Extended RC association models allowing for order restrictions and marginal modeling

• 成果类型: Article
• 署名作者: Bartolucci, F; Forcina, A
• 署名单位: University of Perugia
• 刊物名称: JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
• ISSN/ISSBN: 0162-1459
• DOI: 10.1198/016214502388618988
• 发表日期: 2002
• 页码: 1192-1199
• 关键词: contingency-tables; cross-classifications; canonical correlation; regression-models; bivariate; parameters

摘要

In the context of two-way contingency tables, the RC model may be seen as a way of modeling the set of log-odds ratios for adjacent 2 x 2 subtables through row and column scores. This article presents an extended class of RC models that allows simultaneous modeling of the marginal distributions and the association between two categorical variables; these components are parameterized with logits and log-odds ratios that may be of different types (e.g., global or continuation) to suit the nature of the data and the type of dependence of interest. An algorithm for maximum likelihood estimation under equality and inequality constraints is presented, the asymptotic distribution of the likelihood-ratio statistic to test some interesting hypotheses is derived, and extensions to multiway tables are outlined. Three examples are presented to highlight the features of the new approach.