在线性回归中,当设计矩阵的列向量间存在复共线性时,回归系数的最小二乘估计的性质显著变坏.为了消除或减弱复共线性对参数估计的影响,以获得更高精度的参数估计,在均方误差矩阵意义下,提出了回归系数的一类新的估计,即t—k类估计,它是对最小二乘估计的改进,是一种新的压缩有偏估计.并且与最小二乘(LS)估计、岭估计和主成分估计进行比较,给出了在均方误差矩阵意义下,t—k类估计优于这些估计的充要条件以及这些条件的检验方法.
The properties of LS estimator will be significantly bad when muhicollinearity exists. In this paper, an improved estimator of regression coefficients which named as t - k class estimator is put forward to diminish effects on estimator. The necessary and sufficient conditions for the t - k class estimator to be superior to OLS estimator, ordinary ridge regression (ORR) estimator and principal components regression (PCR) estimator are achieved and then the tests are suggested to verify whether or not the conditions hold in giveu situations by using statistical method.