网络访问固定样本调查的统计推断研究  

Research on Statistical Inference of Web Access Panel Surveys

作　　者：刘展[1] 金勇进[2] 

机构地区：[1]中国人民大学统计学院,北京100872 [2]中国人民大学应用统计科学研究中心,北京100872

出　　处：《统计与信息论坛》2017年第2期3-10,共8页Journal of Statistics and Information

基　　金：国家社会科学基金项目<大数据背景下非概率抽样的统计推断问题研究>(15BTJ014);中国人民大学2015年度拔尖创新人才培育资助计划成果

摘　　要：如何解决网络访问固定样本调查的统计推断问题,是大数据背景下网络调查面临的严重挑战。针对此问题,提出将网络访问固定样本的调查样本与概率样本结合,利用倾向得分逆加权和加权组调整构造伪权数来估计目标总体,进一步采用基于有放回概率抽样的Vwr方法、基于广义回归估计的Vgreg方法与Jackknife方法来估计方差,并比较不同方法估计的效果。研究表明:无论概率样本的样本量较大还是较小,本研究所提出的总体均值估计方法效果较好,并且在方差估计中Jackknife方法的估计效果最好。How to solve the statistical inference problem of web access panel surveys is a serious challenge facing web survey under the background of big data.In order to solve this problem,it is proposed that a survey sample of the web access panel and a probability sample are firstly combined as a sample;then the propensity score inverse weighting and weighting classes adjustment are used to construct pseudo weights for estimating the population;furthermore,the Vwr method based on probability sampling with replacement,the Vgreg method based on generalized regression estimator and the jackknife method are used to estimate the variance;lastly,the effects of different methods for estimating are compared.The research results show that the proposed method for estimating the population mean is better,regardless of the sample size of the probability sample,and the effect of the jackknife method for estimating the variance is best.

关 键 词：网络访问固定样本 倾向得分 逆加权 加权组调整 

分 类 号：C811[社会学—统计学] O212.2[理学—概率论与数理统计]