删失分位数回归模型基于扩展兴趣信息准则的平均估计

本文结合分位数回归技术,基于删失回归模型,把Claeskens和Hjort的传统兴趣信息准侧(focused information criterion,FIC)扩展到兴趣向量的情形,提出扩展的兴趣信息准则(extended focused information criterion,E-FIC),有效解决了同时针对多个兴趣参数的平均估计问题,并且对删失响应变量的不同水平分位数进行建模,以全面反映响应变量分布特征,有效克服异常值和厚尾模型误差的影响.基于扩展的兴趣信息准则给出参数的平均估计方法,证明估计的渐近性质.通过Monte Carlo随机模拟试验比较所提估计方法和最小二乘方法在有限样本量下的表现,用所提方法对原发性胆汁性肝硬化数据集进行数据分析.

Extended focused information criterion for censored quantile regression model and averaging estimation

In this paper, we extend Claeskens and Hjort＇s focused information criterion （FIC） for the censored linear model based on quantile regression, and propose an extended FIC for focused vector. We model different quantiles of the censored response to depict the comprehensive character of the distribution of the response and deal with outliers and heavy tailed model error. We further develop a model selection and averaging estimation procedure based on the extended focused information criterion and show the asymptotical property of the proposed estimators. We conduct a sinmlation study to detect the behavior of the proposed method compared with least square based method with finite sample sizes and analyze the primary biliary cirrhosis data set to illustrate the proposed method in applications.