设计效应分解在复杂样本设计中的应用研究  

The Research on the Application of Design Effects Decomposition in Complex Sample Design

作　　者：罗薇[1,2] LUO Wei(School of Management Guangdong University of Technology,Guangzhou 510006,China;Institute of Big Data Strategic Research,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学管理学院,广东广州510006 [2]广东工业大学大数据战略研究院,广东广州510006

出　　处：《统计与信息论坛》2018年第11期16-23,共8页Journal of Statistics and Information

基　　金：国家社会科学基金项目<住户调查一体化设计方法及其应用研究>(17BTJ037)

摘　　要：在复杂样本设计中,多种抽样方法的结合使得设计效应的直接应用受到限制。通过对设计效应的影响因素进行分解,给出分层、类集、加权调整产生的要素设计效应模型。研究结果表明:复杂样本设计通过影响总体单位的相关性来影响设计效应,分层设计带来的负相关性将有限减少设计效应,类集设计带来的正相关性将显著增大设计效应,加权调整带来的权数变化一般会增大设计效应,所以控制非比例分配和类集的设计效应是进行有效样本设计的关键。对于子群、分析统计量的设计效应,虽然样本设计引起的总体单位间相关性减弱,仍可以通过样本均值的设计效应进行推断。A complex sample that represents the combined effects of a variety of sampling features limits the direct application of design effects.Through decomposition of the component of design effect,it introduces component design effects models such as stratification,clustering and weighting.It suggests that the design of complex samples induce correlations between element values and then affect design effect.In stratification negative correlation reduces design effect modestly.Clustering that induces positive correlations between element values is significant to design effect.Weighting adjustments that induces variable weights generally increase design effect.Therefore,controlling the consequences of a disproportionate allocation of the sample and ofthe effects of clustering is key to effective sample design.Correlation due to sampling designs is reduced for subclass and for analytical statistics.However,their design effects can be inferred by the design effect of sample means.

关 键 词：分层 类集 加权调整 同质系数 

分 类 号：O212.2[理学—概率论与数理统计]