Tuesday, March 20, 2012

Decision Trees and their Nodes


In the paper, Probabilistic Approaches: Scenario Analysis, Decision Trees and Simulations, the author looks at decision trees as an assessment of risk in a sequence. Therefore, the subject in question must pass a series of tests, failure at any point leads to a complete loss of value. The example given is pharmaceutical drugs.


The decision tree is broken down into distinct categories. Root nodes, decision nodes, event nodes and end nodes. A root node is at the beginning and where the decision maker has a decision choice or an uncertain outcome. An event node represents the possible outcomes on a given decision. You have to figure out the possible outcomes and their likelihood for this node. The decision nodes represent choices that can be made by a decision maker. The end nodes represent the final outcomes of the decision tree.


  • By linking actions and choices – decision trees give decision makers a framework and make them think about the consequences.
  • Value of information – having to think through this process gives the decision maker insight on how valuable this information is.
  • Decision trees act as a form of risk management. If the decision doesn’t pass each test – it may be too risky to undertake.
  • They are easy to construct and give definitive answers.


  • There is no wiggle room or room to maneuver.
  • Multiple risks are hard to assess at the same time. This is a linear process at each stage.
  • Event nodes require estimates of outcomes. This is subjective in a lot of cases.
  • The use of the decision tree depends entirely on the decision makers willingness to stick to it strictly.


Decision tree analysis is particularly useful when there are discrete outcomes. However, as a process becomes more complicated and the number of outcomes increases, it is harder to use this tool effectively. Its ease of use and ability to make decision makers think about the choices and consequences makes this a tool that can be applied to many different situations.


Probabilistic Approaches: Scenario Analysis, Decision Trees and Simulations. Stern. NYU. Retrieved from; http://people.stern.nyu.edu/adamodar/pdfiles/papers/probabilistic.pd


  1. It seems like the author thinks that decision trees have a limited utility. I'd be interested in how he applied the pharmaceutical example to an iterated sequence. How did he address its usefulness as the situation becomes more complex?

  2. In your conclusion you said that its harder to use decision trees as the number of outcomes increases. This is probably true of most techniques. It'd be interesting to see if decision trees are more useful in this situation if it was paired with another technique.

  3. To add to the previous thoughts, I'd be interested to know if utilizing a computer program allows for more complex uses of decision trees. I would think with a robust program, decision tree analysis could handle more complicated problems and allow for weighting of options.

  4. Can you explain the author's reasoning of why it is hard to use this with multiple risks? The study I looked at used decision trees because they were less linear than traditional models and were able to incorporate a large number of factors.