响应倾向得分匹配插补法  

Response Propensity Score Matching Imputation

作　　者：杨贵军[1] 孙玲莉[1] 李璐 YANG Gui jun;SUN Ling li;LI Lu(China Center of Economics and Statistics Research,Tianjin University of Finance and Economics,Tianjin 300222,Chin)

机构地区：[1]天津财经大学中国经济统计研究中心,天津300222

出　　处：《统计与信息论坛》2018年第8期3-11,共9页Journal of Statistics and Information

基　　金：国家自然科学基金项目<劣者淘汰两阶段自适应临床试验的设计和分析>(11471239);全国统计科学研究重点项目<大数据背景下住户调查理论与应用研究>(2017LZ25);天津财经大学研究生院科研资助计划项目<大数据背景下住户调查的非抽样误差研究>(2017TCB04)

摘　　要：倾向得分匹配插补法是处理无回答的常用方法,该方法对无回答与回答的样本量差异较为敏感。提出响应倾向得分匹配插补法,建立回答单元响应变量观测值秩变换的响应倾向得分模型,匹配无回答单元和回答单元的响应倾向得分,选择响应倾向得分相近的回答单元作为无回答单元的插补值。模拟研究表明,响应倾向得分匹配插补法的插补效果优于原倾向得分匹配插补法、最近邻插补法和回归插补法。在完全随机无回答机制和随机无回答机制下,随着插补重数增加,采用响应倾向得分匹配插补法的回归系数估计量的偏差绝对值和均方误差呈递增趋势。在实际应用中,插补重数选择不宜过高,建议插补重数为5。针对Sparrows数据集的分析结果显示,采用响应倾向得分匹配插补法处理无回答,能够较好估计回归模型系数。响应倾向得分匹配插补法有效改进原倾向得分匹配插补法,提高了统计分析结果的可靠性。Propensity score matching imputation is used for non response that is more sensitive to the differences of sample size between non response and response units. This paper provides the response propensity score matching imputation. The new method establishes a response propensity score model by transformation of ranks of response units＇ response variable, matching response propensity scores about non response and response units,then choosing response unit of the nearest response propensity score as imputation. Simulation results show that imputation effect on response propensity score matching imputation is superior to the propensity score matching imputation, nearest neighbor imputation and regression imputation. As the imputation multiplicity increasing, the absolute values of the bias and the mean squared errors of regression coefficient estimators show an increasing trend under the completely random mechanism and the random non response mechanism by response propensity score matched imputation. In practice, imputation multiplicity chosen should be small that is suggested as 5. Analysis of the Sparrows dataset shows that could very well estimate regression coefficients via the response propensity score matching imputation for non response. Response propensity score matching imputation effectively modifies the propensity score matching imputation and improves the reliability of the analysis result.

关 键 词：响应倾向得分匹配插补法 插补重数 无回答率 无回答机制 

分 类 号：F224[经济管理—国民经济] O212.1[理学—概率论与数理统计]