Influence Diagrams

Overview
Developing the ID model
Weighting the factors
Rating the factors using a scenario
Reviewing and updating the model

Over the past ten years, Human Reliability has used a methodology called Influence Diagrams (IDs) to model the ways in which a range of factors affect the likelihood of occurrence of a specific outcome. This methodology involves bringing together a group of knowledgeable individuals who have a stake in the outcome, such as marine officers, to participate in an interactive discussion called a Consensus Group. The Consensus Group first develops a representation (called an Influence Diagram) of the factors that have a direct and indirect influence on the likelihood of the outcome (in this case, a high or low workload scenario) being achieved.

The following sections describe the Influence Diagram process in a marine context.

Developing the ID model

The first part of the ID procedure is to develop a seed model. This model is developed separately, and without the input of the consensus group, and is based upon factors that are known to cause feelings of stress or pressure in the Officer of the Watch (OOW). These factors were mainly obtained from the review of shipping accident databases, and the MAIB’s Bridge Watchkeeping Safety Study, but some insights from the theoretical literature review were also included. The seed model used for these interactive sessions included eight primary factors, and twenty-one subfactors.

Click here to view the seed model for the CMWL project.

There are three main reasons for developing a seed model:

• The seed model helps to establish the goal of the Influence Diagram by demonstrating how these factors contribute to overloaded. For example, as the number of distractions increases, the OOW might start to feel increasingly under pressure or overloaded. In contrast, as the degree of visibility improves, this might help to decrease feelings of loading.
  
• The Influence Diagram process can be somewhat difficult to grasp at first, as it requires a different, less traditional way of thinking about the problem area. The seed model helps to demonstrate the conventions of Influence Diagram modelling.
  
• Because IDs require an alternative way of thinking about scenarios, it can be difficult to stimulate thought and discussion when starting with a blank ID. By showing the group a seed model, the group is encouraged to discuss the factors already present, and to expand upon them or delete them as appropriate.

For many participants, the ID session requires a different way of thinking about the problem area, and so it is useful to begin the ID session with a general discussion about the factors that influence workload. The group is then shown the seed model, and the participants are invited to discuss the model and comment upon the factors included.

The group is asked to discuss the seed model to determine whether all of the factors are relevant and whether they are in the correct place. Some factors can be expanded upon (by adding subfactors) to more clearly define how they impact upon workload. Other factors may not need to be decomposed any further, as they are self-explanatory. The group may also decide that some factors should be removed from the model, as they do not have a significant impact upon cognitive workload. At this stage, the group is encouraged to add factors influencing workload from their own experience, to expand upon those included in the seed model.

Weighting the factors

In addition, the group is asked to consider the placement of the factors. Factors that are lower down in the diagram (i.e., subfactors) will have less influence on CMWL than those factors at higher levels. The strength of the factors becomes more diluted towards the bottom of the tree. If the group considers a particular subfactor to be very influential, then that factor may be moved further up the tree, to become a primary factor, which will increase its strength and influence in the overall model.

The next step is to add weights to each of the factors in the model. Not all factors will influence workload to the same degree, and the differences between the strength of the individual factors can be reflected by adding weights to the factors at each level of the tree.

When all of the weights have been assessed, the IDEAS software can use the combined Influence Rankings for all the factors in the model to show which factors will have the highest impact upon the overall outcome if they are changed from their current ratings or states. This is a useful exercise, as displaying the Influence Rankings allows the group to see how the different factors will impact upon the overall outcome at the top of the diagram.

Rating the factors using a scenario

When the group is satisfied with the weights and position of the factors in the model, it should then be tested for accuracy by entering ratings (usually on a scale of 1 to 100) for a particular scenario. The group is asked to think of a real scenario that they have experienced in which they felt highly pressured and on the brink of becoming so overloaded that, had anything else happened, they probably would have made an error or an accident might have occurred.

On the basis of all the ratings entered, combined with ratings, the Seafarer’s Loading Index (SLI) is calculated. This index reflects the overall loading of the individual for the scenarios represented by the ratings and the weights. The SLI is on a scale from 0 to 1, with 0 meaning that, in an overload situation the individual is not suffering from overloading (i.e., best case), and 1 meaning that the individual is experiencing the worst case possible in which he/she is completely overloaded.

An SLI of 1 indicates that all of the factors contributing to loading are at their worst case possible, whereas an SLI of 0 indicated that the factors are at their best case. It should be noted that “best case” may actually represent a moderate level of loading, rather than a minimum level. This is because a moderate level of loading would represent the level at which the probability of human error arising from cognitive loading was also at its minimum.

Reviewing and updating the model

When the model has been completed (i.e., all factors, weights and ratings have been added) and the SLI has been recalculated, the group is then asked whether the resultant SLI reflects their expectations and experience of the overload situation. If the group thinks that the model does not realistically reflect their thoughts then the model can be altered accordingly. For example, weights can be changed to increase or decrease the impact of certain factors, or factors can be moved further up or down the tree to increase or dilute their influence on overload.

It is useful to try out different “what if” scenarios to get a better idea of how accurately the model reflects the group’s beliefs. For example, if the group believes that in a highly stressful situation the addition of an extra qualified person on board the bridge (perhaps as a lookout, or to deal with communications) would make a significant difference to the loading level experienced by the OOW, then this can be tested by increasing the rating of the manning levels factor, and examining the resultant SLI.

By testing the model with different scenarios, the group will eventually come to an agreement of a model structure that best reflects their own beliefs, opinions and experiences about the factors that influence overload. These insights are combined with the results from the literature and from incident investigation data to represent a model that includes all the available knowledge relevant to the domain being addressed.

Back to the top