Restorative practice

Restorative practice covers several approaches. The evidence belongs to named school packages and specific outcomes. Voluntary, accessible participation and case-specific safeguarding are practice conditions, not experimentally proven mechanisms.

Targeted web searching and citation following sought original comparative studies, contradictory/null findings, current identity and professional guidance. Nine retained records include six empirical reports from five distinct evaluations, two original review leads with partial access, and one professional code. Reports S04/S05 share a cohort. No pooled estimate or study-vote count was calculated.

Relevant original methods/results were inspected where accessible. Access is stated per source; full-text availability does not mean every appendix or outcome was extracted. Search breadth does not constitute a systematic search. Source-specific access limits remain important when using these conclusions.

A010-S01 — Augustine et al. (2018), Can Restorative Practices Improve School Climate and Curb Suspensions? RAND RR-2840¹

Role/access: Direct comparative evaluation; Full report; relevant methods, results and limitations inspected. Locator: Summary pp. xiii–xv; Chapters 2–5, especially pp. 9–20,29–30,49–52,69–71.

• Identity/version: Pittsburgh Pursuing Equitable Restorative Communities (PERC); International Institute for Restorative Practices (IIRP) SaferSanerSchools whole-school package, 2015–17.
• Design: Cluster randomised controlled trial (RCT); final 44 schools, 22/22. Two treatment schools excluded after allocation; remaining paired controls re-randomised.
• Population/setting/sample: Urban Pittsburgh schools; outcome-specific student/teacher denominators, no single extracted pupil total.
• Intended/actual delivery: Two years; four training days offered (two requested for all, two voluntary), implementation support.
• Comparator/co-supports: Non-PERC schools; their restorative activity not fully characterised; extra district support accompanied PERC.
• Outcome/reporter/time: Administrative suspensions fell; teacher climate ratings improved. No arrest benefit; no suspension reduction for Individualized Education Program (IEP) pupils or middle grades. Middle-grade achievement worsened.
• Fidelity/adaptation: Staff use varied; individual pupil exposure unknown.
• Burden/unwanted effects: Training/workload; no comprehensive recipient adverse-experience estimate extracted.
• Limits: Single district, allocation changes and incomplete exposure data; staff climate ratings do not substitute for pupil accounts. No general safety or relationship-repair conclusion.

A010-S02 — Bonell et al. (2018), Effects of the Learning Together intervention on bullying and aggression in English secondary schools (INCLUSIVE), Lancet 392:2452–2464²

Role/access: Direct comparative evaluation; Full published paper; university repository. Locator: Methods pp.2453–2455; Tables 2,6; process findings p.2460.

• Identity/version: Learning Together combines restorative staff training, student/staff action groups and social and emotional learning (SEL) lessons.
• Design: Cluster RCT, 40 schools, 20/20, three years.
• Population/setting/sample: Southeast England secondary schools; 6,667 surveyed at baseline, 5,960 at 36 months; no schools withdrew.
• Intended/actual delivery: Selected staff three-day training; action groups twice/term; intended SEL 5–10 hours/year.
• Comparator/co-supports: Usual practice; some controls used similar components.
• Outcome/reporter/time: Student-reported bullying at 36 months: adjusted difference −.03, 95% confidence interval (CI) −.06 to −.001, effect size (ES) −.08. Aggression: −.13, CI −.43 to .18; no clear difference.
• Fidelity/adaptation: Delivery varied; curriculum fidelity weaker; formal activity declined in year 3.
• Burden/unwanted effects: Staff time/training; serious-event counts similar overall but types differed, with severe incidents reported in intervention schools; causation unresolved.
• Limits: Package effects cannot isolate restorative conversations. Self-report, implementation variation and missing responses limit transfer; similar serious-event totals do not establish safety.

A010-S03 — Acosta et al. (2019), Evaluation of a Whole-School Change Intervention, Journal of Youth and Adolescence 48:876–890³

Role/access: Direct comparative evaluation plus nonrandomised exposure analysis; Full public author manuscript inspected; PMC direct access encountered verification challenge. Locator: Manuscript pp.5,7–8,11–15; Tables 3–4.

• Identity/version: IIRP Restorative Practices Intervention, 11-practice whole-school model.
• Design: Two-year cluster RCT; 13 evaluated schools; 2,771 eligible baseline pupils; intention-to-treat (ITT) outcome analysis 1,685.
• Population/setting/sample: Maine middle schools, grades 6/7 initially, 92% White.
• Intended/actual delivery: Intended four training days across two years, consultation, learning groups and coach visits.
• Comparator/co-supports: Control schools; pupils reported substantial restorative exposure there too.
• Outcome/reporter/time: No significant allocation effects across 11 student-reported climate, connection, development or bullying outcomes. Favourable perceived-exposure associations were nonrandomised; some changed under sensitivity analysis.
• Fidelity/adaptation: Only a minority reported high exposure; weak treatment/control separation.
• Burden/unwanted effects: Implementation time and competing demands; no systematic recipient-harms estimate extracted.
• Limits: Few clusters, attrition/listwise deletion assumption, shared self-report method. Null implementation results cannot prove every restorative process ineffective; exposure association cannot establish causality.

A010-S04 — Gregory, Huang & Ward-Seidel (2021), Evaluation of the Whole School Restorative Practices Project: One-Year Implementation and Impact on Discipline Incidents⁴

Role/access: Direct comparative evaluation; Full original technical report inspected. Locator: Study/implementation sections; Exhibits 6–8, pp.19–21.

• Identity/version: Morningside Center Whole School Restorative Practices Project; restorative work, SEL, racial-equity work and coaching.
• Design: Cluster RCT; 18 schools, 9/9; 5,878 pupils analysed; one-year 2018–19 findings.
• Population/setting/sample: Urban US elementary, middle, high and combined schools; disability recorded broadly.
• Intended/actual delivery: Whole-school training/coaching with leadership and student work; individual received dose not extracted.
• Comparator/co-supports: Business-as-usual schools; intervention is a multi-component package.
• Outcome/reporter/time: Administrative discipline-incident likelihood lower: adjusted −4.5 percentage points, p=.03. No detected moderation by race, gender or disability.
• Fidelity/adaptation: Implementation evaluated at school level; student-level receipt not established here.
• Burden/unwanted effects: Staff development and coordination; recipient distress/refusal not a measured primary outcome.
• Limits: Record-writing can change alongside behaviour. No detected subgroup interaction does not establish equity. Same cohort as S05; COVID disrupted longer planned follow-up.

A010-S05 — Huang, Gregory & Ward-Seidel (2023), The Impact of Restorative Practices on the Use of Out-of-School Suspensions, Prevention Science 24:962–973⁵

Role/access: Direct comparative outcome report, same cohort as S04; Full public author manuscript inspected. Locator: Methods pp.11–16; results/discussion pp.17–22.

• Identity/version: Same Morningside Whole School Restorative Practices Project as S04; not another trial.
• Design: 18-school cluster RCT; 5,878 pupils; one-year administrative suspension outcome.
• Population/setting/sample: Same mixed school-level urban cohort; disability status is not a communication-access profile.
• Intended/actual delivery: Training/coaching package; see S04 for shared design.
• Comparator/co-supports: Business-as-usual; four of nine controls reported some restorative training.
• Outcome/reporter/time: Primary model: no overall suspension difference (B −.001, p = .95). Exploratory, non-preregistered prior-suspension interaction favoured intervention (B −.14, p = .04), but logistic sensitivity interaction p = .08 (two-sided).
• Fidelity/adaptation: Training met project threshold, not proof of individual delivery.
• Burden/unwanted effects: Pandemic prevented comparable second-year analysis; adverse experiences not directly quantified.
• Limits: Low baseline suspension rate, limited clusters and outcome/model sensitivity. Do not replace overall null with subgroup success or combine S04/S05 as independent replication.

A010-S06 — Adukia, Feigenberg & Momeni (2025), From Retributive to Restorative: An Alternative Approach to Justice in Schools, American Economic Review 115:2722–2754⁶

Role/access: Quasi-experimental comparative evaluation; Full published paper inspected; author institution copy. Locator: SectionsII–IV; pp.2723–2724,2728–2734; robustness section.

• Identity/version: Chicago restorative-practice rollout: coaching, leadership and peer-council variants.
• Design: Staggered-adoption difference-in-differences; 73 adopting high schools within 239-school sample; 2009–19 records.
• Population/setting/sample: Chicago high schools serving over 100,000 annually; not a fixed unique-participant count.
• Intended/actual delivery: Training intensity varied; schools prioritised partly through interest, discipline and climate indicators.
• Comparator/co-supports: Schools not adopting over relevant comparisons; selection not random.
• Outcome/reporter/time: Administrative suspensions/arrests declined; no significant test-score gains. Climate findings interpreted cautiously because pretrends differed.
• Fidelity/adaptation: Programme adoption identifies exposure, not each pupil’s experience.
• Burden/unwanted effects: Resource intensity varies; individual coercion, distress and acceptability not systematically measured.
• Limits: Depends on parallel-trends and other identification assumptions despite robustness checks; cannot establish an isolated repair mechanism or transfer to all schools.

A010-S07 — Restorative Justice Council (2020), Service Provider Code of Practice⁷

Role/access: Professional guidance, not efficacy evidence; Full official code text inspected. Locator: RequirementsA–J.

• Identity/version: Standards for RJC-registered service providers.
• Design: Normative code; no trial/sample.
• Population/setting/sample: Registered restorative services; not a universal legal requirement for every school conversation.
• Intended/actual delivery: Informed participation, risk assessment, appropriate competence and voluntary agreements.
• Comparator/co-supports: Not applicable.
• Outcome/reporter/time: No measured effectiveness outcome.
• Fidelity/adaptation: Requires provider adherence; does not verify a particular school’s practice.
• Burden/unwanted effects: Consent can be withdrawn; pressure, revictimisation and confidentiality require attention.
• Limits: Supports safeguards and role boundaries; does not prove safe delivery or require Restorative Justice Council (RJC) registration for all informal school support.

A010-S08 — Procter-Legg (2025), Restorative practice and young people with special educational needs and disabilities: a scoping study, Disability & Society 40:3463–3486⁸

Role/access: Original scoping review; transfer/voice question; Repository abstract and publisher-indexed discussion/limitations; full retrieval failed. Locator: Abstract; publisher DOI 10.1080/09687599.2025.2489363, discussion/limitations indexed text.

• Identity/version: School restorative practice and special educational needs and disabilities (SEND) literature, 2000–2023.
• Design: Scoping study; included-study total not verified here.
• Population/setting/sample: School-age educational research; review highlights categorical treatment of SEND and limited youth perspectives.
• Intended/actual delivery: Different practices/programmes; no common tested dose.
• Comparator/co-supports: No common comparator.
• Outcome/reporter/time: Identifies empirical and participant-voice gaps; not a pooled treatment-effect estimate.
• Fidelity/adaptation: Study selection/coding limitations include single-coder risk and exclusion of books.
• Burden/unwanted effects: Ableism and lived experience need investigation; not a measured adverse-event rate.
• Limits: Partial access; do not infer absence of all SEND research or verified equivalence of augmentative and alternative communication (AAC)/access adaptations.

A010-S09 — Marcucci et al. (online 2025; issue 2026), Understanding the Anti-Carceral Possibility of School-Based Restorative Justice, Review of Educational Research 96:1014–1048⁹

Role/access: Original systematic review; conceptual/implementation lead; Official abstract and bibliographic record; main text restricted. Locator: Abstract; first-online 30 April 2025, issue June 2026.

• Identity/version: Methodologically inclusive review of school restorative justice and racialised social control.
• Design: 61 studies in abstract; not 61 randomised trials.
• Population/setting/sample: US schooling focus; corpus boundaries not fully appraised.
• Intended/actual delivery: Examines different operationalisations, not one common programme.
• Comparator/co-supports: No pooled comparator verified.
• Outcome/reporter/time: Abstract supports conceptual question about restorative work becoming entangled with punitive control.
• Fidelity/adaptation: Detailed coding and findings not accessible here.
• Burden/unwanted effects: Coercive implementation is a review question; no measured harm rate extracted.
• Limits: Inherited claims about leadership, student voice and what works best are not certified from the abstract. Retrieve full text before importing those findings.

The sources answer different questions. Some concern school organisation and recorded sanctions; others ask students about bullying, climate or connection. Administrative outcomes are valuable but cannot settle whether a particular harmed person experienced repair. Inferences about fewer incidents must distinguish changes in conduct, reporting and adult decisions.

The family label cannot supply missing delivery information. A short script, community circle, formal conference and multi-year whole-school reform are not interchangeable doses. Fidelity and staff support matter to interpreting what was tested, while null findings must remain visible rather than dismissed as implementation failure. Better implementation as an explanation remains a hypothesis unless directly tested.

No generic safety claim is warranted. Study reports vary in unwanted-effects collection. Locally, record pressure to participate, unwanted disclosure, increased distress, retaliation, lost learning time, unwanted renewed contact and unmet support needs alongside intended benefits. These are Phoenix review questions, not an estimated harm frequency from the trials.

SEND categories in datasets do not establish accessible participation for a particular pupil. Communication arrangements, processing time, private contribution, refusal and trusted support need specialist and recipient review. The access gaps in S08/S09 prevent further claims about their detailed findings.

• Identity and version: Name the component or programme, its edition or protocol where relevant, and the purpose being evaluated.
• Source role: Separate direct comparative evidence, descriptive implementation findings, guidance, theoretical support and firsthand experience.
• Population and setting: Retain the studied age, communication/access characteristics and setting. State when an intended application lies outside them.
• Intended and actual delivery: Record the planned components, frequency and duration (dose) separately from what was received, by whom, and with what support or interruptions.
• Comparison and other support: Name the actual comparison and concurrent changes. If there is no comparison group or condition (comparator) or it is unreported, say so rather than supply an assumed usual-care condition.
• Outcome and reporter: Record the outcome, time point and reporter. Keep the young person’s account separate from someone else’s judgement on their behalf (proxy report), and access or participation separate from task performance.
• Delivery as intended (fidelity) and adaptation: Record whether the core components were delivered and describe adaptations; distinguish an adapted application from the protocol tested.
• Burden and unwanted effects: Record time, effort, practical access, cost or workload where known, alongside refusal, fatigue, distress, restriction and other unwanted consequences. Unreported burden is not evidence of no burden.

This common key applies to the retained source summaries below. Shorter headings such as design, dose or findings are read in that context. The key supplies no missing source data, converts no proxy report into recipient testimony and assigns no new certainty grade.

1. 1.

   Augustine et al. (2018), Can Restorative Practices Improve School Climate and Curb Suspensions? RAND RR-2840. Accessed 12 September 2026; Full report; relevant methods, results and limitations inspected.

   ↩1
2. 2.

   Bonell et al. (2018), Effects of the Learning Together intervention on bullying and aggression in English secondary schools (INCLUSIVE), Lancet 392:2452–2464. Accessed 12 September 2026; Full published paper; university repository.

   ↩1
3. 3.

   Acosta et al. (2019), Evaluation of a Whole-School Change Intervention, Journal of Youth and Adolescence 48:876–890. Accessed 12 September 2026; Full public author manuscript inspected; PMC direct access encountered verification challenge.

   ↩1
4. 4.

   Gregory, Huang & Ward-Seidel (2021), Evaluation of the Whole School Restorative Practices Project: One-Year Implementation and Impact on Discipline Incidents. Accessed 12 September 2026; Full original technical report inspected.

   ↩1
5. 5.

   Huang, Gregory & Ward-Seidel (2023), The Impact of Restorative Practices on the Use of Out-of-School Suspensions, Prevention Science 24:962–973. Accessed 12 September 2026; Full public author manuscript inspected.

   ↩1
6. 6.

   Adukia, Feigenberg & Momeni (2025), From Retributive to Restorative: An Alternative Approach to Justice in Schools, American Economic Review 115:2722–2754. Accessed 12 September 2026; Full published paper inspected; author institution copy.

   ↩1
7. 7.

   Restorative Justice Council (2020), Service Provider Code of Practice. Accessed 12 September 2026; Full official code text inspected.

   ↩1
8. 8.

   Procter-Legg (2025), Restorative practice and young people with special educational needs and disabilities: a scoping study, Disability & Society 40:3463–3486. Accessed 12 September 2026; Repository abstract and publisher-indexed discussion/limitations; full retrieval failed.

   ↩1
9. 9.

   Marcucci et al. (online 2025; issue 2026), Understanding the Anti-Carceral Possibility of School-Based Restorative Justice, Review of Educational Research 96:1014–1048. Accessed 12 September 2026; Official abstract and bibliographic record; main text restricted.

   ↩1