主 题：Optimal Weight Choice for Frequentist Model AverageEstimators
主讲人：Prof. Hua Liang
主持人：林华珍教授
时 间：2015年7月9日下午4：00-5：00
地 点：通博楼B212学术会议室
主办单位：统计研究中心 统计学院 科研处
主讲人简介：
梁华，The GeorgeWashington University生物统计学教授，现任Journal of the American StatisticalAssociation、Journal of Nonparametric Statistics期刊副主编。美国数理统计研究所研究员、美国统计学会会员、国际统计学会的当选委员。美国卫生研究院授予其艾滋病临床研究和流行病学研究组（ACE）、艾滋病研究和发展中心（NIAID）的顾问。主持7项科研课题，参与7项课题研究。已出版《Partially Linear Models》和《RelatedTopics in Partially Linear Models》两本专著。在国内外知名期刊发表学术论文140多篇，其中The Annals of Statistics（AOS）8篇； Journal ofthe American Statistical Association（JASA）7篇；Biometrika（BKA）5篇；Econometric Theory 4篇；Biometrics 3篇；Biostatistics2篇；Journal of the Royal Statistical Society（JRSS）1篇，文章累计引用数千次。
内容提要：
Therehas been increasing interest recently in model averaging within the frequentistparadigm. The main benefit of model averaging over model selection is that itincorporates rather than ignores the uncertainty inherent in the modelselection process. One of the most important, yet challenging, aspects of modelaveraging is how to optimally combine estimates from different models. In thiswork, we suggest a procedure of weight choice for frequentist model averageestimators that exhibits optimality properties with respect to the estimator'smean squared error (MSE). As a basis for demonstrating our idea, we consideraveraging over a sequence of linear regression models. Building on this base, we develop a modelweighting mechanism that involves minimizing the trace of an unbiased estimatorof the model average estimator's MSE. We further obtain results that reflectthe finite sample as well as asymptotic optimality of the proposed mechanism. AMonte Carlo study based on simulated and real data evaluates and compares thefinite sample properties of this mechanism with those of existing methods. The extension of the proposed weightselection scheme to general likelihood models is also considered.
