The difficulty arises due to
the fact that the co efficient estimates of the
first two highly correlated variables is interpreted
as change in the depen dent variable
in question, ‘other things remaining constant’.
However, anytime a given change in
one variable occurs, the corresponding observation
on its highly correlated partner
is likely to change in a predictably similar
manner thereby making interpretation
difficult (Pindyck and Rubinfeld, 1998).