Occupational risks and the value and modelling of a measurement of severity

In studies of occupational risks, severity, which is a component of the est imation of every risk, appears as a multifaceted entity assessable accordin g to numerous criteria. A method of measuring the degree of severity of the consequences of potentially dangerous events would be of undeniable value to organisations seeking to improve their understanding of the complexity o f such events. The need to control severity is highlighted by scientificall y acquired improvements in the understanding of occupational risks, by cert ain new regulatory obligations in Europe, and by some requirements in the f inancial management of organisations. We put forward a statistical way of i ntegrating several constituent elements of severity and hence of determinin g a relevant, synthetic, one-dimensional index. This is achieved by means o f principal component analysis (PCA), which is used here to calculate a res ultant severity, as in some physical measurements. We also investigate how severity may be statistically modelled, with the aim of contributing to the quantitative assessment of occupational risks. The choice of parametric mo dels is detailed and illustrated by the search for a suitable model for wor kplace accidents in an organisational setting. The practical value of model ling severity is twofold. First, one is able to study the distribution of t he numerical values of severity over a continuum (a theoretically infinite numerical set) rather than through a limited number of arbitrarily defined categories. Second, with a generally applicable parametric model, one can e stimate the law of probability of a measurement of severity in a particular situation, notably recent or new. Lastly, the statistical concept of risk curve is defined and discussed. The goal is to incorporate the severity com ponent into the risk assessment in the f'orm of a probability law, thus cir cumventing the difficulties associated with an analysis of scenarios. (C) 1 999 Elsevier Science Ltd. All rights reserved.