复共线性 matlab,消除复共线性影响的一种新的解决办法

【Abstract】 Abstract In this paper, the multivariate linear regression model with multicolinearity in design matrix is considered. On the basis of a new idea that there are random observational errors in both the observation of response variable and of explanatory variables, the model is modified into a multivariate linear errors in variables model. Following this treatment, it can be expected to remove the effect of multicolinearity and to obtain better estimation of regression coefficients. Two simulation examples are given in this paper, they have shown that the estimate values by using the modified model are rather close to the true values of parameters. It can be expected that the technique proposed in this paper is a more effective method for removing the effects of multicolinearities.