Hi,
A reviewer raised some concerns about the design efficiency and power
for a study comparing young and older adults (background info: although
we used optseq to create an efficient design across conditions, our
study uses a subsequent memory design in which items are allocated as
remembered or forgotten based on the individual subject's performance).
Two questions:
1) We've been asked to include "a measure of the relative power (in
percent signal change) of between condition contrasts". I haven't found
anything on the list that indicates how to calculate this in SPM - does
anyone know of any tools??
Verbatim review comment:
With six conditions, all subject determined, there is the possibility of
differential levels of covariance between conditions in the two
different subject populations. In other words, any systematic
differences in design efficiency between the two groups would
significantly affect the ability to detect important signal
changes/differences. (see - Smith, S., Jenkinson, M., Beckmann, C.,
Miller, K., & Woolrich, M. (2007). Meaningful design and contrast
estimability in FMRI. Neuroimage, 34(1), 127-136.). The issue here is
less the number of trials per bin (mentioned as a possibility in the
manuscript), than whether the conditions significantly covary.
The FSL analysis package includes a measure of design efficiency as well
as a measure of the relative power (in percent signal change) of between
condition contrasts. If such a measure is generated for each subject,
then differences between the groups could be tested for significance.
2) I was able to use a formula posted to the list to calculate design
efficiency (thanks!), but I wondered out of curiosity if anyone could
better explain, at a general level, what the formula is doing (e.g., why
invert & multiply matrix elements in this way?):
Eff = trace(C*inv(X’*X)*C’)^-1
Thanks for any assistance!
Angela
--
Angela Gutchess, Ph.D.
Assistant Professor of Psychology and
Volen National Center for Complex Systems
Brandeis University
http://www.brandeis.edu/gutchess/
|