Hi FSL experts
Firstly sorry if this message is a bit long winded.
I have just run a project with 26 subjects where each subject completed two runs of ~30 minutes in the scanner split by approximately 2 minutes in the middle. Without going into too much detail my experiment looks at the neural response to sentences that either confirm or dis-confirm material presented earlier in the study, for the sake of this message I will call the two different types of sentences A and B.
I used the fsl Feat guide to run the analysis. At the first level the entire model was applied to each subjects individual runs creating two .feat folders (runA.feat and runB.feat) - this is a (very) simplified version of the design matrix (obviously in the actual design there is a lot more included):
EV1 EV2
A 1 0
B 0 1
A-B 1 -1
B-A -1 1
At the 'second' level I ran a fixed-effects analysis to create one .gfeat file (combined.gfeat) for each subject - simply by selecting 'inputs are lower-level FEAT directories' (runA and runB) and setting up the model as so:
Mean
Input 1 1
Input 2 1
and a simple group mean in the contrasts:
EV1
Mean 1
The problem comes when looking at the mixed-effects third level analysis. I want to analyse the paired difference between A and B and there are two ways I could do this; either by doing a paired t test between cope 1 (A) and cope 2 (B) (from the 26 combined.gfeats), or by looking at the mean activation across the group for contrasts from the lower level, i.e. cope 3 (A-B), or cope 4 (B-A). I assumed these test would be equivalent, however they are not giving the same results. i.e. I get a result when I bring up the contrasts from the first level (cope 3: A-B) but do not see the same result in the equivalent t-test.
Do you have any idea why this would be the case? I am not sure which analysis to trust in this situation....
Thanks in advance for your help
|