Panel Paper: The Impact of Restorative Practices on Discipline, Climate and Achievement: A Multiple-Site, Multiple-Mediator Instrumental Variable Analysis

Saturday, November 9, 2019
Plaza Building: Concourse Level, Governor's Square 12 (Sheraton Denver Downtown)

*Names in bold indicate Presenter

John Engberg and Catherine Augustine, RAND Corporation

Students who are suspended from school miss instructional time and are less likely to graduate. Restorative practices (RP) are seen as a strategy to reduce suspension rates. By proactively improving relationships among students and staff and by building a sense of community in classrooms and schools, students may be less inclined to misbehave. Although pre/post studies suggest that RP reduce suspensions and increase engagement, randomized control trials (RCT) of RP are only now being completed.

In a recent RCT study of RP in a midsize urban rustbelt district, we found that RP reduced student out-of-school suspensions, improved teachers’ perceptions of school safety and climate, but also led to a small but significant reduction in the test scores of some students (Augustine, Engberg, et al., 2018). Although RP decreased the racial gap in suspensions, it increased the racial achievement gap.

The current study aims to better understand the mechanisms by which RP reduced test scores. Possible mechanisms include (1) reductions in suspensions led to increased classroom disruption, (2) increased class time spent on building relationships reduced instruction time, (3) increased teacher training on RP methods reduced instructional professional development.

We test these hypotheses with a multiple-site multiple-mediator (MSMM) instrumental variables method (Small, 2012; Reardon, Unlu, Zhu & Bloom, 2014) that does not require a sequential ignorability assumption for the mediators. Our preliminary findings reject the first hypothesis but not the latter two. This suggests that the studied RP program was successful in creating a classroom climate that supported learning in the presence of reduced suspensions, but at the cost of reduced instruction.


Treatment schools were chosen randomly from 22 pairs of schools that were matched based on baseline suspension rates, student ratings of classroom control, and teacher ratings of school climate, as well as on grade configuration.

The schools implemented SaferSanerSchools™ Whole-School Change program for two school years (2015-16 and 2016-17), in conjunction with the International Institute for Restorative Practices (IIRP). IIRP provided four days of professional development; all staff were asked to attend two of these days and the other two were voluntary. Each principal was assigned an IIRP coach to support the school during the two-year implementation period. The coaches also visited each of their schools at least twice during a school year. All school staff were asked to participate in monthly professional learning groups (PLGs). The project manager in the district provided additional support to the selected schools including supplementary materials and individualized coaching.

We obtained school, student and staff administrative data from the district and the county for two years prior to implementation and two years during implementation. We estimated impact by regressing student and staff outcomes and mediators on a treatment indicator, baseline outcome values, demographic covariates, and experimental design indicators of matched pairs. Standard errors were calculated using clustering at the school level. These impact estimates were then used in the MSMM analysis.