On Sep 25, 4:01 am, djh <halitsk...@att.net> wrote: > One more question that is orthogonal, i.e. not related, to the three > in my last post. > > Can N be validly regressed on (e,u,L)? (Not the derived factros c/e, > c/u, c/L or their logs - the raw values themselves?) That is, to > remove N as an "interfering" factor, can we treat it as a dependent > variable ?
No. N is design parameter. You choosed its value. It can't be a d.v.
> > If so, can the coefficients of the resulting regressions be compared > so long as the denominator is k-4 where required? > > Or would a "time-series" analysis be required to do this kind of > thing, as you intimated some time ago?
I doubt that traditional time-series models would be appropriate, but then that's only a top-of-the head reaction.