What the research supports, and what it does not

The evidence is real, uneven and routinely overstated by advocates. The honest summary is more useful than the enthusiastic one, and more persuasive to anyone sceptical.

Reasonably well supported

• Satisfaction of harmed parties is consistently higher than in conventional process, across varied settings. This is the most robust finding.
• Participants perceive the process as fairer, including those who caused harm.
• Agreements reached are largely complied with, at rates higher than imposed sanctions.

Mixed or weak

• Effect on repeat behaviour varies by context and by how faithfully the approach was implemented. Pooled effects tend to be modest.
• School exclusion data is confounded by the fact that schools adopting the approach often change recording practice at the same time.
• Long-term organisational effects are thinly studied; most workplace evidence is case reports rather than controlled comparison.

Reading a claim critically

Three questions settle most of it. Was there a comparison group, or only before-and-after? How was fidelity measured — a programme named restorative is not necessarily one practised that way? And how large was the sample, since small studies produce large effects that do not survive replication.

Effect sizes in this literature are often reported without confidence intervals, and a figure quoted as a percentage improvement tells you nothing about the range it sits in. The same gap appears wherever one number stands in for a spread: the range survives in the working paper and vanishes in the press release. A clear illustration of it sits outside this field entirely, in the house edge on screen — about the single figure that decides an outcome and is never put on screen.