Comparison And Application Of The Correction Method Of Multi-collinearity

In order to study methods to remedy multicolinearity,this paper systematically combs methods of correcting multicollinearity: stepwise regression, principal component regression and factor repression, using data about foreign trade and international competitiveness of Guangdong Province,in 2014,and Chinese fiscal revenue data in 1980-2014.In conclusion: stepwise regression can retain variables of the more significant impact, principal component regression and factor repression can be summarized variables systematically.screening independent variables establishment of the principal component regression model is better than contain all independent variables establishment of the principal component regression model.Recommendation: you can use stepwise regression if variables’ multicollinearity is not serious; using principal component regression or factor repression to systematically summarize the complicated relationship between the variables; In principal component regression analysis, using the principle that structure of principal components ’ correlation coefficient matrix achieve simple structure and principal components were significantly associated with variables to select the number of principal components is better than selection bases on cumulative variance proportion what is greater than 80% and characteristic root what is greater than 1.In factor regression analysis, factors were selected in a similar way; when multicollinearity is very serious and variables are so many, we should use stepwise principal component regression、regression factor.