Hallo,

in my previous experience in carrying out PCA on fMRI data (not using the
SPM2 toolbox), I found that there is more than one component in these data.
In general, I would think that there is something wrong with the procedure
you are using since it is just very unlikely that the data happen to lie on
a single line. One possible explanation of this result is that the data were
not centered. In this case, the first eigenvector points to the average of
the data, and the eigenvalue depends on the size of the average vector. This
size could be so large (it will be very large with a large number of
dimensions) that after rounding error it is reported to amount to 100% of
the variance. In this case the other eigenvectors are still there, you just
do not see them with the right eigenvalue.

Hope this helps,
R. Viviani
Dept. Psychiatry
University of Ulm


----- Original Message ----- 
From: "Helmut Laufs" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Thursday, August 04, 2005 4:12 PM
Subject: Re: [SPM] PCA and SPM2


> Dear Irina, dear Ferath,
>
> Finally, I was able to run a PCA on a fingertap using the old MM toolbox
for
> SPM99 and the newer one for SPM2.
>
> Before, I did not use the toolbox on finger tap data, but on some data
where
> I did not know what to expect. This is why I did not notice what you have
> discribed.
>
> This is what I have found now:
>
> 1) I observer the same [problem] that you observe: with SPM2, the 1st
> eigenvariate explains 100% of the variance, the eigenvector looks almost
> like the regressor (convolved boxcar). In fact, it looks slightly
different,
> but this may be due to some filtering?!
>
> 2) Running an analogous SPM99 model and using the older MM version, with
the
> dataset I used, the 2nd eigenvariate explains almost 10% of the data and
> resembles very much a block design, showing 'activations' in the motor
> cortex as expected (and found using the GLM).
>
> In other words: I can confirm but no solve [y]our problem (I do not intend
> to look into the code as I may not be capable to do that with any
success).
> I apologize for my previous, long explications on using MM which was
before
> you had specified your problem. Have you meanwhile found the answer so you
> could share it with us? Thank you for pointing that problem out which
saved
> me some huge problems (because I had planned and started using the
toolbox).
>
> With best wishes,
>
> hoping for help,
>
> Helmut
>
>
>