Dear SPMlers,
I calculated partial correlation coefficients of MRI data with EEG power
in a specific band during sleep (approximately 1 h) in MATLAB. I
transformed these correlation coefficients to Fisher's z values and
would now like to assess group statistics in SPM, as I have seen several
papers describe this.
I implement it the following way in SPM:
One-sample t-test with 1 z-transformed image per subject
no covariates
implicit mask
global calculation: omit
overall grand mean scaling: no
normalisation: none
Then I defined a contrast for the positive correlations [1] and one for
the negative correlations [-1].
Is that the correct way of doing it?
What startles me is that when I look at these correlation data in a very
crude approach (average over z-transformed data of subjects,
backtransform to correlation values), I see large brain areas of
negative correlations (when only looking at correlation coefficients
equivalent to ap-value < 0.05, Bonferroni-corrected for voxel number).
However, in the SPM approach described above, these negative
correlations do not survive correction.
Is there a step that I am missing? Do I need to "convert" the data in
another way? What else could be the reason that these quite robust
looking correlations do not reach significance?
Thank you so much for your help,
it would be highly appreciated,
best,
Laura
--
Dipl.-Biol. Laura Tüshaus
Institut für Pharmakologie und Toxikologie
Winterthurerstrasse 190
8057 Zürich
www.pharma.uzh.ch
Tel.: +41 (0) 44 6355961
|