How does one conduct model selection when there are multiple sessions/subject?
Specifically, Fig. 1 of "Ten simple rules for dynamic causal modeling" (Stephan et al., _NeuroImage_ 49 (2010), p. 3101) has a bifurcation: if "optimal model structure assumed to be identical across subjects," use FFX BMS, otherwise use RFX BMS.
What if one wants the model to be identical across sessions within each subject, but not necessarily across subjects, and the number of sessions is low (not enough to conduct inference on a per-subject basis, e.g. 2 sessions)?
Of course, one could model the entire experiment as one session by concatenating the sessions, but I'm trying to avoid doing that.
TIA,
S
|