Timed written papers fall from 70 to 45 per cent of assessment across the six colleges from the autumn session.
The finding runs against the intuition that has shaped a decade of policy in this area, and the research team were the first to say they had expected the opposite result. What follows is their account of how the study was designed, what it measured, and what they think should now change.
What the study did
Four cohorts were followed over four years, with data collected at six-month intervals rather than at the end. That design is more expensive and slower than the alternative, but it is the only way to distinguish a durable change from an initial burst of enthusiasm that fades by the second year.
Crucially, the comparison group was not left alone. Doing nothing is itself an intervention, and studies that treat it as a neutral baseline consistently overstate the size of whatever they are measuring.
“We went in expecting to confirm what everybody already believed. The value of a four-year design is that it gave the data enough time to tell us we were wrong.”
What it found
- The effect was real in the first six months and had substantially decayed by month eighteen.
- Where the effect persisted, it was associated with the quality of the relationship rather than the size of the incentive.
- Two subgroups showed a reversal — an outcome the original design had not anticipated and which the team have flagged for replication.
- Cost per sustained outcome was roughly four times the figure used in current planning assumptions.
What happens next
- The full dataset is being published under an open licence, with identifying fields removed.
- A replication is planned with two partner institutions, starting in the autumn.
- The team are convening a public seminar in Northcrest to discuss the policy implications.
Anyone wanting the underlying data or the pre-registration document can request both through the school office. The team have committed to answering methodological questions from outside the university on the same terms as questions from inside it.