论文部分内容阅读
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”
This paper mainly studies the detection method of high leverage point in logistic regression, and points out the unreasonableness of hi-detection of high leverage point in traditional methods. It is recommended to use h*i=hi/(pi(1-pi)) as the statistic to detect high leverage points. Based on this, and based on this, the {(h*i, epi)} plot is proposed to detect high leverage and outliers. In this paper, the Monte-Carlo method is also used to illustrate that the {(h*i,epi)} map can not only detect high leverage points and abnormal points, but also distinguish “harmful” and “harmless” from high leverage points.