The purpose behind building the linear regression model is to describe the real linear relation between any explanatory variable in the model and the dependent one, on the basis of the fact that the dependent variable is a linear function of the explanatory variables and one can use it for prediction and control. This purpose does not cometrue without getting significant, stable and reasonable estimatros for the parameters of the model, specifically regression-coefficients. The researcher found that "RUF" the criterian that he had suggested accurate and sufficient to accomplish that purpose when multicollinearity exists provided that the adequate model that satisfies the standard assumpitions of the error-term can be assigned. It is wrong to ignore the assumptions and depend directly on the least "MSE & PRESS" and greatest " " because it satisfies the model with false fit to data, whereas the regession coefficients are still unstable and unreasonable because of the multicollinearity and the effect of the error-term on the explanatory and predicted power. So the researcher has made procedures for using his criterion "RUF" to get the real best subset linear model.
