Dear Colin
If the sphericity correction works you should be all good with bringing the betas to the second level. For a beautiful example of this have a look at chapter 32 section 5 in the SPM manual, found here: [spm('Dir') filesep 'man' filesep 'manual.pdf’]
But there ara assumptions behind the sphericity correction e.g. that the covariance structure is the same for all voxels in the "Effects of Interest" F-test defined mask, which could potentially be a problem.
Best
Torben
> Den 13. jan. 2020 kl. 20.35 skrev Colin Hawco <[log in to unmask]>:
>
> Hi All,
>
> I am trying to set up an analysis in which I used an FIR set to model my events (for various reasons this works better in this case than HRFs... maybe).
>
>
> I am trying to set up contrasts and figure out how to do the group analysis. I have 4 conditions I may want to contrast; conditions A and B are in Run 1, and conditions C and D are in Run 2.
>
> Normally with a t-constrast I'd run the contrasts at the first level and carry the CON files forward. I don't know if that works in this case. Lets say I wanted to contrast event type A and B, from run 1.
>
> For a t-test its just 1 -1 0 0
>
> But here I have to use an f-test, and I'm not entirely sure I'm doing it right.
>
> Lets assume I used 6 FIRs, to contrast conditions A and B I need (and a bunch of zeros padded at the end)?
>
> [ 1 0 0 0 0 0 -1 0 0 0 0 0
> 0 1 0 0 0 0 0 -1 0 0 0 0
> 0 0 1 0 0 0 0 0 -1 0 0 0
> 0 0 0 1 0 0 0 0 0 -1 0 0
> 0 0 0 0 1 0 0 0 0 0 -1 0
> 0 0 0 0 0 1 0 0 0 0 0 -1]
>
> Which looks for differences between A and B at any time point (FIR bin)?
>
> but how do I translate this into the second level? Do I need to carry forward the beta*.nii files, and build the 2nd level matrix like this, stacked by participant? Even with independence=no, seems like I am breaking rules for repeated measures.
>
> Kind of confused, and want to make sure we get it right, advice and help appreciated : )
>
> Colin
|