混合样本的小域模型推断研究

Research on Small Domain Model Inference for Mixed Samples

随着概率抽样成本的日益增加,简单高效且成本低廉的非概率抽样开始被广泛使用。在实际应用中,根据研究的需要和两种抽样方法各自的优势,有时会将两种抽样方法结合使用,从而得到概率样本和非概率样本的混合样本。考虑到小域估计结果可靠、成本低廉等特点,文章提出了对混合样本建立小域模型推断总体的方法:将概率样本和非概率样本分别视为从两个小域中抽取的样本,且假定两个小域的并正好为目标总体,对概率样本和非概率样本建立小域模型,分别得到基于小域模型的概率样本和非概率样本的总体均值估计,进一步为两个总体均值估计构造权数,从而获得最终的总体均值估计。模拟与实证结果表明:对混合样本建立小域模型进行总体推断的效果要比仅使用概率样本进行总体推断的效果更好,所得估计的偏差绝对值、方差和均方误差更小。

With the increasing cost of probability sampling,the simple,efficient and low-cost non-probability sampling is widely used.In practical application,according to the needs of research and their respective advantages,the two sampling methods are sometimes combined to obtain a mixture of probability and non-probability samples.Considering the reliability and low cost of small domain estimation,this paper proposes a method to build a small domain model to infer the population from mixed samples.Taking the probability sample and the non-probability sample as samples extracted from two small domains,and assuming that the union of the two small domains is exactly the target population,the paper establishes a small-domain model for probabilistic and non-probabilistic samples,respectively obtaining the population mean estimates of probabilistic samples and non-probabilistic samples based on the small-domain model.And then,the paper constructs the weights of the two population mean estimators to obtain the final population mean estimators.The simulation and empirical results show that the effect of population inference based on small domain model of mixed sample is better than that of population inference based on probability sample only,with lower absolute bias,and smaller variance and mean square error.