, you need another model that tests the group effect on the average of the two conditions. This would be a one-way ANOVA.

Alternatively, you can use GLM Flex and get the group effects, condition effects, and interaction from the same model.
 

Thank you,
Bedda