Model averaging and weight choice in linear mixed-effects models

• 成果类型: Article
• 署名作者: Zhang, Xinyu; Zou, Guohua; Liang, Hua
• 署名单位: Chinese Academy of Sciences; Academy of Mathematics & System Sciences, CAS; George Washington University
• 刊物名称: BIOMETRIKA
• ISSN/ISSBN: 0006-3444
• DOI: 10.1093/biomet/ast052
• 发表日期: 2014
• 页码: 205218
• 关键词: conditional akaike information; variable selection; regression models; criterion; combination; infection; inference; DYNAMICS

摘要

This article studies model averaging for linear mixed-effects models. We establish an unbiased estimator of the squared risk for the model averaging, and use the estimator as a criterion for choosing weights. The resulting model average estimator is proved to be asymptotically optimal under some regularity conditions. Simulation experiments show it is superior or comparable to estimators based on the final models selected by the commonly-used methods and some existing averaging procedures. The proposed procedure is applied to data from an AIDS clinic trial.