The Role of Protective Factors in the Predictive Accuracy of the Structured Assessment of Violence Risk in Youth (SAVRY)

Abstract

The use of risk assessment instruments has become standard procedure in the juvenile justice system. Most empirical assessments of the predictive validity of these instruments concentrate on the ability of a total risk score, individual risk factors, or risk domains to predict negative juvenile outcomes but fail to consider the utility of the protective factors in influencing or moderating those risks. This study utilizes the Structured Assessment of Violence Risk in Youth to analyze the impact of protective factors on reoffending using a sample (n = 460) of postadjudication juveniles in a southern state. The overall protective domain and two specific protective factors were related to reoffending in bivariate analyses. However, protective factors did not predict reoffending when controlling for risk domains. Rather, further analyses suggest that certain protective factors buffer the effects of some of the risk domains.

Keywords

SAVRY protective factors risk assessment predictive validity

The use of risk assessment instruments (RAIs) by juvenile justice agencies has increased substantially over the past two decades. The general purpose of these tools is to identify the level of risk each youth poses to society and, based on the assessed level of risk, make objective juvenile justice decisions such as community or secure placement. The use of RAIs has improved decision-making by focusing on the youth rather than the offense (Vincent, Chapman, & Cook, 2011), reduced disparities in decision-making across extralegal factors such as race (Chapman, Desai, Falzer, & Borum, 2006), and has led to reductions in out of home placements (Vincent, Guy, Gershenson, & McCabe, 2012). Juvenile courts process over 1 million cases annually (Furdella & Puzzanchera, 2015). Many of these juveniles are confined in residential facilities at various points throughout the process, costing US taxpayers approximately US$241 per day, per juvenile (American Correctional Association, 2008). Based on the available evidence, RAIs carry the potential to significantly reduce these costs by reducing unnecessary placements in secure care which, in turn, will lead to fewer youth exposed to the iatrogenic effects of secure confinement, improved youth outcomes after juvenile justice involvement, and substantial cost-saving by reserving secure confinement for the most at-risk juvenile offenders.

The development and study of RAIs has primarily focused on the types of risk factors included in these instruments and their ability to predict future violence and offending. Overwhelming agreement regarding the importance of both static and dynamic risk factors, as well as moderate support for the predictive validity of risk ratings based on these items, has been provided by a number of studies (Schwalbe, 2007; Singh, Grann, & Fazel, 2011). More recently, there has been a growing recognition that RAIs can also help inform case management and intervention planning (Serin, Chadwick, & Lloyd, 2016). This has led to the acknowledgment of the importance of resilience and protective factors on future offending, and as a result, a number of contemporary RAIs incorporate measures of both risk and protective factors. However, the unique contribution of protective factors, as well as the predictive validity of these items, has received mixed results. Additionally, very little research has examined the hypothesized buffering effect of protective factors on the effects of risk factors (Lodewijks, de Ruiter, & Doreleijers, 2010). Therefore, the overall purpose of this study was to examine the unique contribution of protective factors and the buffering effects of protective factors on risk for reoffending.

Literature Review

RAIs

Risk assessment in the juvenile justice context is defined as the process of estimating a juvenile’s likelihood of continued involvement in delinquent behavior (Baird et al., 2013). Juvenile justice personnel are regularly guided by RAIs to confront the difficult tasks of reducing risk to the community, determine whether a juvenile is amenable to rehabilitation, and distribute resources wisely (Mulvey, 2005). As of 2014, 33 states and the District of Columbia had adopted an RAI at the state level (Wachter, 2014). One study revealed that the RAI was directly or indirectly referred to in 76% of juvenile sentencing hearings or adult transfer proceedings (Urquhart & Viljoen, 2014), indicating the crucial importance of these instruments.

Various methodologies, criteria, and models have been used in the development of RAIs for juveniles (for a review of several RAIs used in juvenile justice systems across the United States, see Baird et al., 2013). Varying RAIs incorporate different approaches to assessing risk including actuarial methods or the use of structured professional judgment (SPJ). Another important variation within RAIs is that some consider only risk factors, while others integrate both risk and protective factors. Risk factors are variables that increase the likelihood of offending. In contrast, protective factors are variables that decrease the likelihood of offending. Empirical research demonstrates that serious violent offending may be directly related to the presence of multiple risk factors (Hawkins et al., 2000; Jennings et al., 2016). However, studies also suggest that the presence of multiple protective factors may reduce the likelihood of antisocial behaviors in juveniles (Andershed, Gibson, & Andershed, 2016; Jennings et al., 2016; Werner, 2000). Accordingly, it is critical to consider both in risk assessment.

However, many studies of juvenile RAIs concentrate on risk factors rather than protective factors. This neglect in the research may be due to the argument by some (Baird, 2009) that protective factors are simply the absence of risk factors. However, other researchers have distinguished between two types of protective factors: negative and positive (see Borum, Bartel, & Forth, 2003). Negative protective factors are notable for the absence of a risk factor, while positive protective factors are notable for their presence. These positive protective factors operate by “buffering” or reducing the effects of risk factors and assist in explaining why some youth exposed to risk factors do not become involved in antisocial behavior (Rennie & Dolan, 2010). Examples of such protective factors include a strong support system, attitudes toward change, and resilient personality traits. Such factors can be particularly important for guiding interventions to reduce the risk of reoffending, especially for youth with risk factors that may not be amenable to change.

Structured Assessment of Violence Risk in Youth (SAVRY)

The SAVRY (Borum et al., 2003) is one of the many RAIs created in recent decades to assist court officials in their decision-making. The SAVRY uses the SPJ approach to guide an evaluator through a checklist of factors with an empirically demonstrated relationship to violent recidivism and includes both risk and protective factors (Borum et al., 2003). The SAVRY takes into consideration the developmental differences of youth when compared to adults and emphasizes dynamic risk/needs factors empirically supported in the literature. The SAVRY has been shown to predict risk for reoffending better than unstructured professional judgment (Hilterman, Nicholls, & van Nieuwenhuizen, 2014; Vincent, Guy, Gershenson et al., 2012) and better than many other structured methods of risk classification including the Psychopathy Checklist: Youth Version (Dolan & Rennie, 2008; Welsh, Schmidt, McKinnon, Chattha, & Meyers, 2008), the Youth Level of Service/Case Management Inventory (YLS/CMI) (Welsh et al., 2008), and other RAIs specifically developed to assess violence risk (Singh et al., 2011).

Previous research has examined the interrater reliability, validity, and predictive accuracy of the SAVRY. Studies indicate that the interrater reliability of the SAVRY falls within acceptable limits (Hilterman et al., 2014; Vincent, Guy, Fusco, & Gershenson, 2012). Research further supports the construct validity of the SAVRY risk indices (Childs, Frick, & Gottlieb, 2016). Meta-analyses of the SAVRY’s predictive accuracy indicate that it significantly predicts general, nonviolent, and violent recidivism (Olver, Stockdale, & Wormith, 2009). Several other studies support the predictive validity of the SAVRY for both violent and nonviolent offending across varying populations (Childs et al., 2013; Lodewijks, Doreleijers, & de Ruiter, 2008; Lodewijks, Doreleijers, de Ruiter, & Borum, 2008; Viljoen et al., 2008; Vincent et al., 2011).

Importantly, the SAVRY includes six positive protective factors: prosocial involvement, strong social support, strong attachments and bonds, positive attitude toward intervention and authority, strong commitment to school, and resilient personality traits (Borum et al., 2003). Prosocial involvement concerns conventional social behaviors of the youth such as cooperation, participation in prosocial activities, appropriate expression of emotions, and helpfulness. In comparison, strong social support refers to both peer-aged and adult individuals providing prosocial support and assistance to the youth in times of distress. The protective factor of strong attachment and bonds is supported by a plethora of research finding that warm and affectionate relationships with adults mitigate against future violence (Hawkins, Catalano, & Miller, 1992; Simons, Paternite, & Shore, 2001). Positive attitude toward intervention and authority represents the youth’s positive attitude toward remediation attempts and authority figures. Strong commitment to school addresses the youth’s involvement in school activities, educational achievement, and a strong bond toward school. Previous studies indicate that both school achievement (Herrenkohl et al., 2000) and a strong bond to school (Hawkins et al., 1998) protect against violence in adolescence. An additional protective factor considered in the SAVRY is resilient personality traits, representing a youth’s ability to overcome adverse conditions. Thus, the SAVRY protective domain aims to conceptualize factors related to commitment to conventional society and against deviant activities and attitudes.

Only a handful of studies have examined the validity and usefulness of the SAVRY protective factors, and the findings are mixed. Studies of the reoffending of male juvenile sex offenders indicate that the SAVRY protective factors do not significantly predict reoffending (Klein, Rettenberger, Yoon, Köhler, & Briken, 2015; Viljoen et al., 2008). Similarly, Penney, Lee, and Moretti (2010) found that the protective factors were not directly associated with decreased odds of offending in a sample of high-risk Canadian youth. A recent study of postadjudication juveniles in Spain revealed that the SAVRY protective factors had no incremental validity over the SAVRY risk factors in predicting reoffending (Hilterman et al., 2014). Additionally, Vincent, Chapman, and Cook (2011) found that the SAVRY protective domain was not significantly related to any rearrest or violent rearrests in a sample of male juvenile offenders

Other studies of the SAVRY protective factors have reported more positive results. An examination of the impact of the protective domain on three separate samples of Dutch male adolescent violent offenders indicated that the overall domain was a significant predictor (negative) of violent recidivism (Lodewijks et al., 2010). However, the area under the curve (AUC) values reported were rather low ranging from .16 to .28. Lodewijks, de Ruiter, and Doreleijers (2010) also assessed the incremental validity of the protective domain and showed that it predicted recidivism above and beyond the contribution of the Dynamic and Static Risk Scales from the SAVRY in all three samples. Similarly, when Lodewijks, Doreleijers, de Ruiter, and Borum (2008) examined the predictive validity of the SAVRY for assessing the risk of engaging in physical violence in an institutional setting, results indicated that the addition of the protective domain significantly improved the predictive validity of the SAVRY (i.e., added incremental validity). Studies of incarcerated British male adolescents indicated that the SAVRY protective domain predicted general recidivism at the bivariate level (Dolan & Rennie, 2008; Rennie & Dolan, 2010). However, when added to the multivariate model that included the SAVRY risk total score, the protective domain did not add incremental validity (Dolan & Rennie, 2008).

A few studies have also examined the predictive validity of the individual protective factors measured on the SAVRY. One study found that four of the SAVRY protective factors (strong social support, strong attachments and bonds, positive attitudes, and strong commitment to school) modestly predicted institutional violence individually (AUC values ranged from .28 to .35; Lodewijks, Doreleijers, de Ruiter et al., 2008). Other research has also found that strong social support and strong attachments and bonds were significant predictors of desistance (Gammelgård, Koivisto, Eronen, & Kaltiala-Heino, 2015; Lodewijks et al., 2010). Additionally, a study involving Finnish youth indicates that the absence of prosocial involvement and positive attitudes was associated with negative outcomes (Gammelgård et al., 2015). In a study of Australian youth, prosocial involvement and positive attitude predicted desistance from general recidivism, while strong commitment to school predicted desistance from violent recidivism (Shepherd, Luebbers, & Ogloff, 2016). In contrast, Rennie and Dolan (2010) found that none of the individual protective factors were predictive of violent reoffending, while only resilient personality traits were predictive of general reoffending.

An important issue when interpreting these past studies is that they typically considered the association between the SAVRY protective factors and recidivism, either alone or in terms of their incremental validity over the SAVRY risk factors. Only one of the aforementioned studies (Lodewijks et al., 2010) specifically investigated a possible buffering effect of the SAVRY protective factors on the risk domains. Results suggested a significant difference in violent recidivism when protective factors were present versus absent. For example, results from the analysis of one of the subsamples indicated that 40% of youth without any protective factors recidivated, compared to 6% of youth with protective factors present. However, due to very small subsample sizes, the study could only examine bivariate associations to test for a buffering effect and did not examine the effects of the individual protective items. Additionally, the sample utilized in the study was highly specific (Dutch, male, violent offenders), and the study did not examine general recidivism.

Current Study

In sum, studies on the usefulness of the SAVRY protective factors have primarily focused on the bivariate association between the SAVRY protective items or domain with recidivism (Lodewijks et al., 2010; Rennie & Dolan, 2010) or the incremental validity of the SAVRY protective domain over the SAVRY risk items (i.e., total score, summary risk rating, domain scores; Rennie & Dolan, 2010). The findings of these studies have been inconsistent and fail to provide a clear picture of the utility of these items in predicting reoffending. A majority of these studies have also failed to examine whether a moderating relationship exists among the SAVRY protective items and the risk domains. This is a major shortcoming to existing research on the utility of the protective items in juvenile risk assessment because the specific relationships hypothesized by the SAVRY authors (see Borum et al., 2003), as well as other juvenile assessment scholars, are a buffering effect of the protective items on the risk items. Lodewijks et al. (2010) conducted the only study to examine the “buffering effect” hypothesized by the SAVRY developers. However, due to some limitations in their data, they were solely able to examine bivariate associations. The current study builds upon the work of Lodewijks and colleagues by using multivariate analyses to detect a buffering effect. This is done using a sample that includes both nonviolent and violent offenders and is derived from an American jurisdiction. This also adds to existing research because a number of existing studies that examine the SAVRY protective factors are based on international samples (Dolan & Rennie, 2008; Gammelgård et al., 2015; Lodewijks et al., 2010; Rennie & Dolan, 2010). Many of the previous studies also suffer the serious limitation of completion of the SAVRY through the retrospective gathering of file information. The SAVRYs utilized in the current study were completed by juvenile probation officers (JPOs) based on multiple sources of information: interviews with the juvenile, interviews with the child’s guardian, and available file information. We examined the SAVRY’s ability to predict reoffending over a 12-month follow-up period after administration using official records. We tested the following research questions:

Do the SAVRY protective factors differ across reoffending and nonreoffending groups of postadjudication delinquents?

Do the SAVRY protective factors add incremental validity to the SAVRY after accounting for the SAVRY risk domains?

Do the SAVRY protective factors moderate the impact of the risk domains on reoffending?

Method

Sample

Data collection was part of a larger John D. and Catherine T. MacArthur Foundation funded project aimed at promoting evidence-based practices in the juvenile justice system. As part of the project, use of the SAVRY was made a mandatory part of predisposition juvenile probation practices in the three sites in a southern state. The three sites were selected due to their representativeness in serving rural, suburban, and urban populations.

The original sample included 505 adjudicated youth placed on juvenile probation in three sites in a southern state. We removed 23 girls from the sample due to the potential for inflation of coefficients resulting from low cell counts across the independent and dependent variables (see Greenland, Mansournia, & Altman, 2016). Additionally, 13 of the boys in the sample were coded as having an age under 10 years old. Since the SAVRY is meant to target youth over the age of 10, these youth were also excluded. This resulted in a sample of 469 adjudicated boys. Finally, 8 cases were missing values on all SAVRY protective items and one boy was missing recidivism data. Since these variables represent the study’s primary focus, we also removed these 9 cases. This left a final sample size of 460 adjudicated boys ages 10–18 at the time of adjudication. A majority of the sample were adjudicated for minor offenses such as theft, misdemeanor assault, or battery. Further description of the sample is provided in Table 1.

Table 1.

Descriptive Statistics.

Race: N (%)
White	110 (24%)
Non-White	350 (76%)
Age: M (SD)	14.79 (1.54)
Original offense type: N (%)
Non-violent offense	213 (46%)
Violent offense	247 (54%)
Number general reoffending: N (%)	167 (36%)
Number violent reoffending: N (%)	56 (12%)
SAVRY domain scores: M (SD, range)
Historical risk domain	5.19 (3.39, 0–19)
Social/contextual risk domain	3.37 (2.52, 0–10)
Individual risk domain	5.03 (3.85, 0–16)
Protective domain	3.67 (1.98, 0–6)
Number of sample presenting with each protective factor: N (%)
Prosocial involvement	194 (42%)
Strong social support	313 (69%)
Strong attachments and bonds	332 (73%)
Positive attitude toward intervention and authority	319 (70%)
Strong commitment to school	210 (46%)
Resilient personality traits	308 (67%)

Note. N = 460. Missing data ranged from no missing values to 12 cases with missing values. SAVRY = Structured Assessment of Violence Risk in Youth.

Measures

Reoffending

The dependent variables measure reoffending which is defined as any new petition to the juvenile or adult court during the 12-month follow-up period. Recidivism data were collected from the court database at each of the three study sites. All participants had at least 12 months after the initial SAVRY administration to reoffend. Four variables were created using the extracted reoffending data. The first is any reoffense during the 12-month follow-up period. This is a dichotomous variable coded as 0 (No) and 1 (Yes). The second variable is time to reoffense, which accounts for the number of days from the original petition to the new petition. Thirty-six percent of the sample was petitioned for a new offense (i.e., violent or nonviolent). We also created two violent reoffense variables to measure violent recidivism only. Violent reoffense is a dichotomous variable coded 0 (No) and 1 (Yes), and time to violent reoffense is the number of days from the original petition to the new petition. A violent reoffense includes all misdemeanor and felony crimes involving the use, attempt, or threat of physical force. Twelve percent of the sample was petitioned for a new violent offense during the follow-up period.

SAVRY risk domains

The SAVRY includes 30 items. Twenty-four of the items are risk factors divided between three risk domains: historical, social/contextual, and individual. The three domains are used individually as independent variables in the current study. The historical domain includes 10 risk factors which are based on past behavior and/or experiences and are generally static. Risk factors include the youth’s various experiences with violence consisting of their history of violence and the possibility of early initiation of violence. The domain also considers family history such as parental/caregiver criminality, early caregiver disruption, exposure to violence in the home, and childhood history of maltreatment. Other miscellaneous historical risk factors include history of nonviolent offending, history of self-harm or suicide attempts, poor school achievement, and past supervision failures. The social/contextual domain scrutinizes interpersonal relationships, connection to social institutions, and the youth’s environment. Risk factors in this domain include peer delinquency, peer rejection, stress and poor coping, lack of personal/social support, poor parental management, and community disorganization. The individual domain focuses on the youth’s attitudes along with psychological and behavioral functioning. These eight risk factors include negative attitudes, anger management problems, poor compliance, risk-taking/impulsivity, substance use difficulties, low empathy/remorse, attention deficit/hyperactivity difficulties, and low interest/commitment to school.

The SAVRY requires a rating for each of the 24 risk items listed above of low, moderate, or high. Trained JPOs administered the SAVRY to the youth postadjudication. JPOs scored each youth based on an interview with the youth, interview with the caregiver, and any additional information available to the JPO. For purposes of the current analyses, numeric values were assigned to item ratings (0 = low, 1= moderate, 2 = high). The independent variables created for each SAVRY risk domain are measured using the sum of the ratings for each item within that domain.

The field reliability of SAVRY ratings among JPOs in two of the three sites was assessed in a previous study (Vincent, Guy, Fusco et al., 2012). Intraclass correlation coefficients (ICCs) were calculated using randomly selected cases of 80 juveniles interviewed by a JPO while observed by a trained research assistant. Both the JPO and research assistant reviewed the juvenile’s information and independently rated the SAVRY. Results indicated that the ICCs were excellent for SAVRY total scores (ICC = .86) and good for overall risk rating (ICC = .71), suggesting that the SAVRY can be used reliably in the field. Additionally, the Cronbach’s α of all three indices indicate acceptable values of internal consistency for all the factors in each risk domain: historical (α = .73), social/contextual (α = .71), and individual (α = .83).

SAVRY protective domain

There are six protective factors included in the SAVRY protective domain. As described above, they include individual and contextual items that can reduce the negative impact of risk factors and lower the probability of violent recidivism. Items from the protective domain are rated differently than the risk factor domains. JPOs rate protective factors as either present or absent. Numeric values were assigned creating a dichotomous variable (1 = present, 0 = absent). An index was created from the sum of the 6 items in the protective domain (α = .79). Additionally, each item from the domain is used independently to examine whether they contribute individually to the predictive validity of the SAVRY.

Control variables

The control variables include age, race, and offense level. Age is a continuous variable representing the youth’s age at the time of adjudication. The variable ranges from age 10 to 18. Numeric values were assigned to the remaining nominal variables. Race is represented by a dichotomous variable (1 = White, 2 = non-White). Offense level is a dichotomous variable distinguishing whether the juveniles’ original offenses were violent (=1) or nonviolent (=0).

Analysis

First, several bivariate relationships were examined. Independent samples t tests were performed to assess differences in mean SAVRY domain scores between reoffending groups. Subsequently, χ² tests of independence were utilized to identify group differences in recidivism among those presenting with each individual protective factor and those scored as absent. Receiver operating characteristics (ROCs) curves were used to estimate the accuracy of the SAVRY scores for predicting general and violent reoffending. AUC values represent the probability that a youth who is randomly selected from the group of youth who reoffended will have a score on the SAVRY measure compared to a randomly selected youth from the group that did not reoffend (Vincent, Guy, Gershenson et al., 2012). AUC values range from 0 (perfect negative prediction) to 1.0 (perfect positive prediction; Douglas, Yeomans, & Boer, 2005).

Then, Cox regression analyses were conducted to investigate the relationship between the independent variables and the event of reoffense, while simultaneously considering the length of time to reoffense. The regression model developed by Cox (1972) was utilized in current analyses due to the inability of ordinary least squares regression methods to analyze survival times. Since the event of reoffense occurs for some youth but not others, the data contain incomplete observations. These incomplete observations are defined as “censored” survival times (Aalen, Borgan, & Gjessing, 2008). Cox regression analysis avoids the problems produced by censoring and time-varying explanatory variables (Allison, 2014). The Cox regression model is viewed as a major bridge between parametric and nonparametric approaches in survival analysis (Allison, 2014). The model is parametric in its specification of a regression model with a specific functional form but nonparametric in that does not specify the form of the distribution of the time events (Allison, 2014).

Prior to conducting the Cox regression analyses, multiple imputation was used to preserve cases where values on all study variables were missing at random (see Statacorp, 2017). Across the study variables, the amount of missing data ranged from no missing values (i.e., recidivism measures, race, offense level) to 12 cases with missing values on 1 or more SAVRY protective items. Then, the Cox regression analyses proceeded in a number of steps. First, we examined the incremental validity of the SAVRY protective domain and each individual protective factor separately. This proceeded in a stepwise fashion where the predictive validity of the SAVRY risk domains was examined first. This served as our baseline model for comparisons across additional analyses. Then each of the relevant protective measures (domain score and each individual item found to be significant in the bivariate analyses) were added to the model separately. All models included age, race, and offense level as control variables. To assess incremental validity, we assessed the change in the log likelihood from Steps 1 and 2 using χ² tests of significance.

Finally, we examined the buffering effect of the SAVRY protective items (i.e., the presence of protective factors mitigates or buffers the effects of risk factors; Borum et al., 2003). The continuous SAVRY domain scores were mean centered, and interaction terms were created (Jaccard & Turrisi, 2003) to appropriately model the hypothesized relationship between risk and protective domains. After creating interaction terms, additional Cox regression models including the SAVRY risk domains and the control variables were estimated to determine whether the SAVRY protective domain moderated the impact of the SAVRY risk domains on reoffending. To examine moderation among the dichotomous protective items, the predictive validity of the SAVRY risk domains was compared between those youth presenting with protective items of interest (i.e., items found to be significant at the bivariate level) versus those without the protective items. These analyses allow for a clear comparison of the effects of the risk domains on reoffending while accounting for the presence or absence of the protective items.

Results

Descriptive statistics are displayed in Table 1. The most commonly present protective factor is strong attachments and bonds (N = 332, 73%), while the least common is prosocial involvement (N = 194, 42%). Additional analyses indicated that the presence of all six protective factors is the most common protective domain score (N = 106, 24%); only 49 juveniles (11%) had a complete absence of protective factors.

Results of independent samples t tests for each of these domains across reoffending groups are listed in Table 2. For general reoffending, significant differences in SAVRY domain scores were found across the reoffending and nonreoffending groups. The reoffending group had significantly higher scores on the historical, t(454) = −3.63, p < .001, social/contextual, t(457) = −3.57, p < .001, and individual risk domains, t(458) = −4.98, p < .001. Results for the protective domain were also in the expected direction. The general reoffending group had a significantly lower mean score (M = 3.41, SD = 2.06) compared to the nonreoffending group, M = 3.82, SD = 1.92; t(446) = 2.09, p < .05. ROC values for all three SAVRY risk domains showed moderate levels of prediction (ROC = .60–0.62, p < .05). However, the ROC value for the protective domain was rather low (.44) but statistically significant (p < .05).

Table 2.

SAVRY Domain Scores Across General and Violent Reoffending.

SAVRY Domains	No General Reoffending M (SD)	General Reoffending M (SD)	t	df	ROC
Historical risk domain	4.75 (3.42)	5.94 (3.22)	−3.63***	454	.62***
Social/contextual risk domain	3.06 (2.42)	3.92 (2.60)	−3.57***	457	.60**
Individual risk domain	4.38 (3.52)	6.19 (4.12)	−4.98***	458	.62***
Protective domain	3.82 (1.92)	3.41 (2.06)	2.09*	446	.44*
SAVRY Domains	No Violent Reoffending M (SD)	Violent Reoffending M (SD)	t	df	ROC
Historical risk domain	5.02 (3.40)	6.45 (3.10)	−2.98**	454	.64**
Social/contextual risk domain	3.29 (2.49)	3.96 (2.64)	−1.88	457	.57
Individual risk domain	4.85 (3.76)	6.36 (4.23)	−2.77**	458	.60*
\Protective domain	3.68 (1.97)	3.62 (2.09)	0.21	446	.50

Note. N = 460. Missing data ranged from no cases with missing values to 12 cases with missing values. SAVRY = Structured Assessment of Violence Risk in Youth; ROC = receiver operating characteristic.

*p < .05. **p < .01. ***p < .001.

For violent reoffending, the historical, t(454) = −2.98, p < .01, and the individual, t(458) = −2.77, p < .01, risk domains were found to be significantly different across youth who had a new petition for a violent offense and those who did not. Two ROC values showed acceptable levels of prediction: the historical risk domain (.64, p < .01) and the individual risk domain (.60, p < .05). The protective domain was not found to significantly differ across violent reoffending groups and did not reveal acceptable levels of prediction (.50, p > .05).

χ² tests of independence were performed to examine the bivariate associations between each individual protective factor and reoffending. Significant differences were found across the general reoffending groups for 2 of the 6 protective items. Juveniles with a positive attitude toward intervention and authority were significantly less likely to reoffend, χ²(1) = 5.18, p < .05. Similarly, juveniles with resilient personality traits were also significantly less likely to reoffend compared to those without these traits, χ²(1) = 4.32, p < .05. None of the ROC values for general reoffending showed sufficient levels of prediction. For violent reoffending, no significant differences in protective items were found across the groups, and all ROC values showed low predictive accuracy.

Based on the results of the bivariate analyses in Tables 2 and 3, we did not conduct Cox regression analyses predicting violent reoffending. Three findings led to this decision. First, none of the SAVRY protective measures were found to significantly differ across the two groups. Second, the ROC values also suggested low predictive accuracy across all of the measures. Third, due to the low number of offenders in the sample that received a new petition for a violent offense (n = 56), low cell counts in the regression models would have increased the odds of biased estimates and inaccurate results. We chose to continue our Cox regression analyses predicting general reoffending because the bivariate results revealed significant differences across the groups for the protective domain, and two of the individual items and no problems with low cell sizes were identified. In addition, the presence of significant differences across groups (i.e., t tests) but low levels of predictive accuracy (i.e., ROC values) suggest that these items may not be directly related to reoffending, but instead, may be performing through the hypothesized buffering effect. Thus, stepwise Cox regression analyses were performed to examine the predictors of time to general reoffending only.

Table 3.

SAVRY Protective Domain Items Across General Reoffending and Violent Reoffending.

Protective Factors Present	Present% Reoffended	Absent% Reoffended	χ²	df	ROC
Prosocial involvement	34	39	1.20	1	.47
Strong social support	35	40	1.10	1	.47
Strong attachments and bonds	35	40	1.00	1	.47
Positive attitude toward intervention and authority	33	44	5.18*	1	.44
Strong commitment to school	35	38	0.48	1	.49
Resilient personality traits	33	43	4.32*	1	.45
Protective Factors Present	Present % Violent Reoffended	Absent % Violent Reoffended	χ²	df	ROC
Prosocial involvement	13	11	0.45	1	.52
Strong social support	12	12	0.01	1	.50
Strong attachments and bonds	12	12	0.01	1	.50
Positive attitude toward intervention and authority	11	15	1.77	1	.45
Strong commitment to school	13	11	0.44	1	.52
Resilient personality traits	12	13	0.23	1	.48

Note. N = 460. Missing data ranged from no cases with missing values to five cases with missing values. SAVRY = Structured Assessment of Violence Risk in Youth; ROC = receiver operating characteristic.

*p < .05.

Results for the SAVRY protective domain are presented in Table 4. In the baseline model, F(6) = 6.13, p < .001, the statistically significant predictors of reoffense included the individual risk domain, Exp (B) = 1.09, p < .01, and age, Exp (B) = .91, p < .05. This means that the hazard of reoffending was 9% higher for adjudicated youth with each unit increase in the individual risk domain score and that younger youth were more likely to reoffend. To evaluate whether the protective domain added any incremental value to the baseline model, the protective domain was added into the second model, F(7) = 4.56, p < .001. Results of the analysis indicated that the protective domain did not significantly add to the model, and similar to previous analyses, only the individual risk domain, Exp (B) = 1.08, p < .01, and age, Exp (B) = .91, p < .05, significantly predicted time to reoffense. For the third model, an interaction term was created to examine whether the protective domain moderated the statistically significant relationship between the individual risk domain and time to reoffense. Model 3, F(8) = 4.05, p < .001, added the product term of the individual risk domain and the protective domain. This interaction was not statistically significant, Exp (B) = .99, p = .95, and did not add to the prediction of reoffending. These results indicate that the protective domain as a whole did not moderate the impact of the individual risk domain.

Table 4.

Results of Cox Regression Analyses With the SAVRY Protective Domain Predicting General Reoffending.

	Model 1		Model 2		Model 3
	Exp (B)	SE	Exp (B)	SE	Exp (B)	SE
SAVRY domains
Historical risk domain	1.01	.03	1.01	.03	1.01	.03
Social/contextual risk domain	1.01	.04	1.01	.05	1.01	.05
Individual risk domain	1.09**	.03	1.08*	.03	1.08**	.03
Protective domain	—	—	0.99	.01	0.99	.01
Protective Domain × Individual Risk Domain	—	—	—	—	0.99	.01
Control variables
Age	0.91*	.04	0.91*	.04	0.91*	.04
Race	0.97	.18	0.96	.18	0.96	.18
Offense level	1.33	.21	1.33	.21	1.33	.21

Note. N = 460. SAVRY = Structured Assessment of Violence Risk in Youth.

*p < .05. **p < .01.

Table 5 presents the results of the Cox regression analyses examining the effects of the 2 protective items found to be significant in the bivariate analyses: positive attitude toward intervention and authority and resilient personality traits. Among the full sample, the inclusion of positive attitude toward intervention and authority did not add incremental validity, F(7) = 5.27, p < .001, nor was the item found to be statistically significant. Interestingly, when the full sample was broken down into youth who did and did not possess a positive attitude, the effects of the SAVRY risk domains were different across the groups. Among youth who did not possess a positive attitude, F(6) = 3.66, p < .01, only the social/contextual domain was found to be a significant predictor of reoffending, Exp (B) = 1.16, p < .05. Thus, with each unit increase in the social/contextual domain, the hazard of reoffending was 16% higher. In contrast, for those youth presenting with a positive attitude, F(6) = 3.56, p < .01, the social/contextual domain was not a statistically significant predictor of time to reoffense, Exp (B) = .90, p > .09, evidencing a buffering effect. However, for those youth presenting with a positive attitude, the historical and individual risk domains were significant predictors of general reoffending. With each unit increase in the SAVRY historical risk domain, the hazard of reoffending was 9% higher for these particular youth. Additionally, with each unit increase in the individual risk domain, the hazard of reoffending was 10% higher. However, the direction of the effects for those presenting with a positive attitude were in the opposite direction of expectations. For youth with a positive attitude, higher scores on the historical and individual risk domains increased the likelihood of recidivism.

Table 5.

Cox Regression Results for Youth Presenting With Protective Items Versus Those Without Protective Items.

					Positive Attitude
	Full Sample (n = 460)				Absent (n = 136)		Present (n = 324)
	Exp (B)	SE	Exp (B)	SE	Exp (B)	SE	Exp (B)	SE
SAVRY risk domains
Historical risk domain	1.01	0.03	1.01	0.03	0.92	0.04	1.09*	0.04
Social/contextual risk domain	1.01	0.04	1.01	0.04	1.16*	0.08	0.90	0.06
Individual risk domain	1.09**	0.03	1.09**	0.03	1.08	0.04	1.10*	0.04
Protective item—positive attitude	—	—	0.94	0.17	—	—	—	—
Control variables
Age	0.91*	0.04	0.91*	0.04	0.87	0.07	0.94	0.06
Race	0.97	0.18	0.97	0.18	0.66	0.23	1.05	0.24
Offense level	1.33	0.21	1.36	0.16	1.66	0.48	1.11	0.22
					Resilient Personality
	Full Sample (n = 460)				Absent (n = 151)		Present (n = 309)
	Exp (B)	SE	Exp (B)	SE	Exp (B)	SE	Exp (B)	SE
SAVRY risk domains
Historical risk domain	1.01	.03	1.01	.03	1.01	.04	1.01	.04
Social/contextual risk domain	1.01	.04	1.01	.04	1.02	.06	1.00	.07
Individual risk domain	1.09**	.03	1.10**	.03	1.11*	.05	1.09	.05
Protective item—resilient personality	—	—	1.10	.20	—	—	—	—
Control variables
Age	0.91*	.04	0.90*	.04	0.92	.07	0.89	.05
Race	0.97	.18	0.97	.18	0.80	.23	1.09	.27
Offense level	1.33	.21	1.33	.21	1.21	.32	1.40	.29

Note. SAVRY = Structured Assessment of Violence Risk in Youth.

*p < .05. **p < .01.

The same procedures were used to examine whether resilient personality traits add incremental validity or moderate the effects of the risk domains on general reoffending. Results from the model based on the full sample suggested that the inclusion of this item did not add incremental validity, F(7) = 5.29, p < .001, and the item was not statistically significant. Among the youth not presenting with resilient personality traits, F(6) = 2.81, p < .05, the individual risk domain remained a statistically significant predictor of time to reoffense, Exp (B) = 1.11, p < .05. However, the subsequent model, F(6) = 2.60, p < .05, using the youth with resilient personality traits provides evidence of a buffering effect, as the individual risk domain was not a statistically significant predictor of time to reoffense, Exp (B) = 1.09, p > .06.

Discussion

This study sought to explore the utility of the SAVRY protective factors in predicting reoffending among a sample of adjudicated offenders. Our initial bivariate analyses suggested that youth who reoffended scored significantly lower on the protective domain (i.e., had a lower number of protective factors present). The two specific protective factors that were found to be related to reoffending were having a positive attitude toward intervention and authority and the presence of resilient personality traits. For both items, a significantly larger proportion of nonreoffenders possessed these traits. However, once entered into the multivariate regression models including the SAVRY risk domains, all three measures failed to reach statistical significance and did not increase the predictive validity of the SAVRY. Instead, the individual risk domain was found to be the strongest, and most robust, predictor of reoffending. In sum, our findings suggest that, when accounting for the SAVRY risk domains, the SAVRY protective factors do not directly predict reoffending or add incremental validity.

These results align with the findings from Vincent, Chapman, and Cook (2011), Penney et al. (2010), and Hilterman, Nicholls, and van Nieuwenhuizen (2014), which all failed to find support for the predictive validity of the protective factors when accounting for SAVRY risk factors. However, our results are inconsistent with the findings from Rennie and Dolan (2010) and Lodewijks et al. (2010), which did support the utility of these factors. One important difference among studies is that Rennie and Dolan (2010) did not control for the effects of the SAVRY risk domains on reoffending in their multivariate models, and Lodewijks et al. (2010) combined the social/contextual and individual risk domains into one overall dynamic risk scale. Thus, when the four individual SAVRY domains are included in multivariate models, low incremental validity of the protective factors has been consistently identified (Penney, Lee, & Moretti, 2010; Vincent et al., 2011). Furthermore, taken together, these studies highlight the importance of the SAVRY dynamic risk domains in predicting reoffending.

The current study also sought to expand on previous findings by using multivariate analyses to better distinguish where and how the buffering effects of the protective factors on the risk domains work. Results did not provide support for a significant moderating effect of the protective domain. However, examinations of the individual protective factors positive attitude toward intervention and authority and resilient personality traits did reveal some buffering effects. The presence of resilient personality traits buffered the effect of the individual risk domain, while the presence of positive attitude buffered the effect of the social/contextual domain. These findings add to the limited research on the buffering effects the SAVRY protective items have on the risk domains.

However, the analyses involving positive attitude provided an additional unexpected result: For youth with higher scores in the historical and individual risk domains, the presence of this protective factor increased the likelihood of reoffending. One possible explanation for this result is that those youth with a higher score on these domains may have more experience with juvenile justice interventions and may better understand the utility of a positive demeanor when being interviewed by JPOs. Previous research has shown that delinquents’ attitudes and demeanor influence practitioners’ decision-making (Bridges & Steen, 1998; Emerson, 1969). Thus, experienced juvenile offenders may have an understanding of how to manipulate their situation to receive better outcomes, and characteristics such as a positive attitude and a respectful demeanor could be essential to persuading JPOs to make positive dispositional recommendations.

Another possible explanation for the unexpected findings is the problematic nature of the item positive attitude toward intervention and authority. Of the six protective factors included in the SAVRY, positive attitude toward intervention and authority presents the most challenges for determining a score. The item allows for more subjectivity on behalf of the rater. For example, how does a JPO determine whether a first-time offender has a positive attitude toward intervention if the youth has never experienced interventions? Would the fact that a youthful offender is polite during probation meetings suffice as a positive attitude toward authority for some JPOs? The other protective factors are clearer and allow for less subjectivity. Retrieving information on items such as prosocial involvement or strong commitment to school should be straightforward. Interestingly, the SAVRY manual is limited in its explanation surrounding the inclusion of positive attitude stating, “Positive attitude toward remediation attempts describe active involvement (i.e., planning, treatment, compliance, openness) by the youth in lessening risk for violence” (Borum et al., 2003, p. 99). Adding to the confusion, the SAVRY manual references a study by Hoge, Andrews, and Leschied (1996) which “found that a positive response to authority was related to lower levels of reoffending during late adolescence” (Borum et al., 2003, p. 99). However, the results of this study indicate that the effect of the variable “positive response to authority” on reoffending was direct and not the buffering effect proposed by the SAVRY creators. Thus, future research should examine the factors JPOs use to determine the presence or absence of positive attitude toward intervention and authority. More clarification is needed regarding the scope of this item and its effects on both the SAVRY risk factors and reoffending.

Although our findings do not provide empirical support for the incremental validity of the SAVRY protective factors for predicting risk for reoffending, they do provide limited support for certain buffering effects. Hence, these factors should be integrated into case management and intervention planning. A growing body of research suggests that intervention plans that consider youth strengths, in combination with dynamic needs, are more likely to produce positive outcomes including treatment completion, desistance from crime, and improved mental health and substance use (Siegal, Li, & Rapp, 2002; Tse et al., 2016). Strength-based approaches take a positive approach to offender rehabilitation by focusing on personal and interpersonal competencies such as personality strengths, capacity for growth, and social support system, which may offset youth needs (Fortune, 2018; Rapp & Sullivan, 2014). Thus, regardless of the ability of protective factors to predict reoffending, their usefulness may become clear when directed toward using them to identify and carry out interventions meant to help youth desist (Serin et al., 2016).

Finally, regardless of the way the protective items were measured (e.g., items, domain) and modeled (e.g., direct effects, moderation), the individual risk domain was found to be the strongest and most consistent predictor of reoffending. The importance of the dynamic, individual-level risk factors has been confirmed in previous studies using the SAVRY (Penney et al., 2010) and other RAIs (Baglivio, Wolff, Piquero, Howell, & Greenwald, 2017; McGrath & Thompson, 2012; Peterson-Badali, Skilling, & Haqanee, 2015). This finding provides further support for the inclusion of dynamic risk factors, or needs factors, in RAIs. A frequent debate in the risk assessment literature is whether risk factors and needs should be combined into one assessment instrument (see Baird, 2009; Schwalbe, 2007). For example, Gottfredson and Moriarty (2006) argue for the distinction between risk assessments (i.e., static factors only) and needs assessments (i.e., dynamic needs only). Our findings align with Vincent et al. (2011) who argue that the inclusion of dynamic risk factors is critical for accurately measuring the likelihood to reoffend, measuring changes over time, and assisting decision makers in identifying areas for case management and intervention planning. Indeed, among the sample included in our study, the failure to measure individual-level, dynamic risk factors would have decreased the predictive accuracy of the SAVRY.

Limitations

There are a number of limitations to the current study. First, the sample is all male and mostly non-White. This limits the generalizability of our findings to more diverse samples of adjudicated offenders as well as justice-involved girls. Second, in the multivariate analyses, it was necessary to eliminate the probation site as a control variable due to a low number of respondents from one of the sites. This led to unreliable estimates for other important variables, such as race, due to low cell counts. Therefore, we were unable to account for any differences in risk assessment or decision-making practices across the three sites. Third, our measures of reoffending were based on official records and, therefore, do not include behaviors that went unrecognized by the juvenile justice system. A wealth of empirical research suggests that certain legal and extralegal factors influence the likelihood of arrests (Bishop, Leiber, & Johnson, 2010; Peck & Jennings, 2016). This has the potential to obscure the true relationship between the protective factors and the engagement in delinquent behavior. Additionally, the current study is unable to account for other protective characteristics shown to impact youth behavior, for example, high IQ (Ttofi et al., 2016) or having an older mother (Jolliffe, Farrington, Loeber, & Pardini, 2016).

Finally, due to missing data on a large number of cases, we were unable to incorporate the SAVRY summary risk rating into our analyses. Exploring the relationship between the protective factors and the summary risk rating may also help to determine the ways in which the protective factors on the SAVRY are used in practice. The summary risk rating is the probation officers’ overall assessment of the offender’s risk level based on each individual item. It is possible that practitioners perceive the presence of protective factors as an indication of lower risk to reoffend (i.e., a perceived buffering effect) and ultimately rate the offenders with protective factors present as lower risk. This could result in less severe or restrictive sanctions which carries a range of potential implications for long-term youth and system outcomes. Thus, a potential avenue for future research on the SAVRY is assessing whether specific SAVRY items or SAVRY domains have a greater influence on raters’ structured professional judgment (i.e., SRR) and whether this is consistent across raters, offenders, and risk level.

Future research should also seek to assess differences in the incremental validity, buffering effect, and utility of the protective factors for intervention planning across different risk levels. It is quite possible that protective factors are relevant at varying levels of risk. For example, the protective factors may be critical in predicting the likelihood of reoffending among high-risk youth where the severity of these risk items is larger and there is more “room” for a buffering effect. This may also be the case for the usefulness of intervention planning—given that most intensive interventions are reserved for higher risk youth. Low risk youth, on the other hand, are often referred to more general, prevention strategies that seek to build skills and develop youth strengths (Howell, Lipsey, & Wilson, 2014).

Most importantly, a critical avenue for future research on the utility of protective, or strength-based, items on RAIs is to assess how often needs and strengths are met in practice and the value that is added by considering protective traits. Specifically, it would be important to examine how the RNR principles work together, in practice, to produce positive outcomes. Prior research has found that a mismatch between treatment recommendations and dynamic risk factors (as assessed by the YLS/CMI) increases the likelihood of recidivism (Peterson-Badali et al., 2015) and that roughly half of treatment recommendations based on risk assessment results were actually matched with services received (Vitopoulos, Peterson-Badali, & Skilling, 2012). Additional research is needed to understand how protective factors, or youth strengths, play a role in treatment recommendations and matching of services.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Melanie F. P. Soderstrom

References

Aalen

O. O.

Borgan

Gjessing

H. K.

(2008). Survival and event history analysis: A process point of view. New York, NY: Springer Science+Business Media, LLC.

Allison

P. D.

(2014). Event history and survival analysis. Thousand Oaks, CA: Sage.

American Correctional Association. (2008). 2008 directory: Adult and juvenile correctional departments, institutions, agencies, and probation and parole authorities. Alexandria, VA: Author.

Andershed

Gibson

C. L.

Andershed

(2016). The role of cumulative risk and protection for violent offending. Journal of Criminal Justice, 45, 78–84.

Baglivio

M. T.

Wolff

K. T.

Piquero

A. R.

Howell

J. C.

Greenwald

M. A.

(2017). Risk assessment trajectories of youth during juvenile justice residential placement: Examining risk, promotive, and “buffer” scores. Criminal Justice and Behavior, 44, 360–394.

Baird

. (2009). A question of evidence: A critique of risk assessment models used in the justice system. Retrieved from National Council on Crime and Delinquency website http://www.nccdglobal.org/publications/a-question-of-evidence-a-critique-of-risk-assessment-models-used-in-the-justice-system

Baird

Healy

Johnson

Bogie

Dankert

E. W.

Scharenbroch

. (2013). A comparison of risk assessment instruments in juvenile justice. Oakland, CA: National Council on Crime and Delinquency. Retrieved from https://www.ncjrs.gov/pdffiles1/ojjdp/grants/244477.pdf

Bishop

D. M.

Leiber

Johnson

(2010). Contexts of decision making in the juvenile justice system: An organizational approach to understanding minority overrepresentation. Youth Violence and Juvenile Justice, 8, 213–233.

Borum

Bartel

Forth

(2003). Manual for the Structured Assessment of Violence Risk in Youth Version 1.1. Tampa: Florida Mental Health Institute, University of South Florida.

10.

Bridges

G. S.

Steen

(1998). Racial disparities in official assessments of juvenile offenders: Attributional stereotypes as mediating mechanisms. American Sociological Review, 63, 554–570.

11.

Chapman

J. F.

Desai

R. A.

Falzer

P. R.

Borum

(2006). Violence risk and race in a sample of youth in juvenile detention: The potential to reduce disproportionate minority confinement. Youth Violence and Juvenile Justice, 4, 170–184.

12.

Childs

K. K.

Frick

P. J.

Gottlieb

(2016). Sex differences in the measurement invariance and factors that influence structured judgments of risk using the Structured Assessment of Violence in Youth (SAVRY). Youth Violence and Juvenile Justice, 14, 76–92.

13.

Childs

K. K.

Ryals

Jr Frick

P. J.

Lawing

Phillippi

Deprato

D. K.

(2013). Examining the validity of the Structured Assessment of Violence Risk in Youth (SAVRY) for predicting probation outcomes among adjudicated juvenile offenders. Behavioral Sciences and the Law, 31, 256–270.

14.

Cox

D. R.

(1972). Regression models and life tables. Journal of the Royal Statistical Society, Series, B34, 187–202.

15.

Dolan

M. C.

Rennie

C. E.

(2008). The Structured Assessment of Violence Risk in Youth as a predictor of recidivism in a United Kingdom cohort of adolescent offenders with conduct disorder. Psychological Assessment, 20, 35–46.

16.

Douglas

K. S.

Yeomans

Boer

D. P.

(2005). Comparative validity analysis of multiple measures of violence risk in a sample of criminal offenders. Criminal Justice and Behavior, 32, 479–510.

17.

Emerson

R. M.

(1969). Judging delinquents: Context and process in juvenile court. Chicago, IL: Aldine.

18.

Fortune

C. A.

(2018). The good lives model: A strength-based approach for youth offenders. Aggression and Violent Behavior, 38, 21–30.

19.

Furdella

Puzzanchera

(2015). Delinquency cases in juvenile court, 2013. Washington, DC: Office of Juvenile Justice and Delinquency Prevention.

20.

Gammelgård

Koivisto

Eronen

Kaltiala-Heino

(2015). Predictive validity of the structured assessment on violence risk in youth: A 4-year follow-up. Criminal Behaviour and Mental Health, 25, 192–206.

21.

Gottfredson

S. D.

Moriarty

L. J.

(2006). Statistical risk assessment: Old problems and new applications. Crime & Delinquency, 52, 178–200.

22.

Greenland

Mansournia

M. A.

Altman

D. G.

(2016). Sparse data bias: A problem hiding in plain sight. BMJ, 352, i1981.

23.

Hawkins

Catalano

Miller

(1992). Risk and protective factors for alcohol and other drug problems in adolescence and early adulthood: Implications for substance abuse prevention. Psychological Bulletin, 112, 64–105.

24.

Hawkins

J. D.

Herrenkohl

T. I.

Farrington

D. P.

Brewer

Catalano

R. F.

Harachi

T. W.

(1998). A review of predictors of youth violence. In Loeber

Farrington

(Eds.), Serious and violent juvenile offenders: Risk factors and successful interventions (pp. 106–146). Thousand Oaks, CA: Sage.

25.

Hawkins

J. D.

Herrenkohl

T. I.

Farrington

D. P.

Brewer

Catalano

R. F.

Harachi

T. W.

(2000). Predictors of youth violence. Juvenile justice bulleting. Washington, DC: U.S. Department of Justice, Office of Justice Programs, Office of Juvenile Justice and Delinquency Prevention.

26.

Herrenkohl

Maguin

Hill

Hawkins

Abbott

Catalano

(2000). Developmental risk factors for youth violence. Journal of Adolescent Health, 26, 176–186.

27.

Hilterman

E. L. B.

Nicholls

T. L.

van Nieuwenhuizen

(2014). Predictive validity of risk assessments in juvenile offenders: Comparing the SAVRY, PCL: YV, and YLS/CMI with unstructured clinical assessments. Assessment, 21, 324–339.

28.

Hoge

R. D.

Andrews

D. A.

Leschied

A. W.

(1996). An investigation of risk and protective factors in a sample of youthful offenders. Journal of Child Psychology and Psychiatry, 37, 419–424.

29.

Howell

J. C.

Lipsey

M. W.

Wilson

J. L.

(2014). A handbook for evidence based juvenile justice systems. London, England: Lexington Books.

30.

Jaccard

Turrisi

(2003). Interaction effects in multiple regression. Thousand Oaks, CA: Sage.

31.

Jennings

W. G.

Gonzalez

J. R.

Piquero

A. R.

Bird

Canino

Maldonado-Molina

(2016). The nature and relevance of risk and protective factors for violence among Hispanic children and adolescents: Results from the Boricua Youth Study. Journal of Criminal Justice, 45, 41–47.

32.

Jolliffe

Farrington

D. P.

Loeber

Pardini

(2016). Protective factors for violence: Results from the Pittsburgh Youth Study. Journal of Criminal Justice, 45, 32–40.

33.

Klein

Rettenberger

Yoon

Köhler

Briken

(2015). Protective factors and recidivism in accused juveniles who sexually offended. Sexual Abuse: A Journal of Research and Treatment, 27, 71–90.

34.

Lodewijks

H. P. B.

de Ruiter

Doreleijers

T. A. H.

(2010). The impact of protective factors in desistance from violent reoffending: A study in three samples of adolescent offenders. Journal of Interpersonal Violence, 25, 568–587.

35.

Lodewijks

H. P. B.

Doreleijers

T. A. H.

de Ruiter

(2008). SAVRY risk assessment in violent Dutch adolescents: Relation to sentencing and recidivism. Criminal Justice and Behavior, 35, 696–709.

36.

Lodewijks

H. P. B.

Doreleijers

T. A. H.

de Ruiter

Borum

(2008). Predictive validity of the Structured Assessment of Violence Risk in Youth (SAVRY) during residential treatment. International Journal of Law and Psychiatry, 31, 263–271.

37.

McGrath

Thompson

A. P.

(2012). The relative predictive validity of the static and dynamic domain scores in risk-need assessment of juvenile offenders. Criminal Justice and Behavior, 39, 250–263.

38.

Mulvey

E. P.

(2005). Risk assessment in juvenile justice policy and practice. In Heilbrun

Sevin Goldstein

Redding

(Eds.), Juvenile delinquency: Prevention, assessment, and intervention (pp. 209–231). New York, NY: Oxford University Press.

39.

Olver

M. E.

Stockdale

K. C.

Wormith

J. S.

(2009). Risk assessment with young offenders: A meta-analysis of three assessment measures. Criminal Justice and Behavior, 36, 329–353.

40.

Peck

J. H.

Jennings

W. G.

(2016). A critical examination of “being Black” in the juvenile justice system. Law and Human Behavior, 40, 219.

41.

Penney

S. R.

Lee

Moretti

M. M.

(2010). Gender differences in risk factors for violence: An examination of the predictive validity of the Structured Assessment of Violence Risk in Youth. Aggressive Behavior, 36, 390–404.

42.

Peterson-Badali

Skilling

Haqanee

(2015). Examining implementation of risk assessment in case management for youth in the justice system. Criminal Justice and Behavior, 42, 304–320.

43.

Rapp

C. A.

Sullivan

W. P.

(2014). The strengths model: Birth to toddlerhood. Advances in Social Work, 15, 129–142.

44.

Rennie

C. E.

Dolan

M. C.

(2010). The significance of protective factors in the assessment of risk. Criminal Behaviour and Mental Health, 20, 8–22.

45.

Schwalbe

C. S.

(2007). Risk assessment for juvenile justice: A meta-analysis. Law and Human Behavior, 31, 449.

46.

Serin

R. C.

Chadwick

Lloyd

C. D.

(2016). Dynamic risk and protective factors. Psychology, Crime & Law, 22, 151–170.

47.

Shepherd

S. M.

Luebbers

Ogloff

J. R.

(2016). The role of protective factors and the relationship with recidivism for high-risk young people in detention. Criminal Justice and Behavior, 43, 863–878.

48.

Siegal

H. A.

Rapp

R. C.

(2002). Case management as a therapeutic enhancement: Impact on post-treatment criminality. Journal of Addictive Diseases, 21, 37–46.

49.

Simons

Paternite

Shore

(2001). Quality of parent/adolescent attachment and aggression in young adolescents. Journal of Early Adolescence, 21, 182–203.

50.

Singh

J. P.

Grann

Fazel

(2011). A comparative study of violence risk assessment tools: A systematic review and metaregression analysis of 68 studies involving 25,980 participants. Clinical Psychology Review, 31, 499–513.

51.

StataCorp. (2017). Stata: Release 15. Statistical Software. College Station, TX: Author.

52.

Tse

Tsoi

E. W.

Hamilton

O’Hagan

Shepherd

Slade

Whitley

Petrakis

(2016). Uses of strength-based interventions for people with serious mental illness: A critical review. International Journal of Social Psychiatry, 62, 281–291.

53.

Ttofi

M. M.

Farrington

D. P.

Piquero

A. R.

Lösel

DeLisi

Murray

(2016). Intelligence as a protective factor against offending: A meta-analytic review of prospective longitudinal studies. Journal of Criminal Justice, 45, 4–18.

54.

Urquhart

T. A.

Viljoen

J. L.

(2014). The use of the SAVRY and YLS/CMI in adolescent court proceedings: A case law review. International Journal of Forensic Mental Health, 13, 47–61.

55.

Viljoen

J. L.

Scalora

Cuadra

Bader

Chavez

Ullman

Lawrence

(2008). Assessing risk for violence in adolescents who have sexually offended: A comparison of the J-SOAP-II, J-SORRAT-II, and SAVRY. Criminal Justice and Behavior, 35, 5–23.

56.

Vincent

G. M.

Chapman

Cook

N. E.

(2011). Risk-needs assessment in juvenile justice: Predictive validity of the SAVRY, racial differences, and the contribution of needs factors. Criminal Justice and Behavior, 38, 42–62.

57.

Vincent

G. M.

Guy

L. S.

Fusco

S. L.

Gershenson

B. G.

(2012). Field reliability of the SAVRY with juvenile probation officers: Implications for training. Law and Human Behavior, 36, 225–236.

58.

Vincent

G. M.

Guy

L.S.

Gershenson

B. G.

McCabe

(2012). Does risk assessment make a difference? Results of implementing the SAVRY in juvenile probation. Behavioral Sciences and the Law, 30, 384–405.

59.

Vitopoulos

N. A.

Peterson-Badali

Skilling

T. A.

(2012). The relationship between matching service to criminogenic need in male and female youth: Examining the RNR principles in practice. Criminal Justice and Behavior, 39, 1025–1041.

60.

Wachter

(2014). Statewide risk assessment in juvenile probation. JJGPS StateScan. Pittsburgh, PA: National Center for Juvenile Justice.

61.

Welsh

J. L.

Schmidt

McKinnon

Chattha

H. K.

Meyers

J. R.

(2008). A comparative study of adolescent risk assessment instruments: Predictive and incremental validity. Assessment, 15, 104–115.

62.

Werner

(2000). Protective factors and individual resilience. In Shonkoff

Meisels

(Eds.), Handbook of early childhood intervention. Cambridge, MA: Cambridge University Press.