Hi Doug,

> 3) In a group X gender X age DODS design, exactly which contrast
> (interaction) has to be examined in order to see if the analysis
> proceeds as DODS or as DOSS? And what would the course of action be
> assuming there are significant clusters for that contrast?

group X gender X age, group X age, gender X age

I've had statisticians tell me that you cannot do anything more if there
is a sig interaction. However, this kind of thing gets published all the
time without people even checking. If you wanted to push ahead, you have
to choose an age where you want to do the test (eg, age=0 which would be
the default or age=mean age). But you have to be aware that a sig effect
found betweeen groups will depend on the age that you test it.

Would it not be possible to decompose such a significant interaction (e.g. group X gender X age) the same way one would decompose any ANOVA interaction, i.e. by doing a post-hoc contrast to directly compare the means corresponding to the levels of the factor of interest (e.g. group A vs group B collapsing across genders) while deciding on an acceptable value for the factors of non-interest, as you suggest for Age?

Since such post-hoc comparisons cannot be done directly in QDEC, would it make sense to instead decompose the interaction by manually repeating the analysis with only the relevant factors/levels left in the design?

Tudor

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