On Oct 10, 9:14 am, djh <halitsk...@att.net> wrote: > Suppose for the sake of discussion that we can establish Re, Ru, Reu > as statistically and empirically meaningful regressions, perhaps in > the new design where the S-C and uL-uH dimensions have been removed > (pushed down to the observation level by taking u as "u for S less u > for C" and e as "e for S less e for C". > > Then we will want to use these regressions on other folds, to generate > predictors for structural alignability logistic regressions. > > Are you saying we have to re-establish the validity of the regressions > on each new set of folds we deal with? That we can't take their > validity for these six as an indication of validity for other sets of > folds ???
From a purely statistical perspective, that's exactly correct. Just as having a drug work well for 6 selected people does not imply that it will work well for other people.