Comparative judgement
Also called: Comparative marking, Pairwise comparison
What it means
Comparative judgement is a way of assessing work by comparing pieces of it with each other. A judge sees two scripts, usually two pupils’ pieces of writing, and picks the better one. The decision takes seconds. The pair goes back into the pool and the judge gets another.
After many decisions by many judges, software turns the results into a rank order from weakest to strongest. Each script is compared several times against different partners, so one odd pairing does not decide where it lands. Teachers then set grade boundaries on the ranking.
What it looks like in daily life
- A primary team sits down after a term and judges every Year 6 writing sample on screen, a few pairs at a time.
- An English department compares Year 9 essays from all its classes to see how each class sits against the others.
- A teacher with a small project runs their own task for internal assessment on the free account of No More Marking, using anonymised scans.
No More Marking is built around writing. Its AI judges make about 90% of decisions. Schools pay a yearly subscription, and the price is not published.
What tends to help
Comparative judgement suits work where quality is hard to pin down, such as extended writing. Agree beforehand what “better” means for the task. Anonymise the scripts so judges do not know whose work they are reading. Judge on overall impression and keep each decision quick.
A rank order tells you where a piece sits. It does not tell a pupil what to fix, so pair it with feedback. Our list of marking and feedback tools covers other options.
If AI judges or AI feedback are involved, read and check the output before pupils see it. Scans of pupil work are pupil data. Follow your school’s data protection policy and use tools the school has approved.
Where the term comes from
The method goes back to the psychologist L. L. Thurstone, who wrote about comparative judgement in the 1920s. Schools took it up much later, once software could handle thousands of pairings.