Dear Frederik, Why don't you just correlate your effect magnitudes across subjects with the parameter values obtained by doing a BMA of all models within subject? This sounds like a more standard strategy. Best, Vladimir On Tue, Dec 1, 2015 at 12:40 PM, Frederik Van de Steen < [log in to unmask]> wrote: > Dear DCM experts, > > I am conducting a DCM study for ERP's . More specifically, one of the > things i would like to do is to link DCM (modulatory) parameters with two > temporal (early and late effect) ERP effects. I already fitted a quite > large model space (400 models). My idea of linking parameter to these two > effects was to calculate a mean (across subject) predicted effect > (channels) for every fitted model for both temporal effects. Then, for each > temporal effect, i would take 10% (or 50%) of the models with the smallest > predicted effect and 10% of the models with the largest predicted effect > (sort of 'data' driven post hoc family partitioning). Than i would use BMA > for these models with lowest and highest effect. After I would simple use > classical paired t-tests on these BMA parameters to see which parameters > are significantly different. > > Is this a valid analysis strategy? I would be grateful for comments/advice > regarding this issue. > > Kind regards, > > Frederik Van de Steen >