Different models can imply different inter-prediction relations. An illustrative example of how different models can imply different relations between predictions. On the top right are the 67% (dashed) and 95% (solid) probability contours of the joint probability density functions of two predictions obtained with model M
1 and M
2, while the other two panels show the distribution of that specific prediction. Note how measuring one of the two predictions would yield no additional discriminatory power while measuring both predictions would.