Prevalence Estimation on CAP Law & Crime Doc-type shift Claude Opus 4.6 (test)
-0.3Prevalence Estimation Bias (pp)Rogan-Gladen
Evaluation Results
| Method | Links | |
|---|---|---|
| Rogan-GladenInput=Binary labels2026.04 | -0.3 | |
| Isotonic regressionInput=Probability scores2026.04 | 0 | |
| MCGrad (binary)Input=Binary labels, Initialization=Base-rate2026.04 | 0 | |
| Importance-weighted estimationInput=Target-specific density ratio2026.04 | 0.1 | |
| MCGrad (scores)Input=Probability scores2026.04 | 0.1 | |
| Classify & CountInput=Fraction of Yes labels2026.04 | 1.6 | |
| Saerens-Latinne-DecaesteckerInput=Probability scores2026.04 | 4.9 |