非概率抽样估计中先验信息的利用——基于贝叶斯模型估计视角

The Use of Prior Information in Non-probability Sampling and Inference--A Research of Bayesian Model-based Inference

非概率样本的估计问题是近年来的研究热点,本文以调查中先验信息的利用作为切入点,在配额抽样下设置贝叶斯形式的超总体模型,使用样本信息与先验信息对总体目标变量进行加权估计,从而解决非概率样本的估计问题。通过对北京市医疗资源调查的实证研究,表明先验信息的准确性和权重的合理分配决定着贝叶斯估计的效果,在合理的模型设置下贝叶斯估计在大量重复抽样下具有更好的稳定性。

The inference of non-probability samples is a popular topic in recent years.To solve this problem,this paper takes the use of prior information as a starting point,sets a Bayesian form super-population model under quota sampling,which leads to a weighted estimation of sample information and prior information.The case study of medical resources in Beijing shows that the accuracy of prior information and the appropriate allocation of weights are key points to the Bayesian model-based estimation for quota sampling.Under an appropriate model setting,Bayesian model-based approach is more stable when repeated numerous sampling is carried out.