Logistic回归模型的影响分析

Influence Analysis for Logistic Regression Model

Logistic回归模型的影响分析是Logistic回归诊断研究中的重要内容。常用的分析方法都是轮换地删除数据点后的逐步判断,而这个判断的过程主要体现在模型的诊断图上。鉴于此,通过构造诊断统计量来有效地开发诊断图成为影响分析的核心内容,并由此能较为准确地探寻出模型的强影响点。本文通过对Logistic回归模型帽子矩阵的分解以及对轮换地删除数据点后的系数估计的相对变化量进行加权,得出Logistic回归模型诊断图使其能比传统的诊断图更准确地判断出模型的强影响点。

Influence analysis for logistic regression model is important content in the process of diagnosis study. The common method of analysis is stepwisely judgment when data points are alternately deleted and the judgment program is mainly reflected diagnosis charts of the model. In view of this, diagnosis charts of influence analysis are effectively developed through the construction of diagnostic statistics that is core content. At the same time, it can accurately find out the strong influential points of the model. This article has got diagnostic charts more accurate than conventional diagnostic charts for determining strong influential points of the model by decomposing the hat matrix for the logistic regression model and alternately deleting the data point estimates of the coefficients of relative changes weighting.