多重填补法与Ad Hoc法对模拟纵向数据集缺失值处理的比较  

COMPARISON BETWEEN MULTIPLE IMPUTATION AND AD HOC METHODS IN SIMULATED LONGITUDINAL DATA WITH MISSING VALUES.

作　　者：茅群霞[1] 李晓松[1] 

机构地区：[1]四川大学华西公共卫生学院卫生统计学教研室,成都610041

出　　处：《现代预防医学》2005年第4期310-312,共3页Modern Preventive Medicine

摘　　要：目的:采用多重填补法(multipleimputation ,MI)和Adhoc法分别对模拟的纵向数据集中的缺失值进行处理,比较两种方法的优劣并探讨其适用性。方法:运用SAS9 0 ,采用数据模拟技术,分别模拟纵向完整数据集和具有各种缺失率的随机缺失数据集,分别用MI和Adhoc法对各缺失数据集进行处理,对结果进行比较和分析。结果:数据缺失率≤10 %时,Adhoc方法有一定优势;数据缺失率在2 0 %～4 0 %时,经MI处理后的分析结果更接近“真实”;数据缺失率≥5 0 %时,两种方法均无效。结论:对不同缺失率的数据集。Objective:To explore the applicability of multiple imputation (MI) and Ad hoc methods in simulated data with missing values.Methods:The simulated datasets with vary missing rates were treated by MI and Ad hoc methods and the results were compared with that of complete dataset by running SAS procedures.Results:Ad hoc methods works as missing rate is less than 10%, and MI is a better choice as missing rates are between 20% to 40%, whereas missing rate is more than 50%, neither is appropriate.Conclusion:In different missing rate data, MI and Ad hoc methods has its own advantages and disadvantages.

关 键 词：多重填补法 数据集 缺失值 Ad 缺失率 纵向数据 SAS9 模拟技术 分析结果 适用性 MI 

分 类 号：R195[医药卫生—卫生统计学] R322[医药卫生—卫生事业管理]