As Johannes mentioned a while ago: It is a priori unclear what the "correct" distribution of overprediction, underprediction and dispersion is. So I think it is technically not correct to talk about a bias towards over / underprediction. Adding here as a comment because it might be relevant throughout. We could look at the bias value from scoringutils.
As Johannes mentioned a while ago: It is a priori unclear what the "correct" distribution of overprediction, underprediction and dispersion is. So I think it is technically not correct to talk about a bias towards over / underprediction. Adding here as a comment because it might be relevant throughout. We could look at the bias value from scoringutils.