基于协方差消除多重共线性的一种方法

A Method of Eliminating Multicollinearity Based on Covariance

多元线性回归模型通常用来研究一个因变量依赖多个解释变量的变化关系,但它有一个前提条件就是解释变量之间不存在相关关系。在实际的应用中,特别是计量经济学中,解释变量之间一般都存在有高度相关关系或近似相关关系,从而使得模型估计不准确。为此,通过协方差计算变换矩阵,提供一种变换矩阵消除随机变量之间相关关系的方法,通过spss25进行实证分析,最后发现通过矩阵变换变换后的数据t检验的显著性值明显降低。

Multi-factor linear regression models are typically used when studying relationship between a dependent variable and multiple explanatory variables,with a prerequisite that there should be no correlation between the explanatory variables.Practically,however,a high correlation or approximate correlation will exist between explanatory variables,especially in econometrics,which affects the model to give out an inaccurate estimation.In this paper,a transformation matrix was derived from a covariance matrix to eliminate linear relationships between random variables.It was proved by spss25 that the P value obtained from the t test obviously decreases after matrix transformation.