Due approcci alla costruzione del modello statistico: confronti ed osservazioni

Abstract

Lauritzen, in his book "Statistical models as extremal families", criticizes the usual procedure of giving in advance the statistical model and then applying the various inference principles, and states that the statistical analysis and the statistical model ought to be jointly considered. The technique which he suggests for the construction of a statistical model, may be traced, although in a simpler context , also in a theorem by Diaconis and Freedman. From a predictive view point the same problem has been also dealt with by Cifarelli and Regazzini. We present a new, and practically more useful version, of the deacons and freedman's Theorem, and analyse in relationship with the predictive viewpoint. Finally we describe a way to combine the two above-mentioned approaches into a unified framework.