| |
Abstract:
We study the combination of experts' probability assessments
in a logarithmic opinion pool. The error of the pool is shown to be
smaller or equal than the average error of individual experts. A
decomposition in terms of individual errors of and distances
between experts leads to a robust and relatively straightforward
procedure for selecting weighting factors. Our general description
includes regression models, models for computing variances, and
classification models.
|
