Comment on ‘The meaning of probability in probabilistic safety analysis’

A recent article in Reliability Engineering and System Safety argues that there is ‘fundamental confusion over how to interpret the numbers which emerge from a Probabilistic Safety Analysis [PSA]’, [Watson, S. R., The meaning of probability in probabilistic safety analysis. Reliab. Engng & System Safety, 45 (1994) 261–269.] As a standard for comparison, the author employs the ‘realist’ interpretation that a PSA output probability should be a ‘physical property’ of the installation being analyzed, ‘objectively measurable’ without controversy. The author finds all the other theories and philosophies discussed wanting by this standard. Ultimately, he argues that the outputs of a PSA should be considered to be no more than constructs of the computational procedure chosen—just an ‘argument’ or a ‘framework for the debate about safety’ rather than a ‘representation of truth’. He even suggests that ‘competing’ PSAʹs be done—each trying to ‘argue’ for a different message. The commentors suggest that the position the author arrives at is an overreaction to the subjectivity which is part of any complex PSA, and that that overreaction could in fact easily lead to the belief that PSAʹs are meaningless. They suggest a broader interpretation, one based strictly on relative frequency—a concept which the commentors believe the author abandoned too quickly. Their interpretation does not require any ‘tests’ to determine whether a statement of likelihood is qualified to be a ‘true’ probability and it applies equally well in pure analytical models. It allows anyoneʹs proper numerical statement of the likelihood of an event to be considered a probability. It recognizes that the quality of PSAʹs and their results will vary. But, unlike the author, the commentors contend that a PSA should always be a search for truth—not a vehicle for adversarial pleadings.