List your company

Prevalence Estimation on CAP Law & Crime Baseline Claude Opus 4.6 (test)

0.1Prevalence Estimation Bias (pp)Importance-weighted estimation
-0.1921.7793.755.721Apr 23, 2026

Evaluation results

MethodLinks
Importance-weighted estimationInput=Target-specific density ratio, Base LLM=Claude Opus 4.62026.040.1
Isotonic regressionInput=Probability scores, Base LLM=Claude Opus 4.62026.040.1
MCGrad (binary)Input=Binary labels, Initialization=Base-rate, Base LLM=Claude Opus 4.62026.040.1
MCGrad (scores)Input=Probability scores, Base LLM=Claude Opus 4.62026.040.2
Rogan-GladenInput=Binary labels, Base LLM=Claude Opus 4.62026.040.5
Classify & CountInput=Fraction of Yes labels, Base LLM=Claude Opus 4.62026.042.2
Saerens-Latinne-DecaesteckerInput=Probability scores, Base LLM=Claude Opus 4.62026.047.4