ABSTRACT
Calculated probabilities of failure commonly have significant uncertainty associated with the possible estimation errors relative to the ‘true’ probabilities. Uncertainty in probability estimates is sometimes expressed in terms of probability distributions of the estimated probability values, or in terms of confidence limits on the estimated probabilities. In order to provide a quantitative measure of uncertainty that reflects not only the expected value but also the variability of the probability of failure, an entropy-based measure of uncertainty has been proposed. The paper describes the use of this entropy-based measure to assess the effect of variability of the probability of failure. It is shown that variability of the probability of failure reduces the overall uncertainty. Accordingly it is concluded that risks may be conservatively assessed with regard to the expected probability, ignoring known variability of the probability of failure.
