Hello,
I am trying to test for the effect of different variables on cortical thickness. The variables (with type) are: index (continuous). Gender(factor), Age(continuous), Years of education (continuous), Depression (factor), Vascular comorbidity (factor), and MMSE (continuous). Later, I need to correlate thickness with my variable of interest (index). Depending on the results of the current analysis, I will choose which of the other variables will be my covariates.
Here is what I attempted to do so far and the error I got:
I have constructed the fsgd file according to the example with one group and covariates (see attahced to this email). My contrasts look like, for example: 0 1 0 0 0 0 0 (to test effects of gender), and 0 0 1 0 0 0 0 (to test for the effects of age). I am following the commands in the Group Analysis tutorial. I pass successfully mris_preproc and mri_surf2surf. When I get to mri_glmfit, I receive the error:
ERROR: matrix is ill-conditioned or badly scaled, condno = 1e+08
Possible problem with experimental design: Check for duplicate entries and/or lack of range of continuous variables within a class.
1. My command line is mri_glmfit --y lh.newindex.thickness.10.mgh --fsgd FSGD.fsgd dods --C index.mtx --C gender.mtx --C age.mtx --C education.mtx --C depression.mtx --C vascularc.mtx --C MMSE.mtx --surf fsaverage lh --cortex --glmdir lh.newindex.glmdir2. Please find attachced my fsgd file 3. Find attached the Design matrix
What is the best way to construct the fsgd and contrasts? In the version I am sending, I have demeaned all continuous variables. Is that something you will recommend to do? (I tried the group analysis with a fsgd file containing the normal, non-demeaned values, but got the same error).
Thank you for the help!
Best,Teodora