Hi everybody,

I'm a postdoc at UC Berkeley's School of Optometry, doing work on
stereovision and eye movements, both of which we're hoping to take
into the brain imaging domain. Since we're just about to start
getting serious about this, and since we're, to my knowledge at
least, the only people on our campus using AFNI, there's a lot of
trial and error and self guided learning going on. Now, however, we
seem to be genuinely stuck. Maybe somebody here can help and point
out our mistake or conceptual error.

We have a stimulus design with 6 different stimuli, with two
parameters, A1B1, A1B2, A2B1, A2B2, A3B1, A3B2. We did a block design
experiment, presenting blocks of each of these stimuli against
fixation perios, randomizing the stimulusblock order. Deconvolving
the response with the time profiles gave us six IRFs. We then ran
pairwise GLTs on these IRFs, important for our problem is the GLT we
ran between A1B1 and A1B2. The matrix C equates all 16 time points of
the two IRFs, not just area/magnitude. The result of this was, that a
fairly widespread cortical area where this GLTs F statistic was
highly significant (after Bonferroni-correction with
nvoxels*ntimelags)

In a different session, we repeated the experiment, randomized block
design, but now we run A1B1 against A1B2. No fixation. Again we
deconvolve, determining the area of significant activation. It's
miniscule (again Bonferroni-corrected, this time using just the
number of voxels).

Any way I think about it, if we find a significant difference in the
experiment with the six stimuli, the experiment where we close in on
just two of them should be more sensitive, not less. Is there a
qualitative difference between comparing two IRFs using GLTs and
comparing them directly? In other words, is there some problem in our
analysis that creates our seemingly paradox result?

Any comments are appreciated,
thanks and best wishes from California
Kai
--
Kai Schreiber PhD, School of Optometry, UC Berkeley
360 Minor Hall, Berkeley, CA 94720-2020, USA
Phone ++1 510 642 7710
[genista.de]

------------------------------------------
Posted to Phorum via PhorumMail