Dear SPM Experts,
I am using a flexible factorial design to run a mixed repeated-measures Anova. I am doing a reward task. I have 3 groups of participants and 2 conditions (gain, lose), which have been done by all groups. I would like to see the interaction Groups*Conditions. I would like to be sure that I am doing the right thing, so any tip and validation/invalidation of the following would be more than welcomed.
So I have 3 factors:
Factor 1: Subject (69 subjects, independance = yes ; variance = unequal)
Factor 2: Groups (3 groups with 25 / 22 / 22 subjects per group ; independance = yes ; variance = unequal)
Factor 3: Conditions (2 conditions, independance = no ; variance = unequal)
Then, under "Specify subjects or All scans and factors", I would choose Specify All. The factor matrix would have 4 columns, where:
Column 1 = 1s
Column 2= subject
Column 3 = groupe
Column 4 = condition
So it would look like (factor matrix = (69x2) x 4):
1 1 1 1
1 1 1 2
...
1 25 1 1
1 25 1 2
1 26 2 1
1 26 2 2
...
1 47 2 1
1 47 2 2
1 48 3 1
1 48 3 2
...
1 69 3 1
1 69 3 2
The order of the scans would be: S1G1C1 ; S1G1C2 ; S2G1C1 ; S2G1C2 etc...
Then, under Main effects and Interactions, I will specify the Interaction. Under Factor numbers, to see the interaction Groupe * condition, I would enter "2 3". I assume that I can't look at the main effects of groups or conditions there, but I don't understand why it is invalid to look at them.
Assuming the column order is Sx G1 G2 G3 C1 C2 G1C1 G1C2 G2C1 G2C2 G3G1 G3C2, then the interaction contrast will be:
[zeros(1,69) 0 0 0 0 0 1 -1 -1 1 0 0 ; zeros(1,69) 0 0 0 0 0 0 0 1 -1 -1 1 ].
This is a naive question, but where do I specify this interaction contrast?
Working from the null hypothesis:
Ho: G1C1-G1C2=G2C1-G2C2=G3C1-G3C2 gets split into:
Ho: G1C1-G1C2=G2C1-G2C2 & G2C1-G2C2=G3C1-G3C2
Are the contrasts vectors in the F-test for the interaction the following?:
1 -1 -1 1 0 0
0 0 1 -1 -1 1
Finally, if I run a One-way Anova looking only at the difference of Condition 1 - Condition 2 between the 3 groups, would the results be different? If no, which model would be the best?
Thank you so much in advance!
Best
Caroline
|