one thing that comes to mind is that the pvalues produced by glmfit are two-sided. What does SPSS use?
Goldman, Aaron (NIH/NIMH) [C] wrote:
Hi, I've been looking at thickness data and Freesurfer, and had a question about the process. I initially compared thickness in our two groups on a node-by-node basis using:
mris_preproc --fsgd_file file_name --target average --hemi rh --meas thickness --out rh.thickness.mgh
mri_surf2surf --hemi rh --s average --sval rh.thickness.mgh --fwhm 10 --tval rh.thick_smooth.mgh
mri_glmfit --y rh.thick_smooth.mgh --fsgd file_name doss --glmdir average/glm/rh --surf average rh --C contrast.m
with the fsgd file containing inputs like:
Input name1 NC-male 31.34
etc.
(the number at the end being age)
I also tried running GLM on the average label thicknesses in the parcellation stat files to see how similar the results would be. I did this by reading the values in the ThickAvg column into SPSS, and using group and gender as fixed factors with age as a covariate. Doing it this way, several labels came up significant that contained few or no significant nodes doing it the other way.
Since this was surprising, I was wondering if there was anything that might be expected to cause this difference. For example, could there somehow be a scaling factor applied to one analysis but not the other?
Thanks,
-Aaron-
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer