重复测量计数资料的随机效应ZINB模型  

Random Effect Zero-inflated Negative Binomial Models of Repeated Measured Count Data

作　　者：赵丽华[1,2] 刘桂芬[1] 田娇妮[1] 

机构地区：[1]山西医科大学卫生统计教研室,030001 [2]太原理工大学理学院数学系,030024

出　　处：《中国卫生统计》2011年第6期665-667,共3页Chinese Journal of Health Statistics

基　　金：山西省自然基金项目(200911005-2);太原理工大学校青年基金项目

摘　　要：目的阐明随机效应ZINB模型原理及其在零过多和过度离散并存时在重复测量计数资料中的应用。方法将重复测量资料中的个体当作随机效应引进ZINB模型,建立随机效应ZINB模型。结果模拟研究和实例分析表明,由于随机效应ZINB模型既考虑了个体在不同时点测量值间的相关性,又解决了零过多以及数据的过度离散问题,分析结果优于固定效应ZIP模型、随机效应ZIP模型以及固定效应ZINB模型。结论对存在零过多,又具有过度离散的重复测量计数资料,更宜采用随机效应ZINB模型分析。Objective To clarify the principle of the random effect zero inflated negative binomial model.If zero excess and over dispersion co-exist,explaining the application of the random effect ZINB model in repeated measured count data.Methods Introduced the individual of repeated measurement into zero inflated negative binomial as a random effect and constructed random effect zero inflated negative binomial model.Results The simulation study and example analysis implied that the random effect ZINB model not only considered individuals＇ correlation of different times,but also solved the problem with zero excess and over dispersion,and the output was superior to the random effect ZIP model,the fixed effect ZIP model and the fixed effect ZINB model.Conclusion It should use random effect ZINB model to analysis repeated measured count data with excess zero and over dispersion.

关 键 词：随机效应 ZINB模型 计数资料 零过多 过度离散 

分 类 号：O212.1[理学—概率论与数理统计]