基于倾向得分多层模型的非概率抽样统计推断

Statistical Inference of Non-probability Sampling Based on Propensity Score Multilevel Model

文章针对具有嵌套结构数据的网络候选者数据库,提出基于倾向得分多层模型的非概率抽样推断方法:根据网络候选者数据库的调查样本和参考样本,构建多层回归模型对倾向得分进行估计,并将倾向得分估计的逆作为网络候选者数据库调查样本的调整权数来估计总体。结果显示,基于倾向得分多层回归模型的总体估计效果较好,比基于倾向得分Logistic模型的总体估计的偏差更小,效率更高。

This paper aims at the web candidate database with nested structure data to propose the inference method of non-probability sampling based on propensity score multilevel model.The multilevel regression model is built to estimate propensity scores according to a survey sample of the web candidate database and a reference sample,and the population is then estimated via using the inverse of propensity scores as adjusted weights of the survey sample of the web candidate database.The results show that the overall estimation effect based on propensity score multilevel model is better,with less deviation but higher efficiency than the overall estimation based on propensity score Logistic model.