02.03.2013 Views

Downloadable - About University

Downloadable - About University

Downloadable - About University

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

292 Methods for eliciting probabilities<br />

Defective<br />

weld<br />

Figure 10.5 – A fault tree<br />

Pipeline<br />

fracture<br />

or<br />

0.001<br />

Safety<br />

valve<br />

failure<br />

0.001 + 0.002 − (0.001 × 0.002)<br />

= 0.002998<br />

Excess<br />

pressure<br />

and<br />

0.02<br />

0.02 × 0.1 = 0.002<br />

Regulator<br />

failure<br />

The decision maker can now assess probabilities for the lowest-level<br />

events (these probabilities are shown on the tree) and then work up<br />

the tree until he eventually obtains the probability of the pipeline<br />

fracturing. Since the safety valve and regulator failures are considered<br />

to be independent their probabilities can be multiplied to obtain the<br />

probability of excess pressure. This probability can then be added to<br />

the probability of a weld being defective to obtain the probability of the<br />

pipe fracture occurring. (Note that since excess pressure and a defective<br />

weld are not mutually exclusive events the very low probability of them<br />

both occurring has been subtracted from the sum – see Chapter 4.) Of<br />

course, in practice most fault trees will be more extensive than this one.<br />

Indeed, the decision maker in this problem may wish to extend the tree<br />

downwards to identify the possible causes of safety valve or regulator<br />

failure in order to make a better assessment of these probabilities.<br />

Using a log-odds scale<br />

Because people generally have problems in distinguishing between<br />

probabilities like 0.001 and 0.0001 some analysts prefer to use what is<br />

known as a log-odds scale to elicit the probabilities of rare events. You<br />

will recall that odds represent the probability that an event will occur<br />

divided by the probability that it will not. By converting probabilities<br />

to odds and then taking the logarithms of the results we arrive at a<br />

scale like that shown in Figure 10.6 (this figure shows the scale only<br />

0.1

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!