The Relationship Between Changes in Dynamic Risk Factors and the Predictive Validity of Risk Assessments Among Youth Offenders

Abstract

The Youth Level of Service/Case Management Inventory (YLS/CMI) is a widely used risk assessment tool for youth offenders. It is intended to be administered regularly to capture changes in criminogenic needs and thus inform case management during a youth’s sentence. However, there is a dearth of research examining whether updated assessments are more predictive of recidivism than initial assessments. We examined whether including dynamic risk factors increased the predictive validity of the YLS/CMI and whether changes in dynamic risk scores improved the prediction of recidivism. Two hundred youth offenders were tracked from their first risk assessment conducted at probation to their most recent risk assessment completed prior to first reoffense or study end date. Inclusion of dynamic risk scores improved predictive accuracy above static risk and updated dynamic risk scores improved accuracy over those obtained from the initial assessment, supporting the utility of the YLS/CMI as a reassessment tool.

Keywords

dynamic risk factors Risk-Need-Responsivity YLS/CMI changes in risk recidivism predictive validity youth offenders probation

Introduction

Given the detrimental social and economic costs of youth crime, its prevention and reduction has been a priority for many jurisdictions in North America, the United Kingdom, Europe, Australia, and New Zealand. Based on a general personality and cognitive social learning theory of criminal conduct (Andrews & Bonta, 2010a), the Risk-Need-Responsivity (RNR) framework is considered one of the most prominent models used in the assessment and treatment of youth offenders. It forms the basis for correctional and community (e.g., probation) youth justice practice in many jurisdictions throughout the world. This theoretical framework encompasses a systematic empirical examination of risk factors associated with criminal behavior and focuses on individual differences within a corrections model while simultaneously incorporating community factors (Andrews & Bonta, 2010a). Both individual studies (Luong & Wormith, 2011; Vieira, Skilling, & Peterson-Badali, 2009) and meta-analytic research (Andrews et al., 1990; Lipsey, 2009) have consistently shown reduced recidivism rates for community and correctional interventions and programs that adhere to the RNR model, with reductions in offender recidivism by up to 35% (Andrews & Bonta, 2010a).

According to the risk principle, the intensity of supervision and services provided in the context of sentencing should be matched to an offender’s identified level of risk to reoffend. That is, more intensive supervision, restrictive placements, and treatment services should be reserved for high-risk youth, and fewer resources should be allocated to low-risk offenders who require less monitoring and case management (Andrews & Bonta, 2010a, 2010b; Andrews & Dowden, 2006; Andrews et al., 1990; Loeber & Ahonen, 2014). Studies have shown that higher risk offenders benefit the most when treatment is of adequate intensity (Bourgon & Armstrong, 2005; Kroner & Takahashi, 2012) and that treatment services provided to low-risk offenders may in fact increase the likelihood of poor outcomes (Andrews & Dowden, 2006; Lowenkamp & Latessa, 2004; Lowenkamp, Latessa, & Holsinger, 2006). In the youth literature, there is strong evidence to support that high-risk youth benefit the most from targeted intervention (Lipsey, 2009; Luong & Wormith, 2011) when compared with low-risk youth (skilling, Vieira et al., 2009). Adhering to this principle requires assessment of factors that strongly and directly predict recidivism, which in the RNR framework are termed “criminogenic needs.” In addition to criminal history, seven criminogenic need domains have been identified as strong predictors of reoffending, including Procriminal Attitudes (thoughts, values, and sentiments supportive of criminal behavior), Antisocial Personality (low self-control, hostility, adventurous pleasure seeking, disregard for others, callousness), Procriminal Associates, Social Achievement (education, employment), Family/Marital (marital instability, poor parenting skills, criminality), Substance Abuse, and Leisure/Recreation (lack of prosocial pursuits; Andrews & Bonta, 2010a).

The need principle describes “what should be treated” and points to the critical importance of focusing intervention on criminogenic needs rather than factors empirically shown to be noncriminogenic (i.e., variables with weak or indirect relationships to reoffending; Andrews & Bonta, 2010b). Furthermore, in contrast to risk classification (and the risk principle), which incorporates both static and dynamic risk factors (Andrews & Bonta, 2010b), the focus of the need principle is on the latter group of variables. While there are a number of static risk factors that are strong, direct predictors of criminal behavior (e.g., age, gender, history of criminal offending), such variables cannot be modified with treatment and are therefore not appropriate targets of intervention. Dynamic risk factors comprise variables that are directly related to offending and are amenable to change through intervention. Thus, according to the need principle, intervention should focus on dynamic risk factors if the goal is to decrease a youth’s risk to reoffend (Andrews & Bonta, 2010a, 2010b). Several studies of justice-involved youth have supported the adherence to the need principle, finding that addressing youths’ individually assessed criminogenic needs with appropriate intervention is associated with reduced reoffending (Peterson-Badali, Skilling, & Haqanee, 2015; Vieira et al., 2009; Vitopoulos, Peterson-Badali, & Skilling, 2012). As such, psychometrically sound risk assessment tools that take dynamic risk factors into account are required to identify and provide effective case management to address these needs and, consequently, reduce recidivism (Andrews & Bonta, 2010a).

Risk Assessment of Youth Offenders

It is widely accepted that risk estimates formed from structured risk assessment tools are more accurate than unstructured professional judgment in predicting recidivism (Andrews, Bonta, & Wormith, 2006; Oleson, VanBenschoten, Robinson, Lowenkamp, & Holsinger, 2012; Perrault, Paiva-Salisbury, & Vincent, 2012), and there is support for the predictive validity of youth risk assessment instruments, including various youth versions of the Level of Service tools (e.g., McGrath & Thompson, 2012; Olver, Stockdale, & Wong, 2012; Schmidt, Hoge, & Gomes, 2005; Schwalbe, 2007; Takahashi, Mori, & Kroner, 2013). However, there has been debate regarding the utility of including dynamic risk factors in risk prediction instruments. In the past, some scholars contended that dynamic risk factors are unnecessary in risk assessment because they add little predictive validity over and above static risk factors (Harris & Rice, 2003; Quinsey, 2009). For example, in their meta-analysis, Cottle, Lee, and Heilbrun (2001) found that the strongest predictors of youth recidivism were static factors, including criminal history and age of first contact with the justice system. Recent research on justice-involved youth indicates that combining both static and dynamic risk factors does in fact improve risk prediction (McGrath & Thompson, 2012; Peterson-Badali et al., 2015; Vincent, Perrault, Guy, & Gershenson, 2012). Equally important, current risk assessment models and tools are designed not only to classify offenders but also to inform treatment to lower risk of future offending. The inclusion of dynamic risk factors is critical to this “case management” aspect of risk assessment frameworks and is a critical feature of the most current “4^th generation” risk assessment tools (Andrews & Bonta, 2010a, 2010b).

The utility of risk assessments is premised on the assumption that dynamic risk factors do indeed change over time; reassessments are therefore seen as essential to update risk status for supervision and better inform ongoing treatment planning (Olver, Beggs Christofferson, & Wong, 2015). Andrews et al. (2006) argued that the reassessment of dynamic risk factors has contributed significantly to the predictive validity of risk assessment tools, and estimated that reassessments double or triple the outcome variance explained by initial assessments. However, most studies to date have relied solely on single-wave, cross-sectional research designs that essentially treat dynamic risk factors as static (Brown, St Amand, & Zamble, 2009; Douglas & Skeem, 2005; Serin, Chadwick, & Lloyd, 2016; Vincent et al., 2012). Indeed, Quinsey, Jones, Book, and Barr (2006) referred to dynamic risk factors as “temporally fixed dynamic variables” (p. 1540) in research because these factors function as static predictors that cannot change during the follow-up period and are used to estimate the likelihood of recidivism as an outcome variable. There is a dearth of research that examines change in these dynamic risk factors over time, and whether changes in risk (or updated risk assessments) are more predictive of future offending than initial risk assessments.

To address these theoretical, methodological, and statistical limitations, Brown et al. (2009) designed a three-wave, prospective study to examine whether the reassessment of dynamic risk factors would improve the prediction of recidivism in a sample of 136 adult male incarcerated offenders over a 10-month period. Static factors were measured prior to release from federal custody facilities in Canada while dynamic risk factors were assessed at three different time points (prerelease, 1 month postrelease, and 3 months postrelease). Cox regression survival analysis with time-dependent covariates and receiver operator characteristic (ROC) analyses were used to assess the predictive validity of static and dynamic risk factors at different time points. Results indicated that (a) the best prediction models included both static and time-dependent dynamic risk factors and (b) the reassessment of specific dynamic risk factors increased the predictive validity of risk assessment. Similar results were found in a more recent study of women adult offenders (Greiner, Law, & Brown, 2014). Examining whether this holds true for risk assessments conducted at youth probation is an important research question that has not been answered in a longitudinal research study.

The Present Study

Based on the principles of the RNR model, the Youth Level of Service/Case Management Inventory (YLS/CMI; Hoge & Andrews, 2002) is one of the most widely used fourth-generation risk assessment tools, demonstrating strong psychometric properties among youth offenders (Welsh, Schmidt, McKinnon, Chattha, & Meyers, 2008). Viljoen, Elkovitch, Scalora, and Ullman (2009) found that the YLS/CMI demonstrated the greatest incremental validity for accurately predicting violent and general recidivism among adolescents who had committed a sexual offense. Aside from predicting risk to reoffend, the YLS/CMI is also useful in formulating effective treatment recommendations and implementing appropriate services for youth (Maurutto & Hannah-Moffat, 2007). Despite the fact that the YLS/CMI is presented and utilized as a risk assessment tool that can detect changes in an offender’s risk for recidivism, there is little empirical evidence to support the usage of this instrument as a reassessment tool. Furthermore, few studies that examine dynamic risk factors have collected data from applied settings compared with research environments, which calls into question the field reliability and validity of these measures in correctional settings (Cording, Beggs Christofferson, & Grace, 2016).

The primary goal of the current study was to empirically examine changes in dynamic risk factors at the individual level, explore youth-specific predictors related to recidivism, and determine whether changes in dynamic risk improved the predictive validity of risk assessment tools among youth offenders in a real-world setting. It was hypothesized that both static (i.e., criminal history) and dynamic scores of the YLS/CMI at the time of the initial assessment would be related to recidivism. It was expected that the inclusion of dynamic risk scores would improve the predictive accuracy of recidivism over and above static factors. It was also hypothesized that the reassessment of both criminal history and dynamic risk factors would contribute more strongly to the prediction of recidivism than scores obtained from the initial assessment.

Method

Participants

The sample consisted of 200 youth (172 males, 28 females) in Toronto, Canada, who received a court-ordered assessment just prior to sentencing through a child and adolescent assessment program at a local mental health agency between July 2001 and February 2008. Youth ranged in age from 12 to 18 years old (M = 15.12, SD = 1.56) at the time of their first assessment. The sample was racially diverse (38% Caucasian, 36% Black, 7% Asian, 2% Hispanic, 3% South Asian, 2% Aboriginal, and 12% Other). Research consent and demographic information were obtained from participants. To be included in the study, each youth must have had at least one YLS/CMI assessment to measure criminal history and dynamic risk factors, as well as reoffense data available from the Royal Canadian Mounted Police (RCMP) to measure recidivism during the follow-up period.¹ There did not appear to be any selection bias between youth included in the study and those who did not meet eligibility criteria based on gender, race, age at initial assessment, or YLS/CMI total score. All youth were on probation and assumedly meeting periodically with their probation officers and receiving case management. Involvement at probation and the types of interventions received varied across youth in accordance with their identified criminogenic needs, treatment goals, and service availability.

Measures and Procedure

YLS/CMI

Developed by Hoge and Andrews (2002), the YLS/CMI is a 42-item standardized instrument used by probation officers and mental health professionals to measure risk level, dynamic risk factors, and variables that may affect responsiveness to treatment in youth aged 12 to 18. When the YLS/CMI is given as a semistructured interview, it generally takes 60 to 90 min to complete and provides individual scores (which range from 3 to 9) and risk categories (i.e., low, moderate, and high) in eight domains—one static (young person’s prior and current offences/dispositions) and seven dynamic (family circumstances/parenting, education/employment, peer relations, substance abuse, leisure/recreation, personality/behavior, attitudes/orientation). Items are summed across domains to provide a total score (ranging from 0 to 42) and an associated risk category (low, moderate, high, and very high). A number of independent studies have shown good predictive validity of the YLS/CMI total score for general, nonviolent, and violent recidivism (Onifade, Nyandoro, Davidson, & Campbell, 2010; Viljoen et al., 2009; Welsh et al., 2008), as well as sound interrater reliability among different professionals (Schmidt et al., 2005). Studies have also indicated that the YLS/CMI has good psychometric properties across gender and different ethnicities (Olver, Stockdale, & Wormith, 2009). Adaptations of the YLS/CMI in countries outside of Canada and the United States also show this instrument is an accurate predictor of risk (e.g., McGrath & Thompson, 2012). In the present study, the static domain “criminal history” refers to the prior and current offences/dispositions of the YLS/CMI in all analyses. The dynamic risk total score was comprised of the scores from the seven dynamic risk domains (i.e., all domains except criminal history), which was calculated by subtracting the criminal history score from the total YLS/CMI score.

Risk scores were collected at two separate time points from a government database, with the first assessment conducted by probation officers during sentencing, and the second time point being the most proximal dynamic risk assessment completed prior to youths’ first reoffense, their departure from the youth justice system, or when the study period ended. All probation officers received formal training in the administration of the YLS/CMI as this tool was the standard risk assessment and case management protocol used in the jurisdiction examined.

Recidivism

Information pertaining to post assessment reoffenses and convictions was obtained from a national criminal record database. The cutoff date for establishing recidivism was January 25, 2010. This date was used to calculate each youth’s “survival time,” defined as the number of days between a young person’s first YLS/CMI assessment date and date of first conviction after the assessment. In 2011-2012, Statistics Canada reported that the average length of time to complete a court case was approximately 3 months (Dauvergne, 2013). As such, a 3-month period was added to the start date of the follow-up period (i.e., date of the youth’s initial assessment) to take into account court processing times. As well, the number of days spent in custody during the follow-up period was deducted from the length of time it took for youth to reoffend to account for incapacitation. Time in custody for the follow-up period ranged from 0 to 1,037 days, with a mean of 49.60 days (SD = 132.91); 110 youth served their entire sentence in the community and none of the youth were in custody during the entire follow-up period. Contingent on these conditions, the follow-up period during which there was opportunity to reoffend after the first YLS/CMI was administered ranged from 1 month to 9 years with a mean of 4.11 years (SD = 1.94). Only four youth had an offending opportunity of less than 1 year and two of these youth had less than 6 months.

Analytic Strategies

A hierarchical Cox regression was used to investigate whether dynamic risk factors, and changes in these domains, improved the predictive validity of recidivism after controlling for static variables. Cox regression is a semiparametric statistical technique commonly used to explore the effects of various covariates on survival (Norušis, 2007). This model allows for the prediction of failure time (i.e., time to reoffend) with a provision for censored data (i.e., youth who did not reoffend), and can handle a mix of continuous, categorical, and ordinal variables. For all analyses, race was categorized as a dichotomous variable that was based on whether youth were Caucasian or belonged to an ethnic minority group. Statistical programs that analyze survival time in Cox regression also protect against problems related to multicollinearity (Tabachnick & Fidell, 2007).

The main assumption of any Cox regression is that the proportionality of hazards has been met (i.e., the ratio of the estimated hazard across time is a constant for any two cases; Tabachnick & Fidell, 2007). To test this assumption, each covariate was tested separately prior to entering these variables in the Cox regression model. The proportionality of hazards assumption was met for all covariates except for age, which significantly interacted with time. As recommended by Singer and Willett (2003), when the proportionality of hazards assumption is violated, age was entered in the model as a time-varying covariate to include the interaction with time as a predictor.

Based on a formula cited in Harrell, Lee, and Mark (1996), the number of covariates entered into all regression models should be less than m / 10, where m is the number of people found in the lowest outcome category. As such, up to six covariates could be entered into each model in the present study when using time to reconviction as the outcome variable (62 nonrecidivists in the sample of 200 youth offenders; 62 / 10 = 6.2). Given that there were seven dynamic risk factors measured at each time point, the dynamic risk total score (ranging from 0 to 37) was used instead of individual risk domains as predictors for regression analysis. As there was no fixed follow-up period, ROC analyses were difficult to interpret and thus not reported in the present study.

Results

Preliminary Analyses

Table 1 presents youths’ mean YLS/CMI scores at the time of initial assessment and the most recent assessment scores prior to reoffense. The number of months between the initial and most recent assessment dates ranged from 0 to 44 (M = 10.83, SD = 9.33). Risk scores from youths’ initial assessment ranged from low to moderate across domains, with the YLS/CMI Total score falling in the moderate range. Compared with these initial scores, criminal history (static risk score) and most dynamic risk domain scores (with the exception of education/employment) increased over time, with the most recent reassessment scores falling in the low- to moderate-risk ranges for each risk domain. When the mean scores at initial and most recent assessment were compared using a series of paired-sample parametric and nonparametric tests, only the YLS/CMI total, criminal history, and substance abuse scores increased significantly (see Table 1).

Table 1:

Descriptive Statistics and Paired-Sample Tests of YLS/CMI and Dynamic Risk Total Scores, and Each Risk Domain at Time of Initial and Most Recent Assessment Before Youth Reoffended or Study Period Ended

	Initial assessment			Most recent assessment			Paired-samples test	p
Variable	M	SD	Minimum-Maximum	M	SD	Minimum-Maximum	t / Z	p
YLS/CMI Total	13.77	7.88	0-32	14.92	8.34	0-40	2.73	.007
Static Risk
Criminal History	0.53	1.01	0-5	1.34	1.52	0-5	−7.90	<.001
Dynamic Risk Total	13.24	7.54	0-30	13.58	7.61	0-36	0.85	.40
Family	2.24	1.66	0-6	2.40	1.63	0-6	−1.67	.10
Education	3.03	2.10	0-7	2.87	2.00	0-7	−0.89	.38
Peer Relations	1.74	1.34	0-4	1.78	1.31	0-4	−0.34	.74
Substance Abuse	0.88	1.30	0-5	1.03	1.28	0-5	−2.43	.02
Leisure	1.36	1.00	0-3	1.37	1.02	0-3	−0.29	.77
Personality	2.70	1.95	0-7	2.71	2.09	0-7	−0.06	.96
Attitudes	1.36	1.48	0-5	1.47	1.56	0-5	−1.01	.31

Note. Paired-sampled t-tests (t statistic) were conducted for continuous variables (YLS/CMI and Dynamic Risk total scores) and Wilcoxon signed ranks tests (Z statistic) were used for interval (ratio) variables that are not normally distributed (each risk domain). YLS/CMI = Youth Level of Service/Case Management Inventory.

In terms of recidivism during the follow-up period, 69% (n = 138) of participants reoffended. Of those, the most serious reconviction was for a violent offense in 43% of cases (n = 59), for a property offense in 26% of cases (n = 36), and for an administration of justice offense (e.g., breach of probation condition) in 9% (n = 12) of cases. Time to reoffense from initial assessment ranged from 1 day to 5 years, with an average of 1.11 years (SD = 1.13).

As seen in Table 2, measures of association were computed between demographic variables (i.e., gender, race, age), criminal history, and seven dynamic risk factors for two outcome recidivism variables (i.e., whether a young person reoffended and time to reoffend) at the time of youths’ initial and most recent assessment. Age at initial assessment was significantly and negatively related to whether or not youths reoffended (r = −.28, p < .001) and time to reoffense (r = −.23, p = .008). Neither gender nor race were associated with reoffending. Given that initial assessments were completed at the time of sentencing for youths’ first offense, it was not surprising that criminal history at initial assessment was not significantly related to whether youth reoffended as this domain score was only comprised of the current charges for which the youth was being sentenced. However, when assessed at the most recent assessment prior to reoffense (when prior and current convictions were both included in the criminal history domain score), criminal history was positively related to whether youth reoffended (r = .27, p < .001) and negatively associated with days to reoffense (r = −.24, p = .004).

Table 2:

Relationship Between Recidivism Variables and Demographic Variables, YLS/CMI and Dynamic Risk Total Scores, and Each Risk Domain At Initial and Most Recent Assessment

Variable	Initial and most recent Ax (r)^a	Reoffended (yes/no)^b			Reoffense time^c
Variable	Initial and most recent Ax (r)^a	Initial Ax^a	Most recent Ax^a	t ^d	Initial Ax^a	Most recent Ax^a	t ^d
Demographic variables
Gender		−.07^e			.05^f
p		.31			.58
Race		.09^e			.03^g
p		.23			.73
Age		−.28			−.23
p		<.001			.008
YLS/CMI Total	0.73	0.29	0.40	−2.29	−0.19	−0.27	1.31
p	<.001	<.001	<.001	.02	.02	<.001	.19
Static Risk: Criminal Hx	0.58	0.10	0.27	−2.71	−0.18	−0.24	0.78
p	<.001	.18	<.001	.007	.03	<.001	.44
Dynamic Risk Total	0.73	0.29	0.38	−1.86	−0.18	−0.25	1.14
p	<.001	<.001	<.001	.06	.04	<.001	.25
Family	0.73	0.10	0.16	−0.0.39	−0.10	−0.14	0.64
p	<.001	.16	.04	.70	.25	.11	.52
Education	0.63	0.22	0.24	−0.34	−0.19	−0.20	0.14
p	<.001	<.001	<.001	.74	.03	.02	.89
Peer Relations	0.60	0.15	0.23	−1.29	−0.06	−0.16	1.32
p	<.001	.03	<.001	.20	.47	.07	.19
Substance Abuse	0.76	0.07	0.14	−1.43	−0.13	−0.14	0.17
p	<.001	.31	.05	.15	.13	.13	.87
Leisure	0.58	0.25	0.36	−1.80	−0.05	−0.06	0.13
p	<.001	<.001	<.001	.07	.56	.88	.90
Personality	0.75	0.35	0.39	−0.87	−0.17	−0.29	2.07
p	<.001	<.001	<.001	.39	.04	<.001	.04
Attitudes	0.67	0.28	0.35	−1.29	−0.10	−0.14	0.58
p	<.001	<.001	<.001	.20	.24	.11	.56

Note. YLS/CMI = Youth Level of Service/Case Management Inventory; Ax = Assessment; Hx = History.

Pearson correlation coefficient (r) was computed (variables are interval [ratio] scales). ^bReoffended (y/n) variable was based on the entire sample (N = 200). ^cReoffense time variable was based on the sample of 138 youth who reoffended. ^dWilliams’s t test of correlation coefficients for repeated measures. ^ePhi coefficient (ϕ) was computed (variables are dichotomous and nominal). ^fPoint-biserial correlation coefficient (rpbi) is computed (one naturally created nominal variable and one interval [ratio] variable). ^gBiserial correlation coefficient (rbi) was computed (one artificially created nominal variable and one interval [ratio] variable).

As predicted, the YLS/CMI total and dynamic risk total scores were positively related to whether youths reoffended and negatively correlated with time to reoffend (see Table 2). Except for Substance Abuse, dynamic risk domain scores (at both initial and most recent assessments) were significantly related to whether youths reoffended but did not necessarily correlate with time to reoffense in the subset of recidivists (see Table 2). Although the associations between domain scores and recidivism were generally stronger when using scores from the most recent assessment, for the most part, the differences in magnitude of the correlations from initial to the most recent assessment were not statistically significant when compared using Williams’s t test. Only the correlations between the YLS/CMI total score and reoffense, t(197) = −2.29, p = .02, and between criminal history and reoffense, t(197) = −2.71, p = .007, were significantly stronger at the most recent assessment compared with the initial assessment. The relationship between Personality/Behavior and time to reoffend was also significantly stronger when the most recent assessment risk scores were used, t(135) = 2.07, p = .04 (see Table 2).

Predictive Validity of Static and Dynamic Risk Factors

A hierarchical Cox regression survival analysis was performed to assess the length of time to reoffend after adjusting for the effects of five covariates found to be predictive of survival. Covariates were entered hierarchically rather than simultaneously to determine whether the reassessment of static and dynamic risk factors added predictive validity over and above initial assessments. Gender and race, which contributed little to the prediction of survival time, were excluded from regression analysis. Age and initial criminal history score were entered in at the first step. Criminal history at the most recent assessment was entered in the second step, and dynamic risk total score at initial and the most recent assessment were entered in the third and fourth steps, respectively. Survival time was measured by the number of months until a youth reoffended during the follow-up period. The end date for nonrecidivists was January 25, 2010, the cutoff date of this study.

Covariates with positive B coefficients are associated with increased hazard (i.e., likelihood of reoffense) and decreased duration of predicted survival. Negative coefficients represent decreased hazard and increased survival times. Wald chi-square tests were used to test significance of the regression coefficient for individual covariates (i.e., coefficient = 0; Norušis, 2007). As seen in Step 1 of Table 3, criminal history significantly predicted time to reoffend after adjusting for differences in age at time of initial assessment (Wald χ² = 7.39, p = .007), and age significantly predicted survival time after controlling for criminal history (Wald χ² = 5.84, p = .02). Step 2 shows that there was no statistically significant effect of age or criminal history at initial assessment after adjusting for criminal history scores from the most recent assessment. Similarly, dynamic risk total score at initial assessment added unique variance to the model after controlling for static variables in Step 3 (Wald χ² = 10.27, p = .001), but did not significantly predict risk of reoffense when the dynamic risk total score from the most recent assessment (Wald χ² = 11.51, p = .001) was added in Step 4 of the model.

Table 3:

Hierarchical Cox Regression Analyses Predicting Time to Reoffense Using Age, Criminal History, and Dynamic Risk Total Scores From Initial and Most Recent Assessment

Variable	2 log likelihood^a (G²)with variable(s)	B b	SE	Wald	p	Exp(B)	95% CIExp(B)
Step 1	1,279.63
Age		−0.14	0.06	5.84	.016	0.87	[0.78, 0.98]
Criminal Hx—Initial Ax		0.20	0.07	7.39	.007	1.22	[1.06, 1.41]
Step 2	1,264.02
Age		−0.11	0.06	3.58	.059	0.90	[0.80, 1.01]
Criminal Hx—Initial Ax		0.04	0.08	0.24	.653	1.04	[0.84, 1.17]
Criminal Hx—Recent Ax		0.26	0.06	19.40	.000	1.30	[1.16, 1.46]
Step 3	1,253.84
Age		−0.09	0.06	2.43	.119	0.92	[0.82, 1.02]
Criminal Hx—Initial Ax		0.04	0.09	0.21	.648	1.04	[0.81, 1.23]
Criminal Hx—Recent Ax		0.23	0.06	13.24	.000	1.25	[1.11, 1.41]
Dynamic Risk—Initial Ax		0.04	0.01	10.27	.001	1.04	[1.02, 1.07]
Step 4	1,242.26
Age		−0.10	0.06	3.22	.073	0.90	[0.80, 1.01]
Criminal Hx—Initial Ax		0.06	0.10	0.43	.513	1.06	[0.88, 1.28]
Criminal Hx—Recent Ax		0.14	0.07	4.20	.041	1.15	[1.01, 1.31]
Dynamic Risk—Initial Ax		0.01	0.02	0.15	.701	1.01	[0.98, 1.04]
Dynamic Risk—Recent Ax		0.06	0.02	11.51	.001	1.06	[1.02, 1.09]

Note. CI = confidence interval; Ax = Assessment; Hx = History.

−2 log likelihood (without variable) = 1,290.40; 2 log likelihood = −2 multiplied by the log likelihood value. ^bB = Unstandardized (χ²), which represents the degree to that the baseline survival function increases or decreases as a function of a unit change in the variable.

Exp(B) is the hazard ratio for each covariate in which a positive sign indicates an increase in the probability of an event (i.e., reoffense). The result is not considered statistically significant if the confidence intervals of the Exp(B) overlap with the value of 1 (Norušis, 2007). After adjusting for all other covariates, for each one-point increase in dynamic risk total score at the most recent assessment, the likelihood of reoffense increased by 6%. The greatest contribution was by the logarithm of criminal history at the most recent assessment; each increase of one-point increased the risk of reoffense by 15% (see Table 3).

Log-likelihood chi-square tests were also used to evaluate whether the regression coefficients for each covariate were equal to zero and to compare differences in models with and without covariates (Norušis, 2007). Although all models were statistically significant in comparison with the null model and changes from the previous step (simpler model with less covariates), the final step that included criminal history and the dynamic risk total score at the most recent assessment contributed the most to the model, −2 log likelihood (−2LL) = 1,242.26, overall χ²(5, N = 200) = 52.73, p < .001. In Step 1, the combination of age and criminal history at initial assessment significantly predicted survival time, χ²(2, N = 200) = 10.77, p = .005. The addition of criminal history at the most recent assessment as a covariate showed a significant change from the previous step, χ²(1, N = 200) = 15.61, p < .001. Similarly, significant changes from the previous step were found when the dynamic risk total score at the initial assessment, χ²(1, N = 200) = 10.18, p = .001, and the most recent assessment, χ²(1, N = 200) = 11.58, p = .001, were added to the model. As such, changes in criminal history and the dynamic risk total score successfully predicted survival time after adjusting for variables from the initial assessment.

Discussion

The primary purpose of the current study was to empirically examine changes in dynamic risk factors over two time points and to determine whether proximal dynamic risk assessment scores would better predict recidivism compared with static variables or risk scores collected from the initial assessment. Survival time was predicted by several covariates including age, criminal history, and the dynamic risk total score from both assessment time points. However, there was no statistically significant effect of age, criminal history, or dynamic risk from the initial assessment once scores of these two risk domains from the most recent assessment prior to reconviction were taken into account. Taken together, these analyses indicate that changes in dynamic risk factors over time added significant predictive power to recidivism beyond that contributed by static risk and dynamic factors obtained from the initial assessment. This study contributes to the existing research by being the first to investigate the longer term predictive ability of the YLS/CMI as a reassessment tool among youth offenders.

In the present study, younger youth at initial assessment were more likely to reoffend and were reconvicted more quickly than older youth. This is concomitant with the well-documented age–crime curve in which crime increases from early- to mid-adolescence and peaks between the age of 15 and 19, before significantly declining and leveling off in young adulthood (Farrington, 2003). These results highlight the possible need to separate the younger youth from the older, high-risk offenders who populate the majority of treatment programs, and divert younger offenders to community programming rather than custodial settings. At the same time, youth convicted of a crime at a younger age in this sample may also represent higher risk life-course persistent offenders rather than adolescent-limited offenders (Moffitt, 1993), which would explain higher rates of recidivism among these youth. Findings from the current study illustrate the importance of the YLS/CMI as a reassessment tool because adolescent offenders, especially younger youth, may require more intense services as particular dynamic risk factors increase over time.

In addition, most dynamic risk factors increased over time and were significantly related to recidivism among adolescent offenders. This finding is perhaps not surprising given that the previous literature has acknowledged a lack of adequate treatment services for youth on probation that could mitigate risk across various dynamic risk domains (e.g., Haqanee, Peterson-Badali, & Skilling, 2015; Peterson-Badali et al., 2015; Vieira et al., 2009). Further inspection of current supervision and case management policies and procedures is required to ensure that youth are receiving appropriate and sufficient services to address their needs (Hannah-Moffat, 2016). As well, allocating greater time and resources toward treatment development and implementation that targets specific dynamic risk factors is essential to maximize rehabilitation efforts needed to reduce reoffense rates.

According to government policy in the location of the present study, risk assessments must be carried out at least once a year or when there is a change in one’s risk to reoffend (L. Freedman, personal communication, April 28, 2017). However, some dynamic risk factors change quickly and should be updated frequently to accurately detect changes in risk and better inform treatment planning and delivery (Quinsey et al., 2006). For example, substance abuse was identified as the dynamic risk factor that changed the most over time and increased as youth grew older. As such, careful monitoring of youths’ substance use is necessary to ascertain whether intervention services are required, and when identified, treatment programs should be readily available. In contrast, education and employment needs remain fairly stable across time, which is likely related to this dynamic risk factor being easy to identify, monitor, and target with appropriate services (Haqanee et al., 2015). Although greater focus on the needs of low-risk offenders is recommended, increasing the level of supervision is not suggested; instead, dynamic risk factors should be monitored more frequently during the course of regular supervision to ensure that high-risk needs are identified and addressed appropriately.

Although some scholars contend that risk assessments should be separated from needs assessments (Gottfredson & Moriarty, 2006; Hannah-Moffat, 2016; Maurutto & Hannah-Moffat, 2007), findings from the present study support the inclusion of dynamic risk factors in risk assessments because such factors improve the predictive accuracy of recidivism over and above static risk variables. These results are consistent with more recent research in the field of risk assessment (Andrews et al., 2006; Brown et al., 2009; McGrath & Thompson, 2012; Peterson-Badali et al., 2015; Vincent et al., 2012). As well, identifying changes in dynamic risk factors over time also improved the predictive validity of the YLS/CMI after controlling for scores from the initial assessment, providing further evidence that this instrument is an effective reassessment tool in determining risk and identifying youth needs.

There were several limitations to the present study that should be taken into account when reviewing the results. First, the sample size was relatively small in comparison with other larger scaled longitudinal studies, and consequently, only the dynamic risk total score, rather than individual criminogenic need domain scores, could be examined in the Cox regression analyses. A larger sample size would provide more statistical power for analyses and allow for the examination of individual dynamic risk factors in relation to the predictive validity of the YLS/CMI, as well as of multiple types of recidivism (e.g., violent as well as general). Second, including a sample comprised of more female offenders in future studies may produce valuable information about possible gender effects on the predictive accuracy of risk assessment that did not emerge in the current study.

A third potential limitation was that data were accessed retrospectively, although the assessments were completed in real time. Prospective studies that reassess dynamic needs within smaller time intervals may uncover additional changes in dynamic risk and provide a better understanding of how quickly these dynamic risk factors change over time. At the same time, prospective research designs predetermine how frequently risk factors are reassessed and create artificial time intervals between assessments that do not necessarily reflect the frequency of risk assessments completed during probation. As such, the design of the present study was also considered a strength because it provided risk scores from assessments in real-world settings (Cording et al., 2016). However, interrater reliability, which is recommended to be reported in predictive validity studies (Singh, Yang, Mulvey, & RAGEE Group, 2014), was not collected on the assessment of youths’ risk and needs among probation officers as a result. Given that there were no significant differences in the magnitude of the correlations from the initial to most recent assessment for most dynamic risk factors at the domain level, further investigation into what initiates reassessments at probation and other correctional settings is greatly needed to ensure that changes in dynamic risk factors are being assessed appropriately. For example, it is important to identify whether risk assessments are driven by changes in dynamic risk or are a reaction to new charges being laid, which becomes an exercise in “postdiction” rather than prediction.

Finally, recidivism data used in the present study were based solely on official conviction records. Several studies have found that self-reported criminal activity generally results in higher rates of offending and is a more valid representation of recidivism than official reports (Mulvey & Schubert, 2012; Vincent et al., 2012). At the same time, the use of conviction data was important in the present study because reassessments by probation officers are supposed to be conducted when there are changes to a young person’s risk to reoffend (e.g., criminal history increased) and not based on criminal activities that were unlikely disclosed at probation. It would be important to examine both arrest and conviction dates as outcome variables to determine whether police involvement or recent convictions initiated a new risk assessment or if changes in dynamic risk factors were the driving force for reassessments among youth offenders.

Although matching dynamic risk factors to appropriate treatment services has been shown to reduce recidivism (Mulvey & Schubert, 2012; Vieira et al., 2009), the present study did not investigate the extent to which youths’ identified dynamic risk factors were addressed via intervention. Future longitudinal research should investigate the relationship between the predictive accuracy of risk assessments and additional predictors, including responsivity factors (e.g., cognitive ability, motivation) and the match of intervention services to identified dynamic risk factors. The inclusion of protective factors (e.g., self-efficacy, social support, life goals) that are also dynamic in nature and promote desistance (Serin et al., 2016) should also be considered when examining the predictive validity of risk assessment tools. Moreover, the YLS/CMI only includes the number of prior and current convictions in its calculation of risk and does not differentiate between violent and nonviolent offenses, which may have important implications in terms of changes in dynamic risk factors over time and youths’ risk to reoffend (Douglas & Ogloff, 2003).

Given increases in the utilization of actuarial risk needs assessments in the youth justice sector (Maurutto & Hannah-Moffat, 2007), empirical insight into factors that promote or impede accurate assessment of risk and the identification of dynamic risk factors will be valuable to youth justice personnel, particularly those involved in the coordination of youth services and sentence implementation. That is, more updated and accurate risk assessments will provide additional information to probation officers about young people’s risk and needs, which can then assist in designing effective intervention plans for youth. Implemented consistently and with fidelity, accurate assessments will better inform treatment implementation, which, in turn, will lead to a more accurate matching of treatment services. Increasing effective case management and intervention programs throughout the legal process will not only lead to more positive legal outcomes for young people, but it will also uphold the goals of youth justice legislation to encourage rehabilitation and provide meaningful dispositions for youth. This, in turn, will lead to reductions in youth crime, promote public safety, and reduce correctional costs associated with reoffending.

Footnotes

This article is based on Maggie Clarke’s doctoral dissertation. The research was conducted at the Ontario Institute for Studies in Education, University of Toronto. Maggie Clarke is currently working for the Waterloo Region District School Board.

This research was supported by Grant number 410101516 from the Social Sciences and Humanities Council of Canada to the second and third authors. The authors would like to express their deep appreciation to Kathy Underhill of the Program Effectiveness, Statistics and Applied Research Unit of the Ontario Ministry of Community Safety and Correctional Services and Mr. Justice Brian Weagant of the Ontario Court of Justice for their contributions to this study.

Notes

Maggie C. Clarke is a psychologist at the Waterloo Region District School Board. She currently works with children and adolescents in both school and private practice settings. Her research interests include examining dynamic risk factors over time in youth offenders and parental involvement in youth justice proceedings.

Michele Peterson-Badali is a professor in the Department of Applied Psychology and Human Development at the Ontario Institute for Studies in Education, University of Toronto. Her research focuses on children’s and adolescents’ knowledge, reasoning, perceptions, and experiences of the youth justice system; their understanding of rights; and their evolving legal capacities. Current foci include implementation of the Risk-Need-Responsivity framework in community-sentenced youth, mental health and youth criminal justice, and indigenous youth.

Tracey A. Skilling is a psychologist and clinician scientist at the Center for Addiction and Mental Health and an assistant professor in the Department of Psychiatry at the University of Toronto, Canada. Her research interests include antisocial behavior, mental health issues, and substance use in adolescents; and risk assessment issues in both male and female young offenders. Her current projects include studies evaluating risk assessment practices, and understanding the desistance process, with adolescent male and female offenders.

References

Andrews

D. A.

Bonta

(2010a). The psychology of criminal conduct (5th ed.). New Providence, NJ: Anderson.

Andrews

D. A.

Bonta

(2010b). Rehabilitating criminal justice policy and practice. Psychology, Public Policy, and Law, 16, 39-55. doi:10.1037/a0018362

Andrews

D. A.

Bonta

Wormith

J. S.

(2006). The recent past and near future of risk and/or need assessment. Crime & Delinquency, 52, 7-27. doi:10.1177/0011128705281756

Andrews

D. A.

Dowden

(2006). Risk principle of case classification in correctional treatment. International Journal of Offender Therapy and Comparative Criminology, 50, 88-100. doi:10.1177/0306624X05282556

Andrews

D. A.

Zinger

Hoge

R. D.

Bonta

Gendreau

Cullen

F. T.

(1990). Does correctional treatment work? A clinically relevant and psychologically informed meta-analysis. Criminology, 28, 369-404. doi:10.1111/j.1745-9125.1990.tb01330.x

Bourgon

Armstrong

(2005). Transferring the principles of effective treatment into a “real world” prison setting. Criminal Justice and Behavior, 32, 3-25. doi:10.1177/0093854804270618

Brown

S. L.

St Amand

M. D.

Zamble

(2009). The dynamic prediction of criminal recidivism: A three-wave prospective study. Law and Human Behavior, 33, 25-45. doi:10.1007/s10979-008-9139-7

Cording

J. R.

Beggs Christofferson

S. M.

Grace

R. C.

(2016). Challenges for the theory and application of dynamic risk factors. Psychology, Crime & Law, 22, 84-103. doi:10.1080/1068316X.2015.1111367

Cottle

C. C.

Lee

R. J.

Heilbrun

(2001). The prediction of criminal recidivism in juveniles: A meta-analysis. Criminal Justice and Behavior, 28, 367-394. doi:10.1177/0093854801028003005

10.

Dauvergne

(2013). Youth court statistics in Canada, 2011/2102 (Issue No. 1209-6393). Retrieved from http://www.statcan.gc.ca/pub/85-002-x/2013001/article/11803-eng.pdf

11.

Douglas

K. S.

Ogloff

J. R.

(2003). Multiple facets of risk for violence: The impact of judgmental specificity on structured decisions about violence risk. International Journal of Forensic Mental Health, 2, 19-34. doi:10.1080/14999013.2003.10471176

12.

Douglas

K. S.

Skeem

J. L.

(2005). Violence risk assessment: Getting specific about being dynamic. Psychology, Public Policy, and Law, 11, 347-383. doi:10.1037/10768971.11.3.347

13.

Farrington

D. P.

(2003). Key results from the first forty years of the Cambridge Study in delinquent development. In Thornberry

T. P.

Krohn

M. D.

(Eds.), Taking stock of delinquency: An overview of findings from contemporary longitudinal studies (Vol. 7, pp. 137-183). Boston, MA: Kluwer Academic.

14.

Gottfredson

S. D.

Moriarty

L. J.

(2006). Statistical risk assessment: Old problems and new applications. Crime & Delinquency, 52, 178-200. doi:10.1177/0011128705281748

15.

Greiner

L. E.

Law

M. A.

Brown

S. L.

(2014). Using dynamic factors to predict recidivism among women: A four-wave prospective study. Criminal Justice and Behavior, 42, 457-480. doi:10.1177/0093854814553222

16.

Hannah-Moffat

(2016). A conceptual kaleidoscope: Contemplating “dynamic structural risk” and an uncoupling of risk from need. Psychology, Crime & Law, 22, 33-46. doi:10.1080/1068316X.2015.1114115

17.

Haqanee

Peterson-Badali

Skilling

(2015). Making “what works” work: Examining probation officers’ experiences addressing the criminogenic needs of juvenile offenders. Journal of Offender Rehabilitation, 54, 37-59. doi:10.1080/10509674.2014.980485

18.

Harrell

F. E.

Lee

K. L.

Mark

D. B.

(1996). Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Statistics in Medicine, 15, 361-387.

19.

Harris

G. T.

Rice

M. E.

(2003). Actuarial assessment of risk among sex offenders. Annals of the New York Academy of Sciences, 989, 198-210. doi:10.1111/j.1749-6632.2003. tb07306.x

20.

Hoge

R. D.

Andrews

D. A.

(2002). Youth Level of Service/Case Management Inventory: YLS/CMI manual and interview guide. Toronto, Ontario, Canada: Multi-Health Systems.

21.

Kroner

D. G.

Takahashi

(2012). Every session counts: The differential impact of previous programmes and current programme dosage on offender recidivism. Legal and Criminological Psychology, 17, 136-150. doi:10.1111/j.2044-8333.2010.02001.x

22.

Lipsey

M. W.

(2009). The primary factors that characterize effective interventions with juvenile offenders: A meta-analytic overview. Victims and Offenders, 4, 124-147. doi:10.1080/15564880802612573

23.

Loeber

Ahonen

(2014). What are the policy implications of our knowledge on serious, violent, and chronic offenders? Criminology and Public Policy, 13, 117-125. doi:10.1111/1745-9133.12072

24.

Lowenkamp

C. T.

Latessa

E. J.

(2004). Understanding the risk principle: How and why correctional interventions can harm low-risk offenders. In Topics in community corrections (pp. 3-8). Retrieved from http://www.yourhonor.com/dwi/sentencing/RiskPrinciple.pdf

25.

Lowenkamp

C. T.

Latessa

E. J.

Holsinger

A. M.

(2006). The risk principle in action: What have we learned from 13, 676 offenders and 97 correctional programs. Crime & Delinquency, 52, 77-93. doi:10.1177/0011128705281747

26.

Luong

Wormith

J. S.

(2011). Applying risk/need assessment to probation practice and its impact on the recidivism of young offenders. Criminal Justice and Behavior, 38, 1177-1199. doi:10.1177/0093854811421596

27.

Maurutto

Hannah-Moffat

(2007). Understanding risk in the context of the Youth Criminal Justice Act. Canadian Journal of Criminology and Criminal Justice, 49, 465-491. doi:10.1353/ccj.2008.0005

28.

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, 393, 250-263. doi:10.1177/0093854811431917

29.

Moffitt

T. E.

(1993). Adolescence-limited and life-course-persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100, 674-701. doi:10.1037//0033295X.100.4.674

30.

Mulvey

E. P.

Schubert

C. A.

(2012). Some initial findings and policy implications of the Pathways to Desistance Study. Victims & Offenders, 7, 407-427. doi:10.1080/15564886.2012.713903

31.

Norušis

(2007). SPSS 15.0 Advanced statistical procedures companion. Upper Saddle River, NJ: Prentice Hall.

32.

Oleson

J. C.

VanBenschoten

Robinson

Lowenkamp

C. T.

Holsinger

A. M.

(2012). Actuarial and clinical assessment of criminogenic needs: Identifying supervision priorities among federal probation officers. Journal of Crime & Justice, 35, 239-248. doi:10.1080/0735648X.2012.674819

33.

Olver

M. E.

Beggs Christofferson

S. M.

Wong

S. C.

(2015). Evaluation and applications of the clinically significant change method with the Violence Risk Scale-Sexual Offender Version: Implications for risk-change communication. Behavioral Sciences & the Law, 33, 92-110. doi:10.1002/bsl.2159

34.

Olver

M. E.

Stockdale

K. C.

Wong

S. C. P.

(2012). Short and long-term prediction of recidivism using the Youth Level of Service/Case Management Inventory in a sample of serious young offenders. Law and Human Behavior, 36, 331-334. doi:10.1037/h0093927

35.

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. doi:10.1177/0093854809331457

36.

Onifade

Nyandoro

A. S.

Davidson

W. S.

Campbell

(2010). Truancy and patterns of criminogenic risk in a young offender population. Youth Violence and Juvenile Justice, 8, 3-18. doi:10.1177/1541204009338251

37.

Perrault

R. T.

Paiva-Salisbury

Vincent

G. M.

(2012). Probation officers’ perceptions of youths’ risk of reoffending and use of risk assessment in case management. Behavioral Sciences & the Law, 30, 487-505. doi:10.1002/bsl.2015

38.

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. doi:10.1177/0093854814549595

39.

Quinsey

V. L.

(2009). Are we there yet? Stasis and progress in forensic psychology. Canadian Psychology, 50, 15-21. doi:10.1037/a0014401

40.

Quinsey

V. L.

Jones

B. G.

Book

A. S.

Barr

K. N.

(2006). The dynamic prediction of antisocial behavior among forensic psychiatric patients: A prospective field study. Journal of Interpersonal Violence, 21, 1539-1565. doi:10.1177/0886260506294238

41.

Schmidt

Hoge

R. D.

Gomes

(2005). Reliability and validity analyses of the Youth Level of Service/Case Management Inventory. Criminal Justice and Behavior, 32, 329-344. doi:10.1177/0093854804274373

42.

Schwalbe

C. S.

(2007). Risk assessment for juvenile justice: A meta-analysis. Law and Human Behavior, 31, 449-462. doi:10.1007/s10979-006-9071-7

43.

Serin

R. C.

Chadwick

Lloyd

C. D.

(2016). Dynamic risk and protective factors. Psychology, Crime & Law, 22, 151-170. doi:10.1080/1068316X.2015.1112013

44.

Singer

J. D.

Willet

J. B.

(2003). Applied longitudinal data analysis: Modeling change and event occurrence. New York, NY: Oxford University Press.

45.

Singh

J. P.

Yang

Mulvey

E. P.

& RAGEE Group. (2014). Reporting guidelines for violence risk assessment predictive validity studies: The RAGEE statement. Law and Human Behavior, 39, 15-22. doi:10.1037/lhb0000090

46.

Tabachnick

B. G.

Fidell

L. S.

(2007). Using multivariate statistics (5th ed.). New York, NY: Allyn & Bacon.

47.

Takahashi

Mori

Kroner

D. G.

(2013). A cross-validation of the Youth Level of Service/Case Management Inventory (YLS/CMI) among Japanese juvenile offenders. Law and Human Behavior, 37, 389-400. doi:10.1037/lhb0000029

48.

Vieira

T. A.

Skilling

T. A.

Peterson-Badali

(2009). Matching court-ordered services with youths’ treatment needs: Predicting treatment success with young offenders. Criminal Justice and Behavior, 36, 385-401. doi:10.1177/0093854808331249

49.

Viljoen

J. L.

Elkovitch

Scalora

M. J.

Ullman

(2009). Assessment of reoffense risk in adolescents who have committed sexual offenses: Predictive validity of the ERASOR, PCL:YV, YLS/CMI, and Static-99. Criminal Justice and Behavior, 36, 981-1000. doi:10.1177/0093854809340991

50.

Vincent

G. M.

Perrault

R. T.

Guy

L. S.

Gershenson

B. G.

(2012). Developmental issues in risk assessment: Implications for juvenile justice. Victims & Offenders, 7, 364-384. doi:10.1080/15564886.2012.713900

51.

Vitopoulos

Peterson-Badali

Skilling

(2012). The efficacy of the risk-need-responsivity framework in guiding intervention for justice system-involved female youth. Criminal Justice & Behavior, 39, 1025-1041. doi:10.1177/0093854812442895

52.

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. doi:10.1177/1073191107307966