模型平均辅助抽样估计方法研究  

Research on Method of Model Averaging Assisted Sampling Estimation

作　　者：陈茜儒 贺建风 Chen Qianru;He Jianfeng(School of Financial Mathematics&Statistics,Guangdong University of Finance,Guangzhou 510630,China;School of Economics and Finance,South China University of Technology,Guangzhou 510006,China)

机构地区：[1]广东金融学院金融数学与统计学院,广州510630 [2]华南理工大学经济与金融学院,广州510006

出　　处：《统计与决策》2023年第9期35-41,共7页Statistics & Decision

基　　金：国家社会科学基金资助项目(22BTJ041);全国统计科学研究重大项目(2020LD02);中央高校基本科研业务费专项资金资助项目(ZDPY202207)。

摘　　要：大数据背景下,数据维度高是辅助变量的重要特征,会在超总体模型建模过程中带来模型的不确定性问题,导致单一模型辅助抽样估计精度降低。为此,文章将模型平均方法引入模型辅助抽样估计的框架中。首先,在线性超总体模型下,结合模型平均方法和广义回归估计思想,提出了模型平均辅助抽样估计量;其次,通过数值模拟验证了模型平均辅助抽样估计优于单一模型辅助抽样估计,尤其是在存在干扰信息时,其估计优势更加明显;最后,用实际数据验证了模型平均辅助抽样估计的优势。In the context of big data,high data dimension is an important feature of auxiliary variables,which will bring mod⁃el uncertainty in the process of super population model modeling,resulting in the reduced accuracy of single model-assisted sam⁃pling estimation.Therefore,this paper introduces the model averaging method into the framework of model-assisted sampling esti⁃mation.Firstly,in the linear hyperpopulation model,the model average auxiliary sampling estimator is proposed by combining the model average method and the idea of generalized regression estimation.Secondly,it is verified by numerical simulation that the model averaging assisted sampling estimation is better than the single model-assisted sampling estimation,and its estimation ad⁃vantage is more obvious especially when there exists interference information.Finally,actual data is employed to verify the advan⁃tage of the model averaging assisted sampling estimation.

关 键 词：抽样估计 模型平均 模型辅助 辅助信息 

分 类 号：O212[理学—概率论与数理统计]