True Negative Rate

Overview

AKA specificity. Calculates TN / (FP + TN) for privileged and unprivileged groups. This is the probability that a negative prediction is truly negative.

Formula

TNR = TN / (FP + TN)

Where:

TN (True Negatives): The number of negative instances correctly predicted by the model
FP (False Positives): The number of positive instances incorrectly predicted as positive by the model

Usage

Manually


tnr_priv, tnr_unpriv = true_negative_rate(df, protected_attribute, privileged_group, labels, positive_label, y_true)

print("True Negative Rate for privileged group:", tnr_priv)
print("True Negative Rate for unprivileged group:", tnr_unpriv)

Using Fairness Object

tnr_priv, tnr_unpriv = (fo.compute(true_negative_rate))

Results

True Negative Rate for privileged group: 0.6249999999921875
True Negative Rate for unprivileged group: 0.24999999999375

These results are obtained by using the input data given in the Create Example Data page under Getting Started

Interpretation

A higher TNR indicates that the negatives are being correctly assigned as such.