By sorting X1 and X2 by Y an interesting pattern is revealed. The extremes of
Y are determined by the random pairings of similar X1 and X2 values. In
other words, X1 and X2 are correlated positively in the extreme ranges of Y.
X1 and X2 are correlated negatively in the mid-range of Y. They cancel one
another to form the mid-range values of Y. Thus the correlations between the
independent vaiables X1 and X2 are polarized across the extreme vs. mid
ranges of Y. Polarization occurs when the independent variables are either
subtracted or multiplied, as well, but not when they are divided by one
another.
Such polarization does not occur across the ranges of either independent
variables, whether we are examining the correlations between IVs or between
the IVs and the DV. Thus, such polarization discloses a formal cause
relationship. Rychlak's work led me up to these insights.
This idea is elaborated in detail in a series of papers in the Journal of
Mind and Behavior , 12,1, 1991. The polarization can be applied using
regression analysis to cases where only two variables are measured.
Imagine being able to say things like:
1.This client's views of himself are determined by his views of God, not the
bandwagon.
2.The movement of this client's constructions is determined by recent visits
to the woods, not to a glamorous therapist.
3. The meaning of integrity is determined by the constructs honesty, courage,
and goodness for this client. Compliance with the staus quo does not effect
his construction of integrity.
4.Sweet smug smiles and authoritative manner does not drive scientific
productivity even though they are correlated with number of publications and
life time service awards.
The list goes on. The thing is, of course, that since Hume it has been
unfashionable to even imagine that causes might someday be measured." Mere
correlation" has been the objection, as well as the license for keeping
things loose and conveniently sophistical.
I dare you to simulate these ideas yourself and to consider the impact if the
method of corresponding regressions is valid. Page 336 of Kelly's PPC may be
an interesting place to start.
William Chambers
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%