Fig. 9
From: Cumulative deviation of a subpopulation from the full population
Eskimo Dog or Husky, with scores (\(S_1\), \(S_2\), ..., \(S_m\)) being the negative log-likelihoods; \(n =\) 1300; Kuiper’s statistic is \(0.08082 / \sigma = 6.363\), Kolmogorov’s and Smirnov’s is \(0.08082 / \sigma = 6.363\). None of the reliability diagrams is able to smooth away the irrelevant variations while simultaneously capturing the severe deviation at the lowest negative log-likelihoods. The scalar summary statistics very successfully detect the statistically highly significant deviation of the subpopulation from the full population