Wiveka & Philippe,

1. For the 2nd ANOVA model (which inlcudes 4 levels of factor b), do we need the "a" factor (which specifies controls and patients) at all or is it redundant with factor b?

Yeah, I was thinking about this too. Factor 'a' does seem to be redundant since it is already coded into the new factor 'b'. See more below.


2. Why does "-dset 1 1 2" intrinsically not exist?

Apparently I was absent-minded due to the Friday syndrome. I meant to say something like "-dset 2 1 1 " since there are half of the combinations among all 24 possible ones that do not exist in the 2nd ANOVA approach.


3. We want to know if the interaction extracted in the first ANOVA model by "-fab GrpxRun" and in the second ANOVA model coded by "Interaction Factor1 by Factor2: 1 -1 -1 1 " should yield the same or different brain maps? Is this 2nd model coding correct for determining the presence of an interaction of Group x Run?

The two should give you the same interaction if coded correctly.

I still think that the 2nd approach with your current script is probably not going to work because you only have half of the combinations, and 3dANOVA3 does not allow unbalanced design.

I was wondering whether you could try something like this:

3dANOVA2 -type 3 -alevels 4 -blevels 3 \
-dset 1 1 ../../ctrl1/analysis/NegvPosRun1+tlrc'[0]' \
-dset 1 2 ../../ctrl2/analysis/NegvPosRun1+tlrc'[0]' \
-dset 1 3 ../../ctrl3/analysis/NegvPos Run1+tlrc'[0]' \

-dset 2 1 ../../ctrl1/analysis/NegvPosRun2+tlrc'[0]' \
-dset 2 2 ../../ctrl2/analysis/NegvPosRun2+tlrc'[0]' \
-dset 2 3 ../../ctrl2/analysis/NegvPosRun2+tlrc'[0]' \

-dset 3 1 ../../patient1/analysis/NegvPosRun1+tlrc'[0]' \
-dset 3 2 ../../patient2/analysis/NegvPosRun1+tlrc'[0]' \
-dset 3 3 ../../patient3/analysis/NegvPosRun1+tlrc'[0]' \

-dset 4 1 ../../patient1/analysis/NegvPosRun2+tlrc'[0]' \
-dset 4 2 ../../patient2/analysis/NegvPosRun2+tlrc'[0]' \
-dset 4 3 ../../patient3/analysis/NegvPosRun2+tlrc'[0]' \

-diskspace \
-amean 1 controlR1 \
-amean 2 controlR2 \
-amean 3 patientsR1 \
-amean 4 patientsR2 \
-acontr 1 1 0 0 ctrl \ /*Control main effect*/
-acontr 0 0 1 1 patient \ /*Patient main effect*/
-acontr 1 0 1 0 R1 \ /*run1 main effect*/
-acontr 0 1 0 1 R2 \ /*run2 main effect*/
-acontr 1 0 -1 0 R1ctrlVSpatient \ /*run1 controls vs patients*/
-acontr 0 1 0 -1 R2ctrlVSpatient \ /*run2 controls vs patients*/
-acontr 1 -1 -1 1 GrpxRun \ /*Interaction Factor 1 by Factor2*/
-bucket ANOVA_SRET_NegvPos


4. Under factor c (random factor for subjects in the study) we were under the impression that if there are 12 subjects per group (24 subject total) we should code this as 1 through 12 once for the controls and 1 through 12 for the patients for each level of factor b in the first model.

It seems to be correct with your first model with 3 subjects.

Gang