Rogan-Gladen corrected estimator with finite-sample adjustment.

Classical misclassification correction using sensitivity/specificity, also known as the Rogan-Gladen estimator.

Usage

rg_point_and_ci(p_hat, q0_hat, q1_hat, n, m0, m1, alpha = 0.1, eps_J = 1e-04)

Arguments

p_hat

Average LLM score on the test set.

q0_hat

Calibration specificity estimate.

q1_hat

Calibration sensitivity estimate.

n

Test set size.

m0, m1

Calibration sample sizes for negatives/positives.

alpha

Miscoverage level.

eps_J

Small threshold to guard against near-random judges.

Value

List with theta, var, and CI endpoints.