What policy substitutes alternative, community-based sanctions for state training schools?

Criminology. Author manuscript; available in PMC 2010 Jan 4.

Published in final edited form as:

PMCID: PMC2801446

NIHMSID: NIHMS157953

Abstract

The effect of sanctions on subsequent criminal activity is of central theoretical importance in criminology. A key question for juvenile justice policy is the degree to which serious juvenile offenders respond to sanctions and/or treatment administered by the juvenile court. The policy question germane to this debate is finding the level of confinement within the juvenile justice system that maximizes the public safety and therapeutic benefits of institutional confinement. Unfortunately, research on this issue has been limited with regard to serious juvenile offenders. We use longitudinal data from a large sample of serious juvenile offenders from two large cities to 1) estimate a causal treatment effect of institutional placement, as opposed to probation, on future rate of rearrest and 2) investigate the existence of a marginal effect (i.e., benefit) for longer length of stay once the institutional placement decision had been made. We accomplish the latter by determining a dose-response relationship between the length of stay and future rates of rearrest and self-reported offending. The results suggest that an overall null effect of placement exists on future rates of rearrest or self-reported offending for serious juvenile offenders. We also find that, for the group placed out of the community, it is apparent that little or no marginal benefit exists for longer lengths of stay. Theoretical, empirical, and policy issues are outlined.

Keywords: juvenile offenders, incarceration, propensity scores, deterrence

Recent policies have narrowed the jurisdiction of juvenile court by removing the most serious offenders through various mechanisms, which include waiver, statutory exclusion, and lowering the age of majority (Fagan, 2008; Feld, 1998, 1999; Griffin, 2003; Zimring, 1998). These changes have been occurring in an ongoing cycle for nearly three decades, capped by a frenzy in the 1990s, when virtually every state acted in some way to “toughen up” laws governing the processing of serious juvenile offenders. The removal of many serious offenders from juvenile court jurisdiction has raised basic questions about the viability of the juvenile court itself (Fagan, Kupchik, and Liberman, 2007). Given this political environment, demonstrating the effectiveness of the juvenile courts’ treatment and confinement practices is essential for informed debate about future juvenile justice policy.

It is important to understand what prompted all this change in the first place. In reaction to increases in rates of youth violence in the 1980s (Blumstein and Wallman, 2000), both policy makers and the public lost confidence in the capacity of the juvenile court to fulfill its mission of effectively treating and/or deterring serious juvenile offenders from committing future crimes (Moon, Cullen, and Wright, 2003). Zimring (1998) suggests that the perception of leniency led to a related perception of increased public safety risks. Perceived short stays in juvenile facilities also offended the popular and political culture of proportionality. The lengths of placement in juvenile facilities were viewed as too short to be effective either in treating youth or in inflicting punishment that would have a residual deterrent effect once they were released.

The juvenile system responded by instituting policy changes aimed mainly at increasing public safety. These changes included mandatory placements in secure confinement for adolescents charged with serious crimes, mandatory minimums once placed, and longer overall stays (Fagan, 2008). Attempts were made to combine the treatment benefits of the juvenile system with the greater emphasis on punishment in the adult system (i.e., a “blended approach”) once in placement (Redding and Howell, 2000). These policy shifts were designed with the expectation that the experience of institutional placement, as opposed to community placement, would reduce the incidence of criminality among juvenile offenders and that the reduction in criminality would be a function not just of the occurrence of placement but also of the duration of the placement. In other words, the probability of future criminality could decline as a function of either 1) the deterrent impact of institutional placement itself or 2) the length of exposure to institutional confinement. Longer stays in institutions provide increased treatment as well as a heightened cost for involvement in crime.

Thus, the policy question germane to this debate is finding the level of punishment and/or treatment within the juvenile justice system that maximizes the public safety benefits of confinement (Bhati and Piquero, 2008). A demonstration of capacity for effective punishment and the efficient use of resources are essential to the survival of the juvenile court (Schneider, 1990; Scott and Steinberg, 2008). If longer stays in institutional facilities are not producing gains in reduced offending, then it is questionable whether this use of resources is either justifiable or politically attractive (Blumstein and Piquero, 2007). The financial cost of placing individuals in institutional care for extended periods is substantial, and high levels of spending on this practice should produce some benefit in terms of increased public safety. Without a demonstration that increased or longer institutional stays provide such a benefit, the argument for incurring these costs is substantially weaker.

This article addresses these questions using data from the Pathways to Desistance Study, which includes a longitudinal sample of serious juvenile offenders who can be thought of as the likely targets of competing policies (i.e., removal from court versus tougher punishment within court). We test two hypotheses. First, we estimate a causal treatment effect of placement into institutional care, as opposed to probation, on future rates of rearrest and self-reported offending among serious juvenile offenders to test whether placement reduces future crime. Second, using the group of youth sent to placement, we estimate the marginal effect on recidivism of longer stays in placement by identifying a dose-response relationship between the length of stay and future rates of both rearrest and self-reported offending, to test whether longer lengths of stay suppress recidivism rates. Our results suggest that an overall null effect of placement exists on future rates of recidivism for serious juvenile offenders. Furthermore, for those individuals receiving placement, it is apparent that little or no marginal benefit exists for longer lengths of stay, in terms of reducing rates of rearrest or self-reported offending, at least for being retained out of the community between 3 and 13 months.

BACKGROUND

At its core, deterrence assumes that sanctions delivered in a certain, swift, and severe manner will serve to increase sanction risk perceptions and subsequently reduce criminal activity (Beccaria, 1986 [1764]; Zimring and Hawkins, 1973). Criminologists long have been interested in the effects of sanctions on subsequent criminal activity, and much empirical research has been devoted to this line of work with regard to adult offenders (Nagin, 1998). Although some detectable effect is apparent regarding the certainty of punishment among certain offenders and certain crime types, this literature shows that increased sanctions do not substantially reduce future recidivism but instead produce only a small (Bhati and Piquero, 2008) or incapacitation effect (Piquero and Blumstein, 2007) on recidivism. In a series of studies, Gendreau and colleagues (Gendreau, Goggin, and Cullen, 1999; Gendreau, Goggin, and Fulton, 2000; Gendreau et al., 2001) conducted meta-analyses of 117 prior studies involving 442,471 offenders, which showed that increased sanctions did not suppress future criminality. Instead, this work found that increased sentence length or institutional, as opposed to community, sentences were often associated with increased recidivism. Similarly, evidence suggests (Austin, 1986) that prisoners who had shorter stays (because of early release caused by over-crowding) had lower rates of rearrest in the 1 year period after release than those individuals who served their full prison terms. This result, however, may be attributable to differences among those prisoners selected for early release as much as to any other factor (Austin, 1986).

THE UNIQUE FEATURES OF THE JUVENILE COURT

Two important caveats prevent us from mapping the above results to the context of juvenile response to sanctions. First, as mentioned, it is impossible to make a causal interpretation of the results above regarding the offenders’ response to harsher sanctions. The observed associations between harsher sanctions and future recidivism could simply be the by-product of harsher sanctions being assigned to those criminals with a higher propensity to reoffend. Second, and more fundamentally, the juvenile system is, by design, qualitatively different from the adult system. As Allen (2000:x) notes, from its inception, the juvenile court was designed to be “wholly distinct from criminal courts in form, procedure and purpose.” Even with the recent changes in favor of more strict sanctions in the juvenile court (Jenson and Howard, 1998), inherent differences between the respective missions of the more punitive-oriented adult court and the more rehabilitation-oriented juvenile court prevent a direct comparison (Cullen, Golden, and Cullen, 1983; Platt, 1977).

In contrast to the formalism of sentencing in the criminal courts, in which offense and offender characteristics often are translated into sentencing systems based on retribution and crime-control interests, dispositions in the juvenile court reflect a complex integration of theoretical, legal, and policy interests (Ainsworth, 1991; Feld, 1991; Stapleton, Aday, and Ito, 1982). In fashioning dispositions for juvenile offenders, juvenile court judges are challenged to balance the interests of rehabilitation and reintegration with both retributive and public safety concerns while managing systemic constraints and externalities in the local context.

These interests bear on decisions in which juvenile offenders are placed in nonsecure or secure confinement and for how long they will stay there. First, demands for penal proportionality create popular and political preferences for lengthy periods of secure placement when minors commit serious crimes (Zimring, 2000a, 2000b). Although this has led to an expansion of juvenile waiver activity (Fagan, 2008), the demand for punishment in the juvenile court also has led to dispositional regimes that include mandatory minimum sentences, sentencing guidelines developed by sentencing commissions or corrections authorities, and determinate or fixed sentences based on a sentencing “grid” (Feld, 1999; Torbet and Szymanski, 1998). Second, crime-control interests, which include deterrence and incapacitation, influence dispositional decisions by challenging judges to calibrate the lengths of placement that can maximize their impacts on future criminal activity (Feld, 1999).

But rehabilitative and therapeutic interests also are part of the calculus of dispositional decisions in the juvenile court and create a tension with punitive interests. The core philosophy of the juvenile justice system historically has been individualized justice, in which judges craft “custom-made” dispositions to youths based on a case-by-case consideration of each youth’s distinct rehabilitative needs (Connell, 1980; Feld, 1991; Horwitz and Wasserman, 1980; Zimring, 2000c). Whether these extralegal factors, such as substance abuse, mental health, school attendance, or family dysfunction, play a primary or secondary role in dispositional decisions depends on their interplay with legal factors (Campbell and Schmidt, 2000; Cauffman et al., 2007; Niarhos and Routh, 1992). When offenses are severe, legal factors may have primacy in dispositional decisions. And judges also may attempt to balance these therapeutic concerns with the risk of future criminality. For example, the severity of a juvenile’s sanctions for prior adjudicated offenses significantly influences the likelihood that a youth is placed in secure confinement (Cauffman et al., 2007).

Externalities, or systemic constraints, also may influence dispositional decisions, with the social context of the court being particularly influential (Britt, 2000). Courts located in communities with strong social control may be more willing to avoid placing individuals. Judges may wish to avoid placement for a youth with specific service needs but find that the absence of local services increases recidivism risks and, thus, skews dispositional decisions toward placement in either secure or nonsecure settings. The reverse also may be true; judges may elect placement precisely because intervention services are available that are not available locally or are inaccessible for administrative reasons.1

ASSESSING THE EFFECTS OF INSTITUTIONAL PLACEMENT IN THE JUVENILE SYSTEM

Few studies have been performed that estimate the impact of sanctioning experiences in the juvenile system on the future behavior of youth. As Levitt (1998:1158) concedes, “in contrast to the well-developed literature on deterrence and incapacitation effects of the adult criminal justice system, there is remarkably little previous academic research on the response of juvenile crime to sanctions.” The work to date on juvenile sanctions often has focused on changes in overall crime patterns concurrent with changes in transfer policies, and this work has sometimes produced conflicting results. Levitt (1998) posits some deterrent effect, as opposed to a mere incapacitation effect, to more serious sanctioning of juveniles, based on the decline in state-level crime associated with respective state differences in the age of majority. Similarly, Mocan and Rees (2005) attribute higher arrest rates across different locales with reducing the probability of engaging in serious crimes, such as assault, theft, and selling drugs. Analyses of longitudinal data involving individuals released from California Youth Authority institutions demonstrate that increased sanctions have had a crime reduction effect (Haapanen, 1990; Haapanen, Britton, and Croisdale, 2007). Conversely, others (Jensen and Howard, 1998; Lee and McCrary, 2005; Steiner, Hemmens, and Bell, 2006) find no significant changes in patterns of violent youth crime over time that correspond with policy changes, which shifts the juvenile court to a more punitive focus.

Few studies have been performed that estimate the impact on rearrest of varying lengths of institutional placement in the juvenile system. Several meta-analyses (Andrews et al., 1990; Garrett, 1985; Lipsey, 1999) have indicated that treatment in the juvenile system can lower the likelihood of rearrest, although institutional programs have only limited impact compared with community-based programs. In a more recent meta-analysis of interventions for serious juvenile offenders based on an analysis of 83 institutional outcome reports, Lipsey, Wilson, and Cothern (2000) found that the length of institutional treatment was related to the size of the treatment effect, with longer treatment (above an average of 25 weeks) producing larger decreases in rearrest. However, their findings also indicate that general program characteristics (e.g., how established the program is and the qualifications of those administering the treatment) were more related to the effect size than to the type or amount of treatment.

CONCEPTUAL ISSUES

The ability of the juvenile system to reduce criminal activity can rest on several possible mechanisms. As a result, any prediction about how increased sanctioning in the juvenile system might affect later criminal involvement is neither conceptually simple nor straightforward. Intervention in this system involves deterrence (rational choice) and developmental (rehabilitative) components, and the effects are likely more complex than a simple additive or multiplicative function. Different theoretical perspectives can lead to different predictions regarding the likely outcomes from increased juvenile justice sanctions (Sherman, 1993).

The rational choice framework would suggest that longer durations of confinement will reduce the probability of reoffending (Clarke and Cornish, 1985; Nagin, 1998), as placement out of the community is a price worth avoiding. Institutional confinement carries the costs of material restrictions and freedom restrictions (Fagan and Piquero, 2007) as well as possible victimization or labeling (Smith and Paternoster, 1990). Similarly, longer durations of exposure to rehabilitative services of the juvenile system should reduce the chances of subsequent crime (Andrews et al., 1990; Lipsey, 1999; Lipsey and Wilson, 1998), with these effects more pronounced in situations where more offenders are placed in private, instead of public, institutions (Pratt and Maahs, 1999; Shichor, 1998). Overall, it is not clear whether these effects interact, but they certainly both point in the same direction of more time in institutional placement lowering subsequent criminal involvement.

However, one can also posit a theoretical argument that longer stays in correctional placement may increase the probability of reoffending. First, longer exposure to noxious environments may be criminogenic (Agnew, 1992), undermining any potential positive effects of treatment. Second, psychological and developmental factors, which are important factors related to the risk or amenability of an adolescent offender (Fagan and Piquero, 2007; Mulvey and Iselin, 2008), may change over an extended period of institutional care, reducing the impact of any intervention efforts. Third, the longer the length of stay, the more tenuous the link between the crime committed and the punishment becomes in the mind of the offender. If the offender loses this important connection, then the moral component of confinement is ultimately lost, which could potentially lead to perverse effects, such as defiance and resistance (Piquero, Langton, and Gomez-Smith, 2004). Without an incremental contingency relationship between action and punishment, the impact of deterrence is lost.

In sum, competing arguments can be mustered to posit either decreases or increases from more or longer institutional placements in juvenile justice. In addition, multiple forces are at work that could interact to produce a positive or negative outcome from longer institutional stays. As noted, the ambiguity produced by the competing arguments in place for prediction is exacerbated by the current lack of empirical results in this literature.

This article aims to advance our knowledge and offer some clarity to this issue. It provides a formal empirical assessment of the impact of both institutional placement itself and varied lengths of institutional placement on the rearrest rate of serious adolescent offenders. We do not provide an explicit test of the deterrence framework, but we do provide a first empirical step in how it may be tested. We consider the implications of our results for criminological theory, as well as the possible next steps for research in the Discussion section.

HYPOTHESES

The goal of the analyses presented here is to identify the following two related yet distinct treatment effects: 1) the effect of placement, as compared with probation, on subsequent rates of offending, and 2) the marginal effect of length of stay in placement. Specifically, we test whether recidivism rates are greater for those youths who receive institutional placement compared with those placed on probation and, among those placed in institutions, whether their subsequent rates of offending decline as a function of the length of stay.

DATA

This article examines a subset of the research participants recruited for the Pathways to Desistance study, which is an ongoing, longitudinal investigation of the transition from adolescence to young adulthood in serious adolescent offenders. Participants in the Pathways to Desistance study are adolescents who were found guilty of a serious offense (almost entirely felony offenses) in the juvenile or adult court systems in Maricopa County, AZ, or Philadelphia County, PA. These youth were ages 14 through 17 years at the time of the study index offense. A total of 1,354 adolescents are enrolled in the study, which represents approximately one in three adolescents adjudicated on the enumerated charges in each locale during the recruitment period (November 2000 through January 2003). The study participants completed follow-up interviews at 6-month intervals for the first 3 years of their study involvement and annually thereafter. This investigation considers information through the first 4 years of follow-up. Overall, the study sample comprises minority (44 percent African American and 29 percent Hispanic) males (86 percent) with an average of two [standard deviation (SD) = 2.4] prior petitions to court. More information regarding the rationale and overall design of the study can be found in Mulvey et al. (2004); details regarding recruitment, a description of the full sample, and the study methodology are discussed in Schubert et al. (2004).

The current analyses exclude cases in which, as a result of the study index petition, either the petition was dismissed or the offender was transferred to adult court. This leaves a sample size of N = 921. We omit cases transferred to the adult system because, as noted, the adult and juvenile systems function differently, and experiences in these systems are qualitatively different. The purpose of the current study is to examine the marginal effect of institutional placement in the juvenile system. Of the 921 cases used for this analysis, 55 percent (nprob = 502) were put on probation, and the remaining (nplace = 419, 45 percent) were sent to an institutional placement.

Most (62 percent) of the placed youth spent most of their initial disposition stay in a contracted residential setting (see Mulvey, Schubert, and Chung, 2007 for setting type definitions). The remaining youth were placed across a range of setting types, which include state-run secure facilities (19 percent), contracted residential settings with a mental health focus (14 percent), and others (5 percent). The mean and median lengths of stay for the 419 placement cases were 11.0 and 10.4 months, respectively, with a standard deviation of 6.7 months, which indicates a fair amount of variability.

MEASURES

OUTCOME VARIABLES

Rate of Rearrest

We use the rate of rearrest as a measure of subsequent engagement in criminal activity. Indicators of arrest prior to age 18 were based on reports of petitions to juvenile court recorded in the Juvenile Online Legal Tracking System used in Maricopa County and the juvenile court database in Philadelphia County. Probation violations before age 18 were excluded because they do not necessarily represent a new offense and may indicate local practices as much as adolescent behavior. Arrests after age 18 were based on merging the court record information from each jurisdiction with nationwide Federal Bureau of Investigation arrest records.

For adolescents given probation at disposition, the rate of rearrest is calculated as the total number of arrests divided by total time on the street for the period after disposition to the end of the 48-month follow-up period. For adolescents who were sent to an institution, the rate of rearrest is the total number of arrests divided by the total time on the street for the period after release from the institutional stay ordered at disposition until the end of the 48-month follow-up period. In each case, the rate was standardized such that the outcome would be in terms of yearly rate. The main benefit of this measure is that it controls for exposure time in the community, as opposed to either a binary marker for rearrest or the total number of arrests postdisposition, which may be confounded with exposure time. The importance of controlling for exposure time has been noted elsewhere (Piquero et al., 2001).

Rate of Self-Reported Offending

Because arrest data are an officially based measure of offending, they will be limited to the events that lead to official detection. Thus, we also investigate whether institutional placement is related to a self-reported offending (SRO) measure (Huizinga, Esbensen, Weiher, 1991). The SRO measure used in this study asks participants to report their involvement in 22 more serious antisocial and illegal behaviors. The information was collected at each follow-up interview to capture illegal activity reported for the recall period. The outcome used for these analyses reflects the rate of antisocial activity across the 48-month follow-up period. It is the summed value of the “variety scores” (Thornberry and Krohn, 2000) from the reports obtained from each time point interview through 4 years divided by the number of days in the community during that time. As with rearrest, this outcome includes only reported antisocial behavior that occurred in the months between the time of release from the initial disposition stay and the end of the 48-month follow-up period. The values were standardized to a yearly rate.

We note the correlation between the SRO and the rearrest outcome is strong and positive (ρ = .47, p < .01 for all cases, and ρ = .46, p < .01 for only placement cases). This demonstrates that the two measures seem to be tapping the same underlying construct of involvement in illegal behavior, as they should given the strong concordance observed in the extant literature between self-report and official offending records (e.g., Brame et al., 2004). However, it also demonstrates that they are not perfect substitutes. Thus, to the extent that we arrive at substantively similar conclusions using these alternative measures, we can be more confident regarding the main findings of the article.

Covariates and Predictors

The main benefit of the Pathways data is it allows us to consider—and subsequently eliminate—a wide range of baseline variables as potential confounders related to selection bias. Specifically, we consider 66 covariates measured at baseline, which include demographic, familial, peer, legal, psychological, mental health, substance abuse, psychosocial maturity, and prior adjustment factors. Each of these was selected specifically to be ruled out as a potential confounder within the analysis, and because they have been found to be related to juvenile offending (Loeber and Farrington, 1998). We consider the potential for each to be a confounder in the next section. These covariates operationalize the influences of factors specified in an integrated theoretical framework for serious juvenile crime that incorporates individual, situational, developmental, and ecological influences on delinquency (Mulvey et al., 2004). More information on each of the 66 covariates, including descriptive statistics, can be found in appendix A.2

APPENDIX A

DESCRIPTIONS OF COVARIATES

DESIGN ISSUES

Two factors complicate research on the question of estimating sanction effects in juvenile justice, and these have in all likelihood contributed to the lack of empirical results on this question. First, not much data are available for making solid estimates of these effects. Specifically, there is a relative dearth of panel data sets appropriate for analyzing serious juvenile crime at an individual level (Lee and McCrary, 2005; Levitt and Lochner, 2000). In addition, most existing longitudinal studies (e.g., National Longitudinal Survey of Youth) often do not fully characterize the intricate features of sanctions, such as type, length, and intensity, nor do they typically quantify the individual offender’s response to them (Mulvey et al., 2004).

Second, even with appropriate data on individual sanctions, any attempt to examine the relations among sanctioning effects and future offending among adolescents in the justice system still faces complicated methodological issues. In particular, selection problems plague basic comparisons of outcomes between offenders with differing sanctions and limit the strength of any inferences that can be made. For instance, certain outcomes such as future recidivism may inherently depend on important, pre-existing characteristics of the individual, which may also be related to disposition decisions. Moreover, it is often difficult to address this selection problem because of the limited amount of information available regarding the case factors related to the type of sanction an individual receives (Mears and Field, 2000).

The Pathways to Desistance study provides a unique data set with which to circumvent both of these issues directly, as the data allow for the proper modeling of both sanctions and outcomes of individuals, as well as providing a wide range of background characteristics collected prior to the individual receiving sanctions. This richness allows us to incorporate propensity score methods (Rosenbaum and Rubin, 1983), as well as extensions to this methodology that allow for multiple treatments resembling a dosage (Lu et al., 2001; Zanutto, Lu, and Hornik, 2005). Using these approaches, we can rigorously control for selection biases, ruling out 66 potential confounders. Such an approach is rare in criminology generally and has not yet been applied to the theoretical and empirical issues addressed here regarding serious juvenile offenders. Nagin, Cullen, and Lero-Jonson (2009) have recently commented that such a dose-response approach is the proper way in which to model the relationship of longer sentence lengths on future recidivism to avert problems associated with traditional regression-based methods, but it has not yet been explored in most prior literature.

ANALYSES

The primary goal of this analysis is to identify two related yet distinct treatment effects—the effect of placement and, conditional on an individual being placed, the marginal effect of length of stay on future criminal offending. This section begins by exploring unadjusted differences in rates of rearrest and SRO between groups of individuals who receive different placements and lengths of stay, as well as preexisting differences in terms of baseline covariates between these same groups. We then introduce the propensity score methodology and relevant extensions necessary to adjust for these existing differences between the groups in calculating the effects for placement and increased length of stay.

UNADJUSTED COMPARISON OF OUTCOMES AND PLACEMENT

We begin by examining the mean outcome of rearrest conditional on placement status. For those individuals who were placed out of the community, the mean rate of rearrest was 1.20 arrests per year, postdisposition. In comparison, those who received probation had a mean rate of about .63 arrests per year. This unadjusted comparison of the probation group versus the placement group in terms of rate of rearrest shows that, on average, the placement group is rearrested at a higher rate than those who received probation, by approximately .57 arrests per year (p < .001). Similarly, for SRO, those individuals who are placed report, on average, about 2.5 more offenses per year of time in the community (10.9 vs 8.3 reported offenses per year) than those individuals who receive probation (p < .05). However, as we mentioned, preexisting differences likely exist between the placement and probation groups, which make a causal interpretation of this difference impossible. We consider these differences in more detail in the next section.

UNADJUSTED COMPARISON OF RATE OF LENGTH OF STAY AND REARREST

We also examine the unadjusted association between length of stay and future rate of rearrest. The plot of these two variables for only those 419 who were sent to institutional placement is shown in figure 1. Two things are apparent during visual inspection. First, there seems to be no strong relation between longer length of stay and future rate of rearrest. For verification, note that ordinary least squares (OLS), where the rate is the dependent variable and length of stay (in months) and its square are explanatory terms, yields neither linear nor quadratic slope terms significantly different from zero.3 Second, figure 1 shows some interesting aspects of the distribution of length of stay, which may limit our ability to generalize any results. Dramatic heterogeneity is observed in the upper tail of the distribution, specifically beyond 15 or 16 months, where a few cases have long lengths compared with the rest of the group. Furthermore, many of these cases with lengths greater than 20 months have rates of rearrest that are zero. Given that we only observe a follow-up period of 48 months, it is likely that the outcome itself is confounded with length of stay for some of these extreme cases; offenders with longer lengths of stay simply cannot resume offending to the likely rate in the brief time they are exposed in the follow-up period. As we will discuss, we consider these data limitations a restriction on our ability to generalize the results on stays with lengths exceeding 15 months, which encompasses a small part of our data.

Rate of Rearrest versus Length of Stay

INITIAL IMBALANCES AMONG COVARIATES

The main problem in comparing the rearrest and self-reported offending outcomes across individuals who receive different sanctions centers on the idea that confounding factors may affect each offender’s likelihood of receiving such sanctions. Furthermore, these factors may not only be associated with placement and/or length of stay but also with the individual’s subsequent arrest or self-reported offending. Thus, the comparison of offenders who receive different sanctions involves a contrast of two groups that are dissimilar in important preexisting ways. In formal statistical parlance, because there is no random assignment to the treatment condition, the treatment and control groups are not directly comparable, unless these initial differences are taken into account.

We start by examining initial covariate balance, that is, differences between groups prior to entering any placement, to determine whether such observed differences must be corrected. To do this, we consider 66 covariates measured at baseline, each of which was selected to be ruled out specifically as a potential confounder within the analysis. We assess the imbalance in this set of covariates in two ways. First, we address imbalances in the placement versus probation groups by examining a standardized bias statistic, as advised by Rosenbaum and Rubin (1985), for a difference in means between these two groups for each of the covariates. The standardized bias statistic is the mean difference as a percentage of the average standard deviation, 100(x̄ T−x̄ C)/[(s2 T+s2 C)/2]1/2, where for each covariate, x̄ T and x̄ C are the sample means in the treated group and the control group, respectively, and s2 T and s2 C are the corresponding sample variances. Rosenbaum and Rubin indicate that a standardized difference percentage value greater than 20 for any covariate would suggest that the covariate is out of balance. Second, in this comparison of probation versus placement, we compare the differences in covariates between each group in terms of an ordinary difference in means test (t-statistic). A t-statistic exceeding an absolute size of 1.645 would correspond to a significant difference in the sample means at α = .10, and for our purposes, it would be considered to be out of balance. Second, to assess initial imbalances among the length of stay, we consider a simple correlation coefficient ρ for each covariate with length of stay (conditional on the placement cases), as well as the associated p-value for a test of no association, where a p-value less than .10 would suggest an initial imbalance.

Table 1 reports these values for each of the 66 covariates considered. Prior to treatment (i.e., placement), 41 of the 66 covariates considered are initially out of balance in the probation versus placement comparison. Furthermore, the imbalance between placement and probation groups is on covariates that are likely related to future offending (e.g., number of priors, exposure to violence, personal cost of punishment, personal certainty of punishment, and psychopathy). In the length of stay comparison, 28 of 66 covariates are out of balance. Again, these preexisting differences are on covariates (e.g., offending history, exposure to violence, legal cynicism, and association with antisocial peers), which are arguably related to both length of stay and future offending. This seems to suggest a high degree of potential selection bias, which may contaminate basic comparisons of effects in both of these cases.

Table 1

Baselines Covariates and Initial Balance

BALANCING TREATMENT AND CONTROL GROUPS—PROPENSITY SCORES

The two estimations for this study require different methods to compute propensity scores. In the case of evaluating the effect of placement, the treatment is binary (i.e., placement versus probation), and thus we can use traditional propensity score methods (Rosenbaum, 2002; Rosenbaum and Rubin, 1983) to eliminate bias caused by observable covariates. In the second case of evaluating a marginal length of stay effect, the treatment, length of stay, is no longer binary, but it can be thought of as resembling a dosage. Thus, we employ recent extensions to propensity score methodology (Lu et al., 2001; Zanutto, Lu, and Hornik, 2005) to account for selection biases in this case. We briefly examine both of these methods in more detail.

PROPENSITY SCORES FOR PLACEMENT

The propensity score represents the probability that an individual receives some treatment conditional on a vector of observed covariates (Rosenbaum, 2002). Rosenbaum and Rubin (1983) show that, conditional on two individuals, one treated and one control, with an identical propensity score, the difference in treatment status becomes independent of all observable characteristics, which is the assumption of a strongly ignorable treatment assignment. The idea is to estimate a propensity score for each individual and in turn to use this estimate as a method for creating balance on key covariates that may be confounding the treatment effect estimate. To estimate the propensity score, we employ a binary logit model, with the binary outcome of institutional placement (versus probation), and some combination of the 66 initial covariates, including squares and interactions, as explanatory terms. We treat model selection as an iterative process, where parsimony is unimportant as compared with achieving a covariate balance afterward (Heckman et al., 1998; Rosenbaum and Rubin, 1984). The predicted probability from the final model for each individual ê (x) is thus that individual’s estimated propensity score.

After estimating a propensity score for each individual, instead of conventional matching, we employ stratification, where subjects are divided into equal-sized subgroups based on the propensity score distribution. Stratification can be thought of as a special form of matching where subjects are grouped, rather than paired, with other individuals within a certain range of propensity scores. Rosenbaum and Rubin (1984) contend that stratification using quintiles (i.e., five equally sized subgroups) can remove approximately 90 percent of the initial imbalance in each baseline covariate. The subsequent average effect of treatment on the treated can be estimated as a weighted average of within-stratum probation-minus-placement mean differences in outcome Y:

ATTplace=∑j=1Jnplace,j nplace[Y¯place,j−Y¯prob,j ]

(1)

where there are j = 1,…, J number of strata, nplace, j and nplace represent the total number of placed individuals in stratum j and overall, respectively, and Ȳplace, j and Ȳprob, j denote the mean rate of reoffending for placed and probation individuals, respectively, in stratum j.

Dosages: Propensity Scores for the Length of Stay

Unlike the initial analysis of placement, in which treatment is binary (i.e., placement versus probation), estimating the effect of longer lengths of stay requires modeling treatment in a form that resembles a dosage. Thus, a modification of traditional propensity score methods becomes necessary. Lu et al. (2001) extend the basic propensity score framework described above to accommodate treatments with multiple doses. Zanutto, Lu, and Hornik (2005) extend this work to show how subclassification (i.e., stratification) can be used with treatment doses to estimate a dose-response curve. In the current application of length of stay, we restrict the analysis to the nplace = 419 cases who actually received placement as a sanction. This is done to differentiate those individuals who received probation, who technically have a length of stay equal to zero, as having received a unique sanction, and thus avoid the introduction of an additional and highly problematic selection problem into the analysis.

When considering the length of stay, an individual receiving a longer length of stay can be thought of as being treated with a larger “dose.” The distribution of dose propensity can be modeled using an ordinal logit model (McCullagh, 1980), generating a single balancing score, similar to the binary treatment case (Joffe and Rosenbaum, 1999). In this model, this distribution of doses for some individual i, Di, conditional on a vector of observed covariates xi, can be written as follows:

log[P(Di≥ d)P(Di<d)]=αd+βxi,ford=2,3,4,…

(2)

In this model, the distribution of doses given covariates depends on the observed covariates only through b(xk) = βxi, so that the observed covariates x and the doses D are conditionally independent given the scalar b(xi). Thus, b(xi) is a balancing score and conditional on b(xk); the dose D is assumed to be strongly ignorable. Therefore, after matching (as described in Lu et al., 2001) or subclassifying (as described by Zanutto, Lu, and Hornik, 2005), which we employ in the current study, we can balance the distribution of a large number of covariates simultaneously as we would in the case of binary treatment. Also, as in the binary case, we use an iterative approach for model selection to maximize covariate balance after stratification. A dose-response curve can then be estimated, as a stratum- weighted mean of rearrest conditional on receiving dose D. After subclassification on balancing score quintiles, the estimated mean response for individuals receiving dosage d can be written as follows:

where Ȳd,j is the observed mean outcome among individuals receiving dosage level d in balancing score quintile i.

DEFINING TREATMENT DOSES

To apply the methods described by Lu et al. (2001) and Zanutto, Lu, and Hornik (2005) to the length of stay analysis, it is necessary to define discrete dosage categories of length of stay, or comparable segments for analysis. Figure 2 shows a histogram of length of stay in months. As an initial cut, we use rough adherence to the quartiles of this distribution, which in turn gives us four dosage categories of 1) 0–6 months, 2) 6–10 months, 3) 10–13 months, and 4) >13 months. However, given that the delineation of these categories is rooted in the data and not necessarily in practice, we also consider an alternative option, which defines doses in five 3-month intervals of 1) 0–3 months, 2) 3–6 months, 3) 6–9 months, 4) 9–12 months, and 5) >12 months. These categories, although more attractive in a practical sense, are not all of similar sizes. Lu et al. (2001), however, propose the use of five discrete dosage groups of unequal size membership, which would justify this choice. We estimate a response curve in each of these instances.

Histogram of Length of Stay (in months)

HETEROGENEITY IN PLACEMENT

An additional yet critical factor to consider when modeling juvenile sanctioning is the heterogeneity among institutional settings. For instance, placement in a state training school is likely a different experience from placement in a contracted residential mental health facility. These differences are likely confounded with both length of stay and outcome.

We cannot completely stratify by facility type and conduct unique analyses by each type because of sample size limitations. We can, however, examine how two broad differences in the characteristics of settings that may potentially affect our results—intensity of services provided and public versus private status of the institution. Intensity of service provision is a relevant consideration, because prior literature indicates that any type of sanctioning without the provision of rehabilitative services does not reduce future recidivism (Andrews et al., 1990; Lipsey, 1999; Lipsey and Wilson, 1998). In addition, there is some debate that economic motivation for profits may lead to quality differences between public and private institutions (Pratt and Maahs, 1999; Shichor, 1998). However, Armstrong and MacKenzie (2003) provide evidence that no such environmental differences exist between public and private institutions when assessing juvenile cognitive outcomes.

Given that exposure to services occurs simultaneously with the statistical treatment effect of interest, the length of stay, we cannot simply treat it as an ordinary covariate over which we may achieve balance and thus rule out as a confounder (e.g., number of prior convictions). Similarly, it could be argued that if material differences were found between environments in public and private facilities, then this would, in effect, constitute different treatments (in a statistical sense) altogether. As such, we consider these differences in a cruder manner than we do with length of stay.

First, for a subset of individuals, we can observe intensity of service exposure while in placement. To measure intensity of services, a modified version of the Child and Adolescent Services Assessment (CASA; Burns et al., 1992) was administered during each follow-up interview with the adolescent. The CASA is designed to assess the use of mental health and social services via self-report from youth ages 8–18 years. The instrument collects information regarding the occurrence of an institutional stay and the length of the institutional stay via self-report (Ascher and Farmer, 1996). If an institutional stay of longer than 7 days was reported, then information is also obtained about the types and frequency of services provided during the stay. A measure of the intensity of services provided during the institution stay ordered at disposition was calculated by dividing the total frequency of the services reported by the total number of days in that institutional stay. Because early versions of the interview did not contain the information needed to generate this score, a value for the intensity of services received was only available for a subset of study participants (n = 289). Second, for the entire placement sample (n = 419) we grouped our facilities into two broad types: public (i.e., jails, prisons, and state-run secure facilities) or private (i.e. contracted residential and drug and alcohol facilities).

Table 2, panel A, reports how each factor is associated with the future rate of rearrest using OLS. Column 1 shows no significant linear or quadratic association of length of stay to future rate of rearrest, as discussed.4 Column 2 shows that after controlling for length of stay, there is similarly no significant association with intensity of service exposure. Columns 3 and 4, however, do yield an interesting result, with an intensity–length of stay interaction significant at α = .05, when also controlling for both the linear and quadratic factors. The striking thing here is the positive sign of the interaction effect, which shows that continued higher exposure over time is actually associated with a higher rate of future rearrest, which suggests that a strong selection mechanism for sentence length must be considered when discussing service exposure. This result is reinforced in table 2, panel B, which reports similar analyses using rate of self-reported offending as the outcome. Again, in this case, we notice a marginally significant and positive effect of the length and intensity interaction. As such, in the context of the current analysis of length of stay, we cannot effectively rule it out as an explanatory term in the process. We consider the implications of this in the Discussion section.

Table 2

Difference Among Institutions Explaining Outcomes

Columns 5 and 6 include the entire sample stratified on the public or private delineation. In neither case is there a significant linear or quadratic relationship with length of stay and either future rate of rearrest or self-reported offending. We assume, then, that this distinction is not relevant in our interpretation of the results.

RESULTS

COVARIATE BALANCE AFTER STRATIFICATION

After stratification on the estimated propensity score, it is necessary to reassess covariate balance in both the placement group comparison and the group stratification by dose of length of stay. Essentially, we need to determine whether, after stratification, any covariates still can strongly predict placement instead of probation (or, in the length of stay case, predict who gets a longer length of stay). If so, there could still be selection bias attributable to out-of-balance confounders, and thus, the propensity score must be reestimated, in a model selection process described above, until balance is finally achieved. Again, it is important to remember that in each case, covariate balance, and not model parsimony, is the primary goal of estimating the propensity score (Rubin and Thomas, 1992).5

To reevaluate balance in the case of placement versus probation, we employ two-way analysis of variance (ANOVA) for each covariate, where binary exposure to treatment (i.e., placement) was one factor, the propensity score quintile was a second factor, and the covariate itself was the dependent variable. If balance is achieved, then there should neither be a statistically significant main effect of placement on the covariate nor a statistically significant interaction effect of exposure by quintile. Any F-statistic exceeding 3.84 would suggest a significant effect and, hence, that the covariate was still out of balance. If these two conditions are not met, the propensity score was reestimated by adding quadratic or interaction terms with those covariates that remained out of balance to the propensity score model specification. In the case of placement versus probation, after stratification on the final propensity score specification, 64 of the 66 covariates balanced, which means that we may reasonably rule these covariates out as confounders. Notice that this is actually better balance among observable covariates than one might expect from randomization, because, at α = .05, we would expect about 3 of these 66 to be out of balance.

We also need to evaluate balance after stratification in the length of stay application. Again, we check this for each covariate by using a two-way ANOVA with dose and propensity score quintile as the two factors and the covariate as the dependent variable. Again, if balance is achieved, then there should neither be a statistically significant main effect of dose on the covariate nor a statistically significant interaction effect of dose by quintile. As with the placement application, the propensity score model was reestimated in an iterative manner until reasonable balance was achieved. Again using the F > 3.84 threshold to gauge balance, after stratification with the final model, 65 of the 66 covariates were balanced, meaning they can be reasonably ruled out as confounders. Once again, this is better balance among observable covariates than we would expect through randomization.

EFFECT OF PLACEMENT ON OUTCOMES

After accounting for potential selection biases in the 66 covariates, the magnitude of the difference in mean rate of rearrest between the placement and probation groups falls considerably, from .570 in the unadjusted comparison to .167, which is a reduction of about 70 percent. Furthermore, the associated p-value for this effect is .115, which suggests that, although not significant at conventional levels, this effect of placement is actually still negative, indicating an increased effect on the future rate of rearrest for placement compared with probation. It is important to note that, even if this difference were statistically significant, the magnitude is so small (that is, we would expect the placed offenders to have an extra rearrest only about once every 6 years compared with offenders on probation) that it is also not significant in a policy sense. However, the more important implication from this null finding would seem to be that, even after accounting for selection effects, the results show no marginal gain from placement in terms of averting future offending. Similarly, the difference in mean rate of SRO falls from about 2.5 in the unadjusted comparison to .31 after stratification (p = .81). Again, we uncover no significant differences in SRO rates across the two groups, replicating the null effect that we observed using the rearrest outcome.

LENGTH OF STAY EFFECT ON OUTCOMES

After stratification, we estimate the conditional expected rate of rear-rest as a stratum-weighted average for each dosage level. Note that it is not straightforward how to conceptualize an average effect of treatment on the treated in the multiple dose case. Rather, a point on this curve can be thought of as a conditional expectation relative to the other dosage levels, and thus, this set of conditional expectations yields what we may interpret as a dose-response curve. Figure 3 reports response curves for both rearrest and self-reported offending outcomes for doses as defined by the quartiles. Notice that the shapes of these curves are both essentially flat, at least through 13 months. Not only are these differences not statistically significant, but also the very subtle magnitude of these differences suggest there is no practical benefit, even if they were statistically significant. 6 This suggests there is little or no marginal benefit, at least in terms of reducing future rate of offending, for retaining an individual in institutional placement longer. Although there does seem to be a slight reduction after 13 months, we are cautious about making an inference for this group, because, as noted, there is large heterogeneity in length as well as potential confounding with the outcome (because of a lack of reasonable exposure time) in this highest dose group.

Response Curves by Quartile Dose Category

Figure 4 reports response curves for the case where doses were defined in terms of 3-month intervals. Again, this more differentiated picture of the length of stay may be a more policy-relevant way in which to consider the dose delineation. As was the case when using the quartiles to define dose groups, there again seems to be flat response curves for both rearrest and SRO in terms of the intermediate lengths. However, notice in this case, there does seem to be somewhat of an effect for staying longer than 3 months, as both the rate of rearrest and SRO drop thereafter by a non-trivial amount. Although it is apparent that there may be some mechanism at work here, we are again restricted in our ability to interpret it fully as the number of observations in the 0–3-month dose group is too small to provide adequate statistical power with which to draw sound inferences.

Response Curves by 3-Month Dose Category

In both these cases, it seems that, for lengths of stay between 3 and 13 months, there is no marginal benefit for retaining an offender in institutional care for longer periods of time. Although we cannot make solid inferences outside this range, this result is nonetheless dramatic.

DISCUSSION

This article extends the discussion of how serious juvenile offenders respond to placement and longer stays out of the community. We show two distinct, yet related, results. First, we find that after rigorously controlling for selection, there is an overall null effect of placement on future rate of rearrest and future rate of self-reported offending. Furthermore, if any effect of placement is present, it is likely slightly negative, on average, raising the likelihood of rearrest. Second, conditional on an individual receiving placement, there appears to be no marginal benefit, in terms of reducing future rate of rearrest or rate of SRO, for additional length of stay; the conditional response curve is essentially flat, at least through 13 months. In addition, we find no significant variation in outcome, at least at a macro level, for public versus private facilities, in contrast to some earlier literature (Pratt and Maahs, 1999). These results certainly raise questions about whether the rehabilitative mission of the juvenile justice system is being met efficiently or effectively.

These results have ramifications on the direction of the juvenile court, which is tending toward a more punitive-oriented orientation (Jenson and Howard, 1998), with “get-tough” policies that are perhaps not as entirely effective as hoped for by proponents. Although this analysis is not presented as a formal test of the deterrence framework, it does align with the thesis that there is little benefit from increased deterrence connected with either placement out of the community or longer stays in institutions. It is still an open question whether this is the result of a belief among potential juvenile criminals that the sanctions they will likely face in the juvenile system are “easy,” as Levitt (1998) argues, but these results do nothing to dispel such a proposition. Moreover, if there is no deterrent effect of the juvenile system, it becomes even more imperative that a discussion of the effects of juvenile sanctioning must be framed in terms of the marginal rehabilitative benefit of the system. Namely, we would expect there to be an additional benefit in terms of normal reintegration into non-criminal activity for those serious juvenile offenders who are exposed to the system to a greater degree.

These findings also have implications for the broader issue of costs and benefits from the use of institutional care in juvenile justice. If there is no marginal rehabilitative effect for placement or longer exposure, it calls into question the need to expend resources on extended institutional care. A full cost–benefit analysis of these results, however, would be a complicated task. As Austin (1986) showed with data regarding the effects of early release in Illinois, a considerable economic gain was realized by downsizing the placement population. Even though this result was applied to the adult prison system, parallels can be drawn to the juvenile justice system in a very literal economic sense. As Austin points out, the huge economic savings are reduced by the cost of the crime attributable to these early release cases. If, as our results suggest, there is no evident gain from reduced crime, then one might argue there would be little cost to offset the economic-operating cost benefits. We urge readers to be cautious of such an interpretation, however, given that, even if there is no marginal rehabilitative benefit to reducing future crime, there is still an inherent incapacitation effect with placing serious adolescent offenders. This effect was not estimated in the current analyses, and any full assessment of the implications of the current findings would have to do so.

It is also important to remember that these analyses only considered outcomes related to reoffending. The effects of placement on many non-crime, developmental outcomes critical to positive adjustment in early adulthood are generally unknown (Scott and Steinberg, 2008). It may be that, even though this analysis attempts to balance over age at placement, there are substantive, underlying differences based on age and maturity that contribute in unique and offsetting ways to the overall null effect observed. Steinberg and Cauffman (2000) argue that, in the context of the boundaries of the juvenile court, developmental differences among offenders may lead to different responses, either positive or negative, to an otherwise uniform treatment. For example, placement may be problematic if it labels, and thus stigmatizes, less serious offenders and prevents their successful reintegration into roles connected with a successful transition into nonoffending adulthood. Furthermore, the prior point of an economic cost–benefit at some level must also consider these developmental consequences, as certain outcomes such as delayed development or labeling issues, which might affect items such as future labor market participation, must also be considered as potential economic costs or even possible dead-weight loss. If there are such developmental costs, then any economic analysis that does not include them is likely too muted an estimate of impact. An interesting extension to the current analysis would be to examine how certain developmental outcomes are affected by placement and longer lengths of stay.

An additional methodological note involves the choice to test the dose-response relation using discrete categories for the length of stay variable, as in the Lu et al. (2001) and Zanutto, Lu, and Hornik (2005) extensions to the propensity score literature. An alternative would be to treat the length of stay as a purely continuous treatment and, thus, to estimate a smooth function as the derivative of the response function, an approach taken by Hirano and Imbens (2004). We choose the discrete approach for two reasons. First, given its general similarity to the basic binary treatment case that we employ in the placement effect analysis, it is a relatively easily understood and straightforward extension. Second, although the idea of modeling a continuous treatment is conceptually appealing, it is unlikely that juvenile justice professionals making determinations about the need for longer institutional stays would be thinking in such a manner, as compared with more discrete chunks—like months. Review hearings regarding extended placements occur at regular intervals, and juvenile justice system professionals often use blunt assessments regarding the need for continued treatment involvement (Mulvey and Iselin, 2008). We must strongly emphasize, however, that given our results show a flat response curve, it is unlikely that a different modeling approach of the length of stay effect would have made a substantial, if any, difference in the substantive interpretation of the results.

One important caveat to our findings is that we cannot rule out intensity of exposure to rehabilitative services as a confounder. Given the contention that the provision of an adequate amount of rehabilitative services is perhaps the most important factor to reduce future recidivism (Andrews et al., 1990; Lipsey, 1999; Lipsey and Wilson, 1998), this is one potential limitation of these results. There is some prior evidence from this study that exposure to substance use services may reduce both crime and self-reported offending (Chassin et al., 2009; Mauricio et al., 2009). Our analyses of this issue, performed with only a subsample of the larger group and a gross measure of treatment involvement, indicate that longer stays with more intensive services seem to increase the risk of rearrest. It is possible that longer exposure to services of poor quality, or services that are poorly matched to the individualized needs of incarcerated offenders, may have iatrogenic effects. But these results could also be the result of a selection effect in which adolescents with a high propensity for future offending spend greater time in high-intensity environments, but to no avail. Considering the complexity of properly conducting an analysis of the effects of treatment intensity as well as its profound implications, the question of the true role of service exposure is a high-priority issue for future research.

We hope that this effort begins a course of research on the effects of institutional placement generally, and the dose of institutional confinement in particular, with respect to policy decisions regarding juvenile offenders. Our research shows a general lack of support for lengthy periods of placement and indirectly underscores the movement toward increased use of non-placement/community-based alternatives, especially for those offenders who do not evince the highest risk (Greenwood, 2005). Given the small increases in crime currently being observed among this population, the growth in the number of such individuals in the coming years, and the significant costs associated with (lengthy) incarceration decisions, a more accurate account of whether such policies lead to more or less subsequent criminal activity is of central theoretical, empirical, and policy significance.

Biographies

• 

Thomas A. Loughran, PhD, is an assistant professor in the Department of Criminology at the University of South Florida. His research interests include deterrence and individuals’ responses to sanctions, as well as methods to infer treatment effects from nonexperimental data. He received his PhD in public policy from Carnegie Mellon University in 2007.

• 

Edward P. Mulvey, PhD, is a professor of psychiatry and Director of the Law and Psychiatry Program at the University of Pittsburgh School of Medicine. Professor Mulvey has performed several large-scale studies regarding the link between mental illness and violence, the accuracy of clinical judgment regarding violence, and the development of serious adolescent offenders. He has published empirical findings and conceptual articles in psychiatry, psychology, and criminology journals, and he has worked with federal agencies and local programs regarding prevention and treatment approaches for violent adolescents and adults.

• 

Carol A. Schubert, MPH, is the Research Program Administrator for the University of Pittsburgh School of Medicine, Law and Psychiatry Program. She has directed several large-scale studies regarding the link between mental illness and violence, the accuracy of clinical judgment regarding violence, and the development of serious adolescent offenders. Her recent publications have focused on services for adolescent offenders and symptom change in relationship to violence among individuals with mental illness.

• 

Jeffrey Fagan is a professor of law and public health at Columbia University and Director of the Center for Crime, Community and Law at Columbia Law School. His research focuses on crime, law, and social policy. His current research examines capital punishment, racial profiling, legal socialization of adolescents, the jurisprudence of adolescent crime, and the legitimacy of the criminal law. He has served on the Committee on Law and Justice of the National Academy of Science and the MacArthur Foundation’s Research Network on Adolescent Development and Juvenile Justice. He is a fellow of the American Society of Criminology.

• 

Alex R. Piquero is a professor in the Department of Criminology & Criminal Justice and an affiliate of the Maryland Population Research Center at the University of Maryland College Park, Member of the Mac-Arthur Foundation’s Research Network on Adolescent Development & Juvenile Justice, Executive Counselor with the American Society of Criminology, and coeditor of the Journal of Quantitative Criminology. His research interests include criminal careers, criminological theory, and quantitative research methods.

• 

Sandra H. Losoya is a research assistant professor of psychology at Arizona State University. She received her PhD at the University of Oregon and studies emotional development and sources of risk and resilience in children and adolescents. A recent publication is as follows: Losoya, S. H., Knight, G. P., Chassin, L., Little, M., Vargas-Chanes, D., Mauricio, A., and Piquero, A. 2008. Trajectories of acculturation and enculturation in relation to binge drinking and marijuana use in a sample of Mexican-American serious juvenile offenders.

Footnotes

*The project described was supported by funds from the following organizations: Office of Juvenile Justice and Delinquency Prevention, National Institute of Justice, John D. and Catherine T. MacArthur Foundation, William T. Grant Foundation, Robert Wood Johnson Foundation, William Penn Foundation, Center for Disease Control, National Institute on Drug Abuse (R01DA019697), Pennsylvania Commission on Crime and Delinquency, and the Arizona Governor’s Justice Commission. We are grateful for their support. The content of this article, however, is solely the responsibility of the authors and does not necessarily represent the official views of these agencies.

1All this information suggests that the juvenile system is more complex than the adult system in that it includes intricacies that affect the outcome of a case (i.e., placement decisions) beyond a verdict of guilt. An anonymous reviewer argued that rarely is it stated that a youth is sent to a residential facility to be deterred, in part because judges are hesitant to remove a youth from a family environment unless the environment itself is unhealthy or the youth can receive greater benefit from placement elsewhere. More often, factors that determine placement outside the home include an objective and subjective determination of the level of needs with respect to treatment, the availability of treatment or other resources in the youth’s community, the status of the youth’s family—which includes the criminal involvement and/or drug use of the parents—the ability of the youth to be transported to treatment programs if they are not placed outside the home, and so on. Thus, the presence or absence of an effect based on the length of stay may have to do more with the type and fidelity of treatment received in an institution rather than whether the youth perceived the placement to be “bad enough” to deter them from future delinquency. In short, the reviewer is correct in highlighting the importance of considering information on the treatment received when studying issues related to placement and length decisions in juvenile decision making, because many factors exist beyond experiencing a sanction that could produce any observed effect.

2More information about the instruments and calculations used to generate these covariates can be obtained from the lead author.

3Although we acknowledge that this simple specification is in no way intended to be an accurate portrayal of reality, we include it to highlight the basic relationship between dose and outcome. Thus, the insignificance of the linear and quadratic relationships of length of stay to rearrest are simply intended to demonstrate a lack of simple association but are not intended to test a de facto model of their relationship.

4Again, we must point out the limitations of these simple OLS models in testing a proper relationship between service exposure and future recidivism. The statistical tests of the coefficients presented in table 2 are merely intended to show that a basic relationship between service exposure and future recidivism, conditional on length of stay, cannot be ruled out, and is not de facto evidence that such a statistically significant relationship actually exists. Indeed, we advocate for future research to develop a more formal, explanatory model of this relationship.

5It is also important to note, however, that no statistical or econometric methodology can match the power of a pure experiment to identify effects; and although we can reasonably eliminate a wide range of observable covariates as confounders, there may still be hidden biases confounding our results (Rosenbaum, 2002). Yet, given the richness of pretreatment covariates, we are confident that we eliminate substantial amounts of likely bias.

6Zanutto, Lu, and Hornik (2005) note that although there is a manner in which to estimate corresponding standard errors for estimated points on the response curve, these estimates are inherently biased because of a dependency of responses based on the nature of the propensity score calculation. However, they may be reasonable approximations (Benjamin, 2003). Thus, we do not report standard errors for the points on the response curve given not only the above caution, but additionally, considering the small magnitude of the differences between point estimates, we believe that it would be highly unlikely to reject a null hypothesis of no difference.

Contributor Information

Thomas A. Loughran, Department of Criminology, University of South Florida.

Edward P. Mulvey, School of Medicine, University of Pittsburgh.

Carol A. Schubert, School of Medicine, University of Pittsburgh.

Jeffrey Fagan, Columbia Law School, Columbia University.

Alex R. Piquero, Department of Criminology & Criminal Justice, University of Maryland—College Park.

Sandra H. Losoya, Department of Psychology, Arizona State University.

REFERENCES

• Agnew RobertA. A foundation for a general stain theory of crime and delinquency. Criminology. 1992;30:47–88. [Google Scholar]
• Ainsworth JanetE. Re-imagining childhood and reconstructing the legal order: The case for abolishing the juvenile court. North Carolina Review. 1991;69:1083–1133. [Google Scholar]
• Allen FrancisA. Foreword. In: Fagan Jeffrey, Zimring FranklinE., editors. The Changing Borders of Juvenile Justice: Transfer of Adolescents to the Criminal Court. Chicago, IL: University of Chicago Press; 2000. [Google Scholar]
• Andrews DonA, Zinger Ivan, Hoge RobertD, Bonta JamesA, Gendreau Paul, Cullen FrancisT. Does correctional treatment work? A clinically relevant and psychologically informed meta-analysis. Criminology. 1990;28:369–404. [Google Scholar]
• Armstrong GayleneS, MacKenzie DorisL. Private versus public juvenile correctional facilities: Do differences in environmental quality exist? Crime & Delinquency. 2003;49:542–563. [Google Scholar]
• Ascher BarbaraH, Farmer ElizabethMZ. The child and adolescent services assessment (CASA): Description and psychometrics. Journal of Emotional and Behavioral Disorders. 1996;4:12–20. [Google Scholar]
• Austin James. Using early release to relieve prison crowding: A dilemma in public policy. Crime & Delinquency. 1986;32:404–502. [Google Scholar]
• Beccaria Cesare. On Crimes and Punishment. Indianapolis, IN: Hacket Publishing; 1986. [1764] Translated by David Young. [Google Scholar]
• Benjamin DanielJ. Does 401(k) eligibility increase saving? Evidence from propensity score subclassification. Journal of Public Economics. 2003;87:1259–1290. [Google Scholar]
• Bhati AvinashSingh, Piquero AlexR. Estimating the impact of incarceration on subsequent offending trajectories: Deterrent, criminogenic, or null effect? Journal of Criminal Law and Criminology. 2008;98:207–254. [Google Scholar]
• Blumstein Alfred, Piquero AlexR. Criminal careers research and rational sentencing policy. Criminology & Public Policy. 2007;6:679–688. [Google Scholar]
• Blumstein Alfred, Wallman Joel., editors. The Crime Drop. New York: Cambridge University Press; 2000. [Google Scholar]
• Brame Robert, Fagan Jeffrey, Piquero AlexR, Schubert CarolA, Steinberg Laurence. Criminal careers of serious delinquents in two cities. Youth Violence and Juvenile Justice. 2004;2:256–272. [PMC free article] [PubMed] [Google Scholar]
• Britt ChesterL. Social context and racial disparities in punishment decisions. Justice Quarterly. 2000;17:707–732. [Google Scholar]
• Burns BarbaraJ, Angold Adrian, Magruder-Habib Kathryn, Costello ElizabethJ, Patrick MorganKateS. The Child and Adolescent Services Assessment (CASA) Durham, NC: Duke University Medical Center; 1992. [Google Scholar]
• Campbell MaryAnn, Schmidt Fred. Comparison of mental health and legal factors in the disposition outcome of young offenders. Criminal Justice and Behavior. 2000;27:688–715. [Google Scholar]
• Cauffman Elizabeth, Piquero AlexR, Kimonis Eva, Steinberg Laurence, Chassin Laurie, Fagan Jeffrey. Legal, individual, and contextual predictors of court disposition in a sample of serious adolescent offenders. Law and Human Behavior. 2007;31:519–535. [PMC free article] [PubMed] [Google Scholar]
• Chassin Laurie, Knight GeorgeP, Vargas-Chanes Delfino, Losoya SandraH, Naranjo Donna. Substance use treatment outcomes in a sample of serious juvenile offenders. Journal of Substance Abuse Treatment. 2009;36:183–194. [PMC free article] [PubMed] [Google Scholar]
• Clarke RonaldV, Cornish DerekB. Modeling offenders’ decisions: A framework for research and policy. In: Tonry MichaelH, Morris Norval., editors. Crime and Justice: An Annual Review of Research. vol. 4. Chicago, IL: University of Chicago Press; 1985. [Google Scholar]
• Connell Patricia. Juvenile justice in the year 2005? Journal of Clinical Child Psychology. 1980;9:113–116. [Google Scholar]
• Cullen FrancisT, Golden KathrynM, Cullen JohnB. Is child saving dead? Juvenile rehabilitation in Illinois. Journal of Criminal Justice. 1983;11:1–13. [Google Scholar]
• Fagan Jeffrey. Juvenile crime and criminal justice: Resolving border disputes. The Future of Children. 2008;18:81–118. [PubMed] [Google Scholar]
• Fagan Jeffrey, Kupchik Aaron, Liberman Akiva. Be careful what you wish for: Legal sanctions and public safety among adolescent offenders in juvenile and criminal court; Paper presented at the Annual Conference on Empirical Legal Studies; New York: New York University; 2007. [Google Scholar]
• Fagan Jeffrey, Piquero AlexR. Rational choice and developmental influences on recidivism among adolescent felony offenders. Journal of Empirical Legal Studies. 2007;4:715–748. [PMC free article] [PubMed] [Google Scholar]
• Feld BarryC. Justice by geography: Urban, suburban, and rural variations in juvenile justice administration. Journal of Criminal Law and Criminology. 1991;82:156–210. [Google Scholar]
• Feld BarryC. Juvenile and criminal justice systems’ responses to youth violence. Youth Violence. 1998;24:189–261. [Google Scholar]
• Feld BarryC. Bad Kids. New York: Oxford; 1999. [Google Scholar]
• Garrett CarolJ. Effect of residential treatment on adjudicated delinquents: A meta-analysis. Journal of Research in Crime and Delinquency. 1985;22:287–308. [Google Scholar]
• Gendreau Paul, Goggin Claire, Cullen FrancisT. A report to the Corrections Research and Development and Aboriginal Policy Branch. Ottawa: Solicitor General of Canada; 1999. The effects of prison sentences on recidivism.
• Gendreau Paul, Goggin Claire, Cullen FrancisT, Andrews DonA. Compendium of Effective Correctional Programs. vol. IV. Ottawa, Ontario: Correctional Service of Canada, Solicitor General of Canada; 2001. The effects of community sanctions and incarceration on recidivism. [Google Scholar]
• Gendreau Paul, Goggin Claire, Fulton Betsy. Intensive probation in probation and parole settings. In: Hollin CliveR., editor. Handbook of Offender Assessment and Treatment. Chichester, UK: Wiley; 2000. [Google Scholar]
• Greenwood PeterW. Changing Lives: Delinquency Prevention as Crime-Control Policy. Chicago, IL: University of Chicago Press; 2005. [Google Scholar]
• Griffin Patrick. National Overviews. State Juvenile Justice State Profiles. Pittsburgh, PA: National Center for Juvenile Justice; 2003. [Google Scholar]
• Haapanen Rudy. Selective Incapacitation and the Serious Offender: A Longitudinal Study of Criminal Career Patterns. New York: Springer-Verlag; 1990. [Google Scholar]
• Haapanen Rudy, Britton Lee, Croisdale Tim. Persistent criminality and career length. Crime & Delinquency. 2007;53:133–155. [Google Scholar]
• Heckman JamesJ, Ichimura Hidehiko, Smith Jeffery, Todd Petra. Characterizing selection bias using experimental data. Econometrica. 1998;66:1017–1098. [Google Scholar]
• Hirano Keisuke, Imbens GuidoW. The propensity score with continuous treatments. In: Gelman Andrew, Meng Xiao-Li., editors. Applied Bayesian Modeling and Causal Inference from Incomplete-Data Perspectives. New York: Wiley; 2004. [Google Scholar]
• Horwitz AllanV, Wasserman Michael. Formal rationality, substantive justice, and discrimination: A study of juvenile court. Law and Human Behavior. 1980;4:103–115. [Google Scholar]
• Huizinga David, Esbensen Finn-Aage, Weiher AnneW. Are there multiple paths to delinquency? Journal of Criminal Law and Criminology. 1991;82:83–118. [Google Scholar]
• Jenson JeffreyM, Howard MathewO. Youth crime, public policy, and practice in the juvenile justice system: Recent trends and needed reforms. Social Work. 1998;43:324–335. [PubMed] [Google Scholar]
• Joffe MarshallM, Rosenbaum PaulR. Propensity scores. American Journal of Epidemiology. 1999;150:327–333. [PubMed] [Google Scholar]
• Lee DavidS, McCrary Justin. Crime, punishment, and myopia; National Bureau of Economic Research Working Paper; 2005. p. 11491. [Google Scholar]
• Levitt StevenD. Juvenile crime and punishment. Journal of Political Economy. 1998;106:1156–1185. [Google Scholar]
• Levitt StevenD, Lochner Lance. The determinants of juvenile crime. In: Gruber Jonathan., editor. Risky Behavior by Youths. Chicago, IL: University of Chicago Press; 2000. [Google Scholar]
• Lipsey Mark. Can rehabilitative programs reduce the recidivism of juvenile offenders? An inquiry into the effectiveness of practical programs. Virginia Journal of Social Policy and Law. 1999;6:611–641. [Google Scholar]
• Lipsey Mark, Wilson DavidB. Effective intervention for serious juvenile offenders: A synthesis of research. In: Loeber Rolf, Farrington David., editors. Serious and Violent Juvenile Offenders: Risk Factors and Successful Interventions. Thousand Oaks, CA: Sage; 1998. [Google Scholar]
• Lipsey Mark, Wilson DavidB, Cothern Lynn. Effective Intervention for Serious Juvenile Offenders. Washington, DC: U.S. Department of Justice, Office of Justice Programs, Office of Juvenile Justice and Delinquency Prevention; 2000. [Google Scholar]
• Loeber Rolf, Farrington DavidP., editors. Serious and Violent Juvenile Offenders. Risk Factors and Successful Interventions. Newbury Park, CA: Sage; 1998. [Google Scholar]
• Lu Bo, Zanutto Elaine, Hornik Robert, Rosenbaum PaulR. Matching with doses in an observational study of a media campaign against drug abuse. Journal of the American Statistical Association. 2001;96:1245–1235. [PMC free article] [PubMed] [Google Scholar]
• Mauricio Anne, Little Michelle, Chassin Laurie, Vargas-Chanes Delfino, Knight GeorgeP, Losoya SandraH, Piquero AlexR. Juvenile offenders’ alcohol and marijuana trajectories: Risk and protective factor effects in the context of supervised time. Journal of Youth & Adolescence. 2009 In press. [PMC free article] [PubMed] [Google Scholar]
• McCullagh Peter. Regression models for ordinal data. Journal of the Royal Statistical Society. 1980;42:109–142. [Google Scholar]
• Mears DanielP, Field SamuelH. Theorizing sanctions in a criminalized juvenile court. Criminology. 2000;38:983–1020. [Google Scholar]
• Mocan NaciH, Rees DanielI. Economic conditions, deterrence and juvenile crime: Evidence from micro data. American Law and Economics Review. 2005;7:319–349. [Google Scholar]
• Moon Melissa, Cullen FrancisT, Wright JohnPaul. It takes a village: Public willingness to help wayward youths. Youth Violence and Juvenile Justice. 2003;1:32–45. [Google Scholar]
• Mulvey EdwardP, Iselin Anne-MarieR. Structuring professional judgments of risk and amenability in juvenile justice. The Future of Children. 2008;18:35–57. [PMC free article] [PubMed] [Google Scholar]
• Mulvey EdwardP, Schubert CarolA, Chung HeLen. Service use after court involvement in a sample of serious adolescent offenders. Child and Youth Services Review. 2007;29:518–544. [PMC free article] [PubMed] [Google Scholar]
• Mulvey EdwardP, Steinberg Laurence, Fagan Jeffery, Cauffman Elizabeth, Piquero AlexR, Chassin Laurie, Knight GeorgeP, Brame Robert, Schubert Carol, Hecker Thomas, Losoya SandraH. Theory and research on desistance from antisocial activity among serious juvenile offenders. Youth Violence and Juvenile Justice. 2004;2:213–236. [PMC free article] [PubMed] [Google Scholar]
• Nagin DanielS. Criminal deterrence research at the outset of the twenty-first century. In: Tonry MichaelH., editor. Crime and Justice: A Review of Research. vol. 23. Chicago, IL: University of Chicago Press; 1998. [Google Scholar]
• Nagin DanielS, Cullen FrancisT, Lero-Jonson Cheryl. Imprisonment and reoffending. In: Tonry MichaelH., editor. Crime and Justice: An Annual Review of Research. vol. 38. Chicago, IL: University of Chicago Press; 2009. [Google Scholar]
• Niarhos FrancesJ, Routh DonaldK. The role of clinical assessment in the juvenile court: Predictors of juvenile dispositions and recidivism. Journal of Clinical Child Psychology. 1992;21:151–159. [Google Scholar]
• Piquero AlexR, Blumstein Alfred. Does incapacitation reduce crime? Journal of Quantitative Criminology. 2007;23:267–286. [Google Scholar]
• Piquero AlexR, Blumstein Alfred, Brame Robert, Haapanen Rudy, Mulvey EdwardP, Nagin DanielS. Assessing the impact of exposure time and incapacitation on longitudinal trajectories of criminal offending. Journal of Adolescent Research. 2001;16:54–74. [Google Scholar]
• Piquero AlexR, Langton Lynn, Gomez-Smith Zenta. Discerning unfairness where others may not: Low self-control and unfair sanction perceptions. Criminology. 2004;42:693–727. [Google Scholar]
• Platt AnthonyM. The Child-Savers: The Invention of Delinquency. Chicago, IL: University of Illinois Press; 1977. [Google Scholar]
• Pratt TravisC, Maahs Jeff. Are private prisons more cost-effective than public prisons? A meta-analysis of evaluation research studies. Crime & Delinquency. 1999;45:358. [Google Scholar]
• Redding RichardE, Howell JamesC. Blended sentencing in American juvenile courts. In: Fagan Jeffrey, Zimring FranklinE., editors. The Changing Borders of Juvenile Justice: Transfer of Adolescents to the Criminal Court. Chicago, IL: University of Chicago Press; 2000. [Google Scholar]
• Rosenbaum PaulR. Observational Studies. 2nd ed. New York: Springer-Verlag; 2002. [Google Scholar]
• Rosenbaum PaulR, Rubin DonaldB. The central role of the propensity score in observational studies for causal effects. Biometrika. 1983;70:41–55. [Google Scholar]
• Rosenbaum PaulR, Rubin DonaldB. Reducing bias in observational studies using subclassification on the propensity score. Journal of the American Statistical Association. 1984;79:516–524. [Google Scholar]
• Rosenbaum PaulR, Rubin DonaldB. Constructing a control group using multivariate matched sampling methods that incorporate the propensity. American Statistician. 1985;39:33–38. [Google Scholar]
• Rubin DonaldB, Thomas Neal. Characterizing the effect of matching using linear propensity score methods with normal distributions. Biometrika. 1992;79:797–809. [Google Scholar]
• Schneider AnneL. Deterrence and Juvenile Crime: Results from a National Policy Experiment. New York: Springer; 1990. [Google Scholar]
• Schubert CarolA, Mulvey EdwardP, Steinberg Laurence, Cauffman Elizabeth, Losoya SandraH, Hecker Thomas, Chassin Laurie, Knight GeorgeP. Operational lessons from the pathways to desistance project. Youth Violence and Juvenile Justice. 2004;2:237–255. [PMC free article] [PubMed] [Google Scholar]
• Scott Elizabeth, Steinberg Laurence. Adolescence and Crime: Rethinking Juvenile Justice. Cambridge, MA: Harvard University Press; 2008. [Google Scholar]
• Sherman Lawrence. Defiance, deterrence and irrelevance: A theory of the criminal sanction. Journal of Research in Crime and Delinquency. 1993;30:445–473. [Google Scholar]
• Shichor David. Private prisons in perspective: Some conceptual issues. The Howard Journal. 1998;37:82–100. [Google Scholar]
• Smith DouglasA, Paternoster Raymond. Formal processing and future delinquency: Deviance amplification as selection artifact. Law & Society Review. 1990;24:1109–1132. [Google Scholar]
• Stapleton VaughanW, Aday DavidP, Ito JeanneA. An empirical typology of American metropolitan juvenile courts. American Journal of Sociology. 1982;88:549–564. [Google Scholar]
• Steinberg Laurence, Cauffman Elizabeth. A developmental perspective on jurisdictional boundary. In: Fagan Jeffrey, Zimring FranklinE., editors. The Changing Borders of Juvenile Justice: Transfer of Adolescents to the Criminal Court. Chicago, IL: University of Chicago Press; 2000. [Google Scholar]
• Steiner Benjamin, Hemmens Craig, Bell Valerie. Legislative waiver reconsidered: General deterrent effects of statutory exclusion laws enacted post-1979. Justice Quarterly. 2006;23:34–59. [Google Scholar]
• Thornberry TerenceP, Krohn MarvinD U.S. National Institute of Justice. Measurement and Analysis of Crime and Justice: Criminal Justice Series. Vol. 4. Washington, DC: National Institute of Justice; 2000. The self-report method of measuring delinquency and crime; pp. 33–83. [Google Scholar]
• Torbet Patricia, Szymanski Linda. State Legislative Responses to Violent Juvenile Crime: 1996–97 Update. Washington, DC: Office of Juvenile Justice and Delinquency Prevention, U.S. Department of Justice; 1998. Juvenile Justice Bulletin, November 1998. [Google Scholar]
• Zanutto Elaine, Lu Bo, Hornik Robert. Using propensity score subclassification for multiple treatment doses to evaluate a national anti-drug media campaign. Journal of Educational and Behavioral Statistics. 2005;30:59–73. [Google Scholar]
• Zimring FranklinE. American Youth Violence. New York: Oxford University Press; 1998. [Google Scholar]
• Zimring FranklinE. Penal proportionality and the young offender: Notes on immaturity, capacity, and diminished responsibility. In: Grisso Thomas, Schwartz RobertG., editors. Youth on Trial: A Developmental Perspective on Juvenile Justice. Chicago, IL: University of Chicago Press; 2000a. [Google Scholar]
• Zimring FranklinE. The punitive necessity of waiver. In: Fagan Jeffrey, Zimring FranklinE., editors. The Changing Borders of Juvenile Justice: Transfer of Adolescents to the Criminal Court. Chicago, IL: University of Chicago Press; 2000b. [Google Scholar]
• Zimring FranklinE. The common thread: Diversion in juvenile justice. California Law Review. 2000c;88:2477. [Google Scholar]
• Zimring FranklinE, Hawkins Gordon. Deterrence: The Legal Threat in Crime Control. Chicago, IL: University of Chicago Press; 1973. [Google Scholar]