Cameron Ellis wrote:
Hi all, I have two questions regarding glmfit and testing for effects of different variables. I have four groups. When testing for effects of age across all of the groups, will there be a difference when assigning each group a class in my fsgd file, with age as a variable, like so:
Class PH Class PL Class VH Class VL
Variables age Input NEAA004 PH 30 Input NEAA006 PH 21 Input NEAA007 PL 42 Input NEAA020 PL 27 Input NEAA024 VH 26 Input NEAA025 VH 24 Input NEAA039 VL 24 Input NEAA052 VL 33 . . etc...
and setting up a contrast matrix: 0 0 0 1; and alternatively setting up my fsgd file with a single group, like so:
Class subj
Variables age Input NEAA004 subj 30 Input NEAA006 subj 21 Input NEAA007 subj 42 Input NEAA020 subj 27 Input NEAA024 subj 26 Input NEAA025 subj 24 Input NEAA039 subj 24 Input NEAA052 subj 33 . . etc...
and setting up a contrast matrix: 0 1? Is the first options controlling for effects of group? Would I want to do this?
The first option does control for group (you'd need a contrast of 0 0 0 0 1 to test age). You'd want to do this if you believe that group affects your response variable. If it does and you do not code for group, the age slope may be off.
My second question is in regards to controling for a discrete variable like gender. Should male and female be coded as dummy variables 0 and 1 in the fsgd file? I have tried this and my results seem to make more sense than when I run the same analysis in qdec where I did not code the variables in this way. Should I be doing so in qdec as well?
You should code them as part of a class. Eg:
Class PHMale Class PLMale Class VHMale Class VLMale Class PHFemale ...
doug
Thanks! Cameron
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer