When considering pipe laying in the North Sea in the mid-1970s, weather was a major source of uncertainty in terms of its direct impact on the ability to operate equipment. A 3-m laying barge was a barge deemed capable of working in wave conditions up to a nominal 3 m maximum, so weather in terms of wave height was a direct source of uncertainty in relation to pipe-laying performance.
In addition to this direct effect, bad weather greatly increased the chance of a buckle, and the probability of a buckle increased significantly as the amount of work in the 'shoulder season' (early spring or late autumn) increased because of the weather implications. It was important to recognize and model this effect.
Often dependence between different sources can be identified in causal terms. Sometimes the relationship is not clearly definable in these terms and may be best described in terms of statistical dependence (see Chapter 11). For example, preliminary 'macro'-level assessments of the relationship between capital cost items and direct cost rates associated with North Sea projects suggested an average of about 70 to 80% dependence (equivalent to a coefficient of correlation of 0.7 to 0.8 approximately—dependence measures and approaches to dependence are discussed in Chapter 11). This level of dependence was driven by the prevailing level of construction activity and other specific market pressures as well as more general economic conditions. Attempts to describe this dependence in causal terms were not fruitful, in the sense that too many different factors were clearly driving a similar joint movement to make individual identification and modelling of the factors worthwhile. However, it became clear that it was essential to model this statistical dependence to avoid bias, which otherwise made cost risk estimates misleading to a dangerous degree.
Statistical or causal dependencies can also be generated by responses. For example, in a construction project based on the use of two cranes, if one crane should fail and the response is to press on using only one crane, a significant increase in use may be required from the surviving crane, possibly increasing its failure probability. In the limit, such dependencies can cause a cascade or domino effect. Reliability engineers are familiar with the need to understand and model such effects, but many project managers are not.
It may be important to address dependence very carefully. Failure to do so, as in using a basic PERT model and assuming independence that does not exist, can be dangerously misleading as well as a complete waste of time. For example, in the context of a basic PERT network, with activity A followed by activity B, the durations of activities A and B may be positively dependent. If A takes longer than expected, B may also take longer than expected. This can arise because the sources for A and B are common or related. Causal relationships underlying this dependence might include:
1. the same contractor is employed for both activities, who if incompetent (or particularly good) on activity A will be the same for activity B;
2. the same equipment is used for both;
3. the same labour force is used for both;
4. the same optimistic (or pessimistic) estimator provided estimates for both activity duration distributions.
An important form of dependency is knock-on or 'ripple' effects. In the simple example above, when things go wrong in activity A the cost of A goes up and the delays impact on B. The cost of B then increases as a consequence of contingency responses to stay on target. As a consequence of contingency responses, which induce negative time dependence, the positive statistical dependence between the durations of A and B tends to disappear from view.
However, the negative dependence introduced into the activity duration relationships by contingency planning induces strong, positive dependence between associated costs. If A costs more than expected, B tends to cost very much more than expected, because of the need to keep the project on target, quite apart from other market-driven sources of dependence. Put another way, cost and duration modelling of uncertainty that does not explicitly consider contingency planning tends to estimate time uncertainty erroneously (usually optimistically) and fails to structure or explain it and tends grossly to underestimate direct cost uncertainty. Considering the impact of contingency planning will clarify apparent time uncertainty and increase apparent direct cost uncertainty.
Common causes of knock-on effects are design changes and delays, which not only have a direct impact but also cause ripple effects termed 'delay and disruption'. Often direct consequences can be assessed fairly readily in terms such as the number of man-hours required to make a change in design drawings and the man-hours needed to implement the immediate change in the project works. Ripple effects are more difficult to assess and may involve 'snowballing' effects such as altered work sequences, conflicting facility and manpower requirements, skill dilution, undetected work errors, and so on.
Was this article helpful?