Thank you for your mail, this is very helpful.

Another second level analysis question:  Do the thresholds that we set at the
second level have any bearing on the third level analyses?  If I set more
stringent thresholds at the second level, does that reduce my ability to look at
the data with more liberal thresholds in the third level?

Thanks,
Dost



-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of
Tim Behrens
Sent: Wednesday, September 08, 2004 4:18 AM
To: [log in to unmask]
Subject: Re: [FSL] second level analysis


Hi -
Unfortunately, as you only have two runs per subject, you will not be able to
estimate a between-session variance separately for each subject.

What we advise under these circumstances is to make the assumption that all
subjects in a given group have the same between-session variance. Then you
estimate a single between-session variance for each group, which will
effectively be added onto the variance associated with the mean of the
measurements, and be taken up to the third level.

To do this, you could eaither have 1 big design matrix with 2 groups in it, or
(easier) one design matrix for each group, each of which will look
like:

     GP EV1 EV2 EV3 ....
     1   1   0   0  ....
     1   1   0   0  ....
     1   0   1   0  ....
     1   0   1   0  ....
     1   0   0   1  ....
     1   0   0   1  ....
     .   .   .   .  ....
     .   .   .   .  ....

con:     1   0   0  ....
         0   1   0  ....
         0   0   1  ....


If you take the two DM approach, then you can run the group without your new
subject in, before you collect data from the new subject.

Hope this is useful

T

-------------------------------------------------------------------------------
Tim Behrens
Centre for Functional MRI of the Brain
The John Radcliffe Hospital
Headley Way Oxford OX3 9DU
Oxford University
Work 01865 222782
Mobile 07980 884537
-------------------------------------------------------------------------------

On Tue, 7 Sep 2004, Dost Ongur wrote:

> Hello,
>
> We have a study with 34 subjects in 2 groups (patient and control).
> Each subject had 2 identical runs.
>
> We are waiting to scan a final patient and have 35 subjects for the
> ultimate analysis.  In the meantime, we would like to go ahead and
> finish the first and second level analyses for all existing data sets.
>
> Is it OK to do the second level analysis (across runs for each
> subject) for the 34 existing subjects in one large batch and then run
> the final subject's data when they become available on its own?  Would
> we have to repeat the second level analysis once we have all 35 scans?
>
> Thanks,
> Dost Ongur
>