I don't think you are modeling gender and scanner appropriately. It looks like you have them as continuous covariates? The right way is to create classes. In your case, you'd have 8 classes, Group1MaleScanner1, Group1FemaleScanner2, etc.
On 09/15/2016 10:48 AM, Eryilmaz, H. Hamdi wrote:
Dear Doug/FreeSurfers,
We are looking at effects of age-squared on cortical thickness in two different groups. We are interested in age-squared by group interactions on cortical thickness across the surface. Nuisance covariates include age, sex, scanner, and total brain volume.
The sig map for the interaction shows numerous highly significant regions. When we extract cortical thickness from individual subjects for these regions, and then plot the relationship between age-squared and cortical thickness (after adjusting for the covariates) in group 1 versus group 2, the interaction is very clear.
However, using linear regression in SPSS, we are not able to reproduce these results. I suspect we are not using the same model in SPSS that FreeSurfer is using to do the interaction analysis.
This is how our fsgd file looks like:
GroupDescriptorFile 1 Title GroupAgesqInteraction Class Group1 Class Group2 Variables Agesq Age Gender Scanner TBV
Input 3662291 Group1 0.607 0.654 -1.061 1.041 0.706 Input 3665043 Group1 1.294 1.237 -1.061 1.041 -0.747 . .
This is the contrast we used:
[0 0 1 -1 0 0 0 0 0 0 0 0]
What would be the translation of this contrast to a linear model, where we are interested in the age -squared by group interaction on cortical thickness after controlling for the four nuisance covariates?
Thanks very much!
Hamdi and Josh
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer