Computational Models of Subjective Probability Calibration
Peter Juslin Henrik Olsson Uppsala University
Regardless of whether you are about to decide what to have for dinner this evening or if you are faced with the crossing of your personal Rubicon, it is a common experience that you do not always get what you want. The outcomes of your decisions are often uncertain. One consequence of the uncertainty that permeates our lives is the conceptual dissociation between outcome and optimality: even the best of decisions may lead to a poor outcome. Whereas the outcome is often all too easy to ascertain, optimality is a more problematic notion. A common solution involves the weighting of the values of outcomes with their probability, with the recommendation to decide on the option with the highest expected value. The present-day formulation of this idea--subjective expected utility (SEU) theory (e.g., Savage, 1954)--imposes constraints on a person's decision making that allow the decisions to be interpreted as optimization in terms of a subjective utility function and a subjective probability measure. The decision problem is therefore dissected into two subcomponents: values or utilities and subjective probabilities. Psychological research has accordingly been concerned with people's ability to make subjective probability assessments. One important aspect of subjective probabilities concern their calibration, or realism--the extent to which events assigned subjective probability .xx tend to occur with relative frequency .xx.
In a recent article we introduced a distinction between two origins of error or uncertainty in judgment and decision making ( Juslin & Olsson, 1997) named after two of the more well-known probabilists in the history