Logistic回归模型异常值的判定原理探讨

笔者综合运用非线性回归诊断分析，建立了Ｌｏｇｉｓｔｉｃ回归模型异常值判定的基本原理和方法。其基本思想是：（１）利用Ｐｅａｒｓｏｎ残差初步确定Ｌｏｇｉｓｔｉｃ回归模型的可疑异常值。（２）利用Ｓｃｏｒｅ检验统计量对可疑异常值进行假设检验。最后，作者将此方法应用于急性淋巴细胞性白血病的预后分析中，并对异常值的处理进行了初步探讨。

The principled research to assess the outliers of Logistic regression model

On the basis of the non-linear regression diagnostic analysis, the authors put forward a diagnostic method to assess the outliers of the Logistic regression model. The basic principle is as follow: (1)To preliminarily assess the doubious outliers with Pearson error; (2) To take hypothesis test to the doubious outliers with Score statistic. Finally it was used in the prognosis for cases of acute lymphatic leukemia and it has been simply discussed that how to deal with the outliers.