Prevalence Estimation on CAP Law & Crime (Belgium TV) Claude Opus 4.6 (test)
-4.5Prevalence Estimation Bias (pp)Importance-weighted estimation
Evaluation results
| Method | Links | |
|---|---|---|
| Importance-weighted estimationInput=Target-specific density ratio2026.04 | -4.5 | |
| MCGrad (binary)Input=Binary labels, Initialization=Base-rate2026.04 | 0.7 | |
| MCGrad (scores)Input=Probability scores2026.04 | 1.6 | |
| Isotonic regressionInput=Probability scores2026.04 | 2.6 | |
| Rogan-GladenInput=Binary labels2026.04 | 3.7 | |
| Classify & CountInput=Fraction of Yes labels2026.04 | 4.8 | |
| Saerens-Latinne-DecaesteckerInput=Probability scores2026.04 | 13.7 |