Premium
Data features 1
Author(s) -
Davies P. L.
Publication year - 1995
Publication title -
statistica neerlandica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.52
H-Index - 39
eISSN - 1467-9574
pISSN - 0039-0402
DOI - 10.1111/j.1467-9574.1995.tb01464.x
Subject(s) - inference , mathematics , variance (accounting) , computer science , parametric statistics , feature (linguistics) , statistical inference , sample (material) , data mining , algorithm , statistics , artificial intelligence , linguistics , philosophy , chemistry , accounting , chromatography , business
This article attempts to provide a formal framework for a data based inference which explicitly and consistently recognizes the approximate nature of probability models. It is based on the idea that a stochastic model is adequate if samples generated under the model are very much like the sample actually obtained. The formalization is based on the concept of data feature. Examples are given of applying the ideas to different areas of statistics including location‐scale models, densities, non‐parametric regression, interlaboratory test, auto‐regressive processes and the analysis of variance. The four cornerstones of the approach are direct comparison, approximation, weak topologies and parsimony. The approach is contrasted to that of much of conventional statistics many of whose concepts are pathologically discontinuous with respect to the topology of data analysis and common sense.