高维纵向数据的惩罚广义估计方程分析  

Penalized Generalized Estimating Equations for High-dimensional Longitudinal Data Analysis

作　　者：尹长明[1] 王亚东 YIN Changming;WANG Yadong(School of Mathematics and Information Science,Guangri University,Nanning 530004,China)

机构地区：[1]广西大学数学与信息科学学院,广西南宁530004

出　　处：《应用数学》2021年第3期657-664,共8页Mathematica Applicata

基　　金：国家自然科学基金(11061002);广西自然科学基金(2015GXNSFAA139006)。

摘　　要：广义估计方程(GEE)是分析纵向数据下响应变量是离散的或非负的回归问题常用方法.本文研究了高维GEE的变量选择,在更弱的条件下证明了相关阵(或协方差)假定不一定正确,只要均值函数假定正确,模型选择是相合的,得到了变量选择的Oracle性质.改进了WANG(2011)和WANG,ZHOU,QU(2012)的结果.Generalized estimating equation(GEE)is widely adopted in analyzing longitudinal(clustered)data with discrete or nonnegative response.Variable selection for GEE with High-dimensional covariates is studied in this paper.Under some mild conditions,we proved that the consistency of model selection holds even when the working correlation structure is misspecified provided that the mean function is correctly specified.Our results extend the elegant work of WANG(2011)and WAND,ZHOU and QU(2012).

关 键 词：纵向数据 广义估计方程 高维协变量 变量选择 

分 类 号：O212.1[理学—概率论与数理统计]