线性模型基于M估计的统计诊断与影响分析  

M Estimators and Influence Diagnostics in Linear Models

作　　者：姜荣[1] 钱伟民[1] 

机构地区：[1]同济大学应用数学系,上海

出　　处：《统计学与应用》2012年第2期31-36,共6页Statistical and Application

摘　　要：为了克服实际观测数据与既定模型之间可能存在的较大偏离,目前有两种常用的处理方法:稳健统计与统计诊断。M方法是最重要的稳健统计之一,也是线性回归分析中是最受重视和研究成果最多的方法之一。所以本文结合M估计方法分析数据的统计诊断,从而得到异常点或强影响点。本文给出了参数估计偏离的表达式及几个诊断统计量,最后通过两个实例验证了本文所提方法的可行性。In order to overcome the large deviation between the actual observed data and established models. There are two common methods: the robust estimation and statistical diagnostics. M estimator is a robust es- timation, it is the method which got the most attention and research results in linear regression. Therefore, in- fluence diagnostics and M estimator are used to judge the impact of outliers or strong influence points in this paper. Moreover, the expression of parameter estimation deviate and diagnostic statistics are given. Finally, the proposed methods are applied to two data sets.

关 键 词：数据删除模型 均值漂移模型 局部影响 M估计 COOK距离 

分 类 号：O21[理学—概率论与数理统计]