I have recently tried to use the command line to examine thickness differences between two groups after controlling for the effects of age and sex. I get the following 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.
If you seek help with this problem, make sure to send:
1. Your command line:
mri_glmfit.bin --y CAA_CTRL_RH.thickness.10.mgh --fsgd CAA_CTRL_RH_Age_Sex.fsgd --C Contrasts/CAA-CTRL_Age_Sex.mtx --surf fsaverage rh --cortex --glmdir CAA_CTRL_RH_Age_Sex_Unix.glmdir
2. The FSGD file (if using one)
3. And the design matrix above
Attempting to diagnose further
SumSq: Min=0.000000 (col 5), Max=611.699463 (col 4)
The scale is much different between columns 5 and 4, you may want to
normalize by subtracting the mean and dividing by the standard deviation.
Column 5, all values are 0
Column 6, all values are 0
Columns 5 and 6 are the same
I did not get this error when I was only looking at the difference in thickness between the two groups, so I am guessing there is not an issue with the file format? Looking at the log, does this mean that I also need to demean the age variable?
I was planning on running some contrasts looking at the interactions between group and age after controlling for sex etc. in order to see if I should go with a DOSS approach, but using those contrasts still gives me the same error.