Hello,
Thank you for your reply.
I just came across this tutorial:
http://www.fmrib.ox.ac.uk/fslcourse/lectures/practicals/feat2/index.htm
and am thinking of going the FEAT route if possible, and using randomise only if necessary. Using your suggestion, would the following analyses be appropriate?
1) For the between-subjects effect of HIV status, use the 3 level approach of:
a) Lower level FEAT for each scan for each subject
b) Between-session analysis creating a mean response for each subject (averaging drug & placebo)
c) Between-subject analysis for HIV effect (fixed-effects)
Yes. This would work.
2) For the within-subjects effect of drug, use the paired t-test approach (FLAME1). HOWEVER, can this be modified to model the interaction of drug & HIV? The interaction effect is crucial, and I'm the most unclear about how to model this. Is the following correct?
EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8
Control1_placebo 0 1 1 0 1 0 0 0
Control2_placebo 0 1 1 0 0 1 0 0
HIV1_placebo 0 1 0 1 0 0 1 0
HIV2_placebo 0 1 0 1 0 0 0 1
Control1_drug 1 0 0 1 1 0 0 0
Control2_drug 1 0 0 1 0 1 0 0
HIV1_drug 1 0 1 0 0 0 1 0
HIV2_drug 1 0 1 0 0 0 0 1
EV3 and EV4 should code the group.
1 0
1 0
0 1
0 1
You should insert 4 columns before EV5 that represent the interactions:
1 0 0 0
1 0 0 0
0 1 0 0
0 1 0 0
0 0 1 0
0 0 1 0
0 0 0 1
0 0 0 1
The contrasts should be:
EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8 ... EV12
C1 drug-placebo 1 -1 0 0 0 0 0 0...
C2 placebo-drug -1 1 0 0 0 0 0 0...
C2 HIV*drug_1(placebo) 0 0 0 0 0 1 0 0...
C2 HIV*drug_2(drug) 0 0 0 0 0 0 0 1...
This example is a simplification, as we currently have unequal group sizes, but would this approach to modeling the interaction effect be correct?
I'd have to check to see if my model was equivalent to your model. I'd model each group*condition with a separate column though - rather than collapsing them as you have done. It'll give you more flexibility.
Thank you,
Ahnate
