Greetings
I've preprocessed MEG data from an odd-ball paradigm using SPM 8
through robust averaging as per instructions in the manual. Firstly,
in the manual it is suggested to carry on sensor level analysis using
unpaired t-test and defining a contrast between trial types (rare vs.
frequent trials). What is the reasoning behind using unpaired as
opposed to paired t-test?
Specifically in my case, after defining the design matrix for a two
sample t-test (frequent vs. rare trial types) and computing estimates
I receive the message 'please check data. There are no significant
voxels' and SPM doesn't allow me to proceed further i.e., defining an
F-contrast. The contrast images were produced using MEG PLANAR as
modality and have dimensions 32x32x361, with bad channel
interpolation. I am wondering whether 32 is the appropriate dimension
for 102 or 306 channels? Review of the image for the 'rare' trial
types reveals activity spread over desired regions, and inspecting the
time course of activity for a single channel (attached) reveals a
clear difference in time between conditions. Visually this difference
is quite large and prolonged for this particular channel, I am having
difficulty believing that SPM is unable to reveal ANY significant
differences for this particular contrast in my dataset. Obviously the
MEG data images are massively smoothed during interpolation to 32x32
voxels. Should I be compensating for this by modifying particular
settings in the design matrix, estimation procedures? Thanks for any
suggestions or advise.
Kambiz
--
------------------------------------------------------------
Kambiz Tavabi PhD
Institute for Learning & Brain Sciences
1715 Columbia Road N
Portage Bay Building
Box 357988
University of Washington
Seattle, WA 98195-7988
Tel: 206-221-6415
------------------------------------------------------------
|