讲座题目： Subject-wise Empirical Likelihood Inference for  Longitudinal Data

讲 座 人： 佛罗里达大西洋大学数理统计系正教授 钱莲芬

讲座时间：2017年6月9日（周五）下午15:00  -16：30

讲座地点：浙江工商大学综合楼601会议室

主讲人简介：

       钱莲芬，现任职佛罗里达大西洋大学数理统计系正教授, 统计专业主任，健康与衰老研究所成员。2012浙江省“千人计划”入选人才，浙江省特聘专家, 温州大学特聘教授和统计研究所所长。1984年本科毕业于浙江大学西溪分校数学系并留校任教。于1991留美, 1996年获美国密歇根州立大学统计系哲学（理学）博士学位。在统计学领域主要研究方向是建构变点理论及变点理论在大数据分析，金融统计, 生物统计, 环境统计, 软件工程，基因分类中的应用。出版专著一部发表学术论文40多篇，包括统计学的顶级期刊Biometrika, 以及其它统计一流期刊等。指导过多位博士和硕士研究生。是Advances  and Applications in Statistics等统计期刊的副主编和评审员, USA 自然科学基金统计专门小组成员，MathSciNet 和 Technometrics评论员。美国棕榈滩中文学校（志愿）校长。

摘要：

       In analyzing longitudinal data,  within-subject correlations are a major factor that affects statistical  efficiency. Working with a partially linear model for longitudinal data, we  consider a subject-wise empirical likelihood based method that takes the  within-subject correlations into consideration to estimate the model parameters  efficiently. A nonparametric version of the Wilks theorem for the limiting  distribution of the empirical likelihood ratio, which relies on a kernel  regression smoothing method to properly center data, is derived. On the other  hand, counter-intuitively, adjusting for within-subject correlations is  unnecessary when estimating the nonparametric baseline function. Robust  estimator of semiparametric covariance matrix is also proposed. A simulation  study and an application are reported to investigate the finite sample  properties of the proposed method and compare it with the block empirical  likelihood method and the normal approximation. These numerical results  demonstrate the usefulness of the proposed method.