Marginal Models by Data Information

• Activity Number: 183
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: International Chinese Statistical Association
• Author(s): Michelle Liou*+ and Juin-Wei Liou and Philip E. Cheng
• Companies: Academia Sinica and Academia Sinica and International Chinese Statistical Association
• Keywords: Mantel-Haenszel test ; Marginal models ; Linear Information Models ; Differential item functioning

Abstract

From the geometry of invariant Pythagorean laws (Stat. Sinica, 2008), a two-step likelihood ratio test (JASA, 2010) is used as a substitute for the commonly used Mantel-Haenszel test (J. Nat. Cancer Res., 1959). As a generalization, linear information models can be used to organize the information structures in a large scale categorical data analysis. A scheme for selecting associated variables for a parsimonious model is formulated and applied to an empirical data analysis. For diagnosis of dementia, individual cognitive abilities screening instrument (CASI) scales have been commonly used for identifying patients from normal subjects. Let normal subjects form a reference group and patients be the focal group, differential items functioning (DIF) of CASI subscales were assessed in a multi-way contingency table. Interactions and partial associations were tested using the information model approach. It effectively examines DIF between the CASI subscales and background variables.