國立東華大學應用數學系

專    題    演    講

主講人：程爾觀  教授

         Institute of Statistical Science, Academia Sinica

講  題： A New Approach to Generalized Linear Models

時  間：105 年 11月 18日 (星期五)  15:20-17:00

地  點：理工一館 A324會議室

摘 要

Information identities based on product multinomial distributions offer elementary analysis of the relationship among discrete multivariate variables. The identities are built upon decomposed terms of mutual information and form the foundation of selecting indispensable predictors and valid generalized linear models, in particular, logistic models. To analyze multivariate contingency data tables, a selection scheme of the information identity is devised to yield proper logistic models in a natural way. The new idea and method are studied using empirical data examples, and compared with the AIC and BIC on model selection showing preferable advantages of basic inference aspects.

Keywords: AIC, BIC, Information identity, Logistic model.

Joint work with Michelle Liou and Jiun-Wei Liou at Academia Sinica

※※※   歡      迎      參      加   ※※※se1051118 附檔: