Title: On model selection via penalized likelihood for square contingency tables
Authors: Kouji Tahata - Tokyo University of Science (Japan) [presenting]
Ukyo Matsushima - Tokyo University of Science (Japan)
Abstract: The issues of symmetry and asymmetry arise naturally for the analysis of square contingency table with ordinal categories. Various types of symmetries have been proposed. When we focus on a problem of model selection, we can use some information criteria, and also restrict to hierarchical log-linear model to use the difference of likelihood ratio chi-square statistics. Recently, for the problem of model selection, penalized likelihood approaches are used in many situations, i.e., the least absolute shrinkage and selection operator. We consider methods of model selection by using the penalized likelihood for the class of certain asymmetry models. The class includes symmetry, quasi-symmetry, conditional symmetry, linear diagonals-parameter symmetry, and ordinal quasi-symmetry models. For some examples, the proposed method may be useful to select a symmetry model in the class.