条件logistic回归模型配合适度统计量的Monte Carlo法模拟结果  

Simulated Results of Goodness-of-fit Statistics of Conditional Logistic Regression by Monte Carlo Method

作　　者：余松林[1] 罗登发 

机构地区：[1]同济医科大学卫生统计教研室 [2]深圳市医疗保险管理局

出　　处：《中国卫生统计》1995年第5期23-26,共4页Chinese Journal of Health Statistics

摘　　要：本文用 Monte Carlo 模拟法对1:3匹配设计的条件 logistic 回归模型的配合适度统计量剩余离差、方差比及确定系数的检验效果进行了探讨,结果显示,用 n—p为自由度时能较好地识别正确模型与不正确模型;用迭代法计算参数估计值的收敛成功率及配合适度检验统计量的稳定性,都与样本含量有关;样本含量(即匹配组数)以引入模型中的自变量个数的20倍以上为宜。Monte Carlo method was used to simulate goodness-of-fit statistics of conditional logistic regression for 1:3 matched data to assess their test effects in practice,which are residual deviance D,variance ratio F and determination coefficient R2. The simulated results showed that n-p is better than n*m- pwas as degrees of freedom of D and F to identify correct mod- els from noncorrect ones and D would be better than F;con- vergence rate of iteration method in pamater estimation and robustness of these goodness-of-fit statistics were relevant to sample size (e.g.number of matched sets).It would be better that sample size reaches 20 times or over of independent vari- ables introduced in regression models.

关 键 词：条件LOGISTIC 回归模型 配合适度统计量 

分 类 号：R195.1[医药卫生—卫生统计学]