删失分位数变系数回归模型的FIC模型平均估计（英文）  

FIC Based Model Averaging for the Censored Quantile Varying Coefficient Regression Model

作　　者：吕晓玲[1,2] 王小宁 孙志猛[3] LV Xiaoling;WANG Xiaoning;SUN Zhimeng(Center for applied statistics,Renmin University of China,Beijing 100872;School of Statistic,Renmin University of China,Beijing 100872;School of Statistic and Mathematics,Central University of Finance and Economics,Beijing 100081)

机构地区：[1]中国人民大学应用统计科学研究中心,北京100872 [2]中国人民大学统计学院,北京100872 [3]中央财经大学统计与数学学院,北京100081

出　　处：《系统科学与数学》2018年第7期746-763,共18页Journal of Systems Science and Mathematical Sciences

基　　金：国家自然科学基金项目(11301561,11871488,61472475);国家社会科学基金重大项目(15ZDC024);中央财经大学青年英才培育支持计划项目(QYP1810);中央财经大学社会调研与科研数据库项目,中央财经大学青年科研创新团队项目,中央财经大学学科建设经费资助课题;全国统计科学研究重点项目(2018LZ28)

摘　　要：考虑了删失分位数变系数回归模型的FIC准则，并基于FIC准则给出了兴趣参数的模型选择和平均估计．为了全面反映响应变量的分布信息，克服异常值和重尾模型误差，文章对响应变量的不同分位数水平进行建模，因此与普通最小二乘方法相比更为稳健．在较为一般的条件下，证明了所提估计的渐近性质，通过模拟实验研究了估计的有限样本性质，用所提方法分析了手机用户的游戏时间数据．We derive the focused information criterion for the censored quantile varying coefficient regression model to run model selection and model averaging for some focused parameters. We model different levels of quantiles of the censored response to depict the comprehensive characteristics of the distribution of the response and deal with outliers and heavy tailed model error, which makes the proposed method robuster than the ordinary least square procedure. Under general conditions, we show the large sample property of the proposed estimator. We conduct simulations to study the finite sample property of the estimator and apply the proposed method to analyze the dataset of the smart phone to model the playing time of the users of a game application in China.

关 键 词：FIC 随机删失 模型不确定性 分位数回归 

分 类 号：O212.1[理学—概率论与数理统计]