On Wed, May 23, 2012 at 4:29 AM, Ahnate Lim <[log in to unmask]> wrote:
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)
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
This example is a simplification, as we currently have unequal group sizes, but would this approach to modeling the interaction effect be correct?
Thank you,
Ahnate