Disparate impact
A discriminatory effect measured on a model's outputs, independent of any intent and of the data's apparent neutrality.
The US Supreme Court laid down the concept in Griggs v. Duke Power Co. on March 8, 1971: a facially neutral hiring requirement is unlawful if it screens out a protected group without business necessity. Applied to models, disparate impact shifts the scrutiny of discrimination from input variables to outputs. A model is judged on what it does, not on what it was fed. That is the only approach that survives reconstruction through proxies. Colorado enshrined it in legislation that requires quantitative testing for disparate impact even where data and model are facially neutral. The rules were extended in October 2025 to private passenger auto and health coverage. The obstacle is close to ironic. Testing impact by origin requires knowing origin, which the law often forbids collecting, so impact-based regulation remains a requirement without an instrument.
Regulators require testing a model's disparate impact by origin while forbidding the collection of origin. The insurer is asked to prove the absence of an effect it has no right to observe.
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.