基于EMB多重插补法的线性模型系数估计量的模拟研究

发布时间：2018-06-18 11:04

本文选题：EMB + 多重插补法　；参考：《数量经济技术经济研究》2016年10期

【摘要】：基于EMB多重插补法的线性模型系数估计量,分析其统计性质,并与PMM多重插补法以及DA插补法进行比较。模拟结果显示,随着无回答率增加,系数估计量的偏差绝对值、均方误差呈递增趋势,估计方差的递增趋势相对更显著。在完全随机无回答机制或随机无回答机制下,建议插补重数为15。在依赖被解释变量的非随机无回答机制下,建议插补重数可适当增大。在依赖其他变量的非随机无回答机制下,估计量的均方误差和估计方差的差异大,使用EMB多重插补法要谨慎。
[Abstract]:The statistical properties of the linear model coefficient estimator based on EMB multiple interpolation method are analyzed and compared with PMM multiple interpolation method and DA interpolation method. The simulation results show that with the increase of the unanswered rate, the deviation absolute value of the coefficient estimator and the mean square error are increasing, and the increasing trend of the estimated variance is more significant. In the case of complete random no-answer mechanism or random no-answer mechanism, it is suggested that the number of interpolation weights be 15. It is suggested that the number of interpolation weights can be increased appropriately under the non-random no-answer mechanism which depends on the explained variables. Under the non-random no-answer mechanism which depends on other variables, the difference between the mean square error and the estimated variance of the estimator is large, so we should be careful to use the EMB multi-interpolation method.
【作者单位】：天津财经大学中国经济统计研究中心;
【基金】：国家社科基金重大项目“国家统计数据质量管理研究”(09&ZD040) 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) 天津财经大学研究生科研资助计划项目(2015TCS034,2014TCB02)的资助
【分类号】：F224.0

【相似文献】