England reverses calculated exam grades

On August 17, 2020, England’s qualifications regulator, Ofqual, reversed the statistical standardisation of school exam grades after A-level results had been issued without students sitting exams. Students would receive the higher of their centre assessment grade—the grade their school or college judged they would most likely have achieved—and their calculated grade. The decision covered AS and A levels and the GCSE results still to come that week.^[1]

From cancelled exams to calculated results

The government announced the cancellation of the summer exam series on March 18. Schools and colleges then supplied grades and ranked their students within each subject. Ofqual’s stated objective was to maintain national standards while allowing students to progress to further education or employment.^[4]

A centre assessment grade was a professional judgment, not an exam mark. For the statistical calculation, exam boards combined the centre’s ranking with its historical subject results and the current cohort’s prior attainment. The model estimated a grade distribution for a subject within a school or college, then allocated grades in rank order. An individual’s ranking mattered, but their own teacher-assessed grade did not determine their calculated result in the usual large-entry case.^[4][2]

Small entries were treated differently because historical evidence was weaker. The model used a combined measure of current and historical entry size, the harmonic mean. Below five, it relied on centre assessment grades; from five to fifteen it blended the teacher-assessed distribution with the statistical prediction; above fifteen the statistical distribution applied. Centres with no historical data also used centre assessment grades. These were subject-within-centre rules, rather than an exemption for every student at a small school.^[4]

This was an administrative statistical allocation system. Its documented mechanism should not be confused with a generative AI model inventing grades or a trained agent pursuing a hidden objective.

Results, appeals and reversal

A-level calculated grades were issued on August 13. Ofqual’s contemporaneous Table 9.8 covers 718,276 A-level grades: 35.6% were one grade below the centre assessment, 3.3% two below and 0.2% three or more below. Together, those rounded categories total 39.1%. Another 58.7% were unchanged. These are percentages of subject grades, not students, and “below” compares two assessments rather than an observed exam result.^[4]

On August 17, chair Roger Taylor acknowledged the distress and loss of public confidence. He said that expecting schools to appeal incorrect grades burdened teachers preparing for the new term and prolonged students’ uncertainty. Ofqual changed the basis of the award to the higher of the two grades, rather than requiring each such difference to be resolved through an appeal. Its announcement applied to England: the devolved nations had their own regulators and decisions. GCSE students in England were never issued the calculated grades before the reversal.^[1][2][5]

The reversal therefore changed grades already received by A-level students and the basis of the forthcoming GCSE award. It did not establish what each student would have scored in the cancelled exams. In the later official study’s filtered sample of 246,110 A-level candidates in England, aged 18 at the end of the school year, 68% had at least one subject upgraded from the calculated grade to the final grade. That candidate-level proportion has a different denominator from the earlier 39.1% of subject grades.^[2]

What teachers described, and what later analysis found

Ofqual’s 54 teaching-staff interviews took place July 28–August 12, before results day and the reversal. Participants described varied processes: centrally supplied spreadsheets, department-level judgments, mock exams and other work, and checks against historical results. Historical distributions could therefore influence the teachers’ submissions as well as the later statistical calculation. The volunteer follow-up sample explains how some staff worked; it does not measure how common each approach was nationally or how teachers reacted to the reversal.^[3]

The 2021 grading-gap study examined differences between the two assessments. Its candidate-level models focused on students taking three A levels, excluding those with any subject in a centre cohort of fifteen or fewer. After accounting for centres and subject choices, the authors found no evidence linking a total gap of at least three grades to socioeconomic background, special educational needs or home language; they reported small associations for sex and Asian ethnicity. These conditional findings do not certify every individual allocation or rule out other forms of disadvantage. The study could neither identify who suffered the greatest consequences nor determine which assessment better represented the exam performance that never occurred.^[2]

An independent review by the Office for Statistics Regulation, published March 2, 2021, considered the four UK systems together. It found that model limitations and uncertainty had not been fully communicated, external technical challenge could have been wider, and quality assurance concentrated on aggregate results with limited individual review before results day. Appeals had been expected to address individual problems. The review drew lessons for future public-sector models; it did not adjudicate every student’s grade or pass a definitive performance verdict on each regulator.^[5]

The case connects Social Alignment with Fairness Tradeoffs through a concrete distinction: maintaining a cohort’s grade distribution does not establish the accuracy or acceptability of an individual award. The incidents guide places it alongside other failures of institutional safeguards. Here, the consequential decision was how statistical estimates, professional judgments and opportunities to challenge a result were combined.