关于数据缺失机制的检验方法探讨

在调查研究中,缺失数据是一个非常普遍的问题,各种处理缺失数据的方法都是建立在数据缺失机制的某种假定上.在总结他人研究成果的基础上,分别给出了MCAR、MAR和NMAR机制的检验识别方法,MCAR机制的检验从分布特征入手,通过比较均值和方差是否一致来判定;MAR机制的检验利用Logit模型刻画缺失指示变量R的分布,通过估计参数的显著性来判定,NMAR机制则通过对数据的缺失模式和原因进行分析来识别.

Discussion on Testing the Mechanism of Missing Data

In survey research,missing data is a very common problem,and the methods to deal with missing data are almost based on some assumption of data missing mechanism.In this paper,based on the previous research achievement,the authors propose the test methods for MCAR,MAR and NMAR.The test for MCAR starts with the distribution characteristics, and it is determined by whether the comparison of mean and variance is consistent.The test for MAR uses Logit model to describe the distribution of missing indicator variable R,and this test is determined by the significance of estimated parameters.Moreover,the test for NAMR is identified by analyzing the missingness pattern and reason.