多元线性模型协差阵基于LSE无偏估计数据点影响分析

本文给出多元线性模型协差阵基于LSE无偏估计的影响函数,得出其对数据点异常值不具有稳健性的结论,为对其进行数据点影响分析奠定了理论基础;提出一种影响度量,对于正态模型,优于已有的似然距离.计算了一个实例.

CASE INFLUENCE ANALYSIS OF LSE-BASED UNBIASED ESTIMATED COVARIANCE MATRIX IN MULTILINEAR MODEL

The necessity to make case - influence analysis is proved for the LSE-based unbiased estimate of covariance matrix in multivari-ate linear model following the derivation of its influence function, -which is never seen elsewhere. A statistic, not only intuitive but also easy to calculate , is suggested to measure the influence exerted by a certain data set-Far a normally distributed model , the new criterion with its distributionknown, is more accurate in discerning powerful influential sets than themeasure likelihood distance, for whose distribution no succinct form isavailable- An illustrative example is given.