广义线性模型_七_陈希孺.pdf

DOI :10.13860/j.cnki .sltj.2003.05.012 58 　　　　　　　　　22 　5 　2003 9 　 　　:1002-1566 (2003)05-0058-07 () 陈希孺 (, 100039) 　:本讲座是广义线性模型这 个题目的一个比较系统的介绍。 主要分3 部分:建模、统计分 与模型选择和诊断。 写作时依据的主要参考资料是L.Fahrmeir 等人的《Multivariate Statistical Modeling Based on Generali ed Linear Models》。 :广义线性模型;建模;统计分 ;模型选择和诊断 :O2 12 　　　　　　　　　　　　:A Generalized Linear Models CHEN Xi-ru (Graduate School of Chinese Academia of Science, Beijing 100039, China) Abstract:This set of articles gives an introduction to generali ed linear models.T hey can be divided into three parts: Model building, Statistical inference and Model diagno stics.The presentation is mainly based on L.Fahrmeir et al. 《Multivariate Statistical Modeling Based on Generali ed linear Models》. Key words:g enerali ed linear models ;model building;statistical inference;model diagnostics 1.7 ()　:※※, §1, 2 () 。 logit , ( r=1 , π …π 0) 1 r-1 πr exp (θr +z (x )′β) = , r =1, …, q (2.88) 1-π1 - …πr-1 1+exp (θr +z (x )′β) q=2, θ, θ, z (x ), β, π, …, π 。 (2.88), 1 2 1 q : θ =2.379 　　θ =1.516 　　β =0.094 　　β =0.882 1 2 1 2 β =-0.356 　　β =-0.601　　β =0.092 3 4 5 ()logit , , , 。, P ( r x ), r=1, 2, (2.88): P (1 () x )=ex p(θ+z ′(x )β)/(1+exp (θ+z ′(x )β)) 1 1 exp (θ2+z ′(x )β) P (2 () x )=(1+exp (θ1+z ′(x )β))(1+exp (θ2+z ′(x )β)) θθ, βθ θ, 。P (3 ( 1 2 1 1 x ) 1。 , , 。: , 。 ()