Dear Sam

I hope that other will comment this very important issue (you may found
some occasional comments in the list archives).
We have done only a little work on it, mainly because we wanted to use fMRI
for presurgical mapping, a field where type 2 error should be minimized.
You probably know all that will follow better than I do, and the approach
is rather naive, but anyway :
In order to evaluate type 2 error we thought of computing a statistical
power map. An other way to think about it is by making voxel by voxel power
study. Importantly to be able to do that, you have to make a strong
assumption about the minimal effect that your study should detect at a
given alfa risk... And the first problem is ... to determined it !
For our purpose (presurgical mapping) we found it reasonable to be able to
determined which regions were to much noisy for being able to see a fixed
percentage of the mean effect (we used 50%) given a determined alfa risk
(classically higher than 0.05) - I realize how unsatisfactory it is and I
hope some proposals.
Could it apply to your case ? If you know about the duration of the
potential your are looking for, you may infer its relative duration during
the paradigm : say 10% for a 100ms duration with a ISI (interstimulus
interval) of 1 sec. If you assume that there shouldn't be a different
spatial distribution than the other contributive regions, this could be use
to propose a minimal effect from the observed mean effect (this should be a
conservative assumption). You may found some results of this approach in
the attached pfd file (poster submitted 1 year ago, however we haven't gone
further). Many regions displayed a too large uncorrected noise for being
able to detect any signal of ~ 50% of the mean, the other were assumed not
being involved in the task (we didn't use all the corrections available
with SPM).
If I understood it well, in SPM an inhomogeneous noise distribution should
not be a problem because of appropriate spatial smoothing (something that
is also necessary for the theory of gaussian random field to apply).
However, I have never looked at the error spatial distribution after
regression in SPM. Someone reported me a poster on this theme by the WDCN
team at HBM98, I hope their contribution.

I may not be the right guy for talking about correction for multiple test.
However, I would be surprised that the classical GRF theory could apply
here, since the approximation that is made is only valid for high p value
(or should I say alfa risk ?). In the above approach you proposed as well
as the above described, the used p are high.

I am looking forward to many other comments on this issue ... thanks for
this question

Sincerely



Jack

PS : I realize that the pdf file is rather heavy (480 ko), so I propose
that interested persons simply ask for. I should warn them that it was my
first work on this field and that I realize how naive was this approach for
a so difficult problem. However, iconography look fine ;-)

________________________________________________________________
| Jack Foucher Universite Louis Pasteur |
| Institut de Physique Biologique UPRES-A 7004 du CNRS |
| 4 rue Kirschleger Tel: 33 (0)3 88 77 89 90 |
| 67085 STRASBOURG Fax: 33 (0)3 88 37 14 97 |
| France |
| Faster E-mail: [log in to unmask] |
| Other [log in to unmask] |
|_______________________________________________________________ |



-----Message d'origine-----
De: Sam Reyes [SMTP:[log in to unmask]]
Date: jeudi 4 novembre 1999 20:56
Cc: [log in to unmask]
Objet: Power Analysis

Greetings,

I am interested in using functional imaging results to constrain EEG
source reconstruction, but I'm concerned that there may be a large type
2 error (the other beta) for short duration sources. If I used a large
enough threshold (say 0.2) I assume that the chance of a type 2 error
would go down, but is there a way to determine the chance of a type 2
error? Has anyone worked on multiple comparisons factors into this?

Thanks for any help you might be able to give.


---sam Reyes

PhD Student
SUNY Buffalo
Hearing Research Lab
e-mail: [log in to unmask]
Phone(s): (716) 829-2001; (716) 862-8790
Fax: (716) 829-2980
Mailing Address: 215 Parker Hall
Buffalo, NY 14214



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%