Evidence · The claim ledger
Fairness & disparate impact2
Every ledgered claim this site makes in this evidence area, with the sources that ground it — or, for a PAN-simulation-derived claim, the run it comes from. Source keys link back to the full reference lists on the Evidence Registry.
EmpiricalWhere base rates differ between groups, a risk instrument cannot be both well calibrated across those groups and equal i…
Where base rates differ between groups, a risk instrument cannot be both well calibrated across those groups and equal in its error rates across them. The volume's criminal-justice chapter restates this result, which the site already carries in its primary form from Chouldechova's 2017 analysis of recidivism instruments. It is a property of the scoring instrument and the population it is applied to, not a defect that a better model removes.
Sources: ahn2026, chouldechova2017
Appears on: /domains/security-operations-fraud
EmpiricalThe volume's criminal-justice chapter reports a statewide pretrial reform, evaluated by Anderson and colleagues in 2019,…
The volume's criminal-justice chapter reports a statewide pretrial reform, evaluated by Anderson and colleagues in 2019, in which the pretrial jail population fell without a rise in crime or in failure to appear. Over the same period the racial composition of those still detained did not materially change. Both halves are the finding: the level moved and the composition did not.
Sources: ahn2026
Appears on: /domains/security-operations-fraud