Print

Print


Hi Mark,

Yes, you're right. However I have specific reasons to create separate
contrast EVs for each subject (i.e. EVx = [1 -1]). My design-matrix becomes
singular when I use a single EV / designmatrix as you specified.

In my design (n = 18, two or three repeated measures (inputs) for each
subject, and 18 + 1 EVs (or more if I want to model covariates)), I may have
too many specified regressors. However the problem might also be a bit more
complicated (since FEAT did not produce a warning for rank-deficiency) and
may lie in an adverse balance in DOF / number of inputs, the number of EVs
that I have to specify, and the amount of information contained in the data.

To circumvent such a complex problem (Too many unknowns? Too little
information?), I tried the approach of creating a single contrast EV for
each subject and to do away with the group mean EVs as described in my
previous mail. This would give a small reduction in the number of EVs and
perhaps solve the problem. A 3rd level analysis (again without group mean
EVs) would ten give me the desired group average of that contrast.

Would this be a correct approach, or is there another way in which I can
prevent my matrix from becoming singular?

Thanks again,

Rutger Goekoop

P.S.
Another reason to use higher level single-subject (contrast) EVs is to
combine lower level cope-images in a statistically meaningful way in case
you forgot to specify a lower level cope for the relevant combination of
PEs..... Although dirty, do you think this approach would be valid?

-----Oorspronkelijk bericht-----
Van: Mark Woolrich [mailto:[log in to unmask]]
Verzonden: Thursday, 22 January, 2004 3:55 PM
Aan: [log in to unmask]
Onderwerp: Re: [FSL] Always include a higher level group mean EV?


Rutger,

I'm assuming:
input 1 is subject1 conditionA
input 2 is subject1 conditionB
input 3 is subject2 conditionA
input 4 is subject2 conditionB etc.

Your right in that you need a group mean EV for each subject. However, you
need just one EV for the paired difference (EV1 = [1 -1 1 -1 1 -1]).
See the Paired Two-Group Difference (Two-Sample Paired T-Test) example in
the FEAT web pages for how to do this:

http://www.fmrib.ox.ac.uk/fsl/feat5/detail.html#higher

Cheers, Mark.

Mark Woolrich.

Oxford University Centre for Functional MRI of the Brain (FMRIB),
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK.

Work: +44-(0)-1865-222782, Mobile: +44-(0)-7808-727745

On Thu, 22 Jan 2004, Goekoop, R. wrote:

> Dear FSL-users,
>
> I'd like to create a single contrast between two paired 1st level
> cope-images using a 2nd-level analysis:
>
> Input:  Group:  EV1     EV2     EV3
> 1       1       1       0       0
> 2       1       -1      0       0
> 3       1       0       1       0
> 4       1       0       -1      0
> 5       1       0       0       1
> 6       1       0       0       -1
> ...
>
> Since each EV only contains data from a single subject, would it still be
> necessary to include group mean EVs, indicating which inputs belong
together
> (i.e. are derived from the same subject, are paired), for example:
>
>         ...EV4  EV5     EV6...
>         1       0       0
>         1       0       0
>         0       1       0
>         0       1       0
>         0       0       1
>         0       0       1
> ...
>
> Or would it suffice just to specify EVs1-3 (see above),
>
> Thanks a lot,
>
> Rutger Goekoop
>
> Drs. R. Goekoop, MD.
> Department of Neurology
> Vrije Universiteit Medical Centre
> De Boelelaan 1117, P.O. Box 7057
> 1007 MB Amsterdam, the Netherlands
> Phone: 0031-20-4440316
> E-mail: <mailto:[log in to unmask]>
>