常用多重插补法的插补重数选择

Selection of Imputation Multiplicity on Multiple Imputation Methods

社会经济调查领域普遍存在无回答现象。目前处理无回答的常用方法是多重插补法。文章重点研究常用的六种多重插补法:PMM多重插补法、DA多重插补法、EMB多重插补法、普通线性回归多重插补法、贝叶斯线性回归多重插补法和自助线性回归多重插补法。首先,比较六种多重插补法理论性质。其次,重点模拟研究三种线性回归多重插补法对回归模型系数估计的影响。最后,对比分析六种多重插补法的应用条件,给出使用不同多重插补法的插补重数建议。

Non-response is common in the social-economic investigation.At present,the common method to deal with non-response is multiple imputations.This paper analyzes six multiple imputations,i.e.PMM multiple imputation,DA multiple imputation,EMB multiple imputation,ordinary linear regression multiple imputation,Bayesian linear regression multiple imputation and Bootstrap linear regression multiple imputation.Firstly,the paper compares the theoretical properties of six kinds of multiple imputations,and secondly,emphatically studies by simulation the influence of three linear regression multiple imputation methods on the estimation of regression model coefficients.Finally,the paper compares and analyzes the application conditions of six kinds of multiple imputations,presenting suggestions on imputation multiplicity of different multiple imputation methods.