Restorative for All? Racial Disproportionality and School Discipline Under Restorative Justice

School discipline practices became increasingly harsh in the 1980s as schools embraced exclusionary discipline (i.e., suspension and expulsion), zero tolerance policies, and security measures (Skiba et al., 2002; Kupchik and Ward, 2014). As these practices proliferated, a large body of research examined racial disproportionality in exposure to harsh disciplinary practices, finding that Black and Latinx youth are more likely to be exposed to schools with punitive disciplinary cultures, and that Black students in particular are suspended at higher rates than their White peers (Anderson and Ritter, 2017; Fenning and Rose, 2007; Rios, 2011; Skiba et al., 2014; U.S. Department of Education Office for Civil Rights, 2014). Given the link between school suspensions and juvenile justice contact (Fabelo et al., 2011), scholars have described punitive disciplinary environments as the school-to-prison pipeline, where Black and Latinx students are funneled into the criminal justice system as a result of school disciplinary practices (Wald & Losen, 2003; Hirschfield, 2009).

To address these problems, a growing number of school districts across the United States have adopted restorative justice programs. At their core, restorative justice practices promote equitable and relational learning environments through policies and practices that support students through conflicts in lieu of exclusionary disciplinary practices (e.g., suspension or expulsion). Restorative justice practices exhibit a continual community orientation that seeks to democratize school environments by equalizing the voices of students, educators, administrators, and staff in the school community (Winn, 2018; Zehr, 2002). Restorative justice has become popular in school districts across the U.S. over the last decade, and a joint 2014 report by the U.S. Departments of Education and Justice touted the use of restorative justice as a viable intervention that would improve equity and address the school-to-prison pipeline.

Despite the descriptive promise of restorative justice, existing research on restorative justice in schools finds that this practice may reduce disciplinary inequity in some contexts (Augustine et al., 2018) but unintentionally exacerbate it in others (Hashim et al., 2018). Because restorative justice programs often coincide with normative disciplinary practices and do not completely replace them (Ispa-Landa, 2017), variation in the social contexts that surround the adoption of such programs may lead to a variety of outcomes that may reduce or exacerbate racial disparities. Moreover, recent work has shown the importance of allowing schools time to implement these programs, as fully transitioning to restorative justice is likely to take a minimum of three to five years (Darling-Hammond et al., 2020; Gonzalez et al., 2018).

This study thus adds to a growing body of literature by examining the use of restorative justice practices in Meadowview Public Schoolsⁱ from 2008 to 2017, investigating whether the implementation of restorative justice was associated with changes in both overall and race-specific discipline rates over time. Relying on administrative records from a large urban school district, this study addresses the following questions: 1) Does the implementation of restorative justice practices in schools change exclusionary discipline outcomes? and 2) Does the use of restorative justice change racial disproportionality in exclusionary discipline? In contrast to earlier work, we capitalize on a unique rollout of restorative justice to focus explicitly on how the efficacy of restorative justice might change as programs mature (Darling-Hammond et al., 2020). Further, we highlight the potential pitfalls of layering restorative justice programs on top of traditional discipline policies, suggesting that increasing the opportunities for educator discretion without also addressing anti-Blackness may contribute to racially divergent outcomes that change over time (cf. Gavrieldes, 2014).

Consistent with prior studies, our difference-in-difference estimates show that restorative justice practices reduced overall discipline and suspension rates in the district. However, the benefits of restorative justice in Meadowview do not materialize immediately and do not benefit all students, as disciplinary outcomes improved for White, Latinx, and Asian students and remained largely unchanged for Black students. While the overall effects of restorative justice in Meadowview are promising, the adoption of these practices unintentionally widened the racial disproportionality in school discipline they were instituted to mitigate. These findings bolster those from prior studies that have demonstrated both the promise and the potential challenges in implementing school-based restorative justice programs and provide a deeper understanding of when (and for whom) restorative justice practices are effective in changing student discipline. Further, our findings suggest that the promise of restorative justice programs should not be considered separately from the racialized contexts that govern their adoption.

Schools as Punitive Spaces

Restorative justice practices in schools are often adopted in response to concerns about the punitive nature and racially disparate outcomes in exclusionary school discipline practices. As the U.S. adopted tough-on-crime policies during the War on Drugs era, school districts across the U.S. began to implement zero-tolerance policies, mandating harsh, unrelenting consequences for violation of certain school policies (Kupchik, 2010; Weissman, 2015). Although these policies were putatively adopted for safety, they were developed in the wake of racial integration and fundamentally driven by anti-Black sentiment (Dumas, 2016). The zero-tolerance era also brought attention to stark racial disparities in school discipline, as Black and Latinx students were more likely to attend schools with the harshest practices and security measures (Fenning and Rose, 2007; Rios, 2011; Skiba et al., 2014). Thus, racial disproportionality in school discipline between Black and White students is striking, with Black students three times as likely to be suspended as White students, despite limited evidence that Black students participate in more delinquent behavior (U.S. Department of Education Office for Civil Rights, 2014; Shollenberger, 2015). By contrast, research on Latinx student discipline has been less conclusive, finding that Latinx students are overrepresented in school suspensions in some contexts (Skiba et al., 2011) and potentially underrepresented in others (Anderson & Ritter, 2017).

The punitive disciplinary environments of the zero-tolerance era have been characterized as an axis of social control that heightens racial inequalities in schools—particularly for Black students. Researchers have consistently found that exclusionary disciplinary practices negatively affect school climate, student engagement, mobility, and academic performance (Welsh & Little, 2018; Anderson, Ritter, & Zamarro, 2019; Gregory et al., 2010). These practices are also linked to negative out-of-school outcomes, being associated with a higher risk of drop-out (Lee et al., 2011), decrease in civic participation (Kupchik & Catlaw, 2014), and increased contact with the criminal justice system (Mittleman, 2018). Because of these outcomes, Wald and Losen (2003) characterized these school environments as part of the school-to-prison pipeline, where punitive disciplinary environments make juvenile justice contact more likely than college graduation for Black and Latinx youth. While the school-to-prison pipeline framework has not been universally accepted (see Sojoyner, 2013), there is a general consensus among scholars of race in education that exclusionary disciplinary practices are inherently rooted in anti-Blackness and are thus ineffective and ultimately exacerbate racial inequality (Fronius et al., 2019).

Restorative Justice as a Response to Discipline Disparities

In contrast to zero-tolerance policies, school-based restorative justice practices seek to build, maintain, and repair relationships to form healthy, supportive, and inclusive communities that facilitate optimal learning environments (Zehr, 2002). When conflicts arise in the classroom or in the community, restorative justice practices provide opportunities for individuals and communities to come together to repair harm in lieu of traditional punishments like suspension or expulsion (Morrison, 2003). Thus, restorative justice programs promote relationship building and rely on different forms of mediation, focus groups, and restorative circle processes to address conflicts or behavioral infractions. Along with facilitating relationship building, restorative justice programs are based on equity and social justice principles that encourage educators to address inequities and ultimately move towards transforming educational environments into democratic spaces that equalize voices of students, educators, and staff (Winn, 2018).

Because a core tenet of restorative justice programs is being responsive to the goals and needs of the local community, their operations differ between school districts and in some cases between schools within one district. While diverting students away from exclusionary discipline is a key component of many restorative justice programs, it is notably just one facet of restorative justice. Most schools adopt a continuum of practices, many of which are not directly aimed at discipline but instead towards facilitating relational and inclusive school environments (Gonzalez, 2015). In addition to specific restorative practices, schools often adopt disciplinary policy and procedural changes, and researchers have yet to disentangle the effectiveness of the individual practices, policies, and trainings associated with restorative justice. As restorative justice programs have gained traction, school districts have relied on internal workshops, community organizations, and national conferences for assistance in designing and training educators to use restorative practices in schools (Gonzalez, 2016).

While restorative justice practices have been used in schools for decades, the practices were seldomly used in schools that served high proportions of Black and Latinx students (Payne & Welch, 2015). Recently, however, many racially diverse urban school districts have embraced restorative justice practices as school districts seek interventions to disrupt the school-to-prison pipeline (Gonzalez, 2016). This recent widespread trend towards utilizing restorative justice practices in diverse urban school districts provides an opportunity for a deeper understanding of how school-based restorative justice programs can interrupt existing processes and racial disparities in schools. Winn (2018) highlights restorative justice as a promising approach for schools to achieve equitable and transformational change. Organizational transformation, however, is rarely straightforward and attempts may encounter structural barriers that hinder seamless implementation. Evans and Lester (2013) suggest that restorative justice practices may not impact observable student outcomes early in the implementation process, and that it may take three to five years for schools to fully transition to a restorative-oriented school culture.

Importantly, schools that integrate restorative justice into their practices typically do so as an alternative that exists alongside normative discipline policies, rather than as a complete replacement for normative discipline policies (Gonzalez, 2012). This distinction is important because integrating restorative practices within the traditional disciplinary system increases opportunities for discretion to determine the outcome of disciplinary cases. Rather than the completely new disciplinary system based on restorative principals that proponents advocate, restorative justice becomes one among many options, and school staff determine where, when, and with whom to utilize restorative justice based on their discretion and zero-tolerance mandates. This increased discretion potentially leads to challenges, as both discretion and subjectivity contribute to racial disproportionality in school discipline and may not result in a decrease in the use of exclusionary discipline (Skiba et al., 2002; Steinberg & Locoe, 2018).

Despite this increased discretion, recent research on restorative justice in schools shows that the practices are promising for addressing disparities in school discipline. Suspension rates in Pittsburgh, Denver, San Francisco, and Oakland schools decreased after the implementation of restorative justice policies (Darling-Hammond et al., 2020; Augustine et al., 2018; Anyon et al., 2016). Whether the potential benefits of restorative justice are experienced equally by students from different racial groups and how restorative justice policies affect racial disproportionality in school discipline is inconclusive.

In the most rigorous experimental study of restorative justice practices, Augustine et al. (2018) randomly assigned restorative practices to 50% of Pittsburgh Public Schools. Examining a variety of student, teacher and school-level outcomes, Augustine et al. find that restorative practices boosted teacher reports of school climate, reduced overall rates of suspension, and reduced the Black-White suspension gap. Importantly, however, unlike many restorative justice implementations, the SaferSanerSchools Whole-School Change Program adopted in Pittsburgh implements restorative practices independent of the traditional school discipline process, so that offenses that would have received a suspension or expulsion prior to restorative justice should still receive a suspension or expulsion. As the program does not change the process once disciplinary referrals occur, we infer that the reductions in suspension rates in schools using the SaferSanerSchools program are likely attributable either to a change in student behavior or how student behavior is perceived and categorized, so that fewer incidents escalated to the point of discipline referrals.

By contrast, Los Angeles Unified School District implemented restorative practices in conjunction with a suspension ban (which aimed to eliminate suspensions for low-level infractions) and other disciplinary reforms. Hashim et al. (2018) show that suspension rates in Los Angeles declined for all racial groups, but they also find that racial gaps in school suspensions persist despite these overall declines in suspension rates. Similarly, research on Denver, focusing on students who were referred for discipline, suggests that students of all racial groups who received a restorative intervention instead of going through the traditional disciplinary process were less likely to be involved in a disciplinary incident the following semester (Anyon et al. 2016). However, as in Los Angeles, Gregory et al. (2018) suggest that restorative interventions in Denver did little to close racial gaps in suspension rates.

Despite the centrality of mitigating racial disparities in school discipline through restorative justice, existing research on restorative justice in schools thus offers conflicting insight on this question. Although restorative justice practices show promise (Gregory et al., 2016; Lewis, 2009), the extent to which they ameliorate school discipline disparities may be contextually dependent and may take time to realize. Particularly because restorative justice practices are often layered on top of other policies and may not drastically interrupt existing practice, how implementers negotiate between restorative and traditional disciplinary options can determine subsequent outcomes (Ispa-Landa, 2017). Further, as punitive discipline practices continue to operate alongside restorative justice, racialized decision-making and discretion remain pertinent – and perhaps increase in importance – in these contexts.

Although restorative justice programs are often positioned as a tool to reduce racial disparities in school discipline, it is unclear if they are ultimately successful in addressing the racialized and anti-Black processes that lead to these disparities. As educators seek to reconcile traditional disciplinary and restorative justice logics, we might expect that creating a restorative school culture will take additional time in contexts where both are operating. We thus build on prior studies by examining the effects of restorative justice on student discipline over a five-year implementation period, paying close attention to how suspension rates shift for different racial groups over this period.

Restorative Justice in Meadowview

Meadowview Public Schools (MPS) is a large urban school district that first implemented restorative justice programs in 2008 after citywide concerns about discipline disparities became a priority for local schools. Black MPS students were suspended at rates that were more than two times that of White students and were more likely to be disciplined for disruptive offenses. In addition to districtwide goals to deemphasize suspensions, restorative justice practices were implemented to reduce the total number of exclusions (through suspension or expulsion) and reduce the number of police and juvenile justice incidents.

The restorative justice programs operate through the Alliance for Restorative Communities (ARC), a non-profit organization that employs a restorative justice coordinator to help implement restorative justice practices in schools. The restorative justice coordinators are a diverse group of non-profit staff who are extensively trained in areas such as conflict mediation, addressing inequalities in the workplace, and developing restorative justice practices for students. Coordinators also assist schools with developing an implementation plan that fit their school needs, and the specific restorative practices implemented at each school are described in the Appendix. Schools typically utilize a combination of restorative circles, training sessions for teachers and students, and in-class coaching, with elementary schools relying more heavily on relationship-building between educators and students, while middle and high schools focused more on student voice and empowerment. The variety of practices is not uncommon, as schools often create mechanisms to promote restorative justice that administrators feel will fit the school’s context and student developmental needs (Gonzalez, 2012).

Despite these differences, all restorative justice schools in MPS used restorative practices as an alternative to traditional disciplinary punishment. Typically, schools developed procedural guidelines that integrated restorative justice and relied on restorative justice coordinators to handle some disciplinary cases, with the goal of resolving issues through a restorative process before considering traditional disciplinary action. Despite the presence of restorative justice, school administrators maintained some autonomy in determining which student discipline cases were referred to restorative justice and which cases were sent through traditional disciplinary processes. In this way, restorative justice operated as a selectively applied filter in a school’s disciplinary process, aimed at avoiding exclusionary student discipline while providing additional services that would facilitate relationship building and student support. The success of this filter, however, depended on the discretion of school staff referring students to restorative justice, the caseload of the restorative justice coordinator, and the willing participation of students and staff.

ARC worked with school district administrators to select schools for initial restorative justice implementation. The schools that were selected by the district were those that district officials deemed in need of restorative justice practices based on the prevalence of exclusionary discipline and racial disproportionality in those schools. The pilot restorative justice program began at a middle school during the 2008–2009 academic year, with substantial resources allocated to the development of restorative practices in the school. Based on reported success of the pilot, ARC received a grant to pilot restorative justice programs in other schools in MPS. One K-8 school began adopting restorative justice policies mid-year during the 2010–2011 school year, with two high schools and a middle school beginning restorative justice practices at the start of the 2011–2012 academic year, and a third high school adopting restorative practices prior to the 2013–2014 school year.ⁱⁱ Importantly for our purposes, the staggered rollout of the program did not depend on factors that are likely to influence the effectiveness of the program (e.g., school-specific incidents or perceptions of the program’s expected effectiveness). Although we are unaware of any contemporaneous policies that would bias our results (e.g., district policy shifts targeting specific schools), as with all observational studies of educational policies, we cannot definitively rule out alternative explanations.

Data and Methods

Data

Our study uses administrative records containing student disciplinary information in MPS from the 2007–08 to 2016–2017 school years. The student-level data include demographic and academic information that are linked with school identifiers, allowing for the comparison of students in schools (and years) with and without restorative justice programs. Additionally, the discipline data include specific information about each student disciplinary incident. Within each disciplinary incident, the data include identifiers for the involved students, the type of infraction (e.g., attendance violation, fighting, behavioral issues), the type of punishment (e.g., in-school suspension, out-of-school suspension) and the length of punishment (calculated in days missing from school).

To facilitate comparisons between Meadowview schools that did and did not use restorative justice practices, we drop all schools that were specialization schools (e.g., alternative schools and learning centers) or served less than two hundred students. For the purpose of this study, we only classify schools as “restorative justice” when they employ a restorative justice coordinator, received funding for restorative justice implementation, and made changes to student discipline policies. We exclude eight schools with limited restorative justice activity (i.e., they used some restorative justice practices without having an official restorative justice coordinator or policies) from our analytic sample, though supplemental analyses including these schools as non-restorative justice schools yield similar results.

Dependent Variables.

Our primary outcome measure is a dichotomous variable indicating whether a student was suspended in a given year.ⁱⁱⁱ We also estimate models that focus specifically on whether a student was suspended for a low-level behavioral infraction (e.g., insubordination, defiance, disruptive behavior), as previous research suggests that educator discretion may play a particularly prominent role in these suspensions (Skiba et al., 2002).

Independent Variables.

Our key independent variable is the number of years that a particular school has been implementing restorative justice, which we measure using a series of dummy variables. Students in schools that never implement restorative justice are coded as zero, as are students in schools that eventually implement restorative justice, but have not done so yet.^iv Although one school was in its ninth year of restorative justice by the end of our study period, we focus on the first five years of restorative justice implementation, as we have five years of implementation data in all of the schools that implemented restorative justice (coefficients for all years of implementation are reported in our Online Appendix tables).

As restorative justice programs were implemented in MPS to address racial disparities in school discipline between White, Hispanic,^v and Black students, we estimate models that interact our restorative justice indicators with dummy variables for race. These models allow us to test whether restorative justice significantly reduced suspension rates for different race groups, and whether the reductions for non-White groups were different from those experienced by their White classmates. Our models include controls for student gender, socioeconomic status (as measured through free and reduced lunch status), grade level (a series of dummy variables), and special education status. As we discuss below, all models also include school fixed effects and year fixed effects. These fixed effects account for time invariant characteristics of schools, as well as districtwide year-to-year changes. Taken together, these fixed effects allow us to compare the changes in disciplinary outcomes that occur within a given school (we compare schools to themselves in years with and without restorative justice) while accounting for shared temporal fluctuations.

Table 1 reports demographic statistics from the 2007–08 school year separately for schools that did and did not eventually implement restorative justice programs. Table 1 also reports the suspension rates among different groups of students in each set of schools. The data suggest that the schools that implemented restorative justice had higher proportions of Black and Hispanic students and experienced higher rates of suspensions, relative to other Meadowview schools.

Table 1.

Demographic Information and Suspension Rates in 2007–08

	% of student body	Suspension Rate (%)
Panel A: Restorative Justice Schools Prior to Program Implementation
White	53.1	10.7
Hispanic	15.4	18.3
Black	14.6	20.4
Asian	13.5	7.0
Other	3.4	16.4
Male	50.3	17.7
Female	49.7	7.4
Free/reduced price lunch	52.7	16.9
Special education student	11.0	23.3
N (students)	7,282	918
N (schools)	6	6
Panel B: Schools that Did Not Implement Restorative Justice
White	64.9	3.9
Hispanic	16.7	4.9
Black	12.5	13.2
Asian	12.3	2.3
Other	5.9	5.9
Male	50.4	7.3
Female	49.6	2.6
Free/reduced price lunch	41.7	7.9
Special education student	12.5	10.1
N (students)	35,778	1,774
N (schools)	74	74

Note: Table displays descriptive information on students in Meadowview Schools prior to restorative justice implementation. Panel A displays students in schools that used restorative justice in those years. Panel B displays schools that never used restorative justice at any point. While these tables include information for Asian and Other race students, subsequent analyses focus on outcomes for White, Black, and Hispanic students as the program focused primarily on these groups.

Methods

We use a multivariate difference-in-difference approach to test for changes in student discipline as a school implements restorative justice practices. Intuitively, this analysis examines the difference in student suspension rates before and after the implementation of restorative justice and compares this difference with the differences in suspension rates observed in schools that did not implement restorative justice over the same period. As we note above, we focus on how suspension rates change over the first five years of restorative justice implementation, and as such, our model does not assume that restorative justice programs at different points in their implementation have similar effects. Because of the emphasis on decreasing suspension rates in Meadowview, suspension rates may have decreased during this time even if schools did not use restorative justice. Our difference-in-difference approach allows us to account for any trends as well as year-specific fluctuations that affect all schools in that year. Further, because our estimates compare students within the same school before and after implementation, they also account for stable, unmeasured characteristics of the school. To estimate the effects of restorative justice, we estimate a series of linear probability models that take the following general form:

$$Y_{ist}=\beta X_{ist}+{\gamma }_s+{\delta }_t+{\epsilon }_{ist}$$ (1)

where Y_ist represents a series of binary indicators for different disciplinary outcomes for individual i in school s at time t, X_ist are our independent variables, including a series of dummy variables indicating the number of years a restorative justice program had been implemented at school s in time t (schools and years without restorative justice programs serve as the omitted category), as well as the control variables described above, γ_s represent fixed effects for school s, δ_t represent fixed effects for year t, and ε_ist is an error term. In models estimating the differential effects of restorative justice by race, X_ist includes interactions of our indicators for restorative justice and race variables. Our difference-in-differences approach estimates an average treatment effect for the treated schools, which is sometimes referred to as treatment on the treated. As such, results should be interpreted as providing information about the treated schools, rather than implying that untreated schools would experience similar declines.

Statistical significance.

We use randomization inference to calculate whether differences are statistically significant (Heß, 2017). Education researchers have long recognized the importance of correcting standard errors to account for the non-independence of students in schools. However, the standard cluster-robust estimators generally employed may not be well-suited for difference-in-difference estimators, particularly when the number of treated clusters is small compared to the total number of clusters (Young, 2017). Likewise, in some contexts wild bootstrapping requires sub-cluster resampling to obtain consistent estimates (Roodman 2018). By contrast, randomization inference works well in such contexts, allowing us to randomly re-assign the treatment (i.e., restorative justice programs) to different cases (i.e., schools) and to compute the probability of obtaining the observed results if restorative justice had no effect on student outcomes. Utilizing this approach allows us to estimate rigorous p-values that account for the clustering of students within schools as well as the other complexities of our case.

Limitations.

Although this study provides novel and important insights into racial disproportionality under restorative justice, it also has several limitations. First, beyond descriptive accounts of restorative justice activity in each school contained in the Appendix, we cannot account for variation in school-level implementation fidelity. This limitation is important because even though we can use fixed effects to compare schools to themselves pre- and post-implementation, we are unable to evaluate differences in how—and to what extent—schools are engaged and integrating restorative justice into their practices, and how this might affect outcomes. Likewise, it is unclear which students received restorative justice interventions and the extent of the interventions they received. If collected, this data would likely provide nuance and specificity to the patterns that we find in Meadowview. Finally, while we have no reason to believe that our results are particularly idiosyncratic, as with all case studies it is unclear how broadly generalizable our findings are. Given the focus on being responsive to local contexts in restorative justice programs, and the case-to-case variation that this implies, understanding how the processes described here play out in other settings is an important avenue for future research.

Results

Figure 1 displays suspension rates in each restorative justice school before and after program implementation, as well as the average suspension rate for schools that did not implement restorative justice during the period covered by our data. Each gray line represents a single school, with the solid gray lines representing the pre-restorative justice observations for that school, and the dashed gray lines representing post-restorative justice suspension rates. The solid black line represents the average suspension rate across schools that did not implement restorative justice between 2008 and 2017. As is evident in Figure 1, we see a relatively uniform decrease in suspension rates across the schools that implement restorative justice, suggesting that suspension rate declines under restorative justice in Meadowview are not unique to specific schools.

Figure 1:

Suspension Rates Before and After Restorative Justice Implementation

Note: Solid gray lines represent schools in years without restorative justice practices, while dashed lines represent years with restorative justice practices. The solid black line represents Meadowview schools that never implemented restorative justice.

It is important to ensure that the results from our difference-in-difference models that follow do not simply reflect the fact that schools that eventually adopt restorative justice differ from the schools that do not, even before the eventual restorative justice adopters begin implementation. Figure 2 thus presents the results from an event study model based on our difference-in-difference specification, which allows us to compare the pre-restorative justice suspension rates in schools that eventually adopted restorative justice with the suspension rates in schools that did not adopt restorative justice. The y-axis indexes the coefficient from our event study model, which represents the degree to which the schools that eventually adopt restorative justice differ from those who never do so (see Online Appendix Table A1 for full model results). The x-axis indexes the year, relative to the implementation of restorative justice.

Figure 2:

Average Difference in Suspension Rates Between Restorative Justice and non-Restorative Justice Schools pre- and post-Implementation (w/95% CI)

Note: Figure 2 displays results from an event study model of restorative justice implementation. The solid black line represents model coefficients (i.e., the average difference between schools that did and did not implement restorative justice) and the grey-shaded region represents the region between the 2.5^th and 97.5^th percentiles of the null distribution. Coefficients that fall outside this region are indicative of differences that are statistically significant (i.e., the coefficients diverge from what one would expect if restorative justice had no effect).

Because the six schools that implemented restorative justice did so at different points in time, we have different numbers of pre- and post-implementation years (e.g., only one school had more than three pre-implementation years, and only two schools had more than five post-implementation years); we thus include the number of restorative justice schools represented by each estimate along the x-axis. The solid black line represents the coefficient from our event study model, and the grey shaded region represents the region between the 2.5^th and 97.5^th percentiles of the null distribution. Coefficients that fall outside this region represent differences that are statistically significant (i.e., the coefficients diverge from what one would expect if restorative justice had no effect). The results presented in Figure 2 suggest that pre-implementation differences are unlikely to be driving our results, as conditional suspension rates in restorative justice schools prior to implementation are not statistically significantly different from schools that never implement restorative justice, and suspension rates only begin to diverge after program implementation.^vi

Our results in Figures 3–5 build on Figures 1 and 2 by examining how suspension rates changed in the first five years of a school’s restorative justice implementation. These figures display predicted suspension rates based on our models; the corresponding model coefficients are reported in Online Appendix Tables A2-A4, which also contain the results from analogous models that separate in-school suspensions and out-of-school suspensions.^vii

Figure 3:

Suspension Rates by Length of Restorative Justice Implementation

Note: Figure 3 displays predicted suspension rates for schools in different years of restorative justice implementation. The predicted probabilities are based on model results reported in Online Appendix Table A2 and control for student socioeconomic status, gender, special education status, grade-level fixed effects, year fixed effects, and school fixed effects. Statistical significance is calculated using randomization inference.

* p<0.05, ^† p<0.10

Figure 5:

Insubordination Rates by Length of Restorative Justice Implementation, Separately by Race

Note: Figure 5 displays predicted suspension rates for insubordination by race for schools in different years of restorative justice implementation. The predicted probabilities are based on model results reported in Online Appendix Table A4, and control for student socioeconomic status, gender, special education status, grade-level fixed effects, year fixed effects, and school fixed effects. Statistical significance is calculated using randomization inference. An uppercase “A” indicates that the difference between the baseline suspension rate (year 0) for a particular race group and their suspension rate in a given year of restorative justice implementation is statistically significant (p<.05). An uppercase “B” indicates that the difference between the baseline suspension rate and the suspension rate in a given year of restorative justice implementation for a particular race group is significantly different (p<.05) from the analogous difference for White students (i.e., the interaction effect in Online Table A4 is statistically significant). Lowercase letters denote marginally significant differences (p<.10).

Figure 3 depicts predicted probabilities from a model that estimates how suspension rates changed over time following the implementation of restorative justice programs. Specifically, the first bar reports the percentage of students who were suspended in schools where restorative justice was not in place, and each subsequent bar reports the predicted suspension rate when restorative justice had been in place for one, two, three, four, and five years. In schools without restorative justice, 5.1% of students were suspended in any given year. We see a similar suspension rate of 4.6% in schools that are in their first year of implementation. In the second and third years of implementation, by contrast, we find suspension rates of 3% and 2.4%; the differences between these suspension rates and the suspension rates of schools without restorative justice are marginally significant (p<.10), suggesting that restorative justice begins to lower suspension rates in its second and third years. The final two bars indicate that suspension rates continue to decline in the fourth and fifth year of restorative justice programs, and these suspension rates are significantly different from the suspension rates in schools without restorative justice programs (p<.05). Our results thus suggest that restorative justice programs do not immediately lower suspension rates, but that over a period of several years, restorative justice programs can result in profound reductions in the suspension rate.^viii

Figure 4 reports analogous findings for models predicting race-specific changes in suspension rates. Presenting race-specific results allows us to examine: 1) whether restorative justice realizes its intended benefits across different groups, and 2) whether the suspension rate reductions associated with restorative justice vary by race. Figure 4 thus reports two sets of significance tests, with an “A” denoting suspension rates that are significantly lower than for same-race students in schools without restorative justice programs, and a “B” denoting that the reduction in suspension rates achieved in a particular year of implementation differs from the reduction for White students in that year. We use lowercase letters to denote marginal statistically significant differences (p<.10). Our results for White students largely follow those observed in Figure 3, with the notable difference that White students’ suspension rates exhibit a marginally significant decline even when schools are in their first year of restorative justice implementation. We see a qualitatively similar pattern for Hispanic students, although the effects of restorative justice for Hispanic students are not statistically significant until years four and five of implementation. Among Asian students, by contrast, we find that suspension rates fell to zero early in the implementation of restorative justice. Thus, although the precise pattern of year-to-year reductions in suspension rates that accompanied the implementation of restorative justice varies across White, Hispanic, and Asian students, suspension rates for these students under restorative justice were substantially (and significantly) reduced from their baseline rates, and suspension rates for these students were near zero by the fifth year of restorative justice implementation.

Figure 4:

Suspension Rates by Length of Restorative Justice Implementation, Separately by Race

Note: Figure 4 displays predicted suspension rates by race for schools in different years of restorative justice implementation. The predicted probabilities are based on model results reported in Online Appendix Table A3, and control for student socioeconomic status, gender, special education status, grade-level fixed effects, year fixed effects, and school fixed effects. Statistical significance is calculated using randomization inference. An uppercase “A” indicates that the difference between the baseline suspension rate (year 0) for a particular race group and their suspension rate in a given year of restorative justice implementation is statistically significant (p<.05). An uppercase “B” indicates that the difference between the baseline suspension rate and the suspension rate in a given year of restorative justice implementation for a particular race group is significantly different (p<.05) from the analogous difference for White students (i.e., the interaction effect in Online Table A3 is statistically significant). Lowercase letters denote marginally significant differences (p<.10).

The results for Black students diverge markedly from those observed for other groups. Notably, although concerns regarding the disproportionate suspension of Black students motivated the adoption of restorative justice programs, we find little evidence that they had their intended effect among Black students. In contrast to other groups, we find that suspension rates for Black students are not statistically significantly different from their baseline levels within the first five years of restorative justice implementation. Indeed, suspension rates for Black students in schools that were in the first three years of implementing a restorative justice program were, if anything, higher than the suspension rates for Black students in schools without restorative justice. As Black students’ suspension rates in the fourth and fifth year of restorative justice implementation are lower (though the difference is again not statistically significant) than in the first three years, it is plausible that Black students may eventually realize statistically significant reductions in suspension rates as restorative justice programs mature. Nonetheless, our results suggest that unlike their White, Hispanic, and Asian classmates, Black students in schools with restorative justice programs do not experience dramatic reductions in their suspension rates.^ix Indeed, it is noteworthy that the expected suspension rate for Black students in a school in its fifth year of restorative justice implementation (6.1%) is still slightly higher than the baseline suspension rates of White (4.6%) and Hispanic (4.9%) students.

If we conceptualize racial disproportionality as the ratio of Black and White suspension rates (i.e., relative risk), we find significant and stark differences. That is, although the five percentage point difference in the suspension rates for White (4.6) and Black (9.6) students at baseline is similar to the 5.9 percentage point gap at schools in their fifth year of implementing a restorative justice program (0.2 vs. 6.1), ^x the changes in the relative risks of suspension are substantial and statistically significant. Although Black students are just over twice as likely as White students to be suspended at baseline (9.6/4.6), in schools in their fifth year of restorative justice implementation, Black students were approximately 30 times more likely to be suspended than White students (6.1/0.2). We use randomization inference to conduct supplemental tests of these relative risk ratios, finding that Black-White disproportionality is significantly higher than one would expect in schools that have had a restorative justice program for three or more years. Likewise, the growth in disproportionality (comparing the relative risk in a given year of restorative justice implementation to the relative risk at baseline) is also statistically significant for schools that have had a restorative justice program for three or more years. These differences are driven by the decreases in the White suspension rate, which means that even if Black students are suspended at lower rates than they are at baseline, the vast majority of suspended students in the schools in their fifth year of restorative justice implementation are Black.

As we elaborate in our discussion, we interpret these results as underscoring the importance of how restorative justice is implemented. In Meadowview schools, restorative justice was implemented as a parallel disciplinary track, which introduced the potential for educator discretion in choosing who was referred to restorative justice. Although it is promising that restorative justice lowers the suspension rates of non-Black students and may eventually begin to lower Black students’ suspension rates, we suspect that because incorporating restorative pathways into an existing discipline system increases the potential for educator discretion, it is unlikely to realize significant gains for Black students. Rather, as we discuss below, we believe that restorative justice is most likely to achieve its transformative potential when implemented as a replacement to the traditional ethos and practice of discipline, rather than as a supplement.

We build on the results presented in Figure 4 by examining whether we find similar patterns for offenses related to insubordination (e.g., defiance and other non-violent behavioral infractions). Although we would expect restorative justice programs to reduce all types of disciplinary incidents, reductions might be most readily apparent in incidents without zero-tolerance mandates where teacher and administrator discretion regarding discipline are greater (cf. Steinberg and Lacoe, 2018). These discretionary incidents are also important given racial disparities in these incidents and concerns that these offenses might contribute to later racial disparities (Fabelo et al., 2011). Figure 5 thus replicates the analyses presented in Figure 4, but focuses specifically on suspensions for insubordination.

We find that White, Hispanic, and Asian students’ suspension rates for insubordination drop under restorative justice implementation, and that these reductions are visible already in the first year of implementation (though for all three groups this reduction is initially only marginally significant). For these students, we see that restorative justice programs eliminate suspensions for insubordination. This suggests that for non-Black students, restorative justice programs can help eliminate suspensions by changing the way that educators use their discretion for lower-level disciplinary incidents.

For Black students, by contrast, we do not observe statistically significant reductions in suspension rates for insubordination. Interestingly, although Figure 4 showed that the initial years of restorative justice implementation were characterized by overall suspensions rates that were if anything higher than baseline (though not significantly so), we see no evidence for an initial rise in insubordination suspensions in Figure 5. Although the reductions for Black students are not statistically different than those observed for White students, it is noteworthy that Black students are the only group that does not experience a statistically significant decrease in insubordination suspensions and are the only group that continues to be suspended for insubordination in schools with restorative justice programs. Thus, while the results for insubordination suspensions are more encouraging than the results for suspensions writ large, particularly in the initial years of restorative justice implementation, they nonetheless suggest that Black students did not realize the benefits of restorative justice programs in Meadowview schools.

Discussion

From one perspective, the restorative justice practices implemented in MPS were successful: schools that implemented restorative justice saw marked decreases in their suspension rates. These findings are consistent with prior studies (Anyon et al., 2016; Augustine et al. 2019; Hashim et al., 2018) and suggest that restorative justice practices in MPS grew more effective as the programs matured. Despite the overall reduction of exclusionary discipline under restorative justice, however, the pattern of reductions observed across racial groups are striking. Although the restorative justice policies were implemented to improve disciplinary practices that were disproportionately harming Black and Latinx students, it failed to significantly reduce exclusionary discipline for Black students. By reducing exclusionary discipline for White but not Black students, restorative justice significantly increased disproportionality. While we cannot test our explanation, informal discussions with ARC staff suggest that the racial equity facet of restorative justice was slower to gain traction in Meadowview schools, which may contribute to the racial disparities we observe.

This pattern of results is perhaps attributable to two key facets of how restorative justice operated in Meadowview schools.^xi First, the autonomy of school administrators to dictate implementation may have introduced an element of discretion that may have led schools to embrace certain facets of restorative justice and not others. While allowing the school community to play an active role in determining how restorative justice will operate in their schools is common (Gonzalez, 2012), this collaborative flexibility might lead to a disconnect between the priorities and capabilities of school administrators and the restorative justice philosophy. For example, restorative justice was instituted in MPS for race-specific reasons, but informal conversations with ARC staff and a review of the district’s restorative justice handbook reveals that the policies included minimal race-specific material. Thus, the extent to which a school engaged with the racial equity components of restorative justice largely depended on the willingness and discretion of school staff.

Though unintentional, using language that is race-neutral and colorblind makes it difficult for schools to implement restorative practices in ways that target the specific needs of the populations that they are being implemented to help. While racial equity was a primary goal of ARC implementors, it is possible that those messages were diluted in the policies and practices that schools embraced. Lewis (2003) details the role of colorblind ideology in furthering the advantage of White students even in schools that claimed to be racially progressive. The extension of advantage is far from intentional in these spaces, but the absence of race-specific language in schools results in a perpetuation of racially disparate treatment and discipline practices (Lewis and Diamond, 2016). Research in other spheres finds that identity-conscious structures — and not identity-blind structures — were positively associated with employment outcomes for people of color (Konrad and Linnehan, 1995).

The framing of restorative justice may have contributed to the structural operation of restorative justice practices in Meadowview. Karp and Breslin (2001) find that schools integrating restorative justice have different languages for restorative practices, with some that even omit the word “justice.” They also found that schools often experience external and internal resistance to restorative justice, leading to substantial variation in implementation (c.f., Dusenbury et al., 2003). Moreover, the relationship between restorative justice consultants (i.e., organizations like ARC) and school implementers should be examined in future research, as the differing goals of these schools and restorative justice organizations may impact successful implementation of restorative justice programs (Song and Swearer, 2016). It is possible that the social justice frameworks typically used by restorative justice advocates like ARC might be misinterpreted, selectively used, or even rejected as schools attempt to integrate restorative justice into their existing processes and culture.

Such selective implementation may result in schools implementing restorative justice practices that differ from the original restorative justice philosophy, or that merely focus on a narrow subset of the broader goals of restorative justice. For example, a school might seek to: 1) reduce exclusionary discipline practices and supplement them with restorative practices; 2) reduce racial disproportionality in discipline and achievement outcomes; and/or 3) improve social relationships and build community through restorative practices. While each of these three goals derive from the restorative justice philosophy, they likely require different strategies, and some goals may be easier to reach than others (e.g., reducing suspensions but not reducing racial disproportionality). As Mcluskey and colleagues (2008) show, variability in restorative justice adoption is possible, as teachers and administrators may vary in their perceptions of what it means to be “restorative.”

In contexts where there is a lack of consensus about the goals of adopting restorative justice, school staff may embrace a variety of approaches that although well-meaning, do not drastically interrupt the systemic racialized processes that produce disproportionality in school disciplinary outcomes. Because anti-Black attitudes in school environments influence disparities in Black student discipline (Chin et al., 2020), taking an indirect approach to addressing racial disparities may continue to reinforce the subordination of Black students. Moreover, while restorative justice is not new, the racial equity components of restorative justice are a relatively new integration into the philosophy. This can result in what has been described as the paradox of restorative justice, whereby aracial approaches may be used in response to race-specific problems (Gavrielides, 2014). Given the pervasiveness of anti-Blackness and its role in shaping educator attitudes and racial disparities in student outcomes (Turetsky et al., 2021), it is likely that interventions such as restorative justice will struggle to help Black students if they do not directly address both systemic and interpersonal anti-Blackness.

A second factor that might help explain the racial disproportionality we observe is the implementation of restorative justice within the existing disciplinary system. Rather than establishing restorative justice as a system that consists of an autonomous set of norms regarding school discipline, restorative justice practices in Meadowview are embedded within the traditional disciplinary system and are likely subject to the same processes that lead to racially disparate school discipline outcomes. The implementation of restorative justice as an appendage to the traditional disciplinary system suggests that in these schools, restorative justice is unlikely to be able to fundamentally transform disciplinary structures. Ispa-Landa (2017) highlights that restorative justice programs often coexist with other initiatives, and among them are normative disciplinary practices, zero-tolerance mandates, and increased police presence. In such contexts, it is possible that restorative justice practices become an alternative disciplinary trajectory, and that who is placed on that trajectory is informed by educator discretion in disciplinary cases.

Introducing restorative justice practices as an alternative disciplinary track likely increases the opportunities for educator discretion. For example, a restorative process discipline chart from an MPS school (see Appendix Figure 1) shows that it is still possible for students to be referred back into the traditional disciplinary system at each step of the restorative process. As increasing opportunities for discretion have been shown to increase Black-White gaps in the criminal justice system (Yang, 2015), it is likely that discretion plays a role in the outcomes that we observe. Even if this restorative disciplinary track functions optimally, the role of individual discretion remains pertinent in the initial referral to the restorative justice process. While restorative justice is a promising alternative to exclusionary discipline, personnel in restorative justice schools are not exempt from the implicit biases, differential perceived threat, and cultural dissonance that all lead to race-based disciplinary disparities in schools and elsewhere (Eberhardt et al., 2004; Levinson, 2007; Okonofua & Eberhardt, 2015; Turetsky et al., 2021).

To this end, it is noteworthy that our findings suggest that the effects of restorative justice programs change over time. While proponents of restorative justice have described the philosophy as a transformational and slow-moving process, much of the research has focused on pre/post implementation differences and examined outcomes within the first three years of implementation. Such approaches may be suitable for other school policies, but they likely do not allow schools enough time to realize the full impacts of transitioning to restorative justice programs.

In this vein, it is perhaps promising that Black students begin to experience notable, though statistically insignificant, reductions in suspension rates beginning in the fourth year of implementation. This may suggest that restorative justice practices are slower to reach Black students, but that longer-term implementations may mitigate discrepancies as schools become more familiar with restorative justice.^xii If equity-focused work manifests at a slower pace than school procedural shifts, it is possible that the final effects of restorative justice on racial disproportionality may lag for a number of years after initial implementation. As Asian and White students experienced the quickest reduction in suspension rates under restorative justice, the potential for restorative justice to mitigate discipline disproportionality in contexts like Meadowview, where implementation increased discretion, may require a longer time horizon than is typical adopted. This echoes work examining school turnaround efforts which indicates that whole-scale changes to school culture and support emerge gradually and take three or more years before improvements are fully realized (Sun, Penner, & Loeb, 2017).

Restorative justice implementers face the challenge of interrupting the decades-long organizational processes that created the school-to-prison pipeline. Any conclusions about the effectiveness of restorative justice practices in schools should thus be considered alongside the malleability of the existing racial structures both within schools and in society more broadly. Further, restorative justice was originally conceptualized as a replacement for existing punitive disciplinary practices, and as such, implementing restorative justice by layering restorative processes alongside a punitive disciplinary structure is likely to pose substantial challenges and yield unintended consequences. Particularly in policy contexts characterized by a legacy of anti-Blackness, incorporating restorative justice interventions into a disciplinary system governed by a willfully colorblind logic (or racial apathy) may struggle to improve outcomes for Black students (Mueller, 2017).

Restorative justice in MPS operated like a filter that was successful in diverting many students away from exclusionary discipline; however, the filter was not successful in transforming the underlying racial logics of punishment and anti-Blackness that place Black students at heightened risk of experiencing exclusionary discipline. While we do not attribute these trends to the core tenants of restorative justice, they are noteworthy in light of recent qualitative work that highlights the harm Black students may experience under seemingly progressive school policies (Shange, 2019). Future work should consider the ways in which restorative justice programs are integrated into the organizational processes that are already at play in schools, as these processes are racialized and may drive inequality with or without interventions (Ray, 2019).

Conclusion

Our findings show that while the overall effects of restorative justice are promising for lowering suspension rates, they were not particularly effective in ameliorating persistent racial discipline gaps in Meadowview, and thus served to increase disproportionality. We interpret these findings not as an indictment of the restorative justice philosophy, but rather as highlighting the challenges of addressing systematic and multi-layered racialized inequalities in school discipline, even when using promising policies. The persistence of racial disproportionality under restorative justice points to the need for further evaluation of how organizational practices can influence both the structural and individual conditions under which reform efforts such as restorative justice operate. As the restorative justice framework is broadly focused on inclusion, implementers should be intentional about maintaining and emphasizing the equity components of the philosophy at the structural level. Our findings thus underscore both the promise of restorative justice practices in schools, as well as the possibility that the racial equity intentions of restorative justice can be diluted as schools integrate restorative justice into the colorblind logics that often govern their day-to-day operations.

Biography

Miles Davison is a Doctoral Candidate at the University of California, Irvine; 4215 Social Science Plaza B, Irvine, CA 92697; mdaviso1@uci.edu. Davison’s research focuses on how school policies and interventions impact racial disparities.

Andrew M. Penner, PhD, is Professor of Sociology at UC Irvine. Penner’s research focuses on inequality, social categorization, and educational policy.

Emily K. Penner is Assistant Professor of Education Policy and Social Context in the School of Education at the University of California, Irvine. Penner is a William T. Grant Scholar whose research focuses on K-12 education policy, and considers the ways that districts, schools, teachers, and families contribute to and ameliorate educational inequality.