UK Treasury scrapped graduate maths test after finding it disadvantaged ethnic minority candidates
The News
The UK Treasury quietly removed numerical reasoning tests from its graduate recruitment process in 2020 after an internal review found the tests had an "adverse impact on candidate diversity," according to board minutes and documents obtained by HR Magazine. The move, which only became public this week, was driven by a stated goal to increase ethnic diversity at assessment centres and reduce "hurdles" that sifted out candidates before they could demonstrate broader potential.
The Treasury's own documents show that after removing the test, the adverse impact on ethnic minority candidates did decline — but the revelation has exposed a sharp fault line among employment and assessment experts. Mark Green, founder of Change Rebellion, defended the decision: "Whenever a recruitment tool consistently disadvantages a particular group, it's absolutely right that organisations ask difficult questions about whether it's measuring capability or simply filtering out talented people."
But Kate Underwood, an HR consultant, pushed back: "A numerical test doesn't flag a problem with the candidates. It flags a problem with who got the tutoring, the practice, the head start." Raj Tulsiani, CEO of Green Park, was even blunter: "Dropping a test because one group scores lower, without proving that what replaces it identifies stronger talent, isn't a fix. It's a very public moving of the goalposts."
Why It Matters
Every TA leader using pre-hire assessments — cognitive tests, skills-based evaluations, gamified screening — should treat this as a live case study. The Treasury didn't just debate whether to keep or drop the test; they acted, and the evidence they generated (reduced adverse impact) made them look reasonable — until expert scrutiny revealed the flaw: they can't prove the replacement process selects better candidates.
The debate captures the core tension of modern assessment: fairness and rigour aren't always in conflict, but when they appear to be, removing a filter without understanding why it produces disparate outcomes solves a PR problem while potentially creating a quality problem. Under the EU AI Act's high-risk classification for employment AI and the UK's own evolving algorithmic transparency requirements, the burden of proof is shifting toward employers to demonstrate that their selection tools — whether human or automated — are both fair and valid.
Taking Action
All action is situated in your corporate context. These are starting points, not prescriptions — choose what fits.
Connect
Who to talk with in your enterprise
Your assessment or psychometrics provider: ask them directly what adverse impact analyses they run, how frequently, and whether they can produce subgroup-level pass rate data for your organisation specifically — not just validation studies from three years ago. Your legal or compliance lead: if you use any form of cognitive or skills-based screening, this Treasury case is a preview of what an ICO or EHRC inquiry could look like. Get ahead of it.
Inspect
What to take a closer look at
Request a four-fifths rule analysis on every pre-hire assessment stage in your process. If any stage shows an adverse impact ratio below 0.80 for a protected characteristic group, flag it immediately — not to remove the stage, but to investigate why before someone else does it for you. Also: review whether your assessment provider offers alternative formats (work-based exercises, case studies) that test the same competency without the adverse impact.
Reflect
What to think about more deeply
If your organisation discovered tomorrow that its most predictive hiring assessment also produced a measurable adverse impact on one demographic group, would your first instinct be to fix the assessment or remove it — and what does that say about whether you believe fairness and standards can coexist?