Hello all,
I had a question regarding between-group comparisons of linear trends. We have two groups who are presented 5 conditions over the course of three runs and we want to test for linear increases in activity from condition 1 to 5. I know that this brings up the problem of including within and between subject effects, however I believe it's possible to test for the interaction effect?
One solution to similar problems in previous posts has been to generate contrast images at the second level such that only one image per subject is entered in to the third level group comparison. I could create linear contrasts within each subject at the second level and pass these up. However, looking at Jeanette Mumford's page on linear contrasts, she recommends testing the linear contrast at the third level.
From previous threads, it seems that this is possible as long as I am only looking at the condition and condition*group interaction effects, but not the group main effect. To do this, would I set up my design as follows?:
EV1: group indicator (1 and -1 for group1 and group2)
EV2: linear trend of condition (-2, -1, 0, 1, 2 for condition 1 through 5)
EV3: interaction (linear trend * group indicator)
EV4 ....: subject indicators
Contrasts:
0 0 1 0 … (G1 slope>G2 slope)
0 0 -1 0 ...(G2 slope>G1 slope)
Otherwise, maybe this would work?:
EV1 linear trend group 1 (-2, 1, 0, 1, 2, for group 1 conditions, 0’s for group 2)
EV2 linear trend group 2 (-2, 1, 0, 1, 2, for group 2 conditions, 0’s for group 1)
EV3 ...: subject indicators
Contrasts:
1 -1 0 … (G1 slope>G2 slope)
1 -1 0 ... (G2 slope>G1 slope)
Thanks so much for your help,
Jeremy
|