Hi Joe,
--- En date de : Mer 20.1.10, Joseph Mettenburg <[log in to unmask]> a écrit :
> De: Joseph Mettenburg <mettenburgj@MIR..WUSTL.EDU>
> Objet: [FSL] regression and randomise
> À: FSL@JISCMAIL..AC.UK
> Date: Mercredi 20 Janvier 2010, 17h45
> Hello all,
>
> The archive is a bit confusing to me regarding setting up
> voxelwise linear
> regression with randomise/tbss. I am using FSL
> 4.1.4.
>
> Is the following set up correct in order to answer the
> question: Does some
> measured value X correlate with FA values, with age as a
> confound variable?
>
> I have run tbss with my subject pool and generated
> all_FA_skeletonised etc.
>
> I have demeaned X and age, and then generated a design.mat
> file, eg:
>
> EV1(X) EV2 (age)
> -1 -2
> 1 2
> 0.25 1.5
> -0.75 0
> -0.5 -1..5
>
> I set up a single contrast:
>
> [1,0]
>
> I submit the following:
>
> randomise -i all_FA_skeletonised -o output -m mask -d
> design.mat -t
> design.con -n 500 -V --T2 -D
>
> I get an ouput_corrp_tstat1 image. thresholding at
> 0.95-1 should give
> clusters with p<0.05.
>
> Prior posts mention using an EV (constant); It's not clear
> to me to what this is
> referring, and when I use EV1 with a constant of 1 for
> everyone (as I don't
> want to split them into separate groups), I get an error
> regarding one of my
> EVs not being demeaned. This just doesn't seem right
> to me, so I'd like to
> know if the above example is correct or if I'm missing
> something obvious..
Everything seems fine to me - except that your EV1 in your example hasn't got a mean of 0 ;-)...
Cheers,
Gwenaelle
|