Risk assessment is in many respects acknowledged as a scientific discipline per se: there are many master and PhD programmes worldwide covering this field, and many scientific journals and conferences highlighting the area. However, there are few books addressing the scientific basis of this discipline, which is unfortunate as the area of risk assessment is growing rapidly and there is an enormous drive and enthusiasm to implement risk assessment methods in organisations. Without a proper basis, risk assessment would fail as a scientific method or activity. Consider the following example, a statement from an experienced risk assessment team about uncertainty in quantitative risk assessments (Aven, 2008a):

Based on such a statement, one may question what the scientific basis of the risk assessment is. Everything is uncertain, but is not risk assessment performed to assess the uncertainties? From the cited statement it looks like the risk assessment generates uncertainty. In any event, does this acknowledgment – that a considerable amount of uncertainty exists – affect the analyses and the conclusions?

In this chapter we present the announced requirements of reliability and validity that will be used to verify that a risk assessment is scientific (Section 3.3). But first in Section 3.1 we give some reflections about risk assessment being a scientific method motivated by two interesting editorials of the first issue of the journal Risk Analysis (Cumming, 1981; Weinberg, 1981), in relation to the establishment of the Society of Risk Analysis. We also provide a brief review of the traditional sciences (Section 3.2), such as the natural sciences, social sciences, mathematics and probability theory, to place risk assessment into a broader scientific context. A key issue is to what extent risk assessment should be judged by reference to these traditional science paradigms, or is a science per se.

Reflections on risk assessment being a scientific method

Cumming (1981) concludes that the process of analysing or assessing risks involves science, and consequently is a scientific activity. However, according to Cumming, risk assessment is not a scientific method per se. He writes:

In Chapters 5–7 we have seen how risk assessments and risk management are influenced by the risk perspectives. Now we would like to go one step further, to provide guidance on what should be the preferred approach to risk. The basis for the guidance is the discussions in the previous chapters. Firstly we need to clarify what we mean by risk. A number of definitions and interpretations of the risk concept exist as discussed in Chapter 2. Many of these are probability-based. Below (Section 8.1) we present and discuss a structure for characterising the definitions, which is founded on a clear distinction between (Aven, 2010f)

1. (a) risk as a concept based on events, consequences and uncertainties;

2. (b) risk as a modelled, quantitative concept; and

3. (c) risk descriptions.

The discussion leads to an approach for conceptualising and assessing risk, which is based on risk defined by (a), i.e. is founded on the (A,C,U) risk perspective, and the probability-based definitions of risk can be viewed as model parameters and/or risk descriptions. The approach provides clear guidance on how to think when conceptualising and assessing risk in practice.

Next in this chapter (Section 8.2) we present and discuss a general model-based framework for risk assessments. Starting from an industry guide to quantitative uncertainty analysis and management, clarifications and simplifications are made to ensure consistency with the (A,C,U) risk perspective. Some simple examples are included to motivate and explain the basic ideas of the framework.

In this chapter we study the scientific platform of risk assessments when the objective of these assessments is accurate risk estimation. We first summarise the key concepts, probability and risk, using the set-up introduced in Chapter 2. We then conduct the assessments for the three cases, and from this basis we study the scientific quality of the risk assessments. Focus is on the scientific requirements of reliability and validity defined in Chapter 3.

The risk assessments presented in Sections 5.2–5.4 are rather comprehensive and detailed, although many simplifications have been made. If the reader is mainly concerned about the discussion of the scientific requirements of reliability and validity, a quick reading of these sections would suffice provided the reader is familiar with the statistical nomenclature and methods used. However, to fully appreciate the discussion in Section 5.5 it is necessary to go into the details of the assessments in Sections 5.2–5.4. For example, we cannot evaluate what the main quantities of interest are in the study or see the importance of key assumptions made in the assessments, without looking into the contexts of the analyses and precisely describing how the analyses are carried out.

Next we will study the scientific platform of risk assessments when the objective of these assessments is to describe uncertainties. We follow the same structure as in the previous chapter. We first summarise the framework introduced in Chapter 2 for assessing risk in such a setting, and clarify key concepts like probability and risk. We then conduct the assessments for the three cases, and from this basis we study the scientific quality of the risk assessments. Focus is again on the scientific requirements reliability and validity defined in Chapter 3. We distinguish between an (A,C,Pf)-based risk perspective (referred to as the probability of frequency approach) and an (A,C,U)-based risk perspective; in the former risk is defined through chances (which is the Bayesian term for frequentist probabilities, i.e. fractions of “successes” in the long run; refer to Chapter 2) and in the latter risk is defined through uncertainties.

Scientific basis

We consider an activity and distinguish between the following two ways of looking at risk:

Risk is defined through chances (frequentist probabilities)

Risk = (A,C,Pf), where Pf is a chance (relative frequency-interpreted probability) or a related parameter such as the expected number of occurrences of the event A per unit of time, where expectation is with respect to the chance distribution (relative frequency-interpreted probability distribution).

This appendix reviews some elementary probability theory and statistical analysis. The presentation is based on Aven (2003, 2008a).

Probability theory

The meaning of a probability

The probability of an event A, P(A), can be defined in different ways. It is common to distinguish between three types of probabilities, or more precisely, three types of interpretations:

• classical probabilities

• frequentist probabilities (relative frequency-interpreted probabilities)

• subjective (knowledge-based) probabilities.

The classical interpretation applies only in situations with a finite number of outcomes which are equally likely to occur. According to the classical interpretation the probability of A is equal to the ratio between the number of outcomes resulting in A and the total number of outcomes, i.e.

P(A) = Number of outcomes resulting in A/Total number of outcomes.

As an example consider the tossing of a die. Here P(the die shows one) = 1/6 since there are six possible outcomes which are equally likely to appear and only one that gives the outcome 1.

Following the relative frequency interpretation, probability is defined as the fraction of times the event A occurs if the situation considered were repeated (for real or hypothetically) an infinite number of times. Thus, if an experiment is performed n times and the event A occurs nA times, the P(A) is equal to the limit of nA/n as n goes to infinity, i.e. the probability of the event A is the limit of the fraction of the number of times event A occurs when the number of experiments increases to infinity.

This chapter provides a broad introduction to risk management and risk asssessment, as a basis for the analyses and dicussions in the coming chapters. The presentation highlights general features but also challenges related to the definitions and use of these tools. Key references for the chapters are Bedford and Cooke (2001), Vose (2008) and Aven and Vinnem (2007). The terminology is to a large extent in line with ISO (2009a). See summary of key definitions in Appendix B.

General features of risk management and risk assessments

Risk management is all coordinated activities to direct and control an organisation with regard to risk. Two main purposes of the risk management are to ensure that adequate measures are taken to protect people, the environment and assets from undesirable consequences of the activities being undertaken, and to balance different concerns, for example safety and costs. Risk management covers both measures to avoid the occurrence of hazards/threats and measures to reduce their potential consequences. In industries like nuclear and oil & gas, risk management was traditionally based on a prescriptive regulating regime, in which detailed requirements for the design and operation of the plant were specified (Kumamoto, 2007; Aven and Vinnem, 2007).

In this book we have investigated to what extent risk assessments and in particular quantitative risk assessments, meet the scientific quality requirements of reliability and validity. While reliability is concerned with the consistency of the “measuring instrument” (analysts, experts, methods, procedures), validity is concerned with the success at “measuring” what one set out to “measure” in the analysis. For each of these main criteria a set of sub-criteria are defined.

To be able to perform these investigations we have to study the scientific building blocks of the assessments (related to probability, risk, uncertainty, models, etc.), but also the role of the assessments in the decision-making process. What type of decision support should the assessments provide? Are the objectives (expectations) accurate risk estimates and/or uncertainty characterisations? The scientific quality of the assessments obviously needs to be seen in relation to the objectives. Also the requirements of reliability and validity depend on these objectives. Using these criteria, we evaluate the quality of the assessments for different objectives of the assessments. Three case studies are used to illustrate the analysis. The first of these examples is related to the analysis of accident data, the second relates to the siting of an LNG plant and the third discusses the design of a safety system.

The investigation shows that the reliability and validity criteria are not in general satisfied. Under certain conditions the criteria are met, and to a varying degree depending on the risk perspective and aim of the risk assessment.

In this chapter we introduce the three case studies which will be pursued through the rest of the book to illustrate concepts, principles and methods. The first of these studies is related to the analysis of working accident data whereas the second relates to the risk assessment of a Liquefied Natural Gas (LNG) plant and the third relates to a design of a safety system. Our starting point for the LNG case is the risk assessment process as carried out recently for the plant (Vinnem, 2010). We have, however, made some adjustments, to be in line with the principles studied in this book. For the purpose of this book we consider different approaches for how to carry out the assessments. The presentation only shows excerpts from the assessment; it is simplified, and many numbers have been changed.

Working accidents

In 1999 the Petroleum Safety Authority Norway (PSA) took the initiative to develop a method in order to assess status and trends for the risk levels in the Norwegian offshore petroleum industry (Vinnem et al., 2006). A method was developed which is based on recording occurrences of near misses and relevant incidents, performance of barriers and results from risk assessments, as well as evaluation of safety culture, motivation, communication and perceived risk. In this book we focus on the analysis related to occupational accidents.

As stressed in Chapter 2, risk assessments provide decision support on the choice of measures and arrangements. Such decisions are risk-informed, not risk-based. There is always a need for seeing beyond the results of the risk assessments. There are two main reasons for this:

1. the limitations and boundaries of the risk assessments

2. the need for taking into account other concerns than risk when making such decisions.

The analysis in the two last chapters has addressed the first point, and in this chapter we will discuss the implications of the findings for risk management and communication where point 2 is a key issue. More specifically we focus on the following topics in this chapter:

• the use of predefined risk criteria

• the use of the ALARP principle and cost–benefit type of analyses

• the role of the cautionary and precautionary principles

• risk communication

• the content and purpose of managerial review and judgement.

The use of predefined risk criteria

It is a common approach to risk management to use predefined risk criteria of the form “risk is unacceptable (intolerable) if P > p0”, where P is a risk-related probability index and p0 is a fixed number.

In this chapter we first review and discuss common definitions of risk (Section 2.1–2.6). From this review and discussion we define the concepts and perspectives of probability and risk that will be used as the basis for the risk assessments studied in Chapters 5–7 (Sections 2.7–2.8). These perspectives specify not only how risk is defined, but also how to express risk. We also include an illustrative example (Section 2.9). Basically we will distinguish between two main lines of thinking, one where risk is considered an “objective” property of the activity studied and the aim of the risk assessment is to accurately estimate this risk, and one where uncertainty is a main component of risk and the aim of the risk assessment is to describe the uncertainties. The final Section 2.10 provides a summary of key concepts and perspectives. Some basic references for this chapter are Aven (2009a,b, 2010a,e).

Risk equals expected value

The concept of risk is defined in many ways. In engineering contexts, risk is often linked to the expected loss; see e.g. Lirer et al. (2001), Mandel (2007), Verma and Verter (2007) and Willis (2007). However, such an understanding of risk means that there is no distinction made between situations involving potential large consequences and associated small probabilities, and frequently occurring events with rather small consequences, as long as the sums of the products of the possible outcomes and the associated probabilities are equal. For risk management these two types of situations normally would require different approaches.

This appendix summarises some risk analysis and management terminology used in the book. Unless stated otherwise, the terminology is in line with the international guideline ISO (2009a).

1. • aleatory (stochastic) uncertainty: variation of quantities in a population.

2. This definition is not given in the ISO guideline.

3. • epistemic uncertainty: lack of knowledge about unknown quantities.

4. This definition is not given in the ISO guideline.

5. • event: occurrence or change of a particular set of circumstances.

6. • frequency: number of events per unit of time or another reference.

7. Often frequency is also used for the expected number of events per unit of time.

8. • managerial review and judgement: process of summarising, interpreting and deliberating over the results of risk assessments and other assessments, as well as of other relevant issues (not covered by the assessments), in order to make a decision.

9. This definition is not given in the ISO guideline.

10. • probability: either a knowledge-based (subjective) measure of uncertainty of an event conditional on the background knowledge or a relative frequency (chance). If a knowledge-based probability is equal to 0.10, it means that the uncertainty (degree of belief) is the same as randomly drawing a specific ball out of an urn. A relative frequency-interpreted probability (chance) is the fraction of events A occurring when the situation considered can be repeated over and over again infinitely.

11. This definition is not given in the ISO guideline.

12. […]