失混合效应模型的分位回归及变量选择

纵向数据常常用正态混合效应模型进行分析.然而，违背正态性的假定往往会导致无效的推断.与传统的均值回归相比较，分位回归可以给出响应变量条件分布的完整刻画，对于非正态误差分布也可以给稳健的估计结果．本文主要考虑右删失响应下纵向混合效应模型的分位回归估计和变量选择问题．首先，逆删失概率加权方法被用来得到模型的参数估计．其次，结合逆删失概率加权和LASSO惩罚变量选择方法考虑了模型的变量选择问题.蒙特卡洛模拟显示所提方法要比直接删除删失数据的估计方法更具优势．最后，分析了一组艾滋病数据集来展示所提方法的实际应用效果．

Quantile Regression for Censored Mixed Effects Models and Variable Selection

Longitudinal data are usually analyzed using mixed effects models under the assumption of normal distributions. A departure from normality may result in invalid inference. Compared with the traditional mean regression, quantile regression can characterize a complete scan of the conditional distribution of the response variable and provide more robust inferences for nonnormal error distributions. In this paper, we focus on the quantile estimation and variable selection of censored mixed effects models. Firstly, the inverse censoring probability weighted (ICPW) method is utilized to obtain parameters estimation. Furthermore, the LASSO penalties are incorporated into the ICPW method to implement variable selection. Monte Carlo simulations demonstrate that the proposed method performs superior to the "naive" method which ignores censored data. Finally, an AIDS data set is analyzed to illustrate the proposed method.