Information criteria for statistical model selection

Abstract

In regard to the selection criteria of various statistical models based on the Kullback-Leibler information content, their characteristics are compared and the importance of distinguishing the use in terms of the objective is pointed out. In regard to references dependent on the asymptotic approximation, it is pointed out by examples that the effectiveness of the approximation is important. In particular, as regards the model for discrete values, the necessity of estimation of the Kullback-Leibler information content is presented based on a method not dependent on an asymptotic approximation such as the bootstrap method. © 2001 Scripta Technica, Electron. Comm. Jpn. Pt.