Hello everyone,
I'm hoping you can help me figure something out. I've run three
first-level analyses for each of 31 subjects, one for each run of data
collection. These analyses have a total of 10 EVs, 2 of which are
behavioral covariates representing trials with and without valid
button press responses. At a second level, I'm averaging across these
three first-level analyses using a fixed effects approach. The second
level analysis works just fine for most subjects, but there are two
for which it shoots out all-zero pe's, copes, zstats, etc. These two
subjects are (I think not coincidentally) those for whom, for at least
one run, I had to use all-zero 3-column format files for the
behavioral covariates (we had no button press responses due to
experimenter error for one, and for the other, for one run, the button
press EVs were too colinear with the EVs of interest). At the first
level, these two subjects look just fine (i.e., I see activation in
expected regions like visual cortex). However, at the second level,
the log indicates "WARNING: The passed in varcope file,
tmpvarcope0000, contains voxels inside the mask with zero (or
negative) values. These voxels will be excluded from the analysis."
This warning is produced repeatedly. I'd thought of perhaps just
leaving the behavioral covariate EVs out when running the second level
analyses but I'd actually like to see the effect of the button press
across subjects at the highest level. Does anyone have a suggestion
for how to handle this problem? I'm running FSL 4.1.0 on Mandriva
linux.
Thanks!
Best,
Heather
--
Heather L. Urry
Department of Psychology
Tufts University
490 Boston Avenue
Medford, MA 02155
email: [log in to unmask]
phone: 617-627-3733
fax: 617-627-3181
|