Affordable Access

Access to the full text

Easy computation of the Bayes factor to fully quantify Occam’s razor in least-squares fitting and to guide actions

Authors
  • Dunstan, D. J.1
  • Crowne, J.1
  • Drew, A. J.1
  • 1 Queen Mary University of London, London, E1 4NS, UK , London (United Kingdom)
Type
Published Article
Journal
Scientific Reports
Publisher
Springer Nature
Publication Date
Jan 19, 2022
Volume
12
Issue
1
Identifiers
DOI: 10.1038/s41598-021-04694-7
Source
Springer Nature
Disciplines
  • article
License
Green

Abstract

The Bayes factor is the gold-standard figure of merit for comparing fits of models to data, for hypothesis selection and parameter estimation. However, it is little-used because it has been considered to be subjective, and to be computationally very intensive. A simple computational method has been known for at least 30 years, but has been dismissed as an approximation. We show here that all three criticisms are misplaced. The method should be used to complement and augment all least-squares fitting, because it can give very different, and better outcomes than classical methods. It can discriminate between models with equal numbers of parameters and equally good fits to data. It quantifies the Occam’s Razor injunction against over-fitting, and it demands that physically-meaningful parameters rejected by classical significance testing be included in the fitting, to avoid spurious precision and incorrect values for the other parameters. It strongly discourages the use of physically-meaningless parameters, thereby satisfying the Occam’s Razor injunction to use existing entities for explanation rather than multiplying new ones. More generally, as a relative probability, the Bayes factor combines naturally with other quantitative information to guide action in the absence of certain knowledge.

Report this publication

Statistics

Seen <100 times