A1278
Title: Capturing asymmetric structures and separability in multivariate contingency tables based on f-divergence
Authors: Hisaya Okahara - Tokyo University of Science (Japan) [presenting]
Kouji Tahata - Tokyo University of Science (Japan)
Abstract: A novel extension of asymmetry models is introduced for multivariate contingency tables with ordinal categories based on f-divergence. The proposed model generalizes existing asymmetry models while maintaining a focus on symmetric structures, offering a flexible approach to capturing complex dependence patterns. Theoretical properties of the model are established, extending known results for existing models. These include the decomposition of the symmetry model and the asymptotic properties of likelihood ratio statistics, reinforcing the natural extension of the proposed framework. By incorporating various divergence measures, this methodology provides a unified and adaptable approach for analyzing multivariate categorical data. The model's performance evaluation using real-world data, along with a comparison with conventional approaches in terms of goodness-of-fit, will be presented.
