Mixed effects model in FSL Randomise

Dear community,

I have a question related to FSL Randomise, I hope you may be able to advise me on this.

We are trying to set up a mixed effects model in FSL Randomise, however it is not very clear to us how to best approach this, and if this is even possible with Randomise.

We have 2 groups, with 4 conditions per subject (negative regulate, negative attend, positive regulate, positive attend). We are interested in multiple things, amongst which the group effects in both the regulate versus attend conditions, and an interaction effect of group x regulate>attend x valence.
The FEAT manual (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Randomise_details-14) indicates that FEAT is not suited for a repeated measures model, and for Randomise it is adviced to run a two-sample t-test on the difference scores instead. However, because we are also interested in a three-way-interaction, the two within-subject difference scores (negative regulate>negative attend and positive regulate>positive attend) would still be correlated.

Would it be possible to create difference maps (negative regulate – negative attend and positive regulate – positive attend), and then create a paired t-test model with both these two contrasts in it? Would we then still be able to test group effects? Would this sound valid, or should we take a different approach?

Best, Rozemarijn

Hi Rozemarijn,

I am currently facing a similar issue with FSL Randomise and I was wondering whether you found a solution for applying mixed models.

Best, Linda

Is it crucial to use FSL Randomise? AFNI offers mixed modelling functions although not permutation-based: AFNI program: 3dLMEr