Allie, to be more precise, you have selected control<patient regressing out the effects of age. This second part is the key. If you were to trace the regression lines back to age=0 (the meaning of "regressing out age"), then control<patient by a wide margin. But you have a problem here in that the regression lines are crossed. This means that you have an interaction between age and patient (no interaction means that the lines would be parallel). An interaction means that you cannot assess whether there is a difference in the thickness because this difference changes depending upon the age. One thing you can try is to actually compute an interaction contrast (something like [0 0 1 -1]). If it is significant in this area, then there's not much you can do (though it may be interesting in itself). If it is not significant, then you can switch to a DOSS model which forces the lines to be parallel and makes the distance between the lines be independent of age. Does this make sense? doug
Allie Rosen wrote:
Hi All,
I'm having trouble reconciling the cluster colours with the graph that can be made when you load the group descriptor file and click on a point. I've attached an example.
In this study, I'm comparing thickness in a patient and control group, with age regressed out. The contrast I've used means that blue means control<patient. Therefore, patients are thicker than controls. But in the graph, patients don't seem to be thicker. I've seen the same results with red clusters (i.e. they should mean that controls are thicker but the graph shows otherwhise). Please let me know how I can interpret my findings given the graph. Which group is actually thicker?
Thank you, Allison
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer