The purposes of this study are: (1) to demonstrate the superiority of simple ridge regression over ordinary least squares regression through theoretical argument and empirical example; (2) to modify ridge regression through use of the variance normalization criterion; and (3) to demonstrate the superiority of simple ridge regression based on the variance normalization criterion over those ridge regression estimates based on minimum mean square errors. A theoretical discussion and an empirical study constitute efforts to fulfill these purposes. The discussion of the properties of ridge regression demonstrates that in ridge regression the total variance of the estimated regression coefficients is greatly reduced by the introduction of a small bias in the estimates, but ridge regression based on the minimum mean square error criterion has four limitations. From this analysis a "unifunctional" ridge regression procedure is developed solely to "normalize" the variance inflated by the multicollinearity problem. The criterion, called the variance normalization criterion, avoids three of the four limitations. This study is not to be interpreted as implying that ridge regression is a solution to the multicollinearity problem; it is just a procedure of last resort for reducing the ill-effects of the problem. (Author/CTM)