Bayesian analysis: Critical issues related to its scope and boundaries in a risk context

Terje Aven

doi:10.1016/j.ress.2020.107209

Abstract

Bayesian analysis constitutes an important pillar for assessing and managing risk, but it also has some weaknesses and limitations. The main aims of the present paper are to summarize the scope and boundaries of Bayesian analysis in a risk setting, point to critical issues and suggest ways of meeting the problems faced. The paper specifically addresses the Bayesian perspective on probability and risk, probability models, the link between probability and knowledge, and Bayesian decision analysis. A main overall conclusion of the paper is that risk analysis has a broader scope and framing than Bayesian analysis, and that it is important for risk assessment and management to acknowledge this and build approaches and methods that extend beyond the Bayesian paradigm. To adequately assess and handle risk it is necessary to see beyond risk as commonly defined in Bayesian analysis.

Highlights

This paper discusses Bayesian analysis in a risk context
Following text books on the topic and well-established nomenclature, Bayesian analysis can be viewed as a method of statistical inference that allows one to combine prior information with new information, using Bayes’ formula to guide the statistical inference process (e.g. [10,12,26,27,38])
Bayesian analysis is a cornerstone in decision analysis, and it is common to refer to Bayesian decision theory and Bayesian decision analysis

Summary

Introduction

This paper discusses Bayesian analysis in a risk context This context captures concepts, theories, principles, frameworks, approaches, methods and models for understanding, assessing, characterizing, communicating, managing and governing risk, referred to for short as ‘concepts for risk analysis’. It is defined by what can be seen as the scope of the risk field and science [2,21,44,45]. The present discussion aims at reviewing current knowledge on the topic and gaining new insights, by addressing some current fundamental issues in risk analysis and risk science, including the acknowledgment of the need to see beyond probability-based perspectives to adequately conduct risk analysis. The paper clarifies what these issues are really about and how they should be dealt with in real life risk contexts

The Bayesian perspective

Challenges in a risk setting

Recommendations and conclusions