Dear Jerry,
> For a motor task (finger tapping), I have data for which the global
> brain mean correlates with the task.
>
> You indicate that global scaling should be avoided since it may
> de-emphasize activations in motor areas and create activations in
> non-motor areas. I concur.
>
> However, if I do nothing, I'm left with a set of data for which the
> global brain mean correlates with the task which yields lots of
> (innapropriately) activated voxels.
I guess I disagree that omitting global scaling in the presence of
such correlation would lead to inapproriately activated voxels.
A correlation between the global brain mean and the task is most typically
due to the fact that a substantial number of voxels in the brain were
influenced by the task. Say for instance, 10% of all voxels in the brain
were indeed activated by the motor task and correlated perfectly with
it. When the global brain mean is computed, these 10% will
contribute to its timecourse, and may cause it to fluctuate together with
the task.
Given that, I don't think omitting global scaling would cause
voxels to be inapproriately activated. The converse, however - including
global scaling in the presence of correlations between the task and the
global mean - may cause activations to be obscured and may cause regions
that are not influenced by the task to appear deactivated). Or that's how
it seems to me anyway - see
http://www-psych.stanford.edu/~kalina/SPM99/Tools/rd_global.html
Best regards,
Kalina
> Jerry Allison, Ph.D.
> Medical College of Georgia
>
> >>> Kalina Christoff <[log in to unmask]> - 1/12/01 1:33 AM >>>
> Dear Stephan,
>
> > Hi Kalina, your page has been very helpful in filling in some gaps in
> > my SPM knowledge. I was so happy to happen upon it while checking out
> > the Gabrieli page. In particular, the global scaling discussion
> > helped to clarify the issue beyond what I got from the Neuroimage
> > note by G. Aguirre. Still, I have a question regarding global
> > scaling. If you do not use global scaling at the 1st level of a
> > random effects analysis, should you also not use it at the second
> > level? I suppose my question boils down to: is it valid to perform an
> > entire analysis, either fixed or random, with no correction for
> > subject differences in overall mean signal? What artifacts does one
> > risk?
>
> To my understanding, there are 2 types of scaling:
>
> 1) Global Scaling - which is done if 'global scaling' is selected. This
> means the voxel values in every V-file get divided by the mean of all
> voxels for this V-file. This scaling can be problematic if the global mean
> correlates with the task; it can introduce changes in the signal.
>
> 2) Grand Mean Scaling - which is done if 'global scaling' is NOT
> selected. During this scaling, the voxel values in every V-file get
> divided by the mean of the global means for all V-files within a
> session. This type of scaling does not introduce changes in the signal.
>
> So, even if you don't select 'global scaling', there will be a
> correction for subject differences in overall mean signal, via the Grand
> Mean Scaling (I believe, but other may correct me on this).
>
> > Thanks for your help. I think the answer would be of interest to the
> > SPM mail list.
>
> I hope this is helping rather than confusing things further,
>
> Kalina
>
> _____________________________________________________________________________
> Kalina Christoff Email: [log in to unmask]
> Office: Rm.455; (650) 725-0797
> Department of Psychology Home: (408) 245-2579
> Jordan Hall, Main Quad Fax: (650) 725-5699
> Stanford, CA 94305-2130 http://www-psych.stanford.edu/~kalina/
> _____________________________________________________________________________
>
_____________________________________________________________________________
Kalina Christoff Email: [log in to unmask]
Office: Rm.455; (650) 725-0797
Department of Psychology Home: (408) 245-2579
Jordan Hall, Main Quad Fax: (650) 725-5699
Stanford, CA 94305-2130 http://www-psych.stanford.edu/~kalina/
_____________________________________________________________________________
|