线性回归分析中异常点的诊断统计量  

Diagnostic Statistic for Detecting Abnomal Data Points IN Linear Regression Analysis

作　　者：余松林[1] 向惠云[1] 

机构地区：[1]同济医科大学卫生统计教研室

出　　处：《中国卫生统计》1993年第3期19-22,共4页Chinese Journal of Health Statistics

摘　　要：在线性回归分析中,可将观察资料中出现的离群点分为高杠杆点和异常点。t分布化普通残差及t分布化预报残差对检出资料中的异常点都不理想。作者提出用残差平方和减少量及由此导出的偏F检验作为检验异常点的统计量。这一统计量也适用于诊断非线性回归中的资料异常点。文中用实例对这一统计量作了详细说明。In linear regression analysis,outliers in data can be divided es high leverage points and abn- ormal points. Studenti zed ordinary residuals and studentized predict residuals as diagnostic statistic are not perfect for detecting abnormal data poi- nts. the authors propose a diagnostic statistic that can be used as diagnostic statistic for detecting abnormal data points the decreasing amount of sum of squares of residuals and the partial F statistic. The statistic can be extended to detect abnormal data points in nonlinear regression analysis.A detailed illustration with simple is included in the oontcxt.

关 键 词：诊断 异常点 观察 线性回归分析 离群点 中用 F检验 统计量 T分布 残差平方和 

分 类 号：R195.1[医药卫生—卫生统计学] O212.1[医药卫生—卫生事业管理]