Impactul factorului de inflamare a erorilor (VIF) asupra valorilor calculate ale testului Student în cazul regresiilor lineare multiple

This paper demonstrates that if we deal with multiple regressions the Student Test statistics are sensibly influenced by VIF (Variance Inflation Factor). The author brings arguments in favour of the idea that the study of (multi) collinearity, has to consider not only critical values of VIF, but also the interactions of VIF with the other modeling factors of coefficients of collinear refraction (Tnk), especially the coefficients of correlation mediated by dependent variable of the explanatory variables (rjk), which emphasize the differentiation of the absolute values of Pearson coefficients of correlation of explanatory variables and the dependent variable. This way, it is possible to reveal the role of different factors in the occurrence of harmful collinearity and the advantages of use of a Transformed Form of Student Test Statistics (TFST). Also, a re-grouping of the modeling factors TFST is proposed, in line with the assumptions related to premises of harmful (multi) collinearity occurrence, presented in Glauber and Farrar (1967), and with the comments of Belsey (1991) related to consequences of (multi) collinearity on Student Test statistics. A numerical example, related to the estimation of the index of the fixed capital formation in Romania, during the period 1991-2009, as been modelated by four explanatory, confirms the assumptions made upon the major impact of VIF on Student Test statistics in case of a multiple linear regression. The respective example shows also that harmful collinearity may occurs even in conditions of relatively low values of VIF, if the differentiation of the coefficients rjk is very high.