Suppressing the Harmful Effects of Key Risk Factors

Abstract

Many programs try to reduce adolescent offending with a risk factor approach in which services target the key causes of crime pertaining to families, peer groups, and schools. These programs often reduce crime, presumably through either prevention (in which exposure to a risk factor is prevented from ever occurring) or reversal (in which an individual possessing a risk factor advances to a state of no longer having it). This study examines an alternative way in which such programs may reduce delinquency: They may achieve the goal of risk factor suppression, whereby a risk factor that is neither prevented nor reversed is rendered inconsequential by program treatment. Thus, the risk factor continues to be present, but by virtue of program treatment, it no longer elevates individual involvement in crime. The authors consider this possibility with evaluation data from the experimental Children at Risk program, a 2-year case management intervention that served high-risk early adolescents.

Keywords

risk factors delinquency suppression Children at Risk case management

In recent decades, researchers have identified many programs that successfully reduce juvenile delinquency. Far from being uniform in nature, these programs use a variety of approaches, including community mentoring (Grossman & Tierney, 1998), parent management training (Piquero, Farrington, Welsh, Tremblay, & Jennings, 2009), cognitive behavioral treatment (Landenberger & Lipsey, 2005), and integrated, multimodal approaches (Hawkins, Kostermann, Catalano, Hill, & Abbott, 2005; Henggeler, Schoenwald, Borduin, Rowland, & Cunningham, 1998). Moreover, the significant effects of these programs on offending are substantively important—high-integrity interventions often reduce delinquency by as much as 30% (Howell, 2009; Lipsey, 1995).

These successes are a notable historic development, given the widely stated claims during the 1970s that the principle of rehabilitation was dead. On the contrary, rehabilitation is in many respects now thriving (Cullen, 2005). In addition to the evidence on delinquency reduction just noted, a steady flow of research suggests that the general public favors rehabilitation efforts (e.g., Mears, Hay, Gertz, & Mancini, 2007; Moon, Sundt, Cullen, & Wright, 2000). There also is clear evidence that successful interventions are cost-effective, especially when compared to incarceration (Barnoski, 2004). Thus, in contrast to the bold attacks on the concept of rehabilitation that were common just a few decades ago, it now is clear that rehabilitative approaches are increasingly possible and popular, and when implemented effectively, they are an efficient use of public money.

Despite this progress, there still is much to learn about the performance of delinquency-reduction programs. One neglected issue involves the question of how these programs are successful. Specifically, when program participation leads to a decrease in offending, what mechanisms explain this change? We lack firm answers to this question because evaluations often neglect it—studies that examine the causal processes that explain program effects on delinquency are in short supply (Gottfredson, Kearley, Najaka, & Rocha, 2007). And yet there is good reason to examine such causal processes—doing so can bolster efforts to identify a program’s strengths, replicate it elsewhere, and inform public policy. Indeed, this more nuanced understanding of program effects is integral to the recent shift toward evidence-based crime policy (Mears, 2007).

This study considers this issue with respect to a comprehensive intervention for youths that has significantly reduced many forms of delinquency. Moreover, we seek to do so in a way that draws attention to a unique way in which such programs can succeed. Specifically, most programs of this kind are based on a risk-factor model in which the program should succeed through either prevention (in which exposure to risk factors is prevented from ever occurring) or reversal (in which participants who already possess risk factors are advanced to a state of no longer having them). These are important routes to program success, but we are especially interested in a more novel possibility that often has been neglected. This involves the goal of risk factor suppression, whereby a risk factor that is neither prevented nor reversed is rendered inconsequential by program treatment, presumably because of some protective factor that has been introduced. Thus, the risk factor continues to be present, but it no longer elevates individual involvement in crime.

This idea of risk factor suppression has received limited attention in the literature on delinquency-reduction programming (Farrington & Welsh, 2007), but it may be important, given that some risk factors are especially difficult to prevent or reverse. This should be true when dealing with high-risk adolescents who have spent much of their lives in criminogenic physical and social environments. Under such circumstances, preventing exposure to basic risk factors is no longer possible. Reversing them may be quite difficult as well. Many risk factors (including such things as cognitive problems or impulsivity) are fairly stable over the life course, whereas others involve events (such as incidents of abuse) that, by definition, cannot be reversed. For these types of enduring risk factors, successful interventions may be those that suppress their harmful effects. For example, although the experience of an arrest cannot be reversed once it has occurred, it may be that various services can help limit its harmful effects on later offending, perhaps by reducing feelings of stigma associated with the arrest.

We empirically assess this idea of risk factor suppression in an evaluation of the Children at Risk (CAR) program, a case management intervention for high-risk adolescents living in distressed urban neighborhoods (Harrell, Cavanagh, & Sridharan, 2000). As we elaborate on below, the CAR program is particularly useful for addressing the questions posed in the current study. Most notably, its case management model and the population that it served (juveniles already possessing many risk factors) make it especially relevant to the goal of risk factor suppression. Moreover, the program was evaluated with a rigorous research design (including random assignment) that produced extensive data to be analyzed.

In the next section, we describe in greater detail the CAR program and the preliminary indications of its success. Next, we consider the logical connection between the CAR program and three distinct mechanisms for program success (prevention, reversal, and suppression), noting the ways in which this should inform an evaluation of the program. We then proceed to our discussion of the CAR data and our evaluation of its effects.

The Car Program and Its Initial Signs of Success

The CAR program is a delinquency and drug use reduction program developed, funded, and monitored by the National Center on Addiction and Substance Abuse (Harrell Cavanagh, & Sridharan, 1999). It was designed to reduce chronic, serious offending among youths at high risk for delinquency based on risk criteria assessed by school, court, and social service officials. To be eligible, youths had to be 11 to 13 years old and living in one of the economically distressed, high-crime neighborhoods targeted by the program.

The CAR program was based on an integrated, risk factor perspective informed by empirical research on a wide array of criminological theories. From that research, program developers concluded that involvement in delinquency is affected by key features of (a) families, (b) schools, (c) peer groups, and (d) background personal characteristics. This view of delinquency causation required a “multimodal” approach (Farrington & Welsh, 2007) that could target the different risk factors. Six different substantive services became part of the CAR intervention: family services (including therapy, skills training, and advocacy with other agencies), after-school and summer activities, educational services (including tutoring and homework assistance), mentoring, a system of behavioral incentives, and community policing activities in CAR neighborhoods and with CAR families.

In addition, because it was expected that each subject would face a unique blend of risks, the CAR program used an individualized approach to treatment. Participants were assigned a case manager who assessed the needs of youths and their families, developed a service plan, coordinated service delivery, and collaborated with other agencies, including criminal and juvenile justice authorities when necessary. With respect to educational services, for example, case managers assessed education-related risks, referred youths to testing or tutoring, and monitored participation in these services. To facilitate case managers’ success, caseloads were small (15 to 18 families), and youths’ participation in the program often spanned multiple years.

An especially strong research design was used to evaluate the program (Harrell et al. 2000). This design is described in detail below when we discuss the sample and measures used for our analysis, but it can be noted here that the design included random assignment of eligible participants to control and treatment groups. Moreover, comprehensive data on criminal involvement and exposure to key risk factors were collected at three points: at the time of random assignment, at the time of program completion 2 years later, and at a follow-up 1 year after that. Although comprehensive analyses of these data have not yet appeared in the peer-reviewed literature, a National Institute of Justice report prepared by the evaluation’s principal investigators suggests the value of the program (Harrell et al., 1999). For example, CAR youths commonly used the program services. Moreover, compared to the control group, both CAR youths and their caregivers participated in more positive social activities, including religious, community, and recreational activities. The program also reduced many forms of delinquency, although this pattern was evident only at the follow-up 1 year after program completion when participants were roughly 15 years old. Specifically, those in the treatment group were less involved in such things as violent attacks on others, illegal substance use, and the sale of drugs. Drug selling in the prior month, for example, was 42% lower among CAR youths, whereas prevalence levels for violence and illegal drug use were roughly 20% lower for the treatment group.

With these results, CAR has become widely recognized as a program that can reduce delinquency. Indeed, its success was lauded by many agencies that track the effectiveness of various interventions. This includes the U.S. Department of Education, the U.S. Surgeon General’s Office, and the University of Colorado’s Center for the Study and Prevention of Violence.

Considering the Possibility of Risk Factor Suppression

A key question about the CAR program remains: How is it successful? In what specific ways does it affect the targeted risk factors, such that later delinquency is reduced? In considering this type of question, two mechanisms receive the most attention in research on delinquency-reduction programming. The first of these is prevention, which occurs if a program blocks a risk factor from ever emerging in the first place. For example, CAR family services could prevent high-risk children from ever being exposed to family violence, which is a key risk factor for delinquency. A second commonly considered mechanism is the idea of reversal (akin to the idea of rehabilitation), which occurs if program services lead youths to no longer possess a risk factor that they previously had. For example, CAR youths who were performing poorly at school might experience—as a result of CAR’s educational services (including tutoring and homework assistance)—a reversal in this area. Under such circumstances, the child has advanced from a state of possessing a risk factor for delinquency (poor school performance) to a state of no longer possessing it.

These two mechanisms of delinquency reduction are of general importance, but there is good reason to believe that they are not responsible for the success of the CAR program in particular. With respect to prevention, any pure form of prevention is impossible for the CAR program—it was designed for adolescents already at high risk for chronic delinquency by virtue of such things as arrests, prior delinquency, or the family’s referral to social service agencies. Thus, for these youths, the window of opportunity for primary prevention had closed prior to entry into the CAR program.

Initial evaluations also cast doubt on the program’s ability to accomplish the goal of reversal. To be clear, reversal was an initial goal of CAR’s developers, who sought to reduce delinquency among at-risk youths “by reducing the number of risk factors to which they were exposed” (Harrell et al., 1999, p. 4). The case manager model was chosen with this in mind—case managers would assess risks and design a service plan to reverse them. However, the CAR outcome evaluation revealed that at Wave 2 (after 2 years of services), none of the risk factors targeted by the program had been significantly reversed (Harrell et al., 1999).

This inability to accomplish reversal, along with a program design that bypassed the goal of pure prevention, points to the possibility that CAR succeeded through a more novel mechanism—one that can be referred to as risk factor suppression. This occurs when risk factors continue to exist, but treatment leads those risk factors to no longer be consequential for delinquency. Thus, the risk factor endures but is stripped of its potency.

The way in which this process could play out is evident, for example, with official labeling among juveniles, including the experience of an arrest. Several recent studies find that official labeling is positively associated with subsequent crime or delinquency, even when controlling for background factors that affect the probability of receiving an official label (Bernburg & Krohn, 2003; Chiricos, Barrick, Bales, & Bontrager, 2007; Johnson, Simons, & Conger, 2004). This result is consistent with the labeling theory position that justice system intervention is a stigmatizing “dramatization of evil” that labels the individual as a deviant (Tannenbaum, 1938). According to this view, a socially imposed label engenders a deviant self-concept that becomes the individual’s master status—a status that overrides other qualities that he or she may have. Through processes of both withdrawal and social exclusion, those who are labeled become isolated from the conventional people, goals, and institutions that normally discourage crime and delinquency (Becker, 1963; Bernburg, 2009; Lemert, 1951). In short, official labeling triggers a series of events that should increase future delinquency.

With risk factor suppression, however, such an increase in delinquency would not occur. CAR programming cannot reverse the arrest experience—it cannot “erase” this event from the adolescent’s history—but it may be able to block the harmful consequences described above from coming to fruition. For example, the positive social support provided by CAR case managers and mentors could reduce the stigma and rejection felt by those experiencing an arrest, allowing the youths to become better reintegrated into the community of law-abiding citizens (Braithwaite, 1989). CAR case managers often developed strong relationships with the youths and their families, and this should advance the goal of offender reintegration. Moreover, the treatment and services coordinated by the case manager—including homework assistance and after-school recreational opportunities—could help youths renew or intensify bonds to conventional people and institutions. Taken together, these experiences should increase the salience of prosocial alternatives to delinquency and make CAR youths less vulnerable to developing a deviant self-concept. Thus, although the arrest experience is not erased, its harmful fallout is suppressed such that the arrest does not increase offending in the normally expected way.

A pattern of this kind could emerge for a wide a variety of risk factors for offending, including such things as a difficult family environment, criminal victimization, and negative experiences at school. CAR may not have been able to prevent or reverse those experiences, but it could provide the prosocial opportunities and positive social support needed to suppress their harmful consequences (including heightened involvement in delinquency). Thus, a child may continue to struggle at school, for example, but the CAR case manager can use services relating to such things as a mentoring, behavioral incentives, and involvement in prosocial after-school activities to suppress the normal link between struggles at school and involvement in delinquency. Such an outcome reflects a pattern in which the risk factor still persists, but its effect on later delinquency is suppressed.

This pattern of risk factor suppression seems not only possible but perhaps probable in light of the process evaluation of the CAR program (Harrell et al., 1999). That evaluation revealed that the multiple challenges faced by CAR participants often meant that managing and suppressing the harmful fallout of key risk factors—rather than reversing them—was the only option. Harrell and her colleagues (1999) concluded that “CAR participants had such serious and multiple needs that their whole lives were bound up with dealing with one crisis after another, making it impossible in many cases to establish . . . a regular pattern of services” (p. 5). Ultimately, “far more time was spent on crisis intervention, and less on ongoing case management, than originally anticipated” (p. 4). Thus, case managers used their strong relationships with the youths and their families and their arsenal of services to reduce the fallout from such things as school problems, family violence, and arrests. As the program evolved, the goal therefore became “to offset, rather than remediate, underlying risk factors” (p. 9).

We see these insights from the CAR process evaluation as having important implications for how this program may reduce delinquency and how it should be evaluated. Specifically, any success of the CAR program may have come from its ability to intervene—with such things as counseling, supervision, or prosocial after-school opportunities—when key risk factors manifested themselves. Reversing these risk factors may not have been possible, but CAR nevertheless may have succeeded by suppressing their harmful consequences.

The Current Study

The purpose of this study is to evaluate the CAR program with respect to its ability to suppress the harmful effects of key risk factors for crime. As described below, we test this hypothesis by examining four established risk factors for delinquency: having a prior arrest, associating with delinquent peers, being exposed to family violence, and being weakly committed to school. These were chosen because they met three important criteria: There is strong theoretical and empirical evidence supporting their status as important risk factors, they are well-measured in the CAR data, and they reflect risks that likely would be prioritized and targeted by CAR case managers.

These risk factors do not exhaust the list of potential risks, but taken together they cover key areas of risk, including justice system involvement and problems in the areas of peers, the family, and school. Moreover, these risks follow from prominent theories of delinquency causation. The experience of an arrest was considered in light of research indicating that juvenile and criminal justice system involvement is associated with increased offending even when accounting for prior behavior that prompts official action (Bernburg & Krohn, 2003; Chiricos et al., 2007). This is consistent with the labeling theory argument that official labeling increases crime by severing conventional social bonds and by stigmatizing offenders in ways that encourage a deviant self-concept. Our attention to association with delinquent peers follows from the scholarship on social learning theory, which emphasizes that delinquent peers are important because of the reinforcements and opportunities for crime that they provide (Akers, 1998; Haynie & Osgood, 2005). Next, our focus on family conflict follows from the research that finds a relationship between exposure to a combative, erratic family environment and involvement in delinquency (Sampson & Laub, 1993). This relationship is predicted by social control theory (with its emphasis on the delinquency-inhibiting role of family social bonds), but also by Agnew’s (1992) general strain theory, which focuses on the delinquency-encouraging role of family stress and strain. Last, our consideration of weak school commitment follows from research supporting the social control theory argument that strong school bonds and performance create stakes in conformity that reduce offending (Hirschi, 1969; Sampson & Laub, 1993).

If the suppression hypothesis is to be supported, effects of these risk factors on later delinquency should be lower for those who received CAR treatment than they are for those who did not. For example, although official labeling may generally predict increases in subsequent offending, this should not be true—or at least should be less true—for those in the CAR treatment group. Such a pattern would reveal that the normally positive relationship between official labeling and future delinquency can be suppressed by the case management services provided by the CAR program.

Method

Procedure

Because a rigorous evaluation of every CAR site was not feasible, the evaluation focused on five cities selected to receive CAR funding and participate in the large-scale evaluation: Austin, Texas; Bridgeport, Connecticut; Memphis, Tennessee; Savannah, Georgia; and Seattle, Washington. These cities were competitively selected based on the strength of their proposed implementation models. Their programs targeted small geographic areas with the highest rates of crime, drug use, and poverty. Within these areas, the programs targeted youths who were 11 to 13 years old and deemed by school, court, and social service officials to be high risk for delinquency.

This screening procedure produced the pool of nearly 700 eligible youths. Prior to receiving services, youths were randomly assigned to the treatment group (which received program services for 2 years) or to a control group of participants who lived in the targeted neighborhoods but received no direct services. For both groups, data were collected with comprehensive face-to-face interviews with the adolescents, with the first interviews occurring during a baseline period between random assignment and the start of the program. Participants were then reinterviewed 2 years later at the time of program completion (when participants were about 14 years old) and 1 year after that (when participants were about 15 years old). Each wave of survey data included validated measures both for key risk factors and for involvement in multiple forms of delinquency. This data collection procedure has an important implication for our analyses—the three-wave panel design with repeated measures allows us to better capture the appropriate causal order by conducting analyses that have a temporal lag between the independent variables (the risk factors) and the dependent variables relating to delinquency.

Participants

The initial sample consists of 664 high-risk youths living in high-risk neighborhoods, with 336 participants in the treatment group and 328 participants in the control group. Table 1 reveals the sample characteristics for the two groups in the Wave 1 data. In both groups, respondents were roughly 12 years old at the beginning of the program, and both samples were evenly divided between males and females. The sample has a heavy concentration of racial and ethnic minorities, with Black and Hispanic participants representing approximately 55% and 35% of both samples.

Table 1:

Descriptive Statistics for Study Variables, by Sample Group

	Total Sample				Experimental Group				Control Group
	M	SD	Min	Max	M	SD	Min	Max	M	SD	Min	Max
Delinquency variables
Wave 1
Delinquency	0.00	0.56	−0.33	3.76	−0.02	0.54	−0.33	3.76	0.03	0.57	−0.33	3.14
Substance	0.00	0.48	−2.24	6.80	−0.02	0.32	−0.12	2.81	0.02	0.60	−2.24	6.80
Wave 2
Delinquency	0.00	0.58	−0.42	3.54	0.00	0.58	−0.42	3.54	0.00	0.58	−0.42	3.20
Substance	0.00	0.52	−2.06	6.83	0.03	0.62	−1.51	6.80	−0.04	0.38	−2.06	2.13
Wave 3
Delinquency	0.00	0.63	−0.38	4.42	−0.03	0.66	−0.38	4.42	0.03	0.59	−0.38	3.06
Substance	0.00	0.51	−0.27	4.49	−0.03	0.49	−0.25	3.46	0.03	0.54	−0.27	4.49
Risk factors
Wave 1
Delinquent peers	0.00	0.67	−0.78	1.35	−0.04	0.64	−0.78	1.35	0.04	0.70	−0.78	1.34
Arrest	0.08	0.27	0.00	1.00	0.09	0.29	0.00	1.00	0.07	0.25	0.00	1.00
Family conflict	0.00	0.61	−0.62	2.59	0.01	0.62	−0.57	2.59	0.00	0.60	−0.62	2.22
Weak school commitment	0.00	0.63	−0.78	3.51	−0.05	0.61	−0.78	3.51	0.05	0.65	−0.78	2.31
Wave 2
Delinquent peers	0.00	0.59	−0.67	1.84	−0.01	0.58	−0.67	1.84	0.01	0.60	−0.67	1.84
Arrest	0.24	0.43	0.00	1.00	0.25	0.44	0.00	1.00	0.23	0.42	0.00	1.00
Family conflict	0.00	0.65	−0.62	3.49	−0.03	0.63	−0.62	3.49	0.03	0.67	−0.62	3.12
Weak school commitment	0.00	0.58	−0.45	2.90	−0.02	0.55	−0.45	2.90	0.02	0.59	−0.44	2.90
Control variables
Age	12.36	0.71	10.00	14.00	12.33	0.71	10.00	14.00	12.39	0.72	10.00	14.00
Male	0.50	0.50	0.00	1.00	0.51	0.50	0.00	1.00	0.50	0.50	0.00	1.00
White or other	0.08	0.27	0.00	1.00	0.09	0.28	0.00	1.00	0.07	0.26	0.00	1.00
Hispanic	0.37	0.48	0.00	1.00	0.37	0.48	0.00	1.00	0.38	0.49	0.00	1.00
Black	0.55	0.50	0.00	1.00	0.55	0.50	0.00	1.00	0.54	0.50	0.00	1.00
Treatment	0.51	0.50	0.00	1.00	1.00	0.00	1.00	1.00	0.00	0.00	0.00	0.00
Control	0.49	0.50	0.00	1.00	0.00	0.00	0.00	0.00	1.00	0.00	1.00	1.00

When participants were reinterviewed at Waves 2 and 3, the response rates were relatively high. Of the original sample of 664 participants, 79% (n = 522) were reinterviewed at Wave 2, 76% (n = 506) were reinterviewed at Wave 3, and 69% (n = 460) had data for all three waves. Also, the attrition that occurred was not selective—comparisons of retained cases and those lost to attrition produced no significant differences for any variable considered, including age, sex, Black, and Hispanic (Harrell et al., 2000).

Measures

Risk factors

The CAR data provided strong measures for each risk factor at both Waves 1 and 2. These measures are described in the appendix, but we briefly summarize them here. To measure the experience of an arrest, we used dichotomous self-reported items quite similar to those used in other recent studies (e.g., Bernburg & Krohn, 2003). These items asked participants whether or not they had been arrested during the prior year (or since the last interview), with respondents coded as 1 if they had been arrested. For association with delinquent peers, our Wave 1 variable was measured with a seven-item scale (alpha = .80) in which high scorers indicate that their peers are involved in various delinquent activities, including shoplifting, fighting, and auto theft. For Wave 2, we were able to use a slightly more elaborate measure (a 12-item scale with an alpha of .83) that includes all items in the Wave 1 measure, plus items measuring peers’ involvement in substance use. (These peer scales—and all other multiple-item scales described below—were created by standardizing the included items and then computing their average.)To measure family conflict, we used seven-item scales at both Wave 1 (alpha = .73) and Wave 2 (alpha = .77) in which high scorers indicate that their family life is marked by frequent screaming, threatening, and hitting. Last, weak school commitment was measured with a four-item scale at Wave 1 (alpha = .48) and a seven-item scale (alpha = .66) at Wave 2. High scorers on these scales do not like school, do not try hard to do well, and do not see school as worthwhile. An alpha of .48 for the measure of Wave 1 weak commitment to school is far from ideal, but this measure is adequate in key respects. Principal components analysis yielded a one-factor solution (with loadings ranging from .51 to .70), and each item’s inclusion increases the scale’s internal reliability; also, the scale is correlated with Wave 1 general delinquency (r = .33) and Wave 2 weak school commitment (r = .26) in the predicted manner. This measure therefore was retained, allowing a similar assessment of all risk factors (with all being examined at Waves 1 and 2).

Delinquency

Delinquency at Waves 1, 2, and 3 were measured with self-reported delinquency items that assessed involvement in violent, property, substance, and status offenses. All items were answered by the adolescents themselves, who chose from ordinal response categories ranging from never to 5 or more times. The reference period for Wave 1 was the prior year; for Waves 2 and 3, it was the period since the last interview. The one exception to this was for substance offenses in particular—given the challenges associated with estimating annual incidence for frequent behaviors, we focused on items inquiring about behavior in the prior 30 days. The delinquency items were used to create separate measures for substance delinquency (any act involving the use, sale, or distribution of alcohol or drugs) and general delinquency (all other forms of delinquency, including such things as assault, robbery, and theft). This was done because the CAR program’s specific attention to reducing substance use—reflected in part by the program’s origins in the National Center on Addiction and Substance Use—calls for an evaluation that provides results specific to this form of delinquency. We created measures of general and substance delinquency for all three waves. The Wave 1 scales were used to test whether randomization produced equivalent treatment and control groups at the start of the program, whereas the scales for Waves 2 and 3 were used as dependent variables. These scales all were strong in terms of key measurement properties—each had a relatively large number of items (ranging between 11 and 13 items) and high levels of internal reliability (ranging from .72 to .87). Also, these scales are significantly correlated in the expected ways with one another and with the risk factors considered in this study (we return to this issue momentarily).

Additional variables

To assess the effects of the CAR program, we used an “intent to treat” approach in which the value for the treatment variable was based on the group—treatment or control—to which a subject had been randomly assigned. This is common in evaluations of case management programs, given that the exact services received will vary because of the varying circumstances of youths. Our dichotomous treatment variable was coded 1 for those assigned to the treatment group and 0 for those assigned to the control group. Also, to protect against spuriousness in considering the effects of the risk factors, our equations included controls for age (measured in years), sex (with males coded as the high category), and race and ethnicity (with dummy variables created for the categories of Black, Hispanic, and White/other).

Analytically Assessing Risk Factor Suppression

The risk factor suppression hypothesis involves a straightforward application of the idea of a moderated relationship (Baron & Kenny, 1986)—we are hypothesizing that the association between two variables (a risk factor and involvement in subsequent delinquency) depends on exposure to a third variable (CAR treatment). Specifically, CAR treatment should moderate that association such that the effect of a risk factor on delinquency is reduced for those exposed to CAR treatment. In this sense, CAR treatment is suppressing the delinquent response that normally would be expected from a given risk factor.

Many approaches are used to study moderated relationships, but product term analysis within ordinary least squares (OLS) regression is the most widely recommended method (Aiken & West, 1991; Jaccard, Turrisi, & Wan, 1990; Tabachnick & Fidell, 2007). This is the method used here. This approach uses an interaction term (x₁ × x₂) that is the product of the predictor (a risk factor) and the hypothesized moderator (treatment group membership). The product term is entered into an equation that includes main effects for the risk factor and CAR treatment and takes delinquency as the dependent variable. The estimated model produces two coefficients that are of special importance for evaluating the suppression hypothesis. The first is the coefficient for the risk factor, which reveals the effect of the risk factor on delinquency for those who did not receive CAR treatment. We expect this coefficient to be positive, indicating that the risk factor generally increases delinquency.

The second important coefficient is for the product term—this coefficient reveals whether the effect of the risk factor depends on participants’ value on the dichotomous treatment variable (0 = control group, 1 = treatment group). To support the suppression hypothesis, the product term coefficient should be significant and negative, therefore revealing that an increase in treatment (from 0 to 1) is associated with decreases in the effects of the risk factor. Full support of the suppression hypothesis is observed when the size of the negative product term coefficient matches the size of the positive main effect of the risk factor. For example, when arrest is considered as a risk factor, a main effect of arrest of .60 would indicate that experiencing an arrest increases delinquency by .60. However, a significant arrest × treatment product term coefficient of –.60 would indicate that the observed effect of arrest is reduced by .60 for those who received treatment. Such a pattern would reveal that the normally positive effect of getting arrested on subsequent delinquency (as indicated by its main effect) is eliminated once we account for the way in which CAR treatment reduces that effect (as indicated by the interaction coefficient).

Results

Preliminary Evaluation Issues

Prior to testing our key hypothesis, there were two preliminary evaluation issues to consider. First, did randomization produce treatment and control groups that were equivalent on key variables at the Wave 1 survey? As demonstrated in Table 1, none of the Wave 1 differences between the control and treatment groups were statistically significant.

Second, we confirmed that the risk factors that are the focus of our analysis are correlated with delinquency and therefore appropriate to consider for this study. Table 2 provides a correlation matrix for the Wave 1 and Wave 2 measures of the risk factors and the Wave 2 and Wave 3 measures of delinquency. All 16 correlations between the risk factors and the delinquency variables for the next subsequent wave are significant. The highest correlation is between Wave 2 delinquent peers and Wave 3 substance delinquency (r = .37), with median correlations of .14 for Wave 1 risk factors predicting Wave 2 outcomes and .24 for Wave 2 risk factors predicting Wave 3 outcomes. Indeed, our Wave 1 risk factors are reasonably good predictors of Wave 3 delinquency, despite the 3-year temporal lag—six of these eight correlations are significant (with a median correlation of .16).

Table 2:

Intercorrelations for the Risk Factors, General Delinquency, and Substance Delinquency

		1	2	3	4	5	6	7	8	9	10	11	12
1	Delinquent peers (W1)	1.00
2	Delinquent peers (W2)	.33	1.00
3	Arrest (W1)	.16	.10	1.00
4	Arrest (W2)	.11	.19	.18	1.00
5	Family conflict (W1)	.21	.22	.11	.09	1.00
6	Family conflict (W2)	.13	.33	−.04	.11	.22	1.00
7	Weak school commitment (W1)	.22	.22	.12	.06	.22	.13	1.00
8	Weak school commitment (W2)	.03	.31	.08	.14	.14	.20	.26	1.00
9	General delinquency (W2)	.21	.51	.15	.33	.09	.29	.16	.43	1.00
10	General delinquency (W3)	.19	.33	.16	.22	.07	.19	.09	.31	.61	1.00
11	Substance delinquency (W2)	.16	.37	.11	.21	.11	.15	.12	.32	.54	.41	1.00
12	Substance delinquency (W3)	.21	.37	.21	.24	.11	.18	.15	.24	.53	.66	.51	1.00

Testing For Risk Factor Suppression

We then tested our main hypothesis—the effects of the risk factors on the delinquency should be moderated by CAR treatment, as indicated by a significant risk factor × treatment interactions. We first considered this in reference to the Wave 1 risk factors and their effects on delinquency at Waves 2 and 3. This gives insight on the circumstances that youths faced as they were starting the program and first receiving services. Given that we have two dependent variables (general and substance delinquency) and two temporal combinations (Wave 1 → Wave 2 and Wave 1 → Wave 3), we estimated four OLS regression equations for each risk factor. Each model included the risk factor in question, treatment, a risk factor × treatment product term, and controls for age, male, White, and Hispanic. And because there are clear expectations regarding the direction of any main or interactive effects, we use one-tailed significance tests.

Table 3 provides the key results for these equations, including the unstandardized coefficient and standard error for the risk factor in question, the treatment variable, and the interaction term. Across the 16 equations presented in Table 3 (four for each of the four risk factors), we observed partial support for the risk factor suppression hypothesis—although 9 of the 16 interactions were nonsignificant, 4 interactions were significant at the p ≤ .05 level, and an additional 3 were significant at the p ≤ .10 level. Although these latter three cases narrowly missed the conventional threshold for statistical significance, we mention them to protect against the risk of Type II error, which takes on special significance when discussing real-world programs that address pressing social problems (Langbein & Felbinger, 2006). Of special importance is the negative direction of the interaction coefficients—these significant interactions reveal that treatment decreased the positive effect of the risk factors on later delinquency. And in terms of magnitude, these interactions often were substantively important. For example, when weak school commitment is considered as a risk factor for Wave 3 general delinquency, its main effect (b = 0.160) is entirely eliminated for those who received CAR treatment—the significant interaction coefficient of −0.193 reduces this effect to marginally below .00.

Table 3:

Ordinary Least Squares Regression Analysis With Product Terms: Effects of Wave 1 Risk Factors, Treatment, and Product Terms on Wave 2 and 3 General and Substance Delinquency

	General Delinquency				Substance Delinquency
	Wave 1 → Wave 2		Wave 1 → Wave 3		Wave 1 → Wave 2		Wave 1 → Wave 3
	b	se	b	se	b	se	b	se
Arrest
Risk factor	0.559***	(0.169)	0.798***	(0.167)	0.376***	(0.154)	0.850***	(0.136)
Treatment	0.006	(0.052)	−0.023	(0.056)	0.078	(0.048)	−0.024	(0.046)
Product term	−0.433**	(0.208)	−0.795***	(0.218)	−0.260*	(0.191)	−0.782***	(0.177)
Delinquent peers
Risk factor	0.179***	(0.051)	0.244***	(0.057)	0.134***	(0.047)	0.184***	(0.047)
Treatment	0.005	(0.050)	−0.069*	(0.054)	0.079	(0.046)	−0.068*	(0.045)
Product term	0.075	(0.074)	−0.057	(0.080)	0.018	(0.069)	0.001	(0.066)
Family conflict
Risk factor	0.043	(0.061)	0.159***	(0.070)	0.051	(0.056)	0.074*	(0.058)
Treatment	−0.013	(0.051)	−0.078*	(0.056)	0.066	(0.046)	−0.075**	(0.046)
Product term	0.089	(0.082)	−0.135*	(0.092)	0.072	(0.075)	0.029	(0.076)
Weak school commitment
Risk factor	0.193***	(0.056)	0.160***	(0.064)	0.062	(0.052)	0.137***	(0.053)
Treatment	0.010	(0.051)	−0.067	(0.056)	0.084	(0.047)	−0.059*	(0.046)
Product term	−0.114*	(0.079)	−0.193**	(0.089)	0.058	(0.073)	−0.077	(0.073)

Note. Control variables also included in each model: age, male, White, Hispanic.

p ≤ .10, one-tailed. **p ≤ .05, one-tailed. ***p ≤ .01, one-tailed.

It bears emphasizing that there was clear variation across the risk factors. No support emerged for the prediction that CAR suppressed the effects of delinquent peers—none of the four models estimated for this risk factor produced a significant interaction. On the other hand, one significant interaction emerged for family conflict and two significant interactions emerged for weak school commitment. For both of these risk factors, the significant interactions emerged with respect to general delinquency.

The clearest support for the suppression hypothesis emerged when prior arrest was considered as a risk factor—a significant and negative arrest × treatment interaction emerged in each of the four models (one of these was significant only at a level of p ≤ .10). Thus, for arrest, CAR treatment suppressed its effects on both general and substance delinquency, and the negative interaction often was sufficiently large to counteract most or all of the positive main effect on delinquency of having a prior arrest. For example, more than 90% of the effect of a Wave 1 arrest on Wave 3 substance delinquency is eliminated by CAR treatment—a positive main effect of 0.850 is counteracted by a negative interaction coefficient of −0.782, leaving an effect on delinquency of 0.068. Thus, one key conclusion from the analysis is that case managers often were able to suppress the harmful implications of a Wave 1 arrest—the effect of that experience on later delinquency was significantly lower for those who received CAR treatment.

Our attention then turned to our final temporal combination, which involves the effects of Wave 2 risk factors on Wave 3 delinquency. In considering this issue, we are considering the possibility of delayed, after-program effects, given that no CAR services were delivered in the time between Waves 2 and 3 (the 2-year program of services ended around the time in which the Wave 2 data were collected). Achieving such effects may be a daunting task—it requires services received at an earlier time to suppress the effects of risk factors on delinquency at a later time when youths were no longer working with the case manager. Nevertheless, we consider this possibility in light of Harrell and her colleagues’ (1999) conclusion that the most notable improvements experienced by the CAR treatment group were in fact delayed improvements.

To consider this issue, we estimated equations similar to those presented in Table 3, although fewer equations were needed in focusing on just one temporal combination (Wave 1 → Wave 3). The key results for these equations are provided in Table 4, which reveals little support for the suppression hypothesis. Of the eight interactions, the only one that emerges as significant is for the equation in which delinquent peers are considered as a risk factor for general delinquency. This equation reveals a significant negative interaction that supports the idea that CAR treatment was able to suppress the effects of delinquent peers. It is important, however, that this interaction involves only the partial suppression of the effects of delinquent peers—the positive main effect of delinquent peers (b = 0.397) is notably higher than the negative interaction coefficient of −0.159. Also, no such pattern of suppression emerges when considering the effects of delinquent peers on substance delinquency. Taken as a whole, these results indicate that favorable effects of CAR treatment on Wave 3 delinquency likely occurred through some mechanism other than risk factor suppression.

Table 4:

Ordinary Least Squares Regression Analysis With Product Terms: Effects of Wave 2 Risk Factors, Treatment, and Product Terms on Wave 3 General and Substance Delinquency

	General Delinquency		Substance Delinquency
	Wave 2 → Wave 3		Wave 2 → Wave 3
	b	se	b	se
Arrest
Risk factor	0.277***	(0.099)	0.298***	(0.080)
Treatment	−0.086*	(0.061)	−0.045	(0.049)
Interaction	0.070	(0.129)	0.037	(0.104)
Delinquent peers
Risk factor	0.397***	(0.062)	0.256***	(0.050)
Treatment	−0.064	(0.053)	−0.048	(0.042)
Interaction	−0.159**	(0.087)	0.043	(0.069)
Family conflict
Risk factor	0.169***	(0.060)	0.061	(0.049)
Treatment	−0.064	(0.055)	−0.048	(0.044)
Interaction	0.046	(0.084)	0.136	(0.068)
Weak school commitment
Risk factor	0.237***	(0.064)	0.169***	(0.053)
Treatment	−0.046	(0.053)	−0.048	(0.043)
Interaction	0.098	(0.091)	0.007	(0.075)

Note. Control variables also included in each model: age, male, White, Hispanic.

p ≤ .10, one-tailed. **p ≤ .05, one-tailed. ***p ≤ .01, one-tailed.

Discussion

Recent research has revealed that many programs successfully reduce juvenile delinquency. In most instances, however, the exact mechanism by which this is accomplished has not been assessed. This study addressed this void in reference to the Children at Risk intervention that has been found to reduce a number of forms of delinquency (Harrell et al., 1999). The program’s design and the results from an initial evaluation suggested a novel way in which it may have succeeded—its case management model may have allowed it to suppress the harmful implications of four well-established risk factors for future offending.

Our analysis revealed partial support for this suppression hypothesis. The most direct test of this hypothesis came when considering the effects of Wave 1 risk factors on delinquency at Waves 2 and 3. This analysis captured the way in which risk factors present at the start of the program predicted offending reported at the end of the program 2 years later and in a follow-up that occurred 1 year after program completion. We found that more than half of the estimated equations (9 out of 16) failed to reveal significant evidence of suppression; thus, in these equations, the risk factors in question had effects on later offending that did not vary between the CAR treatment and control groups.

There were instances, however, where significant treatment × risk factor interactions emerged (some of these were significant only at a level of p ≤ .10) such that the risk factor in question had effects that were reduced for those in the CAR treatment group. Every risk factor except association with delinquent peers had at least one instance in which its Wave 1 measure had an effect on delinquency that was significantly reduced by CAR treatment. Moreover, the observed interactive effects often were sufficiently large in magnitude to counteract much of the main effect of the risk factor. Thus, the risk factors in question often had near-zero effects for the CAR treatment group.

Experience of a prior arrest was the risk factor for which we found the most consistent support for the suppression hypothesis. This variable had significant and positive main effects on general and substance delinquency reported at Waves 2 and 3, but in each instance this main effect largely was counteracted by the negative arrest × treatment interaction, such that an arrest was not associated with later delinquency among those who were assigned to the CAR treatment group.

It bears emphasizing that the findings just noted pertain to our analysis of the effects of Wave 1 risk factors that were present at the start of services. We also considered the CAR program’s ability to suppress the effects of Wave 2 risk factors that were reported at the end of services 2 years later. In doing so, we were testing for the possibility of delayed effects—the services received during the program could act to suppress the harmful effects of risk factors reported at the end of the program on delinquency committed after program completion. We found minimal support for this possibility—the typical result in this analysis was a nonsignificant interaction between the risk factors and CAR treatment.

Implications

Taken together, these results suggest that CAR treatment was sometimes able to suppress the harmful effects of key risk factors for delinquency and that this was principally the case when the risk factor in question involved an arrest reported at the beginning of the program. This pattern raises an important question: What is it about a prior arrest that makes its effects seemingly more amenable to suppression? One possibility is that a prior arrest was seen by case managers as a more serious risk factor than the others and therefore provoked a fuller array of services. After all, a prior arrest provides a tangible, concrete indicator of risk (in the form of an official record); indeed, it represents actual involvement in a crime severe enough to merit official action. And yet in supplemental analysis we found little support for the possibility that those reporting a Wave 1 arrest received disproportionately greater services. Although having a prior arrest at Wave 1 was significantly and positively related to the extent of services received, the magnitude of this relationship was similar to what was observed for each of our other risk factors.

We therefore see another explanation as more promising. That explanation involves the possibility that the relationship between being arrested and engaging in future delinquency is highly conditional in nature, such that various circumstances (including exposure to a comprehensive intervention) can moderate this relationship. As noted earlier, labeling theory emphasizes the idea of a positive association between incidents of official labeling (such as arrest) and later delinquency, but a related line of theory emphasizes that the experience of official labeling will have effects that depend on the circumstances and manner in which the label is experienced. Braithwaite’s (1989) reintegrative shaming theory, for example, argues that sanctions will have less harmful effects—perhaps even beneficial effects—when they are delivered in less stigmatizing ways. Sherman’s (1993) defiance theory makes a similar argument in suggesting that increased offending in the wake of an arrest—that is, “defiance” of the legal system—occurs only under certain conditions, such as when offenders feel stigmatized and are not in a good position to acknowledge the shame that has been experienced (Bouffard & Piquero, 2008). In the context of the current study, a key possibility is that the care, treatment, and services provided by CAR case managers helped those with an arrest feel less stigmatized. This, in turn, would allow CAR treatment to largely eliminate the positive relationship between being arrested and engaging in later delinquency. This explanation cannot be directly tested with the CAR data since it lacks survey items on participants’ perceptions of shame and stigmatization. Our findings on this issue nevertheless are consistent with these arguments that the positive relationship between being arrested and engaging in later delinquency can be diminished. This finding also is consistent with the empirical studies indicating that restorative justice programs—many of which are designed in part to reduce stigmatization of offenders—are sometimes able to reduce later offending (Rodriguez, 2007).

This pattern of findings has clear implications for policy, particularly in reference to a key dilemma that emerges in the response to juvenile crime; specifically, the legal system’s most established initial response to such acts—to make an arrest—has been found in rigorous recent studies, as well as in the current study, to be associated with increases rather than decreases in later crime (Bernburg & Krohn, 2003; Chiricos et al., 2007). Thus, an important risk factor for later offending is in fact introduced by the legal response to crime. In the past, advocates of labeling theory have seen this finding as suggesting the need for decriminalizing minor offenses and diverting a greater number of nonserious juvenile offenders. However, in instances in which a legitimate, actionable offense has occurred, an arrest is likely. The key policy implication of our findings is that the increase in crime that often occurs as a result of this action potentially can be prevented, provided that arrested juveniles receive case management that is tailored to their specific needs and that helps make the experience of an arrest less stigmatizing. Admittedly, translating that idea into specific juvenile and criminal justice procedures is a significant endeavor (Healy, 1999). Nevertheless, in light of rigorous recent research on the role that arrests play in escalating later individual offending, such efforts may bring significant gains.

Limitations and Future Directions

Our findings and their policy implications should be seen in the context of this study’s limitations, the first of which involves the risk factors that were considered. Although the CAR data included strong, repeated measures for four important risk factors, this certainly does not exhaust the list of key risk factors that could be considered. Two especially important risks for offending that could not be considered (because of the absence of comprehensive measures) include participants’ level of self-control (Pratt & Cullen, 2000) and commitment to antisocial or aggressive values (Stewart & Simons, 2006). Moreover, another risk factor for offending—residence in an economically distressed, high-crime neighborhood—could not be considered because it is a constant in the CAR data set (all participants in the control and treatment groups were drawn from such neighborhoods). Thus, there is clear room for future research to consider the suppression hypothesis with a different or more exhaustive list of risk factors.

An additional weakness follows from data limitations that made it impossible to directly measure case manager efforts to suppress the harmful effects of specific risk factors. Instead, evidence of suppression was inferred from instances in which a risk factor had an effect on later offending that was reduced by treatment group membership. Although we see this as a reasonable inference that corresponds closely to the idea of risk factor suppression, a more nuanced understanding of this mechanism would be possible with detailed qualitative data on case managers’ actions and intentions in responding to specific risk factors. Such data would not be easily obtained; in the context of the CAR project, this would have required an additional data collection component—a survey of case managers—on top of the extensive data already collected from parents, participants, and other officials. Nevertheless, such efforts would be valuable for better evaluating the effects of case management programs like the CAR intervention.

Conclusion

Our principal conclusion is that the effects of key risk factors on later crime can in some instances be suppressed by a comprehensive delinquency-reduction intervention, and this may be especially possible for the risk arising from the experience of an arrest. Findings of this kind are important in light of the reality that some risk factors may be especially difficult to prevent or reverse; in such instances, understanding how their harmful effects may be suppressed takes on special importance. A key question, however, is whether future evaluations of similar programs will yield results similar to those reported here. Answering this question requires future studies to devote significant attention not just to identifying the effects that a program may have on offending, but also to the specific causal mechanisms that may explain those effects. Our suggestion is that research of this kind can generate a substantially more nuanced understanding of public policy efforts to reduce crime and delinquency.

Footnotes

Appendix

Variable Description and Reliability Coefficients

	Items
Variable	Delinquency Variables	Alpha
Wave 2
General	# of times in past 2 years . . . ran away from home overnight, took something over $50, took something under $50, went on joyride, tried to buy/sell stolen goods, damaged something not yours, committed arson, got in serious school fight, got in group fight, committed robbery, attacked to hurt someone, forced sex (never, 1–2 times, 3–4 times, 5 times +)	.84
Substance	# of times . . . drunk/high from alcohol, marijuana, inhaled drugs, psychedelics, crack cocaine, cocaine in other form, heroin, steroids w/o Rx, Rx non med, drugs w/needle, helped w/drug sales, sold drugs directly, helped cut drugs (never, 1–2 times, 3–5 times, 6–9 times, 20–39 times,40 times +)	.76
Wave 3
General	Same as Wave 2 but with reference period of “in the last year,” and additional item “carried a weapon”	.87
Substance	Same as Wave 2	.72
	Risk Factors
Wave 1
D elinquent peers	Friends . . . sneak things w/o paying, friends act rowdy in public, friends throw bottles, rocks, things, friends join serious fights, friends take a car, friends take something w/o paying, friends have sex (no, yes)	.80
Arrest	Have you ever been arrested (no, yes)	N/A
Family conflict	# of times cursing/yelling/scream fights, # of times hits/slaps (never, less once/wk, once/wk, several/wk, everyday); caretaker threatens/curses/yells at me, curse/threat/scream at caretaker, get hit/slapped, hit/slap caretaker, get thrown out of home for a while (never, rarely, sometimes, usually)	.73
W eak school commitment	Like being at school, find school work interesting, trouble getting along w/teachers,^a try best in school (most of time, some of time, not often, never)	.48
Wave 2
Delinquent peers	Friends . . . sneak things w/o paying, act rowdy in public, throw bottles, rocks, things, join serious fights, joyride, take something w/o paying, have sex, belong to gang, sell hard drugs, use alcohol, use marijuana, use hard drugs (no, yes)	.83
Arrest	Have you ever been arrested (no, yes)	N/A
Family conflict	Same as Wave 1	.77
Weak school commitment	Think homework is a waste of time, try hard in school, working hard in school is worthwhile, like school generally, try to please teacher, grades are important to me, usually finish homework (agree, disagree)	.66

Note. ^aWhere necessary, response categories have been reversed so all items are in the same direction and descriptive of the variable name.

Authors’ Note:

The authors would like to thank the editor and anonymous reviewers for their helpful comments. An earlier draft of this article was presented at the 2010 meetings of the American Society of Criminology.

Carter Hay is an associate professor in the College of Criminology and Criminal Justice at Florida State University in Tallahassee, Florida. His research examines the causes and prevention of individual involvement in crime and delinquency, with recent or forthcoming publications appearing in such journals as Crime & Delinquency and Youth Violence and Juvenile Justice.

Brian Stults is an assistant professor in the College of Criminology and Criminal Justice at Florida State University in Tallahassee, Florida. His research addresses the general issues of race, crime, and community in urban areas, with a particular focus on segregation, racial threat, and spatial and temporal patterns of crime.

Emily Restivo is an assistant professor of criminal justice at the New York Institute of Technology. Her research interests include the testing of criminological theory and applied statistics.

References

Agnew

(1992). Foundation for a general strain theory of crime and delinquency. Criminology, 30, 47-87.

Aiken

L. S.

West

S. G.

(1991). Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage.

Akers

R. L.

(1998). Social learning and social structure: A general theory of crime and deviance. Boston, MA: Northeastern University Press.

Barnoski

(2004). Outcome evaluation of Washington State’s research-based programs for juvenile offenders. Olympia: Washington State Institute for Public Policy.

Baron

R. M.

Kenny

D. A.

(1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182.

Becker

(1963). Outsiders: Studies in the sociology of deviance. New York, NY: Free Press.

Bernburg

J. G.

(2009). Labeling theory. In Krohn

M. D.

Lizotte

M. J.

(Eds.), Handbook on crime and deviance (pp. 187-207). New York, NY: Springer.

Bernburg

J. G.

Krohn

M. D.

(2003). Labeling, life chances, and adult crime: The direct and indirect effects of official intervention in adolescence on crime in early adulthood. Criminology, 41, 1287-1318.

Bouffard

L. A.

Piquero

N. L.

(2008). Defiance theory and life course explanations of persistent offending. Crime & Delinquency, 56, 227-252.

10.

Braithwaite

(1989). Crime, shame and reintegration. New York, NY: Cambridge University Press.

11.

Chiricos

Barrick

Bales

Bontrager

(2007). The labeling of convicted felons and its consequences for recidivism. Criminology, 45, 547-581.

12.

Cullen

F. T.

(2005). The twelve people who saved rehabilitation: How the science of criminology made a difference. Criminology, 43, 1-42.

13.

Farrington

D. P.

Welsh

B. C.

(2007). Saving children from a life of crime. New York, NY: Oxford University Press.

14.

Gottfredson

D. C.

Kearley

B. W.

Najaka

S. S.

Rocha

C. M.

(2007). How drug treatment courts work: An analysis of mediators. Journal of Research in Crime and Delinquency, 44, 3-35.

15.

Grossman

J. B.

Tierney

J. P.

(1998). Does mentoring work? An impact study of the Big Brothers Big Sisters program. Evaluation Review, 22, 403-426.

16.

Harrell

Cavanagh

Sridharan

(1999). Evaluation of the Children at Risk program: Results 1 year after the end of the program. Research in brief. Washington, DC: National Institute of Justice.

17.

Harrell

Cavanagh

Sridharan

(2000). Evaluation of the Children at Risk program in Austin, Texas, Bridgeport, Connecticut, Memphis, Tennessee, Savannah, Georgia, and Seattle, Washington, 1993–1997 [Computer file, ICPSR version]. Washington, DC: Urban Institute [producer], 1998. Ann Arbor: Inter-University Consortium for Political and Social Research [distributor].

18.

Hawkins

D. J.

Kostermann

Catalano

R. F.

Hill

K. G.

Abbott

R. D.

(2005). Promoting positive adult functioning through social development intervention in childhood: Long-term effects from the Seattle Social Development Project. Archives of Pediatrics & Adolescent Medicine, 159, 25-31.

19.

Haynie

D. L.

Osgood

W. D.

(2005). Reconsidering peers and delinquency: How do peers matter? Social Forces, 84, 1109-1130.

20.

Healy

K. M.

(1999). Case management in the criminal justice system. Research in brief. Washington, DC: National Institute of Justice.

21.

Henggeler

S. W.

Schoenwald

S. K.

Borduin

C. M.

Rowland

M. D.

Cunningham

P. B.

(1998). Multisystemic treatment of antisocial behavior in children and adolescents. New York, NY: Guilford.

22.

Hirschi

(1969). Causes of delinquency. Berkeley: University of California Press.

23.

Howell

J. C.

(2009). Preventing and reducing juvenile delinquency. Thousand Oaks, CA: Sage.

24.

Jaccard

Turrisi

Wan

C. K.

(1990). Interaction effects in multiple regression Newbury Park, CA: Sage.

25.

Johnson

L. M.

Simons

R. L.

Conger

R. D.

(2004). Criminal justice system involvement and continuity of youth crime. Youth & Society, 36, 3-29.

26.

Landenberger

N. A.

Lipsey

M. W.

(2005). The positive effects of cognitive-behavioral programs for offenders: A meta-analysis of factors associated with effective treatment. Journal of Experimental Criminology, 1, 451-476.

27.

Langbein

Felbinger

(2006). Public program evaluation: A statistical guide. Armonk, NY: M.E. Sharpe.

28.

Lemert

E. M.

(1951). Social pathology. New York, NY: McGraw-Hill.

29.

Lipsey

M. W.

(1995). What do we learn from 400 studies on the effectiveness of treatment with juvenile delinquents? In McGuire

(Ed.), What works: Reducing reoffending (pp. 63-78). Chichester, UK: Wiley.

30.

Mears

D. P.

(2007). Towards rational and evidence-based crime policy. Journal of Criminal Justice, 35, 667-682.

31.

Mears

D. P.

Hay

Gertz

Mancini

(2007). Public opinion and the foundation of the juvenile court. Criminology, 45, 223-257.

32.

Moon

M. M.

Sundt

J. L.

Cullen

F. T.

Wright

J. P.

(2000). Is child saving dead? Public support for juvenile rehabilitation. Crime & Delinquency, 46, 38-60.

33.

Piquero

A. R.

Farrington

D. P.

Welsh

B. C.

Tremblay

Jennings

W. G.

(2009). Effects of early family/parent training programs on antisocial behavior and delinquency. Journal of Experimental Criminology, 5, 83-120.

34.

Pratt

T. C.

Cullen

F. T.

(2000). The empirical status of Gottfredson and Hirschi’s general theory of crime: A meta-analysis. Criminology, 38, 931-964.

35.

Rodriguez

(2007). Restorative justice at work: Examining the impact of restorative justice resolutions on juvenile recidivism. Crime & Delinquency, 53, 355-379.

36.

Sampson

R. J.

Laub

J. H.

(1993). Crime in the making: Pathways and turning points through life. Cambridge, MA: Harvard University Press.

37.

Sherman

L. W.

(1993). Defiance, deterrence, and irrelevance: A theory of the criminal sanction. Journal of Research in Crime and Delinquency, 30, 445-473.

38.

Stewart

E. A.

Simons

R. L.

(2006). Structure and culture in African American adolescent violence: A partial test of the “code of the street” thesis. Justice Quarterly, 23, 1-33.

39.

Tabachnick

B. G.

Fidell

L. S.

(2007). Using multivariate statistics (5th ed.). Boston, MA: Allyn & Bacon.

40.

Tannenbaum

(1938). Crime and the community. Boston: Ginn.