概率-非概率混合样本的统计推断问题研究

Statistical Inference for Probability and Non-probability Mixed Sample

为了克服大型概率抽样调查组织实施周期长、主题固定、单位成本高等方面的不足,充分发挥概率样本对目标总体的代表性优势和非概率网络调查快速、便捷、低成本的优势,将来自于总体的一个样本量较小的概率样本,与样本量较大的非概率网络样本融合在一起对目标总体特征进行统计推断在实践中具有一定的应用与发展.本文在同时存在概率、非概率样本的背景下,探讨基于概率-非概率混合样本的统计推断问题,提出基于概率样本的HT估计量和非概率网络样本的逆倾向得分加权估计量的综合估计量构建方法.模拟结果显示,本文提出的综合估计量一方能够降低非概率网络样本的偏差,另一方面能够显著降低总体特征估计的MSE,且随着非概率网络样本样本量和概率样本样本量比值的增加,基于倾向得分加权估计量构建的综合估计量中利用非概率网络样本信息的比重也随之增加.

To tackle the challenges traditional probability sampling are facing,such as organization time inefficiency,fixed survey topics and surging cost etc.,and to make full use of advantages of the representativeness of probability sample,and convenience,low cost,fast of the non-probability web sampling,that implementing a small sample size probability sample with a large sample size non-probability web sample to estimate population totals or means has been explored academically and has been applied in survey practices.This article discusses the statistical inference method for probability and non-probability mixed sample under the setting of probability and non-probability mixed sampling,and proposes a composite estimator which is constructed by HT estimator based on probability sample and IPW(inverse propensity score weighted)estimator based on non-probability web sample.Simulation studies show that the composite estimator we proposed can reduce bias of non-probability sample as well as reduce MSE compared to HT estimator or IPW estimator,furthermore,as the ratio of non-probability sample size and probability sample size increase,the weight for IPW estimator tends to increase.