§ Our Approach · Bias

Fairness is a measurable property — not a promise.

A test that predicts well on average can still work unevenly across candidate groups. Bias auditing is how Xobin catches that — at the item level and at the score level — before it affects a hiring decision.
01

Defining fairness in scope

Fairness at Xobin starts by naming, in writing, the subgroups we are testing against and the fairness definition we are holding the assessment to. Group definitions are set by the customer's compliance framework — typically gender, age band, and region — and are used only in aggregate, never as an input to scoring an individual candidate. The scope is documented in the assessment's evidence dossier so a reviewer can see exactly what was audited and what was not.

02

Running the statistical tests

For every scored item and every full assessment we run adverse-impact analysis against the standard 4/5ths rule, screen items for differential item functioning (DIF) — the case where candidates of equal underlying ability answer an item differently across groups — and track group-mean gaps over time. The results are reported alongside reliability and validity, because a fair test is a joint claim about all three.

03

Remediating what we find

Items that show consistent DIF are pulled from the active item pool and sent back to subject-matter experts for review, rewrite, or retirement. Assessments that fail the 4/5ths ratio in aggregate are flagged to the customer with a re-validation recommendation before continued use. Bias auditing is a loop, not a certificate — the same tests run again after every remediation.

Next in Our ApproachAI Bias
§ From the lab

Read the flagship report, or get in touch with the research team.