Hi,
No need to do anything like this.
FEAT has always been able to handle different design matrices
for different first-level analysis - including different numbers of
EVs. As long as you don't want any of these to enter any contrasts
(and you can't really with the confound set-up) then there is
no problem at the first or any higher-level analysis.
All the best,
Mark
On 27 May 2009, at 21:12, Andreas Bartsch wrote:
> Hi,
>
> just a follow-up question: fsl_motion_outliers is likely to generate
> a different number of confound EVs for every subject. Prior to
> higher level analyses, would we be expected to fill up confounding
> outlier EVs of every subject exhibiting less than the maximum
> outliers by dummy EVs so that every first level design matrix
> contains the same total number of EVs (keeping the DoFs constant
> across subjects)? Does that pertain to fixed as well as mixed effect
> analyses, and also to OLS vs. FLAME?
> Cheers-
> Andreas
>
> ________________________________________
> Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag
> von Jesper Andersson [[log in to unmask]]
> Gesendet: Donnerstag, 21. Mai 2009 20:21
> An: [log in to unmask]
> Betreff: Re: [FSL] FSL-MotionCorrection
>
> Dear Zarrar,
>
>> I was interested in doing something just like this and was glad to
>> see that Klara asked this question. I have another clarification
>> question regarding the use of fsl_motion_outliers. When you add the
>> confound file produced by fsl_motion_outliers to your FEAT model,
>> does this impact the inclusion of the motion parameters as
>> covariates? So if you previously included the motion parameters as
>> covariates, will they no longer be necessary with the inclusion of
>> the confound matrix from fsl_motion_outliers?
>
> No, they are pretty much complementary. Including the motion
> parameters removes (as a first order approximation) effects that
> depend ~linearly on subject position, such as e.g. distortion-by-
> position and dropout-by-position interactions.
>
> The covariates you get from motion_outliers on the other hand removes
> effects related to "subject velocity" (again as a first order
> approximation). For example if someone makes a sudden movement in the
> middle of the acquisition of a volume the rigid-body transformation is
> no longer valid, and such a volume would be picked up by
> motion_outliers.
>
> Good Luck Jesper
|