モデル選択とその周辺

Abstract

There are a number of statistical methodologies, each of which has its own philosophy and mathematical background. A way to understand them is to regard them as tools for defining "possible samples from possible worlds" from a set of real samples from the real world. That is, different methods correspond to different ways to generate imaginary data from real data. In this note, we review the issues on model selection from this point of view. We also give a few remarks on philosophi-mathematical subjects around model selection. In the appendix, we discuss psychopathology of schizophrenia from the viewpoint of statistical sciences.