INTERPRETATION OF LEAST-SQUARES REGRESSION WITH INTERACTION OR POLYNOMIAL TERMS

Discusses some problem-causing properties of least squares regression with interaction or polynomial terms. While the equation as a whole is insensitive to linear transformations of the underlying coefficients. Standard significance tests of estimates of individual coefficients in an equation with interaction or polynomial terms are sensitive to nonhomogeneous scale transformations of the underlying variables. If a complete model is not used, significance tests for the equation itself may be sensitive to such transformations. -Authors