Introduction

Abstract

Statistical modeling or model building is an activity aimed at learning rules and restrictions in a set of observed data, proverbially called “laws” or “the go of it,” as stated by Maxwell. The traditional approach, presumably influenced by physics, is to imagine or assume that the data have been generated as a sample from a population, originally of a parametrically defined probability distribution and, later, more generally, a so-called nonparametric distribution. Then the so-imagined unknown distribution is estimated from the data by use of various principles, such as the least-squares and maximum-likelihood principles, or by minimization of some mean loss function, the mean taken with respect to the “true” data generating distribution imagined.