On 7/7/17 1:58 PM, Yann Ying Chye wrote:

Dear FreeSurfer mailing list,

I've got two questions regarding the freesurfer GLM analysis.

1. I am trying to run mri_glmfit on the cortical thickness of a

dataset that has a few interacting factors of unequal levels. For example, my dataset may have groups A, B, C, and D. Within the groups, subjects are further divided by gender - male and female. However, group A only has males, such that the classes in my FSGD file = A-male, B-male, B-female, C-male, C-female, D-male, and D-female. Will this design be a problem? Will it be statistically sound? Additionally if I was looking for an effect of gender, how would I weigh my contrast? I was thinking 0.25 0.25 -0.33 0.25 -0.33 0.25 -0.33. Would that be valid?

I think I would lean against including A-male. Otherwise, what is the null hypothesis? Without A-male, it would be that males and females differ when regeressing out the effect of group, but you can't say that if you include A-male. Eg, if there were no difference between A-male and A-female but A tended to be larger than the other groups, then including A-male and not A-female would create a confound making males look bigger than females.

2. Does mri_glmfit have a limit to the number of subjects it can run

with? If I have say upwards of a few thousand subjects, will this be possible to run/how long would that take?

I don't think it has a limit.

Thanks for your help!

Regards Yann

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer