Discrete Approximations to Continuous Distributions in Decision Analysis
Robert Hammond, Chevron
December 16, 2015 (SDP Webinar co-sponsored by the Decision Analysis Society - Invited Talk)
Decision analyses often call for discretizing continuous
uncertainties to represent them in decision trees. A few common methods have
been in use for decades in practice, but there are many to choose from. This
talk presents an overview of several discretization methods, their strengths
and weaknesses, and the types of continuous distributions they are best suited
for. All discretization methods have some approximation error, but the quality
of a subjectively assessed distribution itself is often a very real concern as
well. Factors such as cognitive biases can lead an expert to give their P15
value when asked for their P10, for example. The additional errors in
assessments have implications for choosing a discretization method, but do not
eliminate the need for accurate methods.
Click on the file below to hear a sample of the presentation.