The variety of education systems has created a natural experiment: to control that experiment, and learn from it, is a fundamental starting point for improvement. There are now rigorous social science methods, using major advances in statistical analysis and ICT, able to implement large-scale, international and comparative studies to identify mistakes – as well as what appears to work well. The confidence in these new methodologies should not blind us to the potential pitfalls. The calls for a social science approach to education systems is over 100 years old and is strongly supported by social scientists – as one might imagine. The vast majority of educational research carried out in the past is not good enough – parents, educationalists and employers rightly distrust any sentence that begins, ‘Research shows . . .’ because of the poor quality of research used to justify almost anything in education. Against this is the rhetoric of the rigorous research community that is visionary, though the reality has been, at times, disappointing.