Statistical modelling and risk assessment

The notion of "risk curve" has been introduced and defined in a previous article concerned with the problem of measuring the severity. This notion is considered here from the standpoint of how to develop its use in practice. We first consider the conditions under which probability theory can be applied to a particular field of study, such as occupational risks, and then describe in detail the characteristics of a risk typology that would lend itself to statistical modelling. The typology is based on a group of general factors likely to lead to undesired events (UEs) in the workplace, which are themselves associated with the functioning of a particular socio-technical system. The choice and definition of this system is the responsibility of the risk manager according to its objectives. The data associated with the recommended typology are, in principle, suitable for modelling in order to define the corresponding risk curve. The latter includes two components: the probability of a UE occurring and the potential severity of this type of UE, both of which are presented here as being random. Two kinds of situations are distinguished for estimating the probability of a UE's occurrence: situations in which the risk is identified in repetitive cycles of operations and situations in which there is a permanent hazard, for which it is necessary to choose a reference period. The estimation of potential severity requires the choice of a numerical index, preferably a composite one combining various elementary indices, and the determination of its values, based on already recorded data on UEs that have occurred in the system under consideration. This approach to constructing a risk curve for UEs.within a specific system is illustrated by applying it to an accident data base in the construction-public works sector. (C) 2002 Elsevier Science Ltd. All rights reserved.