Cox模型影响点识别诊断统计量的模拟研究  

Simulation Study of Diagnostics to Identify Influential Cases in Cox's Model

作　　者：余红梅[1] 师成虎[1] 何大卫[1] 

机构地区：[1]山西医科大学卫生统计学教研室,030001

出　　处：《中国卫生统计》2006年第4期297-300,共4页Chinese Journal of Health Statistics

基　　金：全国统计科学研究项目;山西省归国人员基金项目

摘　　要：目的探讨Cox比例风险模型影响点的有效识别方法。方法通过Monte-Carlo模拟研究,比较了六种诊断统计量及相应诊断图对影响点的诊断效能。结果模拟研究表明,加权score残差、似然距离和最大影响曲率及其诊断图可从模型拟合不同角度有效识别影响点。结论影响分析应是Cox模型分析的一个重要组成部分,加权score残差、似然距离和最大影响曲率是有效识别影响点的诊断统计量。Objective To explore effective methods to identify influential cases in Cox＇ s proportional hazards regression model. Methods By Monte - Carlo simulation, the power of six diagnostics to identify influential cases was presented. Results The s：mulation showed that weighted score residuals, likelihood displacement, maximum influence curvature and their plots can detect influential cases in Cox＇s model from different points of view. Conclusion Influence analysis should be an important part of the Cox＇ s model. Weighted score residuals, likelihood displacement and maximum influence curvature are effective diagnostics to identify influential cases.

关 键 词：COX模型 回归诊断 影响点 MONTE-CARLO模拟 

分 类 号：R195[医药卫生—卫生统计学]