Risk Assessment and Decision Analysis with Bayesian Networks

31

2

The Need for Causal, Explanatory

Models in Risk Assessment

2.1 Introduction

The aim of this chapter is to show that we can address some of the core

limitations of the traditional statistical approaches exposed in Chapter 1

by introducing causal explanations into the modeling process. The causal

models are examples of Bayesian Networks (BNs).

Although we will not formally dene BNs until Chapter 5 (because

such a denition requires an understanding of Bayesian probability that

we explain in Chapter 4), our intention here is to provide a avor of their

power and exibility in handling a range of risk-assessment problems.

In Section 2.2 we use the automobile crash example from Chapter 1

to explain the need for and structure of a causal BN. In Section 2.3 we

explain why popular methods of risk assessment (such as risk registers

and heat maps) are insufcient to properly handle risk assessment. We

describe the causal approach to risk assessment in Section 2.4, showing

how it overcomes the limitations of the popular methods.

2.2 Are You More Likely to Die in an

Automobile Crash When the Weather

Is Good Compared to Bad?

We saw in Chapter 1, Section 1.4 some data on fatal automobile accidents.

The fewest fatal crashes occur when the weather is at its worst and the high-

ways are at their most dangerous. Using the data alone and applying the

standard statistical regression techniques to that data we ended up with the

simple regression model shown in Figure2.1.

But there is a grave danger of confusing prediction with risk assess-

ment. For risk assessment and management the regression model is use-

less, because it provides no explanatory power at all. In fact, from a risk

perspective this model would provide irrational, and potentially danger-

ous, information. It would suggest that if you want to minimize your

chances of dying in an automobile crash you should do your driving

when the highways are at their most dangerous, in winter.

Visit www.bayesianrisk.com for your free Bayesian network software and models in

this chapter

We discuss a simplied view of

risk assessment and do not cover

decision and utility theory except

in passing and to make the point

that such theory is not enough

without coherent models of the

problem situation. Most other

books try to present decision the-

ory and risk all at once and in a

very mathematical way; this can

be rather overwhelming.