Читать книгу Risk Assessment - Marvin Rausand - Страница 195

All risk analyses require a wide range of data and assumptions that may be more or less uncertain. Whenever possible, the data and the assumptions should reflect reality as closely as possible. This is not always feasible, and some decision‐makers question the validity of the results from risk analysis. A pertinent answer to this type of question is given by Garrick (2008):

[…] there is seldom enough data about future events to be absolutely certain about when and where they will occur and what the consequences might be. But “certainty” is seldom necessary to greatly improve the chances of making good decisions.

Whenever possible, assumptions should be made to err on the side of conservatism. Such assumptions, known as “conservative best estimates,” are to ensure that the assumptions do not result in underestimation of risk and, ultimately, unsafe decisions (NSW 2003). Uncertainty in risk assessment is discussed in Chapter 18.

The quality of the risk analysis should rather be measured in terms of how well it supports the decisions that we want to make based on the analysis. Some general criteria could then be that

 The risk assessment provides support to the decision problem.

 The documentation is such that decision‐makers can understand and use the results in their decision‐making.

 The risk assessment should provide a sound basis for risk management.

 Every reasonable effort has been made to secure the completeness, consistency, and correctness of the analysis.

 The best available and relevant information has been used.

A risk analysis represents a model of a certain phenomenon that we are interested in, namely risk. The objective is to give a description and/or a quantification of what the risk is. Similarly to all models, a risk analysis is based on numerous simplifications. Modeling is always a balance between representing the phenomenon that we are concerned with and the efforts required to develop the model. For that reason, we try to leave out aspects that have limited influence on the results the model produce.

The effect of this simplification is that we always need to be careful if we attempt to apply an existing risk analysis to provide decision support for other decisions than the ones that the analysis originally was developed for. Even if it is the same phenomenon that we describe, there may be other aspects that are relevant to include in the model when the decision is changed. This has among others been observed when risk analyses that originally were developed to support design development of a system later has been used to support operations of the same system. This issue is discussed further in Chapter 16.