Researching the behavior of variables relative contributions to the total determination in regression equation estimated using the method of distorted coefficients straightening

Abstract

To solve the problem of multicollinearity in regression analysis a distorted coefficients straightening method developed by the author and based on the construction of fully connected linear regression model can be used. In the article, to assess the degree of independent variables influence on the dependent variable in the regression equation obtained by using this method, it is proposed to use the variables relative contributions to the total determination. It is proved that in such an equation in the case of linear functional dependence of the input variables their relative contributions to the total determination are equal. Then, with a strong correlation of the input variables, their contributions are distributed approximately in the same way. It is proved that the problem of estimating a fully connected regression does not depend on the choice of connecting variable. The obtained results have been successfully demonstrated using the example of the Russia's GDP modeling.