DECISION TREES AND INFLUENCE DIAGRAMS

Decision trees are a powerful way to encounter multi-stage problems. Decision trees usually provide a graphical overview about different stages of a decision situation in time. A research and development decision of a company to continue, to alter or to stop a certain production development is one example, where the development of a decision tree could be appropriate.

A tree usually consist of all relevant decisions options and the uncertainties associated with the consequences of each decision. Relevant decisions in the first stage of the R&D example could be: stop, alter, continue development; possible consequences (“states-of-the-world”) in the second stage could be: patent awarded, patent not awarded; possible decisions on the third level could be: licence technology to other producers, sell technology directly to others. The final “states-of-the world” usually represent monetary values: for example, demand high ($10m profit), demand medium ($6m profit), demand low ($1m profit).

In order to determine the optimal policy usually the tree has to be “folded back” by calculating expected monetary value of the different decision options. This procedure shows the option with the highest expected (monetary) return.

Decision Trees can be converted in influence diagrams. Influence diagrams are much more compact representations of the events involved in the decision. Just like decision trees, influence diagrams use decision and event nodes. These nodes are connected with arrows – indicating the influence of one node to another.

To many decision makers, influence diagrams offer a graphically more appealing representation of a problem. A further advantage of influence diagrams is that they reveal cases of independencies among different events. I.e. in the case when there is no arrow, there is no dependency.

Decision trees and influence can boost creative thinking of the decision-makers, disclose previously unconsidered options and can be an excellent way to enhance the understanding of a decision problem.

Conditions to use decision trees/influence diagrams to allocate resources:

• Resource allocation task is a multi-stage problem with sequential decisions (“develop”/“do not develop”; if “develop”: “licence”/“sell directly”, etc.)
• The consequences of the different decision options are uncertain (if “sell directly”: demand “high” or “medium” or “low”)