Measurements are evidence, not verdicts.

Adverse impact ratio

The adverse impact ratio compares a selected group’s favorable-outcome rate with a reference group’s rate.

A four-fifths or 0.80 threshold can be used as a screening convention. It does not, by itself, establish unlawful discrimination. Population definition, sample size, selection of reference group, policy context, and statistical uncertainty matter.

Statistical parity difference

Statistical parity difference expresses the absolute gap between favorable-outcome rates. Its direction, magnitude, sampling variability, and business context should be reviewed together.

Adverse-action reason review

Avarent can organize candidate explanations and supporting decision fields for human review. The creditor remains responsible for ensuring disclosed reasons are accurate, specific, and tied to factors actually considered or scored.

Known interpretation limits

• Observed disparity does not identify cause by itself.
• Small samples can produce unstable estimates.
• Missing, inferred, or misclassified demographic attributes can change results.
• Post-hoc explanations can approximate rather than reproduce a model’s internal reasoning.
• Thresholds should reflect institutional policy and qualified review, not software defaults alone.
• Results depend on the completeness and correctness of supplied data.

Primary references