Bayesian Analysis of Agent Opinion
Jul 31 1989
A general framework for agent opinion analysis is developed in the context of forecasting a single uncertain event. A decision maker receives probability forecasts from one or more agents (individuals, experts, or models). New models are developed that provide flexibility in describing the decision maker's views about the agents, and lead to novel rules for updating his/her beliefs in the light of observed forecasts from such sources. Components of the model describe concepts relating to the relevance of the agents' information and experience, the accord between the agents and decision maker in terms of common ro conflicting information, and calibration of probability assessments. In simple event forecasting, these issues are developed for a single agent, and extended to multi-agent problems.
Keywords:Agent opinion, calibration, combining probability forecasts