Hello Charlotte,
In randomise the design.grp file is used to specify "exchangeability-blocks" - as we want permutations to respect the repeated measure structure of the data, in your example below this would be
1
1
2
2
3
3
4
4
and would be passed into randomise with the -e option. Without this option the permuter in randomise will mix subject data.
The jiscmail threads:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;6670671f.1002
and
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;81e4e2bb.1002
have example designs ( and contrasts ) for this kind of analysis.
Many Regards
Matthew
> Dear FSL list,
>
> I am running a TBSS study with 2 groups of subjects (1 vs 2). Each group has a within-subjects factor of two levels (A vs B). I would like to be able to test the main effects of group and condition separately, and more importantly to examine the group by condition interaction.
>
> I have a design matrix in which the groups are specified by “1” and “2” in design.grp, the within-subjects factor are modelled in separate EVs, and the repeated measures specified by subject-specific EVs. There are also demeaned covariates of interest (Cov1, Cov2 etc). My design matrix therefore looks like:
>
> Group 1A 1B 2A 2B Cov1 Cov2 Subj1 Subj2 Subj3 Subj4
> 1 1 0 0 0 -3 0 1 0 0 0
> 1 0 1 0 0 -3 0 1 0 0 0
> 1 1 0 0 0 2 0 0 1 0 0
> 1 0 1 0 0 2 0 0 1 0 0
> ...etc
> 2 0 0 1 0 0 4 0 0 1 0
> 2 0 0 0 1 0 4 0 0 1 0
> 2 0 0 1 0 0 -1 0 0 0 1
> 2 0 0 0 1 0 -1 0 0 0 1
> ...etc
>
> I have called the design in randomise as follows:
> randomise –i all_FA_skeletonised.nii.gz –o FA –d design.mat –t design.con –m mean_FA_skeleton_mask.nii.gz –D –T2 –V
>
> However, after the first permutation, this message is generated:
>
> 1.83449e+25 permutations required for exhaustive test of t-test1
> Doing 5000 random permutations
> Starting permutation 1 (Unpermuted data)
> Starting permutation 2
> Warning: tfce has detected a large number of integral steps. This operation may require a great deal of time to complete.
>
> The permutations then stall and progress no further. Looking at previous posts on this message, others have found that it can be generated by corrupted data (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;b0af05ef.1010) or “an error in the design matrix” (https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;e8b434e6.1008). Our data is not corrupted. In addition, we have found that randomise runs perfectly well if we do not include subject specific EVs. We have applied a similar design matrix to a voxel-based analysis of the data in SPM, and there are no such problems.
>
> (1) My first question is: would FSL users be able to confirm that this is an appropriate way to specify a repeated measures design in a TBSS design matrix?
> (2) Secondly: why does introducing subject-specific EVs modelling repeated measures generate this integral steps message? What action can I take to solve it?
>
> Many thanks,
>
> Charlotte
>
|