Sorry to be a bother, but I seem to be having an issue
with my analysis.
I have a control and a patient group, and I'm trying to covary
gender the majority is skewed towards one gender.
I'm looking across 4 metrics (thickness, surface area,
volume, and lgi) of the cortex.
My continuous variables include Age for all metrics, Total
incracranial volume for the volume analysis, total
intracranial surface area for the area measure (cubed root
squared transform of eTICV), and Mean thickness for the left
and right hemisphere (each separate for their respective
hemispheres) for thickness.
My goal is to look at changes in these as a function of
age, relative to total ICV/SA metrics (because my patient
population has arrested brain development in theory), and
diagnosis. So I'm examining the following interactions:
DiagnosisXAge, AgeXICV, DiagnosisXICV, DiagnosisXAgeXICV
Which originally gave me 6 columns for the "total
vol/area/thick" analyses: Control, Patient, ControlAge,
PaitentAge, ConICV, PatICV if I'm not mistaken and 4 for the
non-tICV lgi analysis.
On the recommendation of the tutorial, I coded my change my
2 class variables to 4 to account for gender (so now its
MaleControl, MalePatient, FemaleControl, Female Patient)
Which should give me 12 columns right?
ConMale, ConFemale, PatMale, PatFemale, ConMaleAge,
ConFemAge, PatMaleAge, PatFemaleAge, ConMaleICV, ConFemICV,
PatMaleICV, PatFemaleICV
This analysis will run for volume, area, and lgi (no icv so
8 columns), but I keep getting the following Message from the
terminal for thickness:
--------------------------------
ERROR: matrix is ill-conditioned or badly scaled, condno =
16115.7
--------------------------------
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 --y
lh.diagnosisxagexleftThickslopetest.thickness.10.mgh --fsgd
GenderGroupingSBM_glmLeftthick.fsgd dods --C
diagnosisnonlGIICmeas.mtx --C agesxdiagnosisnonlGI.mtx --C
diagnosisnonxagexICmeas.mtx --C ICmeasurexdiagnosisnonlGI.mtx
--surf fsaverage lh --cortex --glmdir
GenderGrouplh.Controlled_diagnosis_agethickness.glmdir
2. The FSGD file (if using one)
3. And the design matrix above
I see this has been a problem for others, but I'm not sure
if its a scaling issue, a problem with my FSGD file, or that
mean thicknesses correlate too much with age.
I've attache the FSGD file in question along with the contrast
matrixes and the X file (cue theme) from the analysis.
Any input would be greatly appreciated, thanks.
--
Thomas DeRamus
UAB Department of
Psychology, Behavioral Neuroscience
Graduate Research
Trainee
Civitan
International Research Center
1719 6th Ave S,
Suite 235J, Birmingham, AL 35233