Modelling Under Risk and Uncertainty: An Introduction to Statistical, Phenomenological and Computational Methods

2.3 Modelling Under Uncertainty: Building Separate System and Uncertainty Models

2.3.1 The Need to Go Beyond Direct Statistics

Once the process has clarified the salient goals of the study and defined the appropriate events or variables of interest and the risk measure, it becomes necessary to estimate it effectively, in order to make a decision.

In some cases, it is possible to get direct information on the events or variables of interest, through measurement or expertise, so that the uncertainty in predicting z may be inferred straightforwardly through elementary statistics. Think of standardised light bulbs whose lifetime has been observed in large enough homogeneous samples so that the chance that a new bulb will remain reliable over a given period of time can be directly inferred. Yet, in most practical cases, this is not sufficient. Information such as measurement results or expert knowledge may not be directly available in the z, but rather in different variables characterizing the system state that are more amenable to observation. Or even more importantly, the choice of actions d may change the system so that past records cannot be directly relevant. Some form of modelling then becomes necessary.

Building a model in the context of risk and uncertainty is the central purpose of this book and will receive considerable attention within subsequent chapters. It should be seen fundamentally as inferring the best reflection of the extent and lack of knowledge one has about the ...