关键词: 线性分位回归, 倾向得分, 逆概率多重加权, 随机缺失机制, 模型平均

Abstract: In the practical research, there is always an issue of missing data. This paper proposes a new effective estimation, inverse probability multiple weighted (IPMW) estimator, which is under MAR to deal with the problem of missing data in the linear quantile regression (QR) from the perspective of model inference. This method is based on the traditional inverse probability weighted (IPW) estimator, combined with the propensity score matching and the idea of model average. This method applies to situations where the response variable is independent and identically distributed (IID) or independent and identically distributed (INID) and also applicable to most missing scenarios. Based on the theoretical and simulation study, it is found that IPMW estimator is more robust than the traditional IPW estimator. Finally, by applying the IPMW to missing survey data, this paper analyzes the influence factors of consumption in middle income group and the different features of consumption among middle income group, and finds the SD of IPMW estimator is smaller than IPW estimator, and the estimation results are more robust.

Key words: Linear Quantile Regression, Propensity Score, IPMW, Missing at Random, Model Average