农业经济调查缺失数据的多重插补及应用  

Multiple Imputation and Application of Missing Data in Agricultural Economic Survey

作　　者：潘传快 韩京芳[1] 熊巍 祁春节[2] Pan Chuankuai;Han Jingfang;Xiong Wei;Qi Chunjie(School of Economics, Wuhan Textile University, Wuhan 430200, China;College of Economics and Management, Huazhong Agriculture University, Wuhan 430070, China)

机构地区：[1]武汉纺织大学经济学院,武汉430200 [2]华中农业大学经济管理学院,武汉430070

出　　处：《统计与决策》2018年第11期12-17,共6页Statistics & Decision

基　　金：国家社会科学基金资助项目(16BJY136);国家现代农业(柑橘)产业技术体系(MATS)专项经费资助项目(CARS-27-08B);湖北省教育厅人文社会科学基金资助项目(18Y077);华中农业大学研究生课程建设项目(2015KJ15)

摘　　要：农业经济调查缺失数据是一个很常见而又很容易被忽略的问题。在一般缺失模式下,文章利用多元正态模型下的联合分布法对其进行多重插补,拥有很好的估计检验效果。模拟分析显示,根据该方法多重插补后的汇总估计量跟完整数据的估计量非常接近,只是由于数据缺失造成的误差增加使检验显著性下降。跟成列删除后数据的估计检验结果相比,其估计准确性和检验显著性都更高。Missing data in agricultural economic survey is a normal and ignorable issue. In a general missingness mode, this paper utilizes joint distribution method under multivariate normal model to conduct multiple imputation, and obtains a good result of estimate and test. Simulation analysis shows that, according to this method, the aggregate estimate after multiple interpolation is very close to the estimate of the complete data; but the increase of the error caused by data missing decreases the significance of inspection; the accuracy and significance of the estimation are higher than the estimated test results of the column-deleted data.

关 键 词：农业经济调查 缺失数据 多重插补 模拟分析 

分 类 号：C81[社会学—统计学]