On 11/04/2015 05:10 AM, Erik Lindberg wrote:
Dear Freesurfer experts,
I know this has been an ongoing topic in the mailing list – however I would like to raise the issue one more time. In the choice between doss and dods – I usually check if there is a significant interaction and if so my choice is dods.
As I understand it you do not recommend to use doss if an interaction is present. Is the reason for this that I would “over-correct” for the variable in one group while “under-correcting” for the same variable in the other group?
If the interaction is present, then it (almost) makes no sense to run the DOSS because you would be looking for a difference in group intercepts. With an interaction, the difference in intercepts changes depending upon where the intercept is measured (eg, age=0 vs age=70). If there is an interaction, then the thickness vs age plots for the two groups MUST intersect at some age. If you test for a difference in groups at that age, you will not find one. If you test younger you will see a sign flip relative to older. The DOSS model is not appropriate because it assumes/forces the plots to be parallel.
It is some time tempting to present result both from a dods and a doss – to tell where the interaction is present, and where difference are present while controlling for the variable
But if I understand Doug´s previous comments on this – this would not be a correct approach?
Not sure that would work. Why not just test for a difference between slopes?
Thanks
Erik
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer