Print

Print
Hi Sara, hi Lucia,

Please see below:


On 16 August 2017 at 06:54, Sara De Simoni <[log in to unmask]> wrote:
Dear Anderson,

Lucia and I are working on very similar analyses and I was hoping to ask some further questions with regards to the use of these in randomise & FEAT.

I have used the designs you sent through with randomise, e.g.:

randomise -i all_cope1.nii.gz -o ./InteractionModel_Randomise -d Interaction_30subj_randomise.mat -t Interaction_30subj_randomise.con -e Interaction_30subj_randomise.grp --permuteBlocks -n 5000 –T


For the interaction models, the "--permuteBlocks" isn't used. Each subject would have two measurements in the design, i.e., repeated measures.
 
The 4D file included in this command contains cope1 files for the fmri task I am analysing. I had to combine two runs of this task so I ran a fixed effects analysis to do this.

(1) Am I correct in assuming that the 4D file I use with randomise should contain the following:

FixedEffect_Sub1.gfeat/cope1.feat/stats/cope1.nii.gz
FixedEffect_Sub2.gfeat/cope1.feat/stats/cope1.nii.gz
etc


It can be right or wrong. If the two timepoints were combined, then this is no longer a repeated measures design, and there is no need for exchangeability blocks, and the design becomes a simple two-sample t-test. But if what was combined were multiple runs just to increase SNR within timepoint, and the time effects still being analysed in an interaction model, then it's fine.

 
(2) If I am running a normal group FEAT analysis (FLAME) and want to input individual cope files from only one run of the task, the cope files should be taken from the following directory:

Run1_Sub1.feat/reg_standard/stats/cope1.nii.gz
Run1_Sub2.feat/reg_standard/stats/cope1.nii.gz
etc


Yes, this looks right.
 
(3) We have also run an analysis using your designs in FEAT (i.e. using FLAME 1&2) - however instead of including the exchangeability blocks as done with randomise, we just set the Group column to all 1's. This model has run, however we are not sure if it is actually valid to do this?

Looks right, provided that the design is right, i.e., it includes the subject-specific intercepts for the within-subject effects (time effects and interaction group by time), or the subject-specific slopes for the between-subject effects.

Hope this helps.

All the best,

Anderson

 

Thanks very much for your help with this, we really appreciate it!
Sara & Lucia