Printed from
Bayesiske metoder til risikohåndtering
Printervenlig udgave
Rapport med beskrivelse af bayesiske metoder til håndtering af risici (US, 2005)

This report reviews the technical and interpretational limitations of using Bayesian methods in risk and uncertainty analyses. It argues that the methods produce sometimes extreme overconfidence and arbitrariness in the computed answers. These deficiencies are illustrated using various computational problems including projecting uncertainty through predictive expressions, analyzing sample data, updating estimates based on new information, and accounting for model uncertainty. There are three main causes of this overconfidence and arbitrariness:

(1) misuse of equiprobability as a model for incertitude,
(2) overuse of averaging to aggregate information, which tends to erase variation rather than propagate it, and
(3) reliance on precise values and particular distributions when available information does not justify such specificity.

Download rapport (pdf)