关键词: 数据融合, 混合样本, 伪权数, 倾向得分, 校准法

Abstract: With the current survey environment and the development of the Internet, online and offline mixed survey methods are more and more widely used. How to merge the two kinds of data, reduce the waste of information, make rational use of data resources, and obtain effective and high-precision estimation results is a severe challenge for survey data inference in the context of big data. This article puts forward two basic ideas of data integration in the case that an online sample is a non-probability sample and an offline sample is a probability sample. The first is to perform “ probabilistic testing” on a non-probability sample, and then combine the two types of data for statistics inference; the second is to use the information provided in a probability sample to “pseudo-randomize” a non-probability sample. For the second idea, this paper takes the calibration estimation based on the pseudo-weight of the propensity score as an example, discusses specific solutions and variable selection problems, and verifies them through simulations.

Key words: Data Integration, Mixed Samples, Pseudo Weight, Propensity Score, Calibration