不放回样本追加下基于永久随机数抽样技术的多层次推断方法及应用研究  

Research on Multi-level Inference Method and Application Based on Permanent Random Numbers Sampling Techniques under Sample Complementing without Replacement

作　　者：谭馨 陈光慧[2] Tan Xin;Chen Guanghui

机构地区：[1]浙江财经大学数据科学学院 [2]暨南大学经济学院

出　　处：《统计研究》2024年第8期139-149,共11页Statistical Research

摘　　要：抽样调查如何满足多层次推断的需要是我国政府统计调查在实践应用过程中面临的主要难题之一。本文根据我国现阶段抽样调查工作的实际情况,提出一套完整的适用于多层次推断的抽样调查方法。首先,使用永久随机数对现有抽样框进行更新维护,打破现有方法假设抽样框不变的局限性;其次,采用基于列表–序贯算法的条件Poisson抽样实现多层次推断,消除了固定样本量抽样技术仅适用于一次性抽样调查的限制,同时解决了随机样本量抽样技术样本量不固定的问题;此外,分别使用Sen-Yates-Grundy方法和Deville方法进行方差估计,以提高估计精度;最后,通过数值模拟和应用研究验证所提出方法的有效性。该方法在不显著增加调查经费的情况下,可广泛适用于不同数据类型的目标总体,在企业、住户等经济领域调查中具有可推广性的同时,也可为大数据背景下各级政府在实际解决多层次推断问题方面提供有价值的参考。How to meet the needs of multi-level inference in sampling surveys is one of the main problems faced by China's government statistics in practice.Based on the actual situation of sampling survey work in China at present,a complete set of sampling survey methods applicable to multi-level inference is proposed.Firstly,the permanent random numbers are used to update and maintain the existing sampling frame,which breaks the limitation of existing methods with the assumption of the unchanged sampling frame.Secondly,conditional Poisson sampling based on the list-sequential algorithm is used to realize multi-level inference,which not only eliminates the restriction that the fixed sample size sampling techniques are only applicable to one-time sampling surveys but also solves the problem that the sample size of random sample size sampling techniques is not fixed.In addition,the Sen-Yates-Grundy method and the Devile method are used to estimate the variance respectively to improve the estimation accuracy.Finally,the validity of the proposed method is verified by numerical simulation and application research.This method can be widely applied to target populations of different data types without significantly increasing survey costs.It can be generalized in the surveys of enterprises,households,and other economic fields,and it can also provide a valuable reference for governments at all levels to solve multi-level inference problems in the context of big data.

关 键 词：样本追加 永久随机数 列表–序贯算法 条件Poisson抽样 方差估计 

分 类 号：C811[社会学—统计学]