Estrategia para la selección de modelos gráficos en tablas de contingencia no estructuradas

Abstract

Loglinear models are frequently utilized in the analysis of association patterns between variables in a multidimensional contingency table. Discrete graphical models are fashionable. They are a subset of the hierarchical loglinear models and its conditional independence structures are represented by a graph. Because of that they can be more easily interpreted. In this work four models selection methods are described and compared: stepwise selection (backward, forward and in two steps) and the EH procedure.A strategy for model selection is proposed. Keywords: graphical models, contingency tables, model selection methods