<特集：モデル> 統計学におけるモデル : 情報量基準の観点から

Abstract

Within the framework of statistics, the goodness of statistical models is evaluated by criteria for model selection, such as the Akaike and Bayesian information criteria. Each information criteria is based on likelihoodist’s or Bayesian conception. Here, I analyse the inferences used in the derivation of these criteria, and argue that the goodness, evaluated by the Akaike or Bayesian information criteria reflects frequentist’s conception, which is not explained by likelihoodist or Bayesian.