In my blog last August, posted shortly before the exam results, I suggested that there "may be more marking errors than you think". That all depends, of course, on what you might think; but now that Ofqual have published their statistics on reviews of marking following challenges to the results, we know exactly how many marking errors were discovered.
Ofqual count marking errors not in relation to grades awarded, but in terms of components - a component being, for example, one paper in a multi-paper exam. The number of components is therefore always greater than the corresponding number of grades: for the summer 2025 exams in England, a total of 6,526,780 GCSE, AS and A level grades were awarded, with 17,523,265 underlying components. Of those components, 513,045 were subject to reviews of marking, resulting in the discovery and correction of 198,880 components with marking errors.
Marking errors were discovered in only 1% of components. But...
Out of a population of over 17 million components, the discovery of about 200,000 with marking errors - about 1% - seems very small. But surely a more valid measure is to compare the number of components with marking errors (198,880) to the number of components reviewed (513,045), giving a much larger number, some 40%.
The discrepancy between these measures, 1% of total components, 40% of components reviewed, arises from the fact that only about 3% of components are challenged, and so only 3% are examined for the presence, or indeed absence, of a marking error. 97% of components are not scrutinised, and - in principle - every one of those 17,010,220 unchallenged components could contain an error: no one knows, for no one has looked.
That of course is extremely unlikely. But also unlikely is that they are all error-free. How many marking errors there actually are in the whole population of all components remains unknown. My fear is that this figure might be as high as 40%, as suggested by the sample of challenges.
How many grades have marking errors?
There is another unresolved problem too: the number of grades - not components - associated with marking errors.
Ofqual offer no information relating to marking errors in the context of grades. But their statistics do tell us that, for the summer 2025 exams, the 513,045 component reviews of marking were triggered by challenging 301,980 grades, resulting in 72,875 grade changes.
Since 2016, a grade can be changed only if a review of marking discovers a marking error, which, when corrected, results in a grade different from that originally awarded. Given that there is, in general, more than one component associated with every grade, it is therefore possible that all the 198,880 component marking errors were within the 72,875 grades changed. This would imply that the number of challenged grades with marking errors is 72,875.
Alternatively, at the other extreme, the 72,875 grades changed might account for 72,875 component marking errors, leaving 126,005 component marking errors that were present, but did not result in a change to as many as 126,005 grades.
In the 301,980 grades challenged, the smallest number of grades with marking errors is therefore 72,875 (about 1% of grades awarded, 24% of grades challenged), and the largest is 198,880 (around 3% of grades awarded, 66% of grades challenged).
What's the truth?
The truth is somewhere between these two - say, about half-way, maybe 130,000 or so grades with marking errors, around 2% of grades awarded, 43% of grades challenged.
Once again, extrapolating those percentages - 24%, 66% and 43% - to the whole population of grades awarded is frightening.
But why should I have to guess?
Why doesn't Ofqual publish what we need to know: the number of challenged grades that were changed, and the number that contained marking errors, as well as estimates for the corresponding numbers in the whole population of grades awarded?