问题：An Information Approach to Logit Models
报告人：Philip E. Cheng and Michelle Liou，，Institute of Statistical Science, Academia Sinica
时间：2013年6月6日（周四）15:30-16:30
所在：pg电子模拟器2号楼217室
Abstract
The Pythagorean law of mutual information identity, inherited from the multinomial likelihood of categorical data, provides two-step orthogonal likelihood ratio tests for hypotheses on interactions of odds ratios in multivariate tables (Cheng, et al., 2008, 2010). The information identity generates models of association among categorical variables in contrast to the conventional loglinear model sand logistic regression models. In this study, we review the Pythagorean law and apply real data examples to illustrate a few drawbacks in the classical approach to modeling association between categorical variables without referring to information geometry. This essentially outlines a research agenda bringing new insight into categorical data analysis.
Key Words: Information identity, Mantel-Haenszelstatistic, Logistic regression, Loglinear models, Pythagorean law.
