OUTLIERS AND INFLUENTIAL DATA POINTS IN REGRESSION-ANALYSIS
OUTLIERS AND INFLUENTIAL DATA POINTS IN REGRESSION-ANALYSIS
复制标题
DOI：
10.1037/0033-2909.95.2.334
复制
发表时间：
1984-01-01
影响因子：
22.4
通讯作者：
STEVENS, JP
中科院分区：
文献类型：
--
作者：
STEVENS, JP
由于回归分析的结果可能对异常值（在 y 上或在预测变量的空间中）敏感，因此能够检测到此类点非常重要。作者讨论并关联了以下 4 个有助于识别异常值的诊断方法：学生化残差、帽子元素、库克距离和马哈拉诺比斯距离。给出了诊断解释的指南。异常值不一定会对回归系数产生影响。（27 参考）（PsycINFO 数据库记录 (c) 2016 APA，保留所有权利）
Because the results of a regression analysis can be sensitive to outliers (either on y or in the space of the predictors), it is important to be able to detect such points. The author discusses and interrelates the following 4 diagnostics that are useful in identifying outliers: studentized residuals, the hat elements, Cook's distance, and Mahalanobis distance. Guidelines are given for interpretation of the diagnostics. Outliers will not necessarily be influential in affecting the regression coefficients.(27 ref)(PsycINFO Database Record (c) 2016 APA, all rights reserved)