关键词: Akaike信息准则, 模型选择, 广义线性模型, 广义拟似然

Abstract: In this paper we focus on variable selection and propose a new model selection criterion for joint generalized linear models with structured dispersions. Our proposal is based on the extended quasi-likelihood that only need to specify first two moments of the distribution. The new criterion (EAIC) is an extension of Akaike’s information criterion. Its performance is investigated through simulation studies, and the results support the utility of the methodology.

Key words: AIC, Model selection, Generalized linear models, Extended quasi-likelihood