I think the answer is that because the data are globally scaled to a
mean of 100, the betas should always be in relatively similar units
(roughly percent). This slightly finesses the issue that the mean of
many voxels will be quite different from 100 (since it is normalized
across the whole brain), so this issue still exists to some degree.
I'll be interested to hear if there are other issues that impact on
this.
cheers
russ
On Nov 30, 2003, at 7:44 PM, Jeffrey P Lorberbaum wrote:
> Dear Group
>
> I am asking a basic fMRi stats question which I asked just before the
> Thanksgiving holidays and am trying again to see if I can get help.
>
>
> I have a question about the regression coefficients (beta.imgs)
> produced
> by the spm2 analyses. Is the baseline b0 (constant coefficient)
> image weighted in some fashion by default? If not then if I am thinking
> correctly, this would be a problem for a random effects analysis.
>
> Say one person at a voxel had two conditions cond1 with a beta weight
> of 2 and cond2 with a beta weight of 1 and a baseline b0 of 1000.
>
> (ie. yperson1 = 2*cond1 + 1*cond2 + 1000 + error)
>
> Say another person at a voxel had two conditions cond1 with a beta
> weight
> of 2 and cond2 with a beta weight of 1 and a baseline b0 of 500.
>
> (ie. yperson2 = 2*cond1 + 1*cond2 + 500 + error)
>
>
> The beta weights in the second person accounts for a bigger
> percent change relative to baseline when looking at cond1, cond2,or
> cond1
> - cond2. How does a random effects analysis that looks at mean beta
> weights and the variance among them account for this?
>
> Am I thinking correctly?
>
> Thanks
> Sincerely,
> Jeff Lorberbaum
>
---
Russell A. Poldrack, Ph.D.
Assistant Professor
UCLA Department of Psychology
Franz Hall, Box 951563
Los Angeles, CA 90095-1563
phone: 310-794-1224
fax: 310-206-5895
email: [log in to unmask]
web: www.poldracklab.org
|