“Children crushed by school test stress.” Such headlines are horribly familiar. It turns out, however, that the volume of test data now routinely collected on pupils in UK schools could have an unexpected benefit: it is now easy to conduct natural experiments on school-based interventions because the data is there just waiting to be exploited.

Without defending the current testing system per se, Peter Tymms, from the University of Durham, argued powerfully that innovations in education should be implemented in schools on the basis of random allocation, using the monitoring data as outcome indicators. Natural experiments in criminal justice were advocated earlier in the York conference by Lawrence Sherman from the Un