The UK A-level grading algorithm is withdrawn

With exams cancelled by the pandemic, Ofqual, England’s exams regulator, replaced them with a statistical model that adjusted each school’s teacher-predicted grades against that school’s historical results, intended to prevent the grade inflation ministers feared would follow from unmoderated teacher predictions. When A-level results were issued, roughly 36% of grades came out one grade below the teacher’s prediction and a further 3% two grades below. The effect fell unevenly: because the model relied more heavily on a school’s small-cohort history, and because independent schools have historically smaller, higher-attaining cohorts, downgrades landed disproportionately on state-school pupils, while private-school entrants saw a larger rise in top grades than in any comparable year.

The reaction was immediate — protests outside the Department for Education, students showing up with placards reading “the mutant algorithm”, and university offers withdrawn or endangered on the strength of grades a formula had produced rather than an exam. Prime Minister Boris Johnson initially defended the results as “robust and dependable.” Four days after results day, on 17 August, Ofqual and education secretary Gavin Williamson reversed the policy: every student would instead receive the higher of their teacher-predicted grade or their model-generated grade, in effect discarding the algorithm’s downward adjustments entirely. Ofqual chair Roger Taylor acknowledged the distress the episode had caused.

The reversal pushed the proportion of top grades to its highest level in at least two decades, undercutting the standardisation the exercise had been built to enforce. The episode became a frequently cited case study in what happens when an opaque statistical model is used to allocate a consequential, individual outcome without a transparent account of how it treats any one case — cited since in debates over automated decision-making well beyond education.