Dear SPM experts,
I hope that someone would be able to clarify the following issues that I
have
concerning SPM99 analysis of structural data (PET/SPECT mode). I will be
referring to several mailbase responses sent from Dr. Andrew Holmes and
Jesper
Andersson. Thank you in advance.
I ran SPM99 "Multi-group:conditions and covariates" comparing the diffences
between 2 conditions in 2 different groups as described in
http://www.mailbase.ac.uk/lists/spm/1999-06/0097.html
The study is attempting to identify hemisheric asymmetries in group 1 and
group
2. For each subject in both groups their normal scan is condition 1, and a
scan
flipped about the x-axis is used for condition 2. Therefore, I have come up
with the following:
group 1: differences between flipped vs nonflipped
VERSUS
group2: differences between flipped and nonflipped
The results of contrast [1 -1] and [-1 1] appear to be the mirror image of
each
other about the x-axis. Is this correct? Am I using the correct contrasts
to
look at the differences between group 1 and 2 for flipped vs not flipped?
How
does one interpret the contrasts?
I was expecting to use the contrasts to see hemipheric differences in group
1
that differ from those in group 2 and vice versa. Therefore, contrast [1
-1]
infers that group 1 has asymmetrical differences (clusters) that are
distinct in
respect to group 2 . Instead, it seems that the contrasts only have an
effect
on the conditions, flipped vs nonflipped for both groups combined, thus
giving
me a mirror image in opposite contrasts. Does this issue have to do with
the
previous discussion on fixed effects vs mixed effects analysis?
If this is the case, is it possible for me to run a mixed effects analysis.
I
have attempted to create con*.img images (i.e. difference between flipped
and
non-flipped for each subject) to take to a second level analysis, but after
further investigation I realize that I cannot analyze a single subject since
I
will run out of degrees of freedom as I only have 2 scans per subject (one
flipped and one non-flipped). Are their any other models that can perform
this
analysis using this type of data?
Sincerely
Kevin Tessner and Elizabeth Sowell
|