A forecaster can be calibrated but almost only assign probabilities in the 40--60 % range. This is not as ueful as one assiging calibrated probabilities in the

A forecaster can be calibrated but almost only assign probabilities in the 40--60 % range. This is not as ueful as one assiging calibrated probabilities in the full range. We try to measure the increased usefulness of the latter with proper scoring rules.