Abstract
There is growing concern that suspensions trigger a ‘‘downward spiral,’’ redirecting children’s trajectories away from school success and toward police contact. The current study tests this possibility, analyzing whether and in what ways childhood suspensions increase children’s risk for juvenile arrests. Combining 15 years of data from the Fragile Families and Child Wellbeing Study with contextual information on neighborhoods and schools, I find that suspensions disproportionately affect children already enduring considerable adversity. Even so, suspensions appear to redirect children’s trajectories, more than doubling their risk of arrest. Although suspended children experienced greater escalations in behavioral problems than their peers, post-suspension behavioral changes explained relatively little of the association between early suspension and later arrest. Instead, the most consequential way suspended children diverged from their peers was their heightened risk for repeated school sanction. Suspended children’s risk for repeated school removal explained 52 percent of the association between childhood suspension and juvenile arrest.
Keywords
The “school-to-prison pipeline” has held a prominent place in education and juvenile justice research for nearly two decades, occasioning at least 50 symposia, 16 books, and 294 journal articles (McGrew 2016). Although the school-to-prison pipeline concept has been used to describe a variety of processes, school discipline plays a central role across accounts (Skiba, Arredondo, and Williams 2014). In particular, many argue that school suspensions create a “turning point” in children’s lives, redirecting their trajectories in consequential ways (Mowen and Brent 2016). As one early report put it, suspensions “create a downward spiral in the lives of these children, which ultimately may lead to long-term incarceration” (The Advancement Project and The Civil Rights Project 2000:13).
Prior research has demonstrated that adolescents who experience suspensions also face higher rates of legal sanctions, both as teens (Fabelo et al. 2011; Monahan et al. 2014; Mowen and Brent 2016) and adults (Arum and Beattie 1999; Wolf and Kupchik 2017). Nevertheless, how this relationship comes to emerge in students’ lives is unclear. Does the suspension-arrest relationship simply reflect the growing police presence in schools, such that disciplinary incidents now also involve arrests? Do suspensions promote arrests by temporarily removing the incapacitation effect of school attendance for already delinquent teens? Or—as the “downward spiral” metaphor implies—do suspensions themselves actually have an enduring effect on students’ lives, redirecting their long-term trajectories in ways that ultimately lead to arrest?
From the perspective of wanting to stanch the flow of youth into the juvenile justice system, all of these mechanisms are a cause for concern. It is this last possibility, however, that is the most troubling. If suspensions alone redirect children’s lives away from school success and toward legal entanglement, then this implies that suspensions do more than simply punish delinquency. They actually help produce it.
Existing studies have had trouble assessing this possibility because they only begin tracking students in adolescence. By adolescence, many students would have already faced years of escalating school sanctions. Moreover, the incidents that result in teenage students’ removal from school may themselves involve legal consequences. For teens already caught in a vicious cycle of school and legal sanctions, retrospectively assessing whether suspension itself redirected students’ trajectories may simply not be possible.
The current study avoids these problems by tracking students’ disciplinary trajectories beginning in childhood. By analyzing childhood suspensions that occurred years before teens’ first arrests, I am able to exclude the possibility that the suspensions being studied promoted arrests simply through simultaneous sanctions or short-term shocks. Instead, I analyze whether and in what ways childhood suspensions redirect students’ long-term trajectories toward future juvenile justice contact.
Combining 15 years of data from the Fragile Families and Child Wellbeing Study with contextual information on schools and neighborhoods, I address three key questions. First, I ask: How do individual, school, and neighborhood factors structure children’s risk for experiencing a suspension? Second, using propensity score matching methods, I ask: Among teens who faced the same risk of an early suspension, what is the difference in juvenile arrest rates between those who were and were not suspended in childhood? Finally, applying a recently developed decomposition technique, I ask: Of all the ways that suspended children’s lives diverged from their matched peers, which are most consequential in explaining their rates of arrest? At a time when juvenile arrests can permanently reshape teens’ risks and opportunities, this study’s results have implications not only for our understanding of school discipline but also for broader theories of cumulative disadvantage and the path to imprisonment.
Background
The Criminalization of Childhood in Juvenile Justice and School Discipline
The emergence of the school-to-prison pipeline concept is a reflection of the profound changes that have occurred in both juvenile justice administration and school discipline over the past three decades. Historically, America’s juvenile justice system 1 was built on the belief that children are less culpable for their crimes and more capable of rehabilitation (Feld 1999). If adult criminal proceedings are intended to function as public “status degradation ceremonies” (Garfinkel 1956), then juvenile proceedings were designed to prevent precisely this outcome. Several safeguards were built into the juvenile system to limit the stigmatization of the child and protect their ability to succeed in adulthood.
Reflecting the emerging “tough on crime” political climate, several states began to change their approach to juvenile crime in the late 1970s (National Research Council 2001). The punitive turn in juvenile justice administration truly took hold, however, following the violent crime wave that began in the mid-1980s and peaked in the early 1990s. During that time, a historic surge in teen gun violence—violence that was popularly associated with black boys enlisted in the sale of crack cocaine—produced a potent moral panic (Blumstein 1995; Cook and Laub 1998; Feld 1999). Warnings of a new breed of “superpredators”—“hardened, remorseless juveniles . . . who pack guns instead of lunches” (Dilulio 1995)—captured the public imagination.
In response to these events, every state across the country instituted sweeping changes in their policies governing the arrest, prosecution, and detention of juveniles (National Research Council 2001). By the turn of the century, America’s juvenile incarceration rate was the highest in the world: five times higher than that of the next closest country, South Africa (Aizer and Doyle 2015). Even after juveniles complete their sentences, juvenile legal records—historically sealed to protect teens’ futures—are now publicly available in most states. These records are now used to deny youth access to higher education, employment, and a variety of public benefits (Juvenile Law Center 2016).
At the same time that states were restructuring their juvenile justice systems, school districts were changing the way they defined, detected, and responded to student misbehavior (Kupchik 2014). Many argue that schools now understand and address student behavior through the prism of crime control, focusing less on rehabilitation and more on incapacitation and deterrence (e.g., Hirschfield 2008; Simon 2007). The most visible expression of this trend is the growing police presence in schools. In 1975, only 1 percent of all schools reported having an officer stationed in their building (Na and Gottfredson 2013). By 2014, 24 percent of elementary schools and 42 percent of high schools reported having full-time police officers (U.S. Department of Education 2016). These officers are supported by a growing surveillance architecture, particularly in schools that primarily serve poor students of color (Kupchik and Ward 2014).
School-based arrests represent the hard edge of punitive control in schools. However, the same punitive approach that undergirds school arrests is arguably expressed more broadly in schools’ use of exclusionary discipline. In a recent national cohort, 56 percent of black students, 39 percent of Hispanic students, and 30 percent of white students were suspended at least once during their school careers (Shollenberger 2015). Children’s cycles of disciplinary sanction start early: Among those who ever faced a suspension in this cohort, 27 percent of all children—and 39 percent of black boys—had already been suspended at least once before the age of 12.
Much of the recent public attention to school discipline has centered on the zero-tolerance policies adopted nationwide after the federal Gun-Free Schools Act of 1994 (e.g., Curran forthcoming). However, the automatic punishments specified under zero-tolerance policies are just one part of the broader formalization of disciplinary procedures (Arum 2003). In the same way that the discretion of juvenile court judges has become more constrained by statutory exclusions and sentencing guidelines (Feld 1999), “school punishment is increasingly based on uniform procedural and disciplinary guidelines evolving around the nature of the offense rather than the discretion of teachers” (Hirschfield 2008:81–82).
Building on preexisting trends, the No Child Left Behind Act required all districts to institute formal codes of conduct that serve as sentencing guidelines for student misbehavior (U.S. Department of Education 2007). If zero-tolerance policies reflect the logic of mandatory minimum sentencing, then these broader codes often resemble three-strikes laws: Even minor violations, if repeated, attract increasingly harsh consequences. Students who commit rule violations become institutionally “marked” (Kohler-Hausmann 2013) in official records, making any future indiscretion subject to escalating sanction.
Together, all of these changes have had the effect of dramatically magnifying the consequences of early misbehavior. More than ever before, children’s prior sanctions can continue to structure their treatment across the institutions that govern their lives. All told, Bishop and Feld (2012:2770) conclude, “the harshness with which the United States responded to youthful misconduct was unparalleled.”
Three Paths from Suspension to Arrest
Against this backdrop, recent accounts of school discipline and juvenile policing present a picture of students governed by overlapping systems of penal control. “The lives of impoverished urban students,”Nolan (2011:72) argues, “are managed by a complex interpenetration of systems. The school . . . becomes an auxiliary to the criminal justice system.” Similarly, Shedd (2015) argues that schools are part of a “universal carceral apparatus” impacting all parts of children’s lives. Rios (2011) situates schools within a broader “youth control complex.” Given the situation that these authors describe, it can be difficult to disentangle the overlapping paths connecting discipline and arrest in children’s lives. As a way of integrating current evidence, I propose three basic mechanisms through which suspended children could come to face higher rates of arrests: selection bias, simultaneous sanctions, and downward spirals.
The first challenge that research on the consequences of school discipline must confront is selection bias: the possibility that the same factors that caused the child to be suspended may also be responsible for their other negative outcomes. For many students, school sanctions may simply be a leading indicator of an underlying pattern of behavior that would have resulted in arrest regardless of any discipline decision made by school officials. This individual-level selection problem is compounded by the fact that both exclusionary discipline and police contact concentrate in the same contexts. Whether measured by stated policies (Welch and Payne 2010) or actual rates of discipline (Ramey 2015), there is strong evidence that exclusionary discipline is disproportionately applied in schools that primarily serve poor students of color. In turn, these schools are likely to draw from disadvantaged neighborhoods that disproportionately bear the burden of police contact (Sampson and Loeffler 2010). Therefore, as a result of both individual and contextual factors, selection bias could produce a strong association between suspension and arrest even if suspensions have no causal effect on the child’s life.
Even if the relationship between suspension and arrest is not accounted for by selection bias, this relationship could still reflect simultaneous sanctions resulting from a single incident. By simultaneous sanctions, I mean to describe the institutional processes through which a single event comes to be punished with both school sanctions and legal sanctions. Nolan (2011), for example, documents that in schools where the police have responsibility for student discipline, incidents that result in a suspension also regularly yield an arrest and court summons. Increasingly, though, even when arrests occur outside of school, school sanctions are triggered as well. As restrictions on juvenile record sharing have fallen, the majority of states have come to require that law enforcement inform schools whenever a student has contact with the justice system. Using this information, some districts now mandate suspensions for any student who experiences an arrest (Juvenile Law Center 2016).
From the perspective of understanding the processes that facilitate the flow of youth into the juvenile justice system, documenting these simultaneous sanctions is an important task. Nevertheless, from the perspective of identifying whether suspensions themselves redirect children’s trajectories, it is critical to ensure that the suspension being studied did not also involve an arrest. It is this final possibility—that suspension in and of itself redirects children’s trajectories in enduring ways—that I call downward spirals. Borrowing from the Opportunities Suspended report quoted previously (Advancement Project and The Civil Rights Project 2000), I adopt the image of a downward spiral to signify the dynamic and reinforcing processes that may unfold following a suspension.
The broader literature on exclusionary discipline suggests a number of ways in which suspensions could trigger a downward spiral in students’ lives. For example, the loss of instructional time that a child faces as a result of being removed from the classroom could magnify existing academic problems, causing them to fall behind their peers and disengage from future schoolwork (Morris and Perry 2016). If the child perceives that their suspension was unfair, this could change their beliefs about school authority and undermine their connection to the normative culture of the school, encouraging further disruptive behavior (Preiss et al. 2016). Apart from the immediate punishment of a suspension, the lasting record of being a student in need of discipline could continue to indirectly shape children’s treatment. Histories of behavioral sanction appear to impose salient identities on students within their school community, with “troublemakers” (Ferguson 2000) or “frequent flyers” (Kupchik 2010) being well known and singled out for special scrutiny. In practice, these and other processes are likely to unfold in tandem, dynamically reinforcing one another in a downward spiral over time.
Existing Evidence and Study Contributions
Early studies articulating the idea of a school-to-prison pipeline did not attempt to address selection bias. Instead, they appealed only to the common racial disparities observed in both school discipline and juvenile detention, arguing that the patterns “within the two systems are so similar—and so glaring—that it becomes impossible not to connect them” (Wald and Losen 2003:11). Other studies relied on retrospective reports from teens already under custodial control, identifying prior exclusionary discipline as just one factor among many that may have led to teens’ confinement (Sedlak and McPherson 2010). Currently, only a handful of studies have used the kind of longitudinal data necessary to identify whether suspensions independently increase students’ risk for later legal sanction.
The most commonly cited study in this area is Fabelo et al.’s (2011) analysis of school discipline and juvenile justice records in Texas. In a report sponsored by the Council of State Governments Justice Center, Fabelo et al. follow three cohorts of seventh-grade students (2000, 2001, and 2002) for at least six years. Controlling for prior achievement and other student- and school-level factors, they report that students who experienced a suspension or expulsion were more likely to have subsequent juvenile justice contact than their peers. However, because this report relies exclusively on administrative data, its authors have no direct measures of student behavior. As such, it remains possible that a stable pattern of behavior caused both the discipline and the juvenile justice contact. If discipline did redirect teens’ lives toward juvenile justice contact, Fabelo et al. are unable to address the mechanisms through which this process unfolded.
Other research has examined the relationship between school discipline and legal sanctions using longitudinal surveys of adolescents. Two recent studies address concerns about selection bias by analyzing within-respondent changes in the relationship between adolescent suspensions and arrests. Using four waves of the National Longitudinal Study of Youth 1997, Mowen and Brent (2016) estimate that teens had 157 percent higher odds of reporting an arrest during the survey waves in which they also reported a suspension. Similarly, using monthly surveys from a sample of 1,300 previously incarcerated teens, Monahan et al. (2014) find that respondents had 110 percent higher odds of reporting an arrest during the months when they also reported a suspension/expulsion. Because these studies analyze within-respondent changes while also controlling for time-varying measures of self-reported delinquency, they address selection bias in important ways. Nevertheless, as both studies note, it is possible that their results only reflect simultaneous sanctions. That is, the fact that respondents were more likely to report arrests during the same study waves in which they also reported suspensions could indicate that both sanctions resulted from one underlying incident.
Finally, two other studies extend suspensions’ possible consequences into adulthood. Arum and Beattie (1999), using the National Longitudinal Survey of Youth 1979, and Wolf and Kupchik (2017), using the National Longitudinal Study of Adolescent to Adult Health, both find that teens who reported having been suspended prior to their baseline interviews were more likely to be incarcerated years later in adulthood. By analyzing outcomes in adulthood and controlling for a number of relevant confounders, these studies provide evidence that suspended teens face lasting disadvantages compared to their nonsuspended peers. Nevertheless, as with other studies, they note that they are unable to address the “mechanisms that drive the connection between being suspended and negative future outcomes” (Wolf and Kupchik 2017:425).
Underlying the limitations of all of these studies is the fact they begin tracking students only late into their K–12 school experiences. For example, both Mowen and Brent (2016) and Wolf and Kupchik (2017) report that the students in their samples were already 15 years old, on average, by the time of their baseline interviews. The current study builds on existing research by analyzing children’s trajectories from birth through adolescence, studying the relationship between childhood suspensions and adolescent arrests. By focusing on childhood suspensions, I exclude the possibility that suspensions impact arrests only through simultaneous sanctions. By matching children on context and pre-suspension risk factors, I minimize the effect of selection bias. Taken together, by excluding simultaneous sanctions and minimizing selection bias, I am able to plausibly identify whether suspensions themselves trigger a downward spiral in students’ lives. Given these results, I then provide suggestive evidence on the mechanisms that may explain this downward spiral. Leveraging repeated behavioral, academic, and disciplinary measures, I estimate how much of suspension’s total association with arrest can be explained by changes on each of these measures.
Data and Methods
Fragile Families at Year 15
The Fragile Families and Child Wellbeing Study (FFCWS) is birth cohort study of 4,898 children born in large American cities. Employing a three-stage, probability sampling design, the weighted FFCWS sample is representative of all children born between 1998 and 2000 in American cities with populations of 200,000 or more (Reichman et al. 2001). Since its baseline interviews with children’s mothers and fathers, the FFCWS has conducted five additional waves of data collection, documenting child and family development approximately 1, 3, 5, 9, and 15 years after the child’s birth.
The current study takes advantage of new data from the FFCWS Year 15 follow-up. The Year 15 study included interviews with teens and their primary caregivers (PCGs) as well as home visits for a randomly selected subsample of families. Data collection occurred between February 2014 and October 2016, during which time the participating teens were an average of 15.5 years old. 2 Although the FFCWS data have long been a central source for research on parents’ experiences with the justice system, the Year 15 data collection marks the first time in which children’s contact with the justice system has been assessed.
To address the contextual confounding of exclusionary discipline and police contact, I supplement the FFCWS data with contextual information on children’s schools and neighborhoods. School data come from the U.S. Department of Education’s Civil Rights Data Collection (CRDC) and Common Core of Data (CCD). Tract-level data on neighborhoods come from the U.S. Decennial Census.
Analytic Sample
Of the 4,898 families in the original FFCWS sample, the current study focuses on the 3,429 cases in which both PCGs and teens completed interviews at Year 15. Because this study centrally relies on information from Year 9, the sample was then narrowed to cases in which either the child or their PCG completed an interview at Year 9 (n = 3,242). This group was further restricted to those children who attended public schools at Year 9 (n = 2,942). The sample was restricted to public school students both because children attending private schools may be different than the rest of the sample in ways that are difficult to observe and the CRDC data only cover public schools. Because the analysis is focused on identifying the consequences of suspensions that occur long before children’s first arrests, four children were removed because they reported at Year 15 that their first arrest occurred in the same year as their Year 9 interview. On average, this provides 4.4 years between the child’s age at the time of their Year 9 interview and their age at first arrest. Finally, 14 cases were removed because neither the teen nor the PCG reported the teen’s arrest record at Year 15.
These restrictions yield a final sample of 2,924 teens. Cases with missing information on any covariate were retained by implementing multiple imputation by chained equations, producing 20 imputed data sets (White, Royston, and Wood 2011). All covariates used in this study were included in the imputation equations. To test whether this study’s conclusions are sensitive to the imputation procedure, the main analysis was replicated using only the non-imputed data; results, presented in Online Table 1, indicate larger treatment effects than those calculated from the imputed data.
Descriptive Statistics by Childhood Suspension Status.
Note: Two-tailed tests comparing students suspended by 9 and those not suspended by 9. Range for continuous scales in parentheses. Full descriptions of variables can be found in Online Table 3. Summary statistics are presented from first of 20 imputed data sets. Data from Fragile Families and Child Wellbeing Study, U.S. Department of Education’s Civil Rights Data Collection and Common Core of Data, and 2000 Decennial Census. 5, 9, 15 = Survey Years 5, 9, 15; PCG = primary caregiver; CPS = Child Protective Services; PPVT = Peabody Picture Vocabulary Test.
p < .05. **p < .01. ***p < .001.
Compared to the FFCWS baseline sample, this study’s analytic sample is slightly more disadvantaged. For example, compared to the full sample, the mothers in this sample were 4 percentage points less likely to have a high school diploma at baseline (33 percent vs. 37 percent, p < .05) and 6 percentage points less likely to be married (22 percent vs. 28 percent, p < .001). With these facts in mind, it is important to note that this study’s analytic sample cannot be assumed to be representative of any broader population of children and families. Nevertheless, it is a geographically diverse, population-based sample that is disproportionately constituted by the kind of disadvantaged children of color who face the highest rates of both exclusionary discipline (Ramey 2015) and juvenile justice contact (National Research Council 2001).
Measures
Outcome: Arrest by Year 15
All information about arrests is taken from the Year 15 teen and PCG interviews. Following the standard practice of previous FFCWS research, which uses multiple survey reports to identify rare and sensitive events (e.g., Haskins and Jacobsen 2017), teens are considered to have been arrested if either they or their PCG reports the arrest. Teens were asked “Have you ever been arrested or taken into custody by the police?” PCGs were asked “Has [youth’s name] ever been arrested?” As a robustness check, models using only teen reports, only PCG reports, and only cases where both reports agreed were also estimated. Results, presented in the Online Table 2, are substantively unchanged across all report types.
Treatment: Suspension by Year 9
The “treatment” for the current study is whether the child was suspended by the time of their Year 9 interview. As with arrest, children are marked as suspended if either they or their PCG reports as such. Online Table 2 presents robustness checks confirming the consistency of results across the same alternate specifications as those described for arrests. In a set of questions about early delinquency, children were asked if they had ever “been suspended or expelled from school.” PCGs were asked whether the child had any school absences in the past year and, if so, whether these absences were because the child was “suspended or expelled.” Although these questions do not distinguish between suspension and expulsion, it is extremely unlikely that any child had been expelled by Year 9, when most were in third grade. On average, the elementary schools attended by children in this study reported expelling 0.003 students across all grade levels in the CRDC data. Given this fact, I make the assumption that this variable is measuring suspensions.
Risk Factors for Childhood Suspension
Drawing on research regarding the risk factors for child behavioral problems (e.g., National Institute of Mental Health 2001) and later juvenile delinquency (e.g., Sampson and Laub 2003), this study focuses on three sets of factors: family background and home environment, child behavior and temperament, and school and neighborhood context. The full list of variables can be found in Table 1, with complete details on their operationalization in Online Table 3. With the exception of the contextual measures, all risk factors are measured “pretreatment,” using reports from baseline through Year 5. All measures of child behavior and temperament were reported by the PCG at Year 5. The Peabody Picture Vocabulary Test, a test of verbal ability, was administered during the Year 5 home visit. Because of the importance of ensuring that matched children were in similar school and neighborhood contexts, contextual measures were constructed using information from Year 9.
For school context, the average rate of suspensions and expulsions reported by the school across three waves of the CRDC was calculated. Importantly, matching students on their school-level disciplinary climate allows me to distinguish the effects of personally being suspended from the broader negative effects of attending a school in which suspensions are common (Perry and Morris 2014). Five other measures of school context were also included: whether the school employs at least one guidance counselor, the composition of the student body (percentage black, percentage receiving free or reduced-price lunch), the student-to-teacher ratio, and whether the school is a charter school. For neighborhood context, a measure of concentrated disadvantage was constructed based on the index defined by the Project on Human Development in Chicago Neighborhoods (e.g., Sampson, Raudenbush, and Earls 1997). For this index, five measures of tract-level disadvantage were averaged across the tracts inhabited by the child in their first nine years of life.
Mechanisms Linking Childhood Suspension and Adolescent Arrest
Finally, to understand what post-suspension changes in children’s lives may have made them more likely to be arrested, a variety of measures reported at both Year 9 and Year 15 were used. Informed by the studies of exclusionary discipline described previously, I focus on three areas: problematic behavior, school disengagement, and additional exclusionary discipline.
For problematic behavior, I calculate children’s scores on two well-validated scales: Maumary-Gremaud’s (2000) Things That You Have Done delinquency scale (e.g., “I took something from a store without paying for it”) and the aggressive behavior subscale from Achenbach and Rescorla’s (2001) Child Behavior Checklist (CBCL-6/18) (e.g., “Child has temper tantrums or a hot temper”). Delinquency is self-reported; aggression is reported by PCGs. For school disengagement, I consider three measures: the child’s score on a scale measuring their affective connection to school (e.g., “I feel like I am part of my school”), their reports of recent truancy, and PCG reports of whether the child was required to repeat a grade. Finally, for repeated discipline, I constructed a measure of whether either the teen or their PCG reports that the teen was suspended or expelled in the two years prior to the Year 15 interview. Complete details on these variables’ operationalization can be found in Online Table 3.
Analytic Approach
This study seeks to understand (1) whether and (2) in what ways childhood suspensions increase children’s risk of later arrests. To address the first part of this question, I use propensity score matching methods to construct comparison groups that are indistinguishable along observable risk factors. To address the latter part of the question, I start with the matched sample constructed in the first part of the analysis and then apply a counterfactual decomposition technique, estimating how much of suspension’s total association with arrest can be explained by post-suspension changes in a variety of possible mechanisms.
The propensity score matching analysis proceeds in four steps. First, I generate propensity scores for every child in the sample using a logistic regression. I model the log-odds of having been suspended by Year 9 as a linear function of the contextual and pre-suspension risk factors described previously. Second, I match children using their estimated propensity scores, using one-to-one matching and imposing a caliper of 0.001 (Caliendo and Kopeinig 2008). 3 Third, I confirm that this matching process produced treatment and control cases that are statistically indistinguishable and substantively identical in terms of all the covariates used to estimate the propensity score. Finally, I compare the arrest rates of suspended and nonsuspended children, reporting average treatment effects for suspended children.
For these results to be interpreted as reflecting the causal effect of suspension on arrest, one needs to assume that, conditional on the estimated propensity score, suspensions were administered as if random. This assumption—often referred to as the “ignorability” or “selection on observables” assumption—is likely to be too strong in this context. Even with the variety of risk factors considered here, there are still likely to be unobserved factors that govern children’s risk for both suspensions and arrests. If suspensions indicate the crossing of some unmeasured behavioral threshold that divides seemingly similar children, then the suspension itself could still carry few consequences.
To test the sensitivity of these results to an unmeasured factor of this kind, I conduct the test proposed by Rosenbaum (2002) and implemented by DiPrete and Gangl (2004). This test simulates a “worst case scenario” to estimate how consequential an unmeasured confounder would have to be to render the estimated treatment effect insignificant. In this context, this test proposes that there is an unmeasured childhood factor, U, that is a near perfect predictor of whether the child will be arrested by Year 15. It then varies the degree to which U increases the child’s odds of being suspended compared to their matched control case, specifying the increased odds of suspension with Γ. At each level of selection bias, Γ, Wilcoxon signed-rank tests are computed. I summarize these analyses with the p values of tests of the hypothesis that suspension has no causal effect on arrest.
After estimating and bounding the total effect of childhood suspension on adolescent arrest, I conduct supplementary analyses to decompose this effect into two parts: the proportion of suspension’s effect explained by a hypothesized mechanism and the proportion explained by all other factors. To do this, I apply Imai, Keele, and Tingley’s (2010) mediation analysis framework. The intuition behind this framework is as follows: Suppose, for example, that childhood suspensions partially promote later arrests by triggering increases in delinquency. To determine how much of suspension’s effect on arrest is explained by delinquency, one can conduct a decomposition exercise similar to those more commonly used by sociologists and demographers (e.g., Kitagawa 1955), decomposing the total difference into two parts by varying one factor and fixing the other. Here, by predicting arrest rates with and without suspension-related changes in delinquency, one can estimate how much of the total difference in arrest rates between suspended and nonsuspended children is explained by delinquency.
More formally, Imai et al.’s (2010) framework applies the counterfactual model of causality to the study of mechanisms. It proposes that individuals’ potential outcomes depend both on their treatment status and the way in which that treatment comes to affect a hypothesized mechanism. Imai et al. (2010) model an individual i’s outcome, Y, as a function of their treatment, t, and mechanism, m:
Within this framework, every individual has two potential values for the hypothesized mechanism: one that is realized under treatment,
The direct effect of the treatment—the proportion of the treatment operating through all factors other than the hypothesized mechanism—is defined similarly, by fixing the mechanism but varying treatment status:
Together, these two effects,
The forgoing discussion implies that there are two sets of predicted values needed to calculate mediation effects in the current study: the level of the mechanism (predicted with and without a childhood suspension) and the level of the outcome (predicted with and without the post-suspension changes in the mechanism). To generate these predictions, I estimate two sets of regression models, one modeling the mechanisms and another modeling whether the teen was arrested. For binary mechanisms, I estimate logistic regressions; for continuous mechanisms, I use ordinary least squares. For all regressions, I model the outcome as a linear function of: Year 9 suspension status, the child’s propensity for suspension, and the level of the mechanism at Year 9. 4 Importantly, for all models, I use only the matched subsample generated from the propensity score matching analyses.
The predictions generated by these models are used as inputs into an algorithm implemented by Hicks and Tingley (2011). Using delinquency as an example mechanism, the algorithm first uses the mediator model to generate counterfactual delinquency rates for every teen in the sample: one rate as if the teen had faced a childhood suspension, one as if they had not. Second, the algorithm inputs these two counterfactual delinquency rates into the arrest model, generating two counterfactual arrest rates. By predicting arrest rates with the direct effect of the suspension held constant and only delinquency rates varying, this step calculates the individual-level mediation effects defined by Equation 2. Third, the algorithm computes the average causal mediation effect of delinquency by averaging across all of the predicted individual-level differences in arrest rates, as in Equation 4. Finally, it repeats the prior steps 500 times, each time drawing a different set of model parameters from their sampling distribution.
Although this analysis framework is designed with reference to a counterfactual framework of causality, it is important to note this study’s data do not support causal claims about mechanisms. Because both mechanisms and arrests are reported at the same time, the causal ordering between mechanisms and arrests is unclear; it is possible that the mechanisms’ levels themselves are affected by prior arrest. However, because this set of analyses is purely descriptive, the underlying causal ordering between mechanism and arrest is immaterial to the descriptive task at hand. Focusing only on the conditional association between mechanism and arrest, these analyses answer the question: Of the factors in teens’ lives associated with having experienced a childhood suspension, which are best able to explain their rates of arrest?
Results
Descriptive Statistics
Descriptive statistics on the analytic sample are presented in Table 1. Overall, 8 percent of the sample had been arrested by Year 15. Arrest rates, however, are markedly different by suspension status. Teens who were suspended in childhood experienced arrest rates that were four times higher than their nonsuspended peers (20 percent vs. 5 percent, p < .001). Nevertheless, it is unclear whether these differences reflect the consequences of suspension itself as suspended and nonsuspended children also differed across virtually every other factor. For example, at Year 5, suspended children’s PCGs were about twice as likely to report that the child “physically attacks people” (18 percent vs. 7 percent, p < .001) and that they were “disobedient at school or child care” (45 percent vs. 25 percent, p < .001). Suspended and nonsuspended children also inhabit considerably different schools and neighborhoods. For example, suspended children attended elementary schools with suspension rates that were nearly double those in the schools attended by nonsuspended students (9 percent vs. 5 percent, p < .001). At the same time, suspended children grew up in census tracts that were 0.59 standard deviations higher (–0.11 vs. 0.48, p < .001) on the concentrated disadvantage scale.
All of these differences highlight the difficulty of identifying how, if at all, suspension itself actually intervened in children’s lives to trigger a downward spiral. They also underscore an important advantage that propensity score matching provides over traditional regression methods for this study. Because suspended and nonsuspended students have such dramatically different distributions of risk factors, it is critical to identify that subset of students who could have plausibly been observed in either group (Morgan and Harding 2006).
Estimating Children’s Risk for Suspension by Year 9
Given the large differences between suspended and nonsuspended children, what factors are most consequential in shaping children’s risk for suspension? Table 2 presents these results. Although this study is primarily focused on the consequences of suspensions, these results also have important implications for understanding the determinants of suspensions.
Modeling the Predictors of Suspension by Year 9.
Note: Reported coefficients are odds ratios, with 95 percent confidence intervals in brackets. To allow comparisons across different underlying scales, all continuous variables are standardized to a mean of zero and standard deviation of one. Results pooled across 20 imputed data sets. Data from Fragile Families and Child Wellbeing Study, U.S. Department of Education’s Civil Rights Data Collection and Common Core of Data, and 2000 Decennial Census. 5, 9 = Survey Years 5 and 9; PCG = primary caregiver; CPS = Child Protective Services; PPVT = Peabody Picture Vocabulary Test.
p < .05. **p < .01. ***p < .001, two-tailed tests.
A few specific results stand out. Perhaps most striking is that even with the wide range of family, behavioral, and contextual factors included here, black students still have substantially higher odds of having been subject to exclusionary discipline. Apart from whether the child is male (odds ratio [OR] = 5.13, p < .001), the main effect of the child being black is the single strongest predictor of suspension in these data (OR = 3.28, p < .01). This result provides new evidence that differences in behavior and context are unlikely to fully explain racial disparities in discipline rates, despite recent claims to the contrary (Wright et al. 2014).
Three other results should be briefly noted. First, attending an elementary school that employs at least one guidance counselor is associated with 22 percent lower odds of receiving a suspension (p < .05). At a moment when approximately 1.6 million students attend schools that employ a police officer but no guidance counselor (U.S. Department of Education 2016), this result suggests that guidance counselors may play an important role in shaping how schools address behavior problems. Second, given concerns about school discipline practices in charter schools (e.g., Golann 2015), it is notable that children attending charter schools had 60 percent higher odds of being suspended (p < .05). Finally, the large risk associated with lead poisoning (OR = 1.98, p < .05) reinforces recent quasi-experimental evidence that lead exposure increases children’s risk for school discipline and juvenile justice involvement (Aizer and Currie 2017).
Apart from these specific results, there is an important broader story provided by this analysis. Much of the previous literature on school discipline has focused on the question of whether differences in behavior explain disparities in discipline outcomes (e.g., Rocque 2010). The results presented here extend that literature by providing information not just on behavior itself but also on the larger context out of which such behavior emerges. Across multiple measures, I find that children enduring adversity—and, potentially, abuse—face higher rates of suspensions. For example, children in households that have been inspected by Child Protective Services had 50 percent higher odds of suspension (p < .05). Children whose primary caregivers report hitting them with “a belt, hairbrush, a stick or some other hard object” had 44 percent higher odds of suspension (p < .05). Children who grew up in neighborhoods characterized by concentrated disadvantage faced higher rates of suspension, with a one standard deviation increase in disadvantage being associated with 25 percent higher odds of suspension (p < .01).
All of these results are consistent with research showing that children enduring chronic, toxic stressors display higher levels of externalizing behavioral problems (e.g., National Institute of Mental Health 2001). For children in more advantaged school contexts, such problems might be treated as a medical issue in need of extra support (Ramey 2015). However, for children in schools where behavior problems are met with sanctions rather than supports, whatever negative consequences suspensions do have will only exacerbate the considerable disadvantages these children already face.
Propensity Score Matching Estimates of Juvenile Arrest Rates
Given the structure of suspension risk estimated previously, do suspended children still face higher rates of arrests when compared to other children who displayed the same levels of observable risk factors? Table 3 presents results for this question. Teens who experienced a childhood suspension were more than twice as likely to be arrested than those who had the same observable risk for a suspension but who somehow avoided being sanctioned. In all, 18.5 percent of teens suspended by Year 9 were arrested by Year 15, compared to only 8.8 percent of their matched peers.
Juvenile Arrests: Propensity Score Matching Results by Childhood Suspension Status.
Note: Standard errors in parentheses. Matched estimates reflect average treatment effects for suspended students. Teens matched on propensity for suspension by Year 9, modeled as a function of contextual and pre-suspension risk factors listed in Table 2. One-to-one matching without replacement (caliper = 0.001) was used. Results pooled across 20 imputed data sets. Data from Fragile Families and Child Wellbeing Study, U.S. Department of Education’s Civil Rights Data Collection and Common Core of Data, and 2000 Decennial Census.
p < .001, two-tailed tests.
This is a substantively large effect, stratifying students’ risk for arrest in a consequential way. One way to index the magnitude of this difference in arrest rates is to note that the relative risk of arrest between matched suspended and nonsuspended teens (2.11: 18.5 percent vs. 8.8 percent) is the same as the total, unadjusted relative risk of arrest observed between males and females in the full sample (2.12: 10.6 percent vs. 5.0 percent). Supplementary analyses reported in Online Table 5 demonstrate that this pattern of results holds across student subgroups. Estimating propensity scores and matching within separate samples of only black students, only non-black students, only male students, and only female students revealed that suspended students in every subgroup had significantly higher rates of arrest than their matched peers.
Despite the magnitude of the difference reported in Table 3, these results still indicate that almost half of the observed association between childhood suspension and adolescent arrest was the result of selection bias: A naïve comparison suggested that suspended children were nearly four times more likely to be arrested. The selection bias inherent in relying on observed differences between suspended and nonsuspended children is further underscored by the fact that almost 30 percent of suspended children could not be matched to any observably similar control child. Among those who could be matched, matching on the propensity score successfully balanced all covariates. Balance tests indicating that the matched subsamples were statistically indistinguishable and substantively identical are included in Online Table 6.
Sensitivity Test
These results suggest that childhood suspensions in and of themselves substantially increase children’s risk for later arrest. Nevertheless, it is possible that I have not been able to fully model children’s risk for suspension, such that an unmeasured factor is still producing a spurious association between suspension and arrest. To address this possibility, I conducted the sensitivity analysis described previously. Results are presented in Online Table 7.
The results indicate that to render the association between suspension and arrest insignificant, an unmeasured confounder would have to be a near perfect predictor of arrest and also increase the odds of suspension by about 60 percent. It is plausible that there is an unmeasured factor that could increase children’s odds of suspension by 60 percent; this is, for example, slightly less than the risk associated with being disobedient in preschool (69 percent). However, it is more difficult to argue that this unmeasured childhood factor would also be a near perfect predictor of adolescent arrest. Adolescent arrests contain a large degree of randomness and are difficult to predict (Kirk and Sampson 2013; Liberman, Kirk, and Kim 2014). For example, despite being a strong predictor of suspension, the zero-order correlation between preschool disobedience and arrest is only 0.08. Even reports of serious delinquency at Year 15 only weakly predict arrest in these data. For example, using a weapon to get something from someone is correlated with arrest at 0.11 and selling drugs at 0.15.
Therefore, to render the relationship between childhood suspension and adolescent arrest insignificant, the unobserved childhood factor would have to (1) predict arrest much more strongly than any available variable while also (2) increasing the odds of suspension by roughly the same magnitude as the strongest behavioral predictor in the data. Although this is possible, it seems unlikely. What is more likely is that additional unmeasured factors would further decrease the causal effect of a childhood suspension while not eliminating that effect entirely.
Mediation Analysis
Thus far, the results have provided strong evidence consistent with the interpretation that childhood suspensions triggered a downward spiral in these children’s lives. If suspensions did trigger such a spiral, what mechanisms might explain children’s path from suspension to arrest? Table 4 presents these results. The primary estimates of interest are in the last two columns, where suspension’s total association with arrest is decomposed into two parts: the percent explained by changes in the hypothesized mechanism and the percent explained by all other factors. To put results into a consistent, intuitive metric, I present the percent mediated rather than the average mediation effect itself. The first two columns present the key parameter from the mechanism and arrest models. For the mechanism model, this parameter is the change in the mechanism associated with having received a childhood suspension; for the arrest model, it is the odds of arrest associated with a unit change in that mechanism. Although the algorithm uses the full set of model parameters to generate counterfactual predictions and calculate results, these two parameters provide an intuitive basis for understanding the decomposition results and are substantively interesting in their own right. The complete regression results used in the decomposition algorithm are presented in Online Table 8.
Mechanism Decomposition Results.
Note: Standard errors in parentheses. Each parameter estimate is from a separate regression. Parameters in mechanism model are z scores for continuous mechanisms and odds ratios for binary mechanisms. All parameters in arrest model are odds ratios. Parameter estimate in first column is the conditional association between childhood suspension and the mechanism at 15, controlling for propensity for suspension and the mechanism’s value at 9. Parameter estimate in second column is the conditional association between mechanism at 15 and the odds of arrest, controlling for childhood suspension, the propensity for suspension, and mechanism’s level at 9. Complete regression results presented in Online Table 7A. Models run only on matched sample constructed from Table 3. Mediation percentages estimated using 500 simulations of counterfactual decomposition outlined in Hicks and Tingley (2011). Results pooled across 20 imputed data sets. Data from Fragile Families and Child Wellbeing Study, U.S. Department of Education’s Civil Rights Data Collection and Common Core of Data, and 2000 Decennial Census.
p < .05. **p < .01. ***p < .001, two-tailed tests.
The results indicate that compared to their matched peers, suspended children experienced significantly greater escalations in behavioral and disciplinary problems. Suspended children also faced increasing school-related issues, but these estimates did not reach statistical significance. At the same time, every mechanism tested was a significant risk factor for experiencing an adolescent arrest. For example, suspended children had 0.33 standard deviations greater escalations in delinquency than their matched peers (p < .001). In turn, changes in delinquency consequentially shape teens’ risk for arrest: a one standard deviation increase in delinquency was associated with 90 percent higher odds of arrest (p < .001). The decomposition algorithm combines the predictions from these two models, estimating that 31.7 percent of suspension’s total association with arrest can be explained by increases in delinquency.
Comparing across mechanisms, the pattern is clear. School-related problems have little direct explanatory power, explaining only between 2.9 percent and 5.2 percent of the total association between suspension and arrest. This may not be surprising since one might imagine that school problems would increase teens’ risk for arrest only insofar as they manifest in behavioral issues. Correspondingly, escalating behavioral problems have more explanatory power. In addition to the delinquency results described previously, 18.2 percent of the association between suspension and arrest can be explained by escalations in aggressive (but not necessarily illegal) behavior. By far, though, the single factor that best explains suspended children’s arrest rates is their elevated risk for repeated school sanction.
Compared to their matched peers, suspended children had 180 percent greater odds of facing at least one more suspension/expulsion prior to their Year 15 interview (p < .001). In turn, experiencing an adolescent suspension/expulsion is an extremely strong risk factor for arrest: controlling for prior discipline and children’s propensity for such discipline, an adolescent suspension/expulsion is associated with 652 percent greater odds of arrest (p < .001). Because adolescent school sanctions stratify teens’ risk for arrest in such a dramatic way, the increased risk for repeated removal from school that suspended children face is able to explain 52.2 percent of the total association between early suspension and later arrest.
As in prior studies, I am not able to further disentangle the underlying processes connecting school and legal sanctions among teenagers. At least part of this relationship is likely to be the result of simultaneous sanctions. It could also be the case that teens were arrested on the specific days during which they were removed from school (e.g., Cuellar and Markowitz 2015). Whatever the processes, the key fact is not that adolescent school sanctions are tightly linked with arrests. Instead, it is that children with histories of suspension face substantially higher risks for later school sanctions. Moreover, suspended children’s increased risk for repeated sanction cannot be easily explained by their behavior alone. Even if one additionally controls for delinquency at both 9 and 15, suspended children still face 127 percent greater odds of repeated removal than their peers (p < .001; model results in Online Table 9). In terms of understanding suspended children’s risk for juvenile arrest, it is these repeated school sanctions—over and above behavior itself—that are best able to explain the path from first suspension to later arrest.
Although I cannot exclude the possibility that these repeated sanctions reflect unmeasured aspects of children’s behavioral trajectories, these results are consistent with the possibility that the mark of a disciplinary record itself encourages increasingly harsh treatment from school officials. Having faced a first sanction, children’s subsequent margin for error in terms of regulating their behavior may be smaller than that experienced by students who are not institutionally marked. This process could unfold informally, by shaping teachers’ and administrators’ perceptions of students (e.g., Ferguson 2000). That kind of informal process could also be reinforced by schools’ own formal codes of conduct, which typically mandate more severe punishments for students with histories of rule violations (Hirschfield 2008). Through both of these reinforcing processes, early suspensions could trigger a cycle of escalating sanctions. For children facing an “unforgiving world” (Allen 2017) both inside and outside of school, the snares of institutional sanction emerge early, narrowing opportunities in adulthood before adolescence has even begun.
Discussion
In her ethnography of elementary school discipline, Ferguson (2000:1) describes a teacher pointing to a 10-year-old black boy, barely four feet tall. The boy was on his way out of his classroom and into the school’s “punishing room.”“That one,” the teacher says, “has a jail cell with his name on it.” Reflecting on such students’ experiences, Ferguson (2000:230) warns that “there are serious, long-term effects of being labeled a Troublemaker that substantially increase one’s chances of ending up in jail.”
This study’s results are consistent with Ferguson’s (2000) conclusions. Like the children Ferguson describes, many of the students in this study began being removed from their classrooms at a young age. Of those ever subject to exclusionary discipline in this sample, 45 percent had already been suspended by Year 9, when most were only in third grade. Although these children were already more disadvantaged than their peers, early suspensions still appear to have redirected their lives in consequential ways.
Figure 1 depicts the diverging paths taken by suspended children and their matched peers. Having faced a childhood suspension, few suspended children made it to Year 15 without facing some form of additional sanction: Only 36 percent of suspended children faced no further sanction by the time of the Year 15 interview. By Year 15, nearly one in five suspended children (19 percent) had already acquired an arrest record. Even though most suspended children managed to avoid arrest, far fewer escaped additional school sanctions. Combining those who faced only school sanctions (45 percent) with those who faced both school and legal sanctions, 62 percent of suspended children went on to face at least one more suspension/expulsion. The paths taken by the matched, nonsuspended children provide an almost mirror image. Whereas just 36 percent of suspended children escaped additional sanction, only 39 percent of nonsuspended children ever went on to experience any sanction. Only 9 percent of matched control children were arrested by Year 15.

Adolescent school and legal sanctions by childhood suspension status.
It is possible that the diverging paths taken by suspended children and their matched peers purely reflect unmeasured aspects of suspended children’s behavioral trajectories. However, the balance of the evidence presented previously suggests another possibility: that childhood suspensions initiate a process of cumulative disadvantage of the kind outlined in Sampson and Laub’s (1997) life course theory of crime. Sampson and Laub suggest that criminal behavior persists across the life course not because certain people have permanently high propensities for crime. Instead, they argue that juvenile delinquency continues into adult crime because the consequences of teenage misbehavior create a “snowball” effect: “adolescent delinquency and its negative consequences (e.g., arrest, official labeling, incarceration) increasingly ‘mortgage’ one’s future, especially later life chances molded by schooling and employment” (Sampson and Laub 1997:15). As the consequences of prior events build up, the ability to break with the past is diminished, and the calculus of future decision making is tilted toward continued criminality.
Contemporary school discipline policies now appear to start this process of cumulative disadvantage at even younger ages. Well before children come into contact with the juvenile justice system, many now have their behavior officially sanctioned by public officials in the form of school suspensions. Among arrested teens in the sample, suspensions were a nearly universal experience: 89 percent of arrested teens had also experienced at least one suspension. By law (if not always in practice), these suspensions would have required the production of a record that is designed to follow the child for the rest of their educational career (Education Law Center 2012). Indeed, states are now required to ensure that students’ disciplinary records follow them across schools, even for students transferring to private schools (U.S. Congress 2002). Once arrested, school records are one of the key pieces of evidence that juvenile intake officers consider when deciding whether to formally petition the court or informally divert the case (Mears 2011; National Research Council 2013). Later in life, teens applying to college will face a Common Application that includes a mandatory question about whether they have experienced a “suspension, removal, dismissal or expulsion” (The Common Application 2017).
In all of these ways, formal school sanctions magnify the consequences of early misbehavior, allowing them to continue to reverberate across the life course and the institutions that govern it. For children from more advantaged backgrounds, a single suspension may not be enough to trigger these kinds of cascading consequences. However, this study shows that it is the most disadvantaged children who face the greatest rates of suspension. For these children—whose connection to school and academic self-concept may already be precarious—suspensions further constrain the already narrow path to adulthood that they face.
These conclusions come with important caveats that suggest lines for future research. First, I cannot be sure how the results observed in this sample would generalize to the broader student population. Compared to teens nationwide, the FFCWS sample is comparatively disadvantaged, constituted only by teens who were born in cities. My analytic sample is further disadvantaged still, constituted exclusively by public school students and primarily by students of color. Future research should work to replicate these findings using more nationally representative data. Nevertheless, to my knowledge, the FFCWS is the only study currently available that has information on childhood suspensions, pre-suspension risk factors, and juvenile justice outcomes.
Second, I note again that I rely on survey reports for both the treatment and the outcome of this study. I am not able to independently verify the existence and timing of the key covariates that underlie these results. Even so, this limitation is shared by all survey-based research on this topic; the current study is in a stronger position than most because it uses reports from parents rather than relying solely on child reports. Nevertheless, the limitations of the survey data used in this study reinforce recent calls for greater linkages between survey results and administrative records on children and schools (Schneider, Saw, and Broda 2016).
Finally, the central limitation of the current study is that I cannot definitively demonstrate that its core results reflect the causal effect of suspensions rather than unmeasured aspects of the child’s behavior, context, or experiences. Given the variety of contextual and pre-suspension risk factors considered, the magnitude of the estimated treatment effect, and the results of the sensitivity analysis, the absence of a true causal effect is unlikely. Nevertheless, without some institutional mechanism guaranteeing that the administration of suspensions was effectively random, any causal claims have to be carefully qualified.
Conclusion
Schools face genuine challenges in creating safe and orderly environments. The best way to address these challenges is not obvious. Since the 1990s, however, the balance has shifted decisively in the direction of more punitive approaches, particularly in schools that primarily serve poor students of color. The results of this study suggest that a punitive approach to school discipline promotes rather than prevents juvenile delinquency. Just as detaining juveniles seems to promote adult incarceration (Aizer and Doyle 2015) and arresting juveniles seems to encourage future arrests (Liberman et al. 2014), this study suggests that suspending students promotes repeated school sanctions and ultimately, a higher risk of arrest. These results should be considered alongside evidence suggesting that discipline decisions reflect racial bias (Okonofua and Eberhardt 2015), that schools emphasizing exclusionary discipline experience worse outcomes for nondisciplined students (Perry and Morris 2014), and that cognitive behavioral therapy interventions for delinquent teens produce large positive effects on school engagement, violent crime, and recidivism (Heller et al. 2017).
A new approach to school discipline and juvenile justice administration has begun to take hold across the country. Since 2011, 22 states and the District of Columbia have revised their laws governing school discipline; 23 of the 100 largest school districts have implemented reforms limiting suspensions, often by ending or restricting the use of suspensions in elementary schools. These reforms have caused sharp declines in suspension rates in cities like Los Angeles, New York City, and Baltimore (Steinberg and Lacoe 2017). At the same time, states have begun to reform their juvenile justice systems, limiting the jurisdiction of the criminal system over juveniles, emphasizing diversion, and investing in mental health services for adolescent offenders (National Conference of State Legislatures 2015).
During the Obama administration, these state and local efforts were supported by federal programs at the Departments of Education and Justice. In the past year, however, newly vocal critics of discipline reform have emerged. At a moment when the future of federal support for discipline reform is uncertain, this study provides reason for state and local officials to persist in the work of developing less punitive approaches to school discipline in the post-Obama era.
Supplemental Material
Supplemental Material, DS_10.1177_0038040718784603 – A Downward Spiral? Childhood Suspension and the Path to Juvenile Arrest
Supplemental Material, DS_10.1177_0038040718784603 for A Downward Spiral? Childhood Suspension and the Path to Juvenile Arrest by Joel Mittleman in Sociology of Education
Footnotes
Acknowledgements
Marta Tienda, Sara McLanahan, Jen Jennings, Ian Lundberg, and anonymous reviewers provided valuable feedback on earlier drafts. Sara McLanahan, Kate Jaeger, and Louis Donnelly were instrumental in facilitating early access to the Fragile Families Year 15 data and enabling the appending of contextual information. All mistakes are my own.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) of the National Institutes of Health under award numbers R01HD36916, R01HD39135, and R01HD40421, as well as a consortium of private foundations. The content is solely the responsibility of the author and does not necessarily represent the official views of the National Institutes of Health.
Research Ethics
All research presented here was conducted using secondary data sources. Each wave of the Fragile Families and Child Wellbeing Study was approved by the Institutional Review Boards of the participating institutions. The appending of contextual data was conducted in such a way that the author of this study had no access to personally identifying information on study participants at any time.
Notes
Supplemental Material
The supplemental material are available in the online version of the article.
Author Biography
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
