Hi everybody,
I know that this kind of post are probably the most abundant in the mailing list, but I would really like to have your two cents about this.
I have a study in which I have three treatments groups, one covariable of interest, and two "nuisance" covariables.
The dependent variable is FA (the skeletonised mask coming from TBSS masked with an a priori region of interest).
I am interested in the main effect of the covariable of interest and to the interaction between groups and the covariable of interest.
I set up the design as following
EV1 EV2 EV3 EV4 EV5 EV6 EV7 EV8
1 0 0 4 0 0 12 32
1 0 0 5 0 0 35 52
0 1 0 0 2 0 5 15
0 1 0 0 6 0 2 22
0 0 1 0 0 3 23 39
0 0 1 0 0 4 13 60
EV1 = group 1
EV2 = group 2
EV3 = group 3
EV4 = Covavriable of interest X group 1
EV5 = Covavriable of interest X group 2
EV6 = Covavriable of interest X group 3
EV7 = nuisance 1
EV8 = nuisance 2
I set up the following t contrast
c1 = [0 0 0 1 1 1 0 0]
c2 = [0 0 0 1 -1 0 0 0]
c3 = [0 0 0 0 1 -1 0 0]
c4 = [0 0 0 1 0 -1 0 0
F1 = [c1]
F2 = [c1;c2]
F1 should be the main effect of the covariable and F2 should test the interaction.
So I run randomise (randomise_parallel -i all_FA_skeletonised.nii.gz -o OP -m my_ROI -d design.mat -t design.con -f design.fts --T2) and I peeked to the results. The interaction was not significant, but the main effect was highly significant (in the order of p < .001 tfce corrected). Note that I repeated the analysis for a total of 3 covariables of interest, always finding the same pattern of signifiances.
Now, the problem come from the fact that checking back my behavioral data, I realized that the order in which I enter the behavioral variables was not the same order in which I entered the original FA for the TBSS.
In other words: I should have not find any significance. Do the significance that I found stem from some horrible error in the specification of the model or of the contrasts ? And if not, what could possible happening here ?
Any help or comment would be greatly appreciated.
Alain
|