logistic 回归中高杠杆点的检测

The Detecting of the High Leverage Cases in logistic Regression

本文主要研究了ｌｏｇｉｓｔｉｃ回归中高杠杆点的检测方法，指出了传统用ｈｉｉ检测高杠杆点的不合理性，建议用ｈ＊ｉ＝ｈｉ／（ｐｉ（１－ｐｉ））作为检测高杠杆点的统计量，并在此基础上提出｛（ｈ＊ｉ，ｅｐｉ）｝图用于检测高杠杆点和异常点。本文还通过Ｍｏｎｔｅ－Ｃａｒｌｏ方法说明了｛（ｈ＊ｉ，ｅｐｉ）｝图不但能检测出高杠杆点和异常点，还可区分高杠杆点的“有害”与“无害”

In this paper,we studied mainly the detecting methods of the high leverage cases in logistic regression.We pointed out the unreasonablity of h ii for detecting high leverage cases.In this paper,a graphical method-{(h ii *,e p i)} plot was proposed for detecting high leverage cases and outliers.And the practicality of the {(h ii *,e p i)} plot was verified by Monte-Carlo method.