主题：加拿大Waterloo大学宋学坤教授讲学

题目: TitleVector Generalized Linear Models: A Gaussian Copula Approach

报告人: 宋学坤教授

(加拿大Waterloo大学统计与精算科学系)

时间: 2006年4月7日(周五)上午10点

地点: 理科楼207

内容简介: A multivariate extension to the classic generalized linearmodels using the approach of the Gaussian copula. This extensionfurnishes a unified framework of analyzing multi-dimensional non normaldata in the context of regression analysis(The proposed models canaccommodate discrete, continuous and mixed vector outcomes arising inmany practical studies, including longitudinal, clustered or spatial data.With the availability of joint probability distributions, we develop asimultaneous maximum likelihood inference in the class of models, wherea Gauss-Newton type algorithm is implemented to obtain numerical resultsfor the parameter estimation. I will illustrate vector logistic (or probit)models for multivariate binary outcomes, vector log-linear models formultivariate count outcomes, and a vector GLM for normal and binary mixedoutcomes. Comparisons to the GEEs (Generalized Estimating Equations) approachwill be discussed in detail through both simulation and data analysis examples.

欢迎大家参加（统计方向的研究生必须参加）。

数理学院

2006.4.5