We have scanned 19 patients belonging to 2 nosological groups (10 vs. 9 subjects). 16 of the patients (9 vs. 7 subjects) have been examined twice at 2 conditions of their illness, 3 of them (1 vs. 2) have just completed the examination at condition 1.

(1) For group 1, a previous study using Xe133-cerebrographies (the old "gold standard" to determine absolute perfusion values) has indicated a significant global bihemispheric hyperperfusion at cond 1 as compared to cond 2 manifest in 7 out of 8 patients. That may confound our rCBF anaylsis. Thus, I have extracted the globals scaled to the subject-specific means by ANCOVA. In neither of the 2 groups nor the total population there was any significant difference detectable between these globals of cond 1 versus 2 (t- & sign-test). Thus, I would like to use proportional scaling for the final design setup (particulary since I do not want to waste my sparse degrees of freedom in the analysis). However, individual differences of the estimated globals between conditions may amount up to 40 %. Is proprotional scaling proper for such kind of data, or are there other alternatives? After all, I could also test for condition effects or group per cond interactions allowing for global changes. However, I feel my data do not support the notion of consistent gCBF changes between conditions and therefore I would prefer proprtional scaling.

(2) Is there any way to set up a single design matrix allowing to test for not only condition effects within groups and the group per cond interaction but also for the effects of groups within & across conditions? Can I include subjects who were only scanned at cond 1?