摘要
本文对于线性回归诊断提出了几种新的模型和方法。我们首次研究了方差加权和均值漂移的混合模型,得到了相应的诊断统计量。本文还引入了罚函数方法,并以此为工具,讨论了若干有偏估计的影响度量,最后,本文提出了基于重心的诊断统计量,对于识别异常点有较好的效果。
In this paper, some new models and methods are presented for linear regression diagnostics. We study the diagnostic model with case-weights and mean-shift simultaneously, and some new diagnostic statistics are derived. The penalty function method is introduced, and we discuss the influential measures for some biased estimates. The diagnostic statistics based on gravity center are also proposed. It seems efficient for identification of outliers. Two numerical examples are given to illustrate our results.
出处
《高校应用数学学报（A辑）》
CSCD
北大核心
1993年第3期279-289,共11页
Applied Mathematics A Journal of Chinese Universities(Ser.A)
基金
国家自然科学基金
关键词
回归诊断
线性回归
参数估计
Regression Diagnostics, Cook Distance, Case-weighted Model, Mean-shift Model, Outliers.
