This design titled "longitudinal" will be fine if you also include the subject-specific intercepts (like in designs for paired t-tests).

The contrasts would then be:

C1: [1 1 1 1 0 0 0 ...] - Effect of age/time, within-subject.

C2: [1 -1 1 -1 0 0 0 ...] - Interaction group by age/time.

This design doesn't allow testing group or site effects.

The second design isn't correct and shouldn't be used.

All the best,

Anderson

On Friday, 29 September 2017, pauline bezivin <[log in to unmask]> wrote:

Dear FSLers,

I want to conduct a longitudinal analysis with diffusion images using randomise.

My sample has 150 subjects  with 2 groups (G) on 2 sites (S), who underwent 2 mri exams at approximatively 2 years.

I want to study the interaction group*age on changes between time 1 and time 2. I ran TBSS on each subject’s image.

Can I use a design matrix with 2 images per subject or a design matrix with a difference image (timepoint 2 – timepoint 1) as explained in fslwiki? Are they both of them valid? 

 

Model longitudinal:

	G1S1	G2S1	G1S2	G2S2
Subject1	Age_time1	0	0	0
Subject1	Age_time2	0	0	0
Subject2	0	Age_time1	0	0
Subject2	0	Age_time2	0	0
Subject3	0	0	Age_time1	0
Subject3	0	0	Age_time2	0
Subject4	0	0	0	Age_time1
Subject4	0	0	0	Age_time2

               

Model substract:

	G1S1	G2S1	G1S2	G2S2	age_time1
Subject1	Age_time2-Age_time1	0	0	0	Age_time1
Subject2	0	Age_time2-Age_time1	0	0	Age_time1
Subject3	0	0	Age_time2-Age_time1	0	Age_time1
Subject4	0	0	0	Age_time2-Age_time1	Age_time1