Use gender as a discrete variable in the FSGD, ie, a class for M and a class for F. If you have two other groups (say, normals and patients), then use four classes (Mnormals, Mpatients, etc). Then test for the normals vs patients with (Mnorm+Fnorm) - (Mpat+Fpat). This "regresses out" gender.
doug

On 9/5/12 8:10 PM, Jeff Sadino wrote:

Hello,

I think our emails were getting crossed in cyberspace. I have a group that has a disproportionate number of females, and so there is a significant difference in the number of females as compared to males (chi-square p=.01). This must be accounted for. If I find that one group has a smaller brain volume, etc., then it could be attributed to one group having fewer females than the others. My ultimate goal is to run an analysis on the thickness, taking into account these sex differences (and ICV and age). My plan was to covary for sex, icv, and age. I have covaried for icv and age lots of times, but I am having trouble covarying (if that is the right word) for this categorical sex variable.

Any help would be greatly appreciated!

Jeff

On Wed, Sep 5, 2012 at 10:44 AM, Jeff Sadino <jsadino.queens@gmail.com> wrote:

Hello,

I am trying to covary for Gender in mri_glmfit, but I am having errors pop up. I tried to follow Doug's suggesting in this post:
http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg12272.html
but after modifying my Xg.dat file and rerunning mri_glmfit, it says:
ERROR: matrix is ill-conditioned or badly scaled, condno = 1.53319e+07

I can run it using the --illcond flag, but then I get a bunch of:
***
CEPHES ERROR: fdtrc domain error
***

Can anyone please give advice on how to covary for Gender in mri_glmfit (or qdec)?

As always, thank you for all of your great help!!
Jeff Sadino
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer