Subjective probabilities for earthquake damage and loss

This paper presents a method for developing consensus probability distributions of damage from subjective probability densities provided by individual experts. The method employs Bayesian statistical theory and incorporates weighting of expert judgements in evaluating the consessus distributions. Earthquake ground motion vs. damage relationships are developed in the form of damage probability matrices using Gaussian quadrature methods. The main advantage of this approach over existing distribution discretization techniques is that no error is introduced in the mean value of the random variable and errors in the variance are reduced significantly with even a small number of value-probability pairs. As a consequence, expected loss estimates obtained by the proposed method contain no error due to discretization. The results from this paper are particularly useful for forecasting damage and losses in geographical areas with large numbers of structures and with a variety of engineering characteristics.