Decision Analysis Working Paper Abstract
Archive
WP030005
|
Title: Debiasing Expert Overconfidence: A Bayesian Calibration
Model
Authors: Robert
T. Clemen and Kenneth C. Lichtendahl
Jr,
Duke University
Date: June 2002
Status: working paper
In a decision and risk analysis, experts may provide subjective probability
distributions that encode their beliefs about future uncertain events.
For continuous variables, experts often provide these judgments in the
form of quantiles of the distribution (e.g., 5th, 50th, and 95th percentiles).
Psychologists have shown, though, that such subjective distributions tend
to be too narrow, representing overconfidence on the part of the expert.
We propose an approach for modeling and debiasing expert overconfidence.
Based on past performance data (previous assessments and realizations for
a number of uncertain variables), and using Bayesian methods to update
prior distributions on the model parameters, we show how our model can
be used to debias expert probabilities. We develop and demonstrate both
a single-expert model and a multiple-expert hierarchical model.
Click here
to visit Bob Clemen's web site to download this paper.
Back to the Decision Analysis Working Paper
Index
