抽样调查领域分层模型相关理论及估计方法研究  

Research on Statistical Inference Methods Based on Multilevel Models in Sample Survey

作　　者：于力超 YU Li-chao(College of Science,Minzu University of China,Beijing 100081,China)

机构地区：[1]中央民族大学理学院

出　　处：《数学的实践与认识》2018年第16期185-195,共11页Mathematics in Practice and Theory

基　　金：2017-2018年度全国统计科学研究项目(重点项目)“相依抽样调查数据模型的统计推断及在民族人口和教育领域的应用”(2017LZ01)

摘　　要：大规模抽样调查常采用整群抽样、多阶段抽样等复杂抽样设计,纵向调查也越来越得到广泛应用,得到的调查数据具有分层嵌套结构,传统回归分析方法关于样本变量间独立的假定不再满足,因此对分层结构数据集进行数据分析时,应采用分层模型建模,以综合考虑个体层级变量和群组层级变量对响应变量的影响.本文旨在全面介绍分层模型理论,研究包括固定效应模型、随机效应模型等分层模型的参数估计方法,并比较分层模型与传统模型的异同.为分层结构数据分析提供工具和建议.Complicated sampling design like cluster sampling and multi-stage sampling are usually used in sample surveys, longitudinal survey is also widely used. We get multilevel survey data from these kind of surveys. The traditional regression analysis method is no longer satisfied with the assumption that the independence between samples is not satisfied. Therefore, we need to concern the cluster effect to the value of the variables, the multilevel model （like fixed effect model and random effect model） should be used for data analysis of multilevel data sets. This paper will comprehensively introduce the theory of multilevel modeling, including the parameter estimation methods of multilevel model, and will also compare the similarity and differences of multilevel model and traditional model so as to provide tools and suggestions for multilevel data analysis.

关 键 词：抽样调查 分层模型 固定效应模型 随机效应模型 贝叶斯估计 

分 类 号：O212[理学—概率论与数理统计]