Dear Thomas,
The answer here is simple - just do the triple t-test as this is already accounting for any offset within each individual subject and so the age and sex covariates are completely redundant (since these are fixed values for each subject).
If you want to look at differences between groups then just have separate versions of EV1 and EV2 for each group and then construct the appropriate differences in the contrasts.
All the best,
Mark
"Goldschmidt, Thomas" <[log in to unmask]> wrote:
Dear fsl-experts,
I have some questions regarding the higher-feat analysis of an experiment (n-back-task) with 3 different sessions (BL, placebo, treatment) and two groups (15patients, 10HC). I am interested if there is a treatment effect on brain activation during task performance in patients and/or controls. Thus, I want to know if there is a difference in brain activation after treatment (compared to baseline) and if this differs to placebo. Later, I would also be interested in a difference between patients and controls.
Based on the Feat-practical and earlier questions in this list: I made up the following design:
1. Individual Feat-analysis (activation during n1n2n3>n0)
2. 2nd-level Feat: Mixed Effect analysis (Flame 1 with automated outlier de-weighting)
Matrix (here only 5patients and 5 HC for the sake of simplicity):
Group
BL
placebo
treatment
demeaned age
demeaned sex
1
1
0
0
-9.62
0.41
1
1
0
0
2.12
0.41
1
1
0
0
6.74
0.41
1
1
0
0
-22.75
0.41
1
1
0
0
14.84
-0.59
2
1
0
0
2.24
0.41
2
1
0
0
2.91
-0.59
2
1
0
0
-2.38
0.41
2
1
0
0
-1.91
-0.59
2
1
0
0
-2.01
0.41
1
0
1
0
-9.62
0.41
1
0
1
0
2.12
0.41
1
0
1
0
6.74
0.41
1
0
1
0
-22.75
0.41
1
0
1
0
14.84
-0.59
2
0
1
0
2.24
0.41
2
0
1
0
2.91
-0.59
2
0
1
0
-2.38
0.41
2
0
1
0
-1.91
-0.59
2
0
1
0
-2.01
0.41
1
0
0
1
-9.62
0.41
1
0
0
1
2.12
0.41
1
0
0
1
6.74
0.41
1
0
0
1
-22.75
0.41
1
0
0
1
14.84
-0.59
2
0
0
1
2.24
0.41
2
0
0
1
2.91
-0.59
2
0
0
1
-2.38
0.41
2
0
0
1
-1.91
-0.59
2
0
0
1
-2.01
0.41
Contrasts:
EV1
EV2
EV3
EV4
EV5
BL:
1
0
0
0
0
Placebo:
0
1
0
0
0
Treatment:
0
0
1
0
0
This matrix doesnt work because of the different groups (what would I need to change?). It does work however, if I do the analysis separately for patients and controls.
3. 3rd level feat (mixed effect, flame1 without outlier de-weighting) with the 3 3D copes from the 2nd level, done as a triple t-test
Matrix (only for patients)
group
EV1
EV2
1
1
1
1
-1
0
1
0
-1
Contrasts:
EV1
EV2
placebo - baseline
-2
-1
treatment - baseline
-1
-2
treatment - placebo
1
-1
placebo - treatment
-1
1
QUESTION: Is this design ok or should I try to incorporate 2nd and 3rd level feat in one higher-feat-analysis such as in a triple t-test (mixed effect, flame1 with outlier de-weighting):
This matrix only works without the EVs for age and sex (otherways it states that at least 1 EV is (close to) a linear combination of the others).
EV1
EV2
EV3
..
age
sex
1
1
1
0
0
0
0
-2.435
0.41
1
1
0
1
0
0
0
-1.965
-0.59
1
1
0
0
1
0
0
2.065
0.41
1
1
0
0
0
1
0
-2.775
-0.59
1
1
0
0
0
0
1
-9.67039
0.41
-1
0
1
0
0
0
0
-2.435
0.41
-1
0
0
1
0
0
0
-1.965
-0.59
-1
0
0
0
1
0
0
2.065
0.41
-1
0
0
0
0
1
0
-2.775
-0.59
-1
0
0
0
0
0
1
-9.67039
0.41
0
-1
1
0
0
0
0
-2.435
0.41
0
-1
0
1
0
0
0
-1.965
-0.59
0
-1
0
0
1
0
0
2.065
0.41
0
-1
0
0
0
1
0
-2.775
-0.59
0
-1
0
0
0
0
1
-9.67039
0.41
With contrasts:
EV1
EV2
placebo - baseline
-2
-1
0
0
0
0
0
0
0
treatment - baseline
-1
-2
0
0
0
0
0
0
0
treatment - placebo
1
-1
0
0
0
0
0
0
0
placebo - treatment
-1
1
0
0
0
0
0
0
0
Which design is better (and did I choose the right higher level-modelling?) and in case its the first (multi-level) how could I incorporate the different groups and in case its the second (only two levels) how could I incorporate different groups and add age and sex as EVs?
I would be very pleased to get some help with these questions!
Best,
Thomas
