Authors
I.V. ORLOVA
Abstract
This article is devoted to the analysis of methods aimed at solving the problem of multicollinearity of data arising due to the high information redundancy of metric data. Decision making in modern conditions is based on the analysis of huge volumes of data, often having only a small amount of information content, which means that information redundancy is high. In the case of the linear regression model, multicollinearity can be interpreted as a type of redundancy. The possibility of using the PETRES Red test, a red indicator for measuring the proportion of useful content in assessing linear regression parameters, is considered. to quantify the degree of redundancy. The article provides a comparative analysis of the PETRES Red test with the most used multicollinearity detection procedures among regressors, which are implemented in the mctest R-package: Farrara-Glouber test, VIF (dispersion inflation factor) and Bellsley method implemented in Gretl. A comparative analysis was performed on data from 186 enterprises related to the type of crude oil production for 2016. The problem of identifying and eliminating multicollinearity was solved. It is concluded that it is advisable to use an integrated approach to testing multicollinearity using the mctest R-package, which calculates general and individual diagnostic tests of multicollinearity.
Keywords
multicollinearity, information redundancy, red indicator, mctest R-packet, Gretl.
