Predictive Validity of the Youth Level of Service/Case Management Inventory with Youth who have Committed Sexual and Non-Sexual Offenses

Abstract

The predictive validity of the Youth Level of Service/Case Management Inventory (YLS/CMI) and the use of professional override were examined in a matched sample of youth who committed sexual (n = 204) and non-sexual (n = 185) offenses. Based on the actuarial score, the YLS/CMI obtained moderate to strong levels of predictive validity for non-violent, violent, sexual, and technical recidivism in both samples of youth. Probation officers always used override to increase risk level classification and did so at a high level for both sexual (n = 151; 74.0%) and non-sexual (n = 77; 41.6%) offending youth. There was a detrimental impact on the predictive validity of the YLS/CMI for youth who received an override adjustment, regardless of offending category. These preliminary findings suggest that the application of override should be carefully considered on instruments such as the YLS/CMI.

Keywords

risk youth professional override recidivism sexual

Assessing risk for criminal and violent behavior is a common role of professionals who work in the youth justice system. Historically, professionals relied on unstructured clinical judgment to determine risk. This form of evaluation, however, was often incomplete or focused on irrelevant factors (Douglas & Kropp, 2002; see also Borum, Otto, & Golding, 1993, for a discussion of problems associated with clinical judgment in risk prediction). Research has consistently demonstrated, across a variety of decision-making contexts, the poorer predictive validity of unstructured clinical judgment when compared with structured empirically based methods of risk assessment (Ægisdóttit et al., 2006; Grove, Zald, Lebow, Snitz, & Nelson, 2000). Moreover, Grove et al. (2000) found that experienced or highly trained professionals were no more accurate than students at predicting future criminal behavior. They concluded that the use of structured actuarial instruments is generally superior to unstructured clinical judgment, regardless of the rater’s experience or training.

The shift toward standardized and evidence-based risk tools is part of a larger theoretical shift in correctional case management and intervention practices as outlined in the risk–need–responsivity (RNR) model (Andrews & Bonta, 2010a, 2010b). This model, first espoused by Andrews, Bonta, and Hoge (1990), demonstrated that empirically identified risk, need, and responsivity factors were linked to successful rehabilitation. The first principle of “risk” refers to an estimation of the probability that an individual will re-offend, and that the level of service provided should match the identified level of risk. This principle suggests that as risk level increases, the intensity of interventions and resources used for rehabilitation should also increase. In fact, when low-risk individuals are provided with high levels of intervention, it can lead to an increased rate of recidivism (Andrews & Bonta, 2010a). The second principle, referred to as “need,” guides the targets of intervention by distinguishing between criminogenic and non-criminogenic needs. Research has established eight criminogenic need domains (e.g., antisocial attitudes, antisocial associations, and substance abuse) that link empirically to the origin of criminal behavior (Andrews & Bonta, 2010a). These criminogenic factors are distinct from other possible non-criminogenic treatment factors such as self-esteem, anxiety, or fear of punishment, which do not relate to recidivism. The need principle is critical for successful intervention because it clearly identifies the targets for effective rehabilitation and reduction of risk. The final RNR principle of “responsivity” refers to the importance of matching treatment strategies with each individual’s unique characteristics, such as mental health issues, motivation level, and cognitive capacity, to maximize their response to interventions. The use of treatment approaches that are supported by research to reduce risk is also inherent to the responsivity principle.

Empirically based risk tools provide a foundation upon which to apply the RNR model. In the youth justice system, a number of risk instruments have been developed for sexual, violent, and general criminal behavior. Perhaps the most prominent and well studied is the Youth Level of Service/Case Management Inventory (YLS/CMI; Hoge & Andrews, 2006). Adapted from the adult Level of Service/Case Management Inventory (LS/CMI; Andrews, Bonta, & Wormith, 2004), the YLS/CMI is an adolescent version designed to assess risk and need for general recidivism in 12- to 17-year-old youth. This instrument was recently updated and now covers a broader age range, from 12 to 18 (YLS/CMI 2.0; Hoge & Andrews, 2011). The 42 empirically based risk items of the YLS/CMI are divided into eight sub-scales, which correspond to the eight domains of the RNR model. In addition, a section on case planning is used to note the targets for intervention and guide community supervision. Finally, there is an option for professional override, whereby the professional can adjust the risk level obtained by the actuarial score.

Substantial psychometric data on the reliability and validity of the YLS/CMI are accumulating and are more advanced than other risk tools used with youth (see Olver, Stockdale, & Wormith, 2009). Based on aggregate weighted correlations across studies, the overall predictive validity of the YLS/CMI was found to range from a low of .20 for sexual recidivism to a high of .32 for general recidivism (Olver et al., 2009). While limitations have been noted in a few studies with minority and female youth (Marczyk, Heilbrun, Lander, & DeMatteo, 2003; Onifade et al., 2008), a recent and comprehensive meta-analysis found that the cumulative body of YLS/CMI research performed well with females and ethnic minorities (Olver, Stockdale, & Wormith, 2014). Importantly, the YLS/CMI has been implemented beyond Canada and the United States, with a culturally adapted version used in Australia (YLS/CMI-AA; Thompson & Pope, 2005; Upperton & Thompson, 2007). It has also been used internationally in Japan (Takahashi, Mori, & Kroner, 2013), Singapore (Chu, Yu, Lee, & Zeng, 2014), Spain (Hilterman, Nicholls, & Nieuwenhuizen, 2014), and the United Kingdom (Marshall, Egan, English, & Jones, 2006).

YLS/CMI Professional Override

Designed primarily as an actuarial risk tool, the YLS/CMI and its adult version, the LS/CMI (Andrews et al., 2004), provide the clinical practitioner some discretion to override or adjust the risk level obtained through the actuarial score. Use of clinical judgment on the YLS/CMI differs from other risk tools that have adopted a structured professional judgment (SPJ) approach. SPJ instruments use empirically derived risk factors that are integrated collectively to identify overall risk level through clinical judgment; they do not rely on an actuarial score to determine risk. The SPJ approach is used in many common risk guidelines for adults, like the Historical, Clinical, Risk Management–20, Version 3 (HCR-20V3; Douglas, Hart, Webster, & Belfrage, 2013), and with adolescents, including the Structured Assessment of Violence Risk in Youth (SAVRY; Borum, Bartel, & Forth, 2006) and the Estimate of Risk of Adolescent Sexual Offender Recidivism (ERASOR; Worling & Curwen, 2001). While some similarities exist, there are important differences between professional override and SPJ approaches to risk assessment which warrant independent investigation.

The rationale for a professional override option on the YLS/CMI stems from the potential insensitivities that actuarial approaches have to changing environments, the presence of protective factors, and the unique characteristics of each individual that are not captured by aggregate scores (Childs, Frick, Ryals, Lingonblad, & Villo, 2014). The manuals for the YLS/CMI (Hoge & Andrews, 2006) and the YLS/CMI 2.0 (Hoge & Andrews, 2011) provide some guidance about professional override. For example, the manual states that the assessor should consider all of the information about the youth and then provide a risk estimate, which in “some circumstances” might be different than the actuarial score (Hoge & Andrews, 2011, p. 7). The manual authors also state that override should only be used with caution and in “rare circumstances” supported by logical argument and reasonable evidence (p. 9). Despite the significant influence that override adjustments could have on case management practices or the predictive validity of the YLS/CMI, there is very little research regarding how frequently override is used or the effect it might have on predictive validity. Moreover, there is little known about whether override varies by population groups such as adults versus adolescents, sexual versus non-sexual offending youth, or the impact it may have on specific types of risk prediction such as general, violent, or sexual recidivism.

A recent study examined the influence of LS/CMI override on risk prediction for adult offenders (Wormith, Hogg, & Guzzo, 2012). In this study, Wormith et al. (2012) examined the recidivism rates of 1,905 sexual and 24,545 non-sexual offenders following release from a correctional facility after serving a minimum 1-month sentence. The predictive validity for the LS/CMI was examined for a 1,650-day follow-up period. The authors found that override was used more frequently with sexual (35.1%) than non-sexual (15.1%) offenders, and ostensibly used to increase (14.9%) rather than decrease (1.6%) risk. While the LS/CMI initially demonstrated sound predictive validity values for both the sexual and non-sexual offending samples, override was found to decrease the predictive validity of the LS/CMI in both samples and across all three outcome measures of general, violent, and sexual recidivism. The drop-in predictive validity was more pronounced in the sexual offending sample, which had a higher rate of override when compared with the non-sexual offending group. The results of this study raised questions about the utility or benefit of override, at least with respect to the LS/CMI.

Studies examining override on the YLS/CMI with adolescents, similar to the adult study by Wormith et al. (2012), have not been published. However, several studies contrasting the actuarial and SPJ methods of risk determination on various adolescent risk instruments have been completed. While actuarial scores are typically not calculated using the SPJ approach, research in this area has contrasted SPJ risk level determinations by a practitioner with a total actuarial score using a summation of the same SPJ empirically based risk factors. For example, based on a meta-analysis of various risk tools with adolescent sexual offenders, Viljoen, Mordell, and Beneteau (2012) found that use of SPJ with the ERASOR did not increase or decrease performance; it seemed to perform equally well as the actuarial method. In another study, Storey, Watt, Jackson, and Hart (2012) recommended caution when using override with sexually offending adults as they found decreased accuracy on the STATIC-99 when compared with a purely actuarial approach to risk determination.

A study by Childs et al. (2014) examined how probation officers determined risk using the SAVRY, an SPJ risk tool. Childs et al. (2014) compared risk level using the SPJ method with an empirically derived risk level using statistical latent class analyses on the same SAVRY risk profiles. They found that probation officers obtained higher risk ratings for youth when compared with those obtained through statistical analysis. While the authors of the study continued to recommend an SPJ approach (Childs et al., 2014), the results of the study highlighted the potential differences that could occur with SPJ judgments when practitioner decision making is compared with an objective empirical approach. That is, practitioner risk classifications were in the direction of being higher when compared with a statistical method of risk determination. Similar possible inflation of risk was identified in a qualitative study within the youth justice system where probation officers were found to consistently raise, not lower, risk classification as a means of managing community safety and personal liability issues (Ballucci, 2012).

Within the juvenile detention system, override on detention risk assessments was used disproportionately more for some youth depending on key demographic variables, including gender and race (Chappell, Maggard, & Higgins, 2012). Chappell et al. (2012) did not look at the effect of clinical override on predictive validity. However, the biases observed with professional override in the detention setting could possibly affect risk decision making in a non-detention context. When taken together, this small body of research examining override and risk determination with adults and adolescents highlights the need for further study in this area of correctional practice.

The Present Study

Professional override is embedded within the risk assessment field. The recent study by Wormith et al. (2012) found a relatively high level of LS/CMI override by probation officers with some detriment to the observed predictive validity. These results have important implications for field practice with adult offenders. Following a thorough literature search, a similar study with adolescent offenders examining override with the YLS/CMI was not found. Given the developmental differences in maturation and unique adolescent offending trajectories (Dyck, Campbell, Schmidt, & Wershler, 2013), it is not clear whether override by probation officers would be different for adolescents when compared with adults or how override might affect the predictive validity of a risk instrument with this developmentally younger group. The current study attempted to address this gap in the field by examining the YLS/CMI override feature as used by probation officers in usual practice for a sample of adolescents who committed sexual and non-sexual offenses. While the prevalence of override adjustments by probation officers was unknown and exploratory with adolescents in the youth justice system, it was expected that override would decrease the predictive validity of the YLS/CMI for both sexual and non-sexually offending youth.

Method

Participants

De-identified data for two adolescent offender samples were obtained from the Ontario Ministry of Children and Youth Services. As the purpose of the study was to compare both sexual and non-sexual offending youth, a request was made to obtain samples for both sub-groups. Given the relatively less frequent occurrence of sexual offending behavior, the full sample of youth convicted of sexual offense in the years 2009 and 2010 in the province of Ontario (n = 271) was obtained. To achieve a greater level of similarity with respect to risk level and probation practice/monitoring requirements, a second random sample of non-sexual offenders matched by disposition type (i.e., custody placement, length of probation) and over the same time period were also obtained (n = 276). To ensure comparability of the two samples, further data checking was done. For example, as only one female youth was convicted of a sexual offense, all females were excluded from the sexual (n = 1) and non-sexual offending (n = 57) samples to allow for balanced comparisons. This resulted in 219 male youth in the non-sexual offending sample and 270 male youth in the sexual offending sample. Between the time period 2009 and 2010, the first edition of the YLS/CMI, which was validated for ages 12 to 17, was being used. Thus, youth in both samples who were 18 or older were excluded. This resulted in a further 33 non-sexual and 62 sexual offending youth being excluded, leaving 186 and 208 youth in the non-sexual and sexual offending samples, respectively. Finally, following a calculation of each youth’s time in the community prior to recidivism (i.e., re-offense date), it was noted that five youth would have been in custody at the time of their re-offense. One of these instances occurred in the non-sexual offending sample and four occurred in the sexual offending sample. Once these 5 youth were excluded, the final samples included 185 non-sexual and 204 sexual offending youth.

The mean age of youth in the sexual and non-sexual offending samples was 15.16 (SD = 1.36) and 15.83 (SD = 1.10), respectively. This age difference was significant: t(387) = 5.31, p < .001. While this small age difference of roughly one-half year was statistically different, there is no theoretical reason to expect that this would affect interpretation of the YLS/CMI profiles or case management practices of probation officers. Age for both groups ranged from 12 to 17. Ethnic background revealed similarities in composition with the majority of youth having a Caucasian background in both the sexual (n = 81, 39.7%) and non-sexual (n = 78, 42.2%) offending samples. A small minority of Aboriginal-Canadian youth were also found in the sexual (n = 12, 5.9%) and non-sexual (n = 20, 10.8%) samples. Unfortunately, significant amounts of data were missing for the ethnic background for both the sexual (n = 100, 49.0%) and non-sexual (n = 74, 40.0%) offending youth. Other demographic information was not available in the dataset.

Measures

YLS/CMI

The YLS/CMI is a standardized risk assessment tool designed to identify recidivism risk in youth (Hoge & Andrews, 2006). It identifies the most important risk and need factors for each youth, and aids with the development of a case management plan. The instrument is a 42-item checklist on which each item is marked as either present or absent. The items are divided into eight categories that correspond to the eight domains of the RNR model: Prior and Current Offenses/Dispositions, Family Circumstance/Parenting, Education/Employment, Peer Relations, Substance Abuse, Leisure/Recreation, Personality/Behavior, and Attitudes/Orientation. The risk level is calculated by adding up all the items that were marked as present. A total score of 0 to 8 is considered low risk, 9 to 22 considered moderate, 23 to 34 considered high, and 35 to 42 considered very high. If the assessor does not believe that the risk score correctly represents the youth, he or she has the discretion to alter the score through professional override. Research findings have reported adequate to strong internal consistency (Catchpole & Gretton, 2003; Thompson & Putnins, 2003), test–retest reliability (Thompson & Putnins, 2003), inter-rater reliability (Schmidt, Hoge, & Gomes, 2005), and concurrent validity with other behavioral measures of functioning (Jung & Rawana, 1999; Schmidt et al., 2005). It has also obtained strong overall predictive validity in a recent meta-analysis and has been the most frequently studied youth risk assessment measure (Olver et al., 2009). The youth included in this dataset were assessed prior to the publication of the YLS/CMI 2.0 (Hoge & Andrews, 2011).

Recidivism Data

In addition to the YLS/CMI scores, recidivism data for each youth, ending on March 21, 2012, were included. Recidivism data were obtained from an existing ministry database. Thus, minimal coding was required. Existing recidivism information in the dataset was initially coded by offense, with 26 possible offense types (e.g., drug possession, weapons offenses, breach, etc.). These 26 offense types were recoded into four offense categories, including violent, non-violent, sexual, and technical offenses. Two independent raters completed the data coding for the entire sample with discrepancies resolved through consensus agreement. For coding purposes, violent offenses included murder, manslaughter, attempted murder, assault, assault causing bodily harm, aggravated assault, assault with weapon, assault of a Peace Officer, robbery, kidnapping, possession and/or use of a weapon, arson, criminal harassment, and uttering threats. This definition is consistent with that used by the Uniform Crime Reporting survey across Canada and has been used in previous research (Schmidt et al., 2005). Sexual offenses consisted of sexual assault, sexual interference, or sexual offense (any form), including the making or possession of child pornography and related offenses. Technical offenses included breach or failure to comply, obstruction of justice, and traffic offenses. All other offenses, such as break and enter, causing damage to property, and drug offenses were categorized as non-violent. As the individual convictions for each youth were already coded and within the dataset provided by the ministry, a high rate of reliability (absolute agreement of 99%) was obtained for grouping these offenses into the four categories used in this study.

Time to first re-offense was also calculated for these samples. To control for the effect of incarceration on the community time available to re-offend, all custody time during the follow-up period was deducted from the number of follow-up days. The time to re-offense and length of time in custody were from official ministry records and did not need to be coded for the purpose of this study. According to the Youth Criminal Justice Act (YCJA), which is the current protocol for youth involved in the justice system in Canada, once a youth has served two thirds of their custodial sentence, they are released for re-integration into the community. Given the early release and re-integration practice for youth, each custody sentence was reduced by one third to take into account this practice and calculate time in the community. The mean follow-up time to re-offense for youth in the sexual offending sample was 957.41 days (SD = 127.86, range = 721-1,168) and for youth in the non-sexual offending sample was 937 days (SD = 136.53, range = 586-1,164).

Procedures

Probation officers completed the YLS/CMI for each youth as mandated by Ontario youth correctional services. In Ontario, each probation officer is trained in the use of the YLS/CMI and completes this instrument as part of routine case management protocols. This dataset contained information on each youth’s YLS/CMI scores (individual items, sub-scale scores, and total scores), risk level, and use of the override feature. It also contained basic demographic information (age, gender) and offense characteristics, including length of disposition. In addition, unique variables regarding the presence of specific concerns related to mental health, suicide risk, and substance abuse, as determined by probation officer’s clinical judgment, were also available. These latter special case management items are independent of the YLS/CMI items and were flagged at roughly the following prevalence rates for youth: 12% (n = 45) with mental health issues, 13% (n = 49) with suicide risk issues, and 20% (n = 79) with substance abuse needs. Other case management items available on the YLS/CMI were not present (e.g., such as protective factors), and there were no repeat YLS/CMI assessments in the dataset. Research approval for this study was obtained from the Ministry of Children and Youth Services, and ethics approval received from Lakehead University.

Results

Recidivism Profiles and Overall Risk Level

All data were screened for normality and outliers. For all variables, standard error of kurtosis was within ± 2, and skewness within ±3. Youth in the sexual offending sample re-offended at a significantly lower rate, χ²(1, N = 389) = 31.42, p < .01, than youth in the non-sexual offending sample (34.8% vs. 63.2%). The lower recidivism rate in the sexually offending youth was consistent across violent, non-violent, and technical offenses. The exception to this pattern occurred with respect to sexual recidivism. While youth in the sexual offending sample committed a higher number of sexual re-offenses, this result was not statistically significant (p = .20). Sample characteristics and recidivism profiles, including number of re-offenses and percent of youth re-offending in each offense category, are displayed in Table 1.

Table 1:

Comparison of Recidivism Characteristics Between the Sexual and Non-Sexual Offending Samples

	Sexual Offenders (n = 204)			Non-Sexual Offenders (n = 185)
	M	SD	%	M	SD	%
Days until re-offense***	390.11	302.44		183.77	180.32
Violent re-offenses**	0.69	2.05	18.1	1.41	3.12	30.8
Non-violent re-offenses***	0.65	2.13	19.1	2.22	3.85	45.4
Sexual re-offenses*	0.25	0.86	10.8	0.14	0.89	3.8
Technical re-offenses***	1.34	3.47	28.4	3.93	6.13	57.8

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

Consistent with recidivism profiles, youth in the sexual offending sample obtained significantly lower YLS/CMI scores when compared with youth in the non-sexual offending sample. The YLS/CMI total score of 10.95 (SD = 7.68) for the non-sexual offending sample was significantly higher than the YLS/CMI total score of 6.96 (SD = 6.58) for the sexual offending sample. A multivariate test indicated that the YLS/CMI total and sub-scale scores were significantly higher in the non-sexual offending sample, Hotelling’s Trace (1, 8) = 6.54, p < .001. Univariate comparisons across all eight sub-scales of the YLS/CMI were significantly different (p < .05), with the non-sexual offending sample consistently scoring higher. When compared with previous studies using the YLS/CMI, the average overall risk scores of the current samples were lower than a sample of youth referred for court-ordered assessments (M = 16.9, SD = 9.3; Schmidt et al., 2005), a sample of high-risk youth (M = 25.1, SD = 7.7; Olver, Stockdale, & Wong, 2012), and a general community sample of youth (M = 20.34, SD = 9.28; Luong & Wormith, 2011).

Predictive Validity of the YLS/CMI

A receiver operating characteristic (ROC), or area under the curve, analysis was performed for both samples using the YLS/CMI total score as the predictor for all possible recidivism categories. ROC values of 1.00 represent perfect prediction, whereas values around .50 represent poor or chance prediction (Rice & Harris, 1995). Rice and Harris (2005) described ROC values of .556 as small, .639 as moderate, and .714 as representing large predictive validity effect sizes.

In both samples, the ROC values were at .70 or above for all offense types. Table 2 displays the respective ROC values along with point-biserial correlation coefficients for each sample. To determine whether the YLS/CMI was better at predicting some forms of recidivism than others across the two samples, ROC predictive values for the sexual offending sample were statistically tested against the ROC values for the non-sexual offending sample. The formula for ROC comparisons across independent samples recommended by Hanley and McNeil (1983) was used to test for group differences. This formula provides a critical z score which must reach ±1.96 to achieve statistical significance at the .05 level. No statistically significant differences were found between the two samples on any of the ROC values with the following z scores obtained for violent (1.83), non-violent (0.35), sexual (0.17), and technical (0.86) re-offenses.

Table 2:

Correlation and ROC Values for the YLS/CMI Total Score With Various Forms of Recidivism in the Sexual and Non-Sexual Offending Samples

	Sexual Offenders (n = 204)			Non-Sexual Offenders (n = 185)
Recidivism Type	rpb	ROC	95% CI	rpb	ROC	95% CI
Violent	.48	.80	[.72, .89]	.30	.70	[.62, .78]
Non-violent	.38	.73	[.63, .82]	.32	.70	[.63, .78]
Sexual	.22	.74	[.65, .83]	.16	.72	[.54, .90]
Technical	.53	.81	[.75, .88]	.40	.75	[.68, .83]

Note. No significant differences observed between samples. ROC = receiver operating characteristic; YLS/CMI = Youth Level of Service/Case Management Inventory; r_pb = point-biserial correlation; CI = confidence interval.

YLS/CMI Professional Override

The override feature of the YLS/CMI was used to adjust risk levels for 74% (n = 151) of the youth in the sexual offending group and 41.6% (n = 77) of the non-sexual offending youth. In all cases, the risk level was increased. Table 3 shows in more detail the distribution of youth assigned to each risk level. Based on the YLS/CMI total score, the majority of sexual offending youth (71.1%, n = 145) fell in the low-risk category, while none fell in the very high–risk category. After override, 12.7% (n = 26) remained low risk, 28.9% (n = 59) moderate risk, roughly half (51%, n = 104) were in the high-risk category, and 7.4% (n = 15) in the very high–risk category. Moreover, following override adjustment, 26% (n = 53) of youth remained in the same risk level classification, whereas 30.4% (n = 62) and 39.7% (n = 81), respectively, were placed in risk categories that were one step or two steps higher. A total of 3.9% (n = 8) of youth were moved from the lowest risk category to the highest risk classification. A Wilcoxon signed-rank test was completed to determine whether there was a significant change in risk level classification following override. The Wilcoxon signed-rank test was conducted rather than chi-square as initial risk and final override risk levels were not independent and multiple cells contained less than five participants. A significant difference was obtained between initial risk levels based on the YLS/CMI total score and the final override risk level classification for the sexual offending sample, Z = 10.97, p < .001.

Table 3:

YLS/CMI Initial and Override Risk Level Classifications for the Sexual and Non-Sexual Offending Samples

	Sexual Offenders (n = 204)		Non-Sexual Offenders (n = 185)
	Initial Risk Level	Override Risk Level	Initial Risk Level	Override Risk Level
Risk Level	n (%)	n (%)	n (%)	n (%)
Low	145 (71.1)	26 (12.7)	84 (45.4)	38 (20.5)
Moderate	54 (26.5)	59 (28.9)	94 (50.8)	107 (57.8)
High	5 (2.5)	104 (51.0)	7 (3.8)	36 (19.5)
Very high	0 (0.0)	15 (7.4)	0 (0.0)	4 (2.2)

Note. YLS/CMI = Youth Level of Service/Case Management Inventory.

Professional override was less frequent, but still high, within the non-sexual offending sample. Using the YLS/CMI total score, roughly one half of the non-sexual offending youth (50.8%, n = 94) fell within the moderate risk category. As with the sexual offending youth, no cases were considered very high risk before adjustment. More than half of the non-sexual offending youth (58.4%, n = 108) did not receive any override adjustments. A total of 38.4% individuals (n = 71) received a risk level classification increase of one level, whereas 3.2% (n = 6) were increased by two risk levels. A Wilcoxon signed-rank test indicated a significant difference between the initial and final override risk level classifications in the non-sexual offending sample, Z = 8.48, p < .001.

To allow for the comparison of risk level classifications attained before and after override, the actuarial total score of the YLS/CMI was converted to numerical risk categories of 1 (low), 2 (moderate), 3 (high), and 4 (very high). Based upon these risk categories, the mean risk level classification for the non-sexual offending youth was 1.58 (SD = 0.57). After the override adjustment, the mean risk level increased to 2.03 (SD = 0.70). The initial risk level for the sexual offending youth of 1.31 (SD = 0.52) was increased to 2.53 (SD = 0.81) after override. The risk levels of non-sexual offending youth were significantly higher than those of sexual offending youth before override, t(387) = 4.93, p < .001, but significantly lower after override, t(387) = 6.45, p < .001.

Effect of Professional Override on Predictive Validity

Table 4 displays ROC values for the sexual offending sample based on the predictive accuracy of initial risk levels (derived from the YLS/CMI total score as described in the manual) and the override risk levels (adjusted by probation officers). For comparison purposes, ROC calculations were based on the overall risk categories, and not an actuarial score, both before and after override. This was necessary because risk classification of youth after override was only available in the form of the four main risk classifications and not through an actuarial score. To test for significant differences in ROC values within each sample after override, the procedures of Hanley and McNeil (1983) were followed. For the sexual offending sample, initial and override ROC values for violent (z = 4.85), non-violent (z = 3.85), and technical (z = 3.81) re-offenses were significantly different. ROC values based on initial risk level classifications were significantly more accurate than the ROC values obtained following override risk adjustments. This pattern was not observed for sexual recidivism (z = 1.67).

Table 4:

Comparison of YLS/CMI Predictive Accuracy Using the Initial Risk Level Classification for the Full Sample to the Sub-Samples of Unadjusted and Adjusted Risk Level With Override

Recidivism Type	Full Sample ROC (CI)	Unadjusted^a ROC (CI)	Adjusted^b ROC (CI)
Sexual sample (n = 204)
Violent	.78 [.69, .87]	.71 [.56, .86]	.59 [.46, .72]
Non-violent	.71 [.61, .81]	.62 [.46, .78]	.58 [.45, .72]
Sexual	.64 [.52, .77]	.80 [.66, .94]	.44 [.29, .59]
Technical	.74 [.66, .82]	.72 [.58, .87]	.58 [.45, .66]
Non-sexual sample (n = 185)
Violent	.65 [.56, .73]	.69 [.58, .79]	.60 [.47, .74]
Non-violent	.65 [.57, .73]	.69 [.59, .79]	.62 [.49, .74]
Sexual	.61 [.40, .83]	.82 [.53, 1.0]	.57 [.32, .81]
Technical	.70 [.62, .78]	.77 [.68, .86]	.64 [.51, .77]

Note. YLS/CMI = Youth Level of Service/Case Management Inventory; ROC = receiver operating characteristic; CI = confidence interval.

n = 53 for sexual offender and n = 108 for non-sexual offender samples.

n = 151 for sexual offender and n = 77 for non-sexual offender samples.

To better understand the influence of override on predictive validity, additional supplementary ROC analyses were completed on those youth who did and did not receive an override adjustment. For the 151 sexual offending youth who received an override risk adjustment, ROC values and resulting confidence intervals for all forms of recidivism overlapped with chance levels of predictive validity: violent (ROC = .59, CI = [.46, .72]), non-violent (ROC = .58, CI = [.45 to .72]), sexual (ROC = .44, CI = [.29, .59]), and technical (ROC = .56, CI = [.45, .66]) recidivism. The ROC values for sexual offending youth who received override adjustments are displayed in Table 4. For the 53 sexual offending youth where override was not used, ROC values continued to reflect significant values of predictive validity, above chance levels, with the following values for violent (ROC = .71, CI = [.56, .86]), sexual (ROC = .80, CI = [.66, .94]), and technical (ROC = .72, CI = [58, .87]) recidivism. In this latter sub-sample, prediction of non-violent offenses was no longer significant (ROC = .62, CI = [.46, .78]) as the confidence interval overlapped with chance prediction values.

Comparison of ROC values before and after override adjustment was non-significant for the full non-sexual offending sample, with resulting z values for violent, non-violent, sexual, and technical re-offenses of −.04, −.08, −.83, and −.60, respectively. This suggested that the predictive validity of the YLS/CMI, for the full sample, was not significantly poorer after override. However, supplementary ROC comparisons for the non-sexual offending youth who received override (n = 77) with the youth who received no override adjustment (n = 108) revealed important differences. For non-sexual offending youth who received an override adjustment, ROC values were found to overlap with chance levels of prediction across violent (ROC = .60, CI = [.47, .74]), non-violent (ROC = .62, CI = [.49, .74]), and sexual (ROC = .57, CI = [.32, .81]) re-offense outcomes. The exception to this pattern occurred with technical re-offenses, where a significant effect was still found with a ROC value of .64 (CI = [.51, .77]). In contrast, the 108 non-sexual offending youth who did not receive any override adjustment obtained significant ROC values, above chance levels, for violent (ROC = .69, CI = [.58, .79]), non-violent (ROC = .69, CI = [.59, .79]), sexual (ROC = .82, CI = [.53, 1.0]), and technical (ROC = .77, CI = [.68, .86]) recidivism. These results are summarized in Table 4. Thus, in the sub-sample of non-sexual offending youth who did not receive override adjustment, the YLS/CMI predictive validity remained significant. However, the non-sexual offending youth who did receive override adjustment obtained non-significant ROC values on violent, non-violent, and sexual recidivism outcomes.

Predictors of Professional Override

Due to the high but different levels of override in the two samples, separate predictor analyses were conducted for the sexual and non-sexual offending groups. While limited, the available predictors in the dataset were used. These included the initial YLS/CMI risk level classification and three unique dichotomous indicators of functioning as rated by probation officers indicating the presence of serious mental health problems, suicide concerns, and substance abuse. For both analyses, the initial YLS/CMI risk category (i.e., low, moderate, high, and very high) was entered in the first step, and the three additional variables were entered in the second step. Consideration was given to include the eight YLS/CMI sub-domains in the second step of the regression. However, significant problems with multicollinearity invalidated the analyses. This appeared to stem from the fact that the initial risk level score was based on the cumulative score of the YLS/CMI sub-domain scores. As a result, the YLS/CMI sub-domains could not be included. As override was always used to increase risk, youth with lower risk levels were more likely to receive override when compared with youth in high-risk categories. Of interest was whether other available factors would predict override above and beyond the initial risk level classification. For this reason, a hierarchical logistic regression with the initial risk category in the first step was used. When using this more limited set of predictors, diagnostic tests for multicollinearity and tolerance were within acceptable limits for both samples.

Bivariate correlations between predictors and override for the sexual offending sample was non-significant for mental health (r = −.06, p = .38), substance abuse (r = .05, p = .52), and suicide (r = .008, p = .91), but significant for initial risk level (r = −.28, p < .001). Model results for the sexual offending sample were significant for Step 1 when using only the initial risk category, χ²(1, N = 204) = 13.74, p < .001, but not for Step 2, χ²(3, N = 204) = 4.39, p = .22. Explained variance of 73.5% from Block 1 (Nagelkerke R² = .096) was unchanged to Block 2 (Nagelkerke R² = .125), with a final explained variance of 73%. None of the three individual predictors, beyond the initial risk level category, were significant in the logistic regression model for sexual offending youth.

A similar hierarchical logistic regression using the same predictors was also completed with the sample of non-sexual offending youth. Bivariate correlations obtained between predictors and override were significant for mental health (r = .18, p = .01), substance abuse (r = .17, p = .02), and initial risk level (r = −.25, p = .001), but not significant for suicide (r = .05, p = .52). Regression results were significant for Step 1, χ²(1, N = 185) = 12.082, p = .001, and for Step 2, χ²(3, N = 185) = 21.95, p < .001. Predictive accuracy regarding override increased from 62.7% using only the initial risk category (Block 1; Nagelkerke R² = .085), to 70.3% when the additional three predictors were included in Step 2 (Block 2; Nagelkerke R² = .226). Examination of the three individual predictors revealed that both the mental health and substance abuse indicators were significant predictors, while suicidality was not. Full results for the sexual and non-sexual offending samples are displayed in Table 5.

Table 5:

Results for Logistic Regression Predicting Override With Sexual (n = 204) and Non-Sexual (n = 185) Offending Youth

Predictor	B	SE	Wald	OR	95% CI	p value
Sexual sample
Step 1
Constant	2.56	.46	31.15	12.90		.000
Initial risk	−1.10	.30	13.36	0.33	[0.19, 0.60]	.000
Step 2
Constant	2.78	.49	32.74	16.15		.000
Initial risk	−1.39	.38	15.98	0.25	[0.17, 0.49]	.000
Mental health	1.01	.64	2.51	2.75	[0.79, 9.63]	.113
Suicide	0.28	.54	0.27	1.33	[0.46, 3.83]	.601
Substance	0.48	.67	0.54	1.64	[0.44, 6.16]	.461
Non-sexual sample
Step 1
Constant	1.15	.46	6.21	3.16		.013
Initial risk	−0.96	.29	11.28	0.38	[0.22, 0.67]	.001
Step 2
Constant	1.39	.50	7.84	4.01		.005
Initial risk	−1.52	.34	19.59	0.22	[0.11, 0.43]	.000
Mental health	1.72	.56	9.63	5.59	[1.89, 16.58]	.002
Suicide	0.23	.53	0.18	1.25	[0.44, 3.56]	.671
Substance	1.10	.37	9.03	3.00	[1.47, 6.16]	.003

Note. OR = odds ratio; CI = confidence interval.

Discussion

The current study extends and builds upon a small body of research examining professional override with sexual and non-sexual offending youth. The frequent use of override in these samples was always in the direction of increased risk classification and had a similar effect on the predictive validity of the YLS/CMI, regardless of offending history. In both samples, override decreased the predictive validity of the YLS/CMI to chance levels across violent, non-violent, and sexual recidivism outcomes. The only exception to this pattern occurred on the prediction of technical recidivism for non-sexual offending youth. Several important factors appeared to influence probation officer’s decision to use override with non-sexual offending youth, including the presence of significant mental health problems and substance abuse. Many of the findings obtained in this study are consistent with previous research and add to a growing knowledge base regarding professional override.

As expected, and consistent with Olver et al. (2009), the YLS/CMI demonstrated strong predictive validity for non-violent, violent, and technical recidivism. This was true in both the sexual and non-sexual offending samples, where ROC confidence intervals spanned across moderate to large effect sizes. Importantly, the ROC confidence interval for sexual recidivism in the sexual offending sample was also within the moderate to large effect size range. Although developed for general recidivism, the YLS/CMI appeared to perform relatively well in the prediction of sexual recidivism. This goes contrary to expectations as the YLS/CMI does not contain risk items related to sexual deviance, which is thought to be critical to sexual offending prediction (see Hanson & Morton-Bourgon, 2009). However, a meta-analysis of youth versions of the Level of Service Inventory found a small, but significant, weighted correlation of .20 with sexual recidivism (Olver et al., 2009). Moreover, Wormith et al. (2012) noted that the YLS/CMI should be given some consideration as a risk tool with sexual offending youth, given the performance of the LS/CMI with sexually offending adults. The current results suggest that the YLS/CMI could be used to complement other sexual offending risk instruments. However, within the non-sexual offending sample, the ROC confidence interval for sexual recidivism was very large and almost bordered on chance predictions (i.e., CI = [.54, .90]). This demonstrated greater unreliability and belies the relatively strong ROC value of .72 for sexual recidivism found within the non-sexual offending sample. This suggests that some caution should be exercised in the use of the YLS/CMI for the prediction of sexual recidivism with this latter group.

Perhaps surprising was the frequency of override by probation officers, contrary to that described in the YLS/CMI manual. The high use of override with the sexual offending sample (i.e., 74.0%; n = 151) may make some intuitive sense. The YLS/CMI was not developed specifically as a risk tool for sexual recidivism, and the majority of the sexual offending youth fell within the low-risk category. Given public safety concerns regarding sexual recidivism, it may not be surprising that probation officers would use override to increase risk levels as a means of raising supervision requirements to manage liability and maintain public safety. This approach to override with sexual offending youth is often encouraged as part of regular probation practices given the perceived limitations of the YLS/CMI as a sexual recidivism risk instrument. However, in doing so, the predictive validity of the YLS/CMI dropped considerably, similar to the findings of Storey et al. (2012) when override was used on the STATIC-99 with adults. In the current study, predictive validity dropped from strong levels of prediction to chance levels across all recidivism outcomes, including sexual recidivism. The implications of this finding on case management and long-term rehabilitation of youth who sexually offend are unknown. As there was no control or randomization of youth who did and did not receive override, it is not possible to accurately determine the effect of this practice on outcomes. However, supplementary analyses with the sexual offending sample, whereby ROC values were determined for sub-samples of youth who did and did not receive override, confirmed the same overall effect. The YLS/CMI performed at chance levels for those sexual offending youth who received override adjustments, while the YLS/CMI maintained significant levels of predictive validity for unadjusted risk scores with sexual offending youth. Professional override had a detrimental effect on the predictive validity of the YLS/CMI for this population group and suggests that caution should be exercised in the use of this feature.

In addition to affecting the predictive validity of the YLS/CMI, override can have a number of important implications for case management and rehabilitation of youth. According to the extensive RNR rehabilitation research, increasing a youth’s risk level without supporting evidence could have a detrimental effect on low-risk offenders due to increased probation involvement, supervision, and treatment. For low-risk youth, this mismatch between risk classification and treatment intensity may actually result in poorer outcomes (Andrews & Bonta, 2010a). It is also important to consider the possibility in this sample that the increase in risk level classification led to enhanced supervision and higher levels of interventions, which may have altered the recidivism of sexual offending youth and decreased predictive validity of the YLS/CMI. This would require further investigation of clinical practice following override use to untangle factors related to the decrease in the predictive validity of the YLS/CMI.

It was also noted that many sexual offending youth (43.6%, n = 89) received increases of two or more risk level classifications on the YLS/CMI. This differed from the non-sexual offending youth, where only 3.2% (n = 6) received such dramatic changes of more than one level in their risk level classification. The impact of such significant adjustments to risk level on supervision and case management practices also warrants investigation. While the bulk of RNR research is based on general and violent adult and youth offenders, much less has been published on sexual offending youth. However, the application of RNR principles to sexual offenders has been strongly endorsed (Hanson, Bourgon, Helmus, & Hodgson, 2009). Given the current preliminary findings and the application of RNR principles, the use of override as a case management strategy requires further replication with youth who commit sexual offenses.

More unexpected may have been the relatively high frequency of override with youth who committed non-sexual offenses. As the YLS/CMI was specifically developed and validated on this population group, it was expected that override would have been used more selectively. This is based on the recommendations in the YLS/CMI manual, which suggests using override in only “some circumstances” (Hoge & Andrews, 2011, p. 7). Despite the appropriateness of the YLS/CMI with the non-sexual offending group, probation officers used the override feature with 41.6% (n = 77) of the youth. Similar to the sexual offending group, the override feature was always used to increase risk level. While Wormith et al. (2012) found a lower rate of override with a comparable adult population group (i.e., 15.1%), override was ostensibly used to increase risk with adults in their sample also. This may reflect a bias or reluctance for professionals to decrease risk level determination due to multiple factors, including a need to manage personal liability (Ballucci, 2012) and/or a general tendency to inflate risk level estimations (Childs et al., 2014). While override is promoted as a case management prerogative with the YLS/CMI (Hoge & Andrews, 2006), the expectation is that its use would be on a more selective basis. This did not appear to be the case with this youth justice sample. It is important to note than an additional factor in the decision to always increase override classification may have stemmed from the fact that this was a relatively low-risk sample of youth when compared with the general youth justice population. A different pattern of results with less override use may have been obtained if the sample had included a greater proportion of high-risk youth.

Similar to the findings with the sexual offending sample, the predictive validity of the YLS/CMI decreased for non-sexually offending youth when override risk classification adjustments were made. These findings run parallel to those obtained by Wormith et al. (2012) with adult offenders. Again, similar to the sexual offending sample, it is important to acknowledge that this was not a randomized study and there was no information regarding the actual case management of youth within these samples. It is not possible to reach conclusions regarding how override affected the supervision or case management processes of probation officers. Given the limited research on youth justice rehabilitation, there is a need for further replication before reaching any conclusions about how these tentative findings might influence field practice. When used frequently, override can have a significant and negative effect on the predictive validity of the YLS/CMI across both sexual and non-sexual offending youth. These findings would suggest that greater guidance and direction regarding the use of override may be warranted.

Efforts to explore possible predictors of override were limited by available variables within the existing dataset. However, several possible predictors were available, including probation officers’ ratings of significant mental health, suicide, and substance abuse problems. After controlling for initial risk levels as a contributing factor for override use, the presence of serious mental health and substance abuse, but not suicide, was associated with an increased likelihood of override for the non-sexual offending sample. This would suggest that the presence of these clinical variables might have influenced probation officers’ decisions to alter risk classification. Including substance abuse as a factor related to increased risk is consistent with the criminogenic factors identified by the RNR model and fits with the available literature (Andrews & Bonta, 2010a, 2010b). However, substance use is already captured by a sub-domain of the YLS/CMI (i.e., substance abuse). It is unknown if this may have led to an over estimation of risk due to the redundancy of this particular variable with the YLS/CMI criminogenic domains. The presence of mental health issues, however, is not a criminogenic factor related to recidivism or risk (Andrews & Bonta, 2010a). Thus, use of this factor to adjust risk level through override would appear to be counter to the principles of the RNR model. This is a relevant and important issue as adherence to the RNR model improves outcomes for offenders. For example, Luong and Wormith (2011) found that a match between need and interventions for high-risk adolescent offenders resulted in a 37.9% decrease in recidivism, while an under-identification of risk and needs resulted in an 81.7% increase in recidivism. Similarly, with adolescents, Vieira, Skilling, and Peterson-Badali (2009) found improved outcomes for youth with increased adherence to the RNR model. The RNR body of research is considerable and demonstrates improved outcomes when adhering to and targeting criminogenic factors for rehabilitation (Andrews & Bonta, 2010a). It is important to note, however, that mental health issues are considered to be a responsivity factor (Andrews & Bonta, 2010a). As such, probation officers may have identified this responsivity issue in youth as a factor in adjusting risk classifications for the non-sexual offending youth. Further study of how responsivity factors influence case management practices with youth is warranted.

Study Limitations

The limitations of this study should be considered when interpreting the results. The data were obtained from an existing government dataset that could not be checked for reliability or accuracy. Second, the recidivism data for predictive validity analysis were based on existing provincial criminal records. This may have excluded criminal behavior and recidivism that occurred outside of the province. In addition, this may have introduced some biases or limitations that exist when official criminal records are used to gauge recidivism, as opposed to other means of collecting criminal behavior such as self-report data. Due to the intention to compare override with both sexual and non-sexual offending youth, females were excluded from analyses. As a result, generalization of these results to female youth should not be attempted. Further research should replicate the current study with females and more diverse samples of youth. Moreover, the non-sexual offending sample was matched on disposition outcome characteristics with the sexual offending sample. Thus, the non-sexual offending sample, which appeared to be a lower risk sample of youth, may not be fully representative of all youth who commit non-sexual offenses. The data analyses were also limited to one risk assessment instrument. This did not allow for any comparative analyses with other risk tools, which would allow for some validation of the obtained results across risk instruments or whether the use of override adjustment differed with a higher risk sample of youth. Finally, without the use of a randomized control group where override was not used, it is difficult to ascertain the influence of override on subsequent case management/supervision practices, recidivism, and resulting predictive validity results.

Conclusion

Notwithstanding the described limitations, the results of this study, which included a relatively large sample of sexual and non-sexual offending youth, raise important questions regarding case management practices. Professional override with the YLS/CMI was used frequently in practice for both sexual and non-sexual offending youth. This negatively affected the predictive validity of the YLS/CMI. As override with the YLS/CMI has not been investigated for youth, this requires further replication with this particular risk instrument, as well as with other risk measures. However, the similarity of these results to those obtained in an adult sample (Wormith et al., 2012) suggests that this may be a meaningful finding that has implications regarding the risk assessment and case management of both youth and adults within the justice system.

Footnotes

Acknowledgements

The authors wish to express their appreciation to the Ontario Ministry of Children and Youth Services (MCYS) for permission to use the data described in this study. In particular, a special thank you is extended to Ms. Nadia Mazaheri and Ms. Kathy Hill for their considerable assistance.

The views expressed in this paper do not necessarily reflect the views of the MCYS.

Fred Schmidt is the associate director for Clinical Development, Evaluation, and Research at the Children’s Center Thunder Bay where he consults to the children’s mental health, child welfare, and youth justice systems and conducts program evaluation and research. He also holds an adjunct professor position in the Department of Psychology at Lakehead University, Thunder Bay, Ontario, Canada.

Sarah M. Sinclair is a doctoral student at Lakehead University. She is currently completing her pre-doctoral residency in clinical-forensic psychology. Her research and clinical work involves development of reliable and valid measures of substance-use consequences, assessment of risk and risk classification, and evidence-based service provision for individuals involved in the justice system, in both correctional and community settings, and for those found not criminally responsible on account of mental disorder. She was supported by a Joseph-Armand Bombardier Canada Graduate Scholarship from the Social Sciences and Humanities Research Council while conducting this research. This funding source had no involvement in the study design, data collection, analysis or interpretation of data, writing the manuscript, or decision to submit the manuscript for publication.

Sólveig Thomasdóttir completed her honors bachelor degree in psychology at Lakehead University and has been actively working with and assisting the homeless population in Thunder Bay.

References

Ægisdóttit

White

M. J.

Spengler

P. M.

Maugherman

A. S.

Anderson

L. A.

Cook

R. S.

. . . Rush

J. D.

(2006). The meta-analysis of clinical judgment project: Fifty-six years of accumulated research on clinical versus statistical prediction. The Counseling Psychologist, 34, 341-382. doi:10.1177/0011000005285875

Andrews

D. A.

Bonta

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

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

Hoge

R. D.

(1990). Classification for effective rehabilitation: Rediscovering psychology. Criminal Justice and Behavior, 17, 19-52. doi:10.1177/0093854890017001004

Andrews

D. A.

Bonta

Wormith

J. S.

(2004). The Level of Service/Case Management Inventory. Toronto, Ontario, Canada: Multi-Health Systems.

Ballucci

(2012). Subverting and negotiating risk assessment: A case study of the LSI in a Canadian youth custody facility. Canadian Journal of Criminology and Criminal Justice, 54, 203-228. doi:10.1353/ccj.2012.0008

Borum

Bartel

Forth

(2006). Structured Assessment of Violence Risk in Youth (SAVRY). Odessa, FL: Psychological Assessment Resources.

Borum

Otto

Golding

(1993). Improving clinical judgment and decision making in forensic evaluation. Journal of Psychiatry & Law, 21, 35-76. Retrieved from http://scholarcommons.usf.edu/cgi/viewcontent.cgi?article=1540&context=mhlp_facpub

Catchpole

R. E.

Gretton

H. M.

(2003). The predictive validity of risk assessment with violent offenders: A 1-year examination of criminal outcome. Criminal Justice and Behavior, 30, 688-708. doi:10.1177/0093854803256455

10.

Chappell

A. T.

Maggard

S. R.

Higgins

J. L.

(2012). Exceptions to the rule? Exploring the use of overrides in detention risk assessment. Youth Violence and Juvenile Justice, 11, 332-348. doi:10.1177/1541204012470849

11.

Childs

Frick

P. J.

Ryals

J. S.

Jr. Lingonblad

Villo

M. J.

(2014). A comparison of empirically based and structured professional judgment estimation of risk using the Structured Assessment of Violence Risk in Youth. Youth Violence and Juvenile Justice, 12, 40-57. doi:10.1177/1541204013480368

12.

Chu

C. M.

Lee

Zeng

(2014). The utility of the YLS/CMI- SV for assessing youth offenders in Singapore. Criminal Justice and Behavior, 41, 1437-1457. doi:10.1177/0093854814537626

13.

Douglas

K. S.

Hart

S. D.

Webster

C. D.

Belfrage

(2013). HCR-20V3: Assessing risk of violence: User guide. Burnaby, British Columbia, Canada: Mental Health, Law, and Policy Institute, Simon Fraser University.

14.

Douglas

K. S.

Kropp

P. R.

(2002). A prevention-based paradigm for violence risk assessment. Criminal Justice and Behavior, 29, 617-638. doi:10.1177/009385402236735

15.

Dyck

H. L.

Campbell

M. A.

Schmidt

Wershler

J. L.

(2013). Youth psychopathic traits and their impact on long-term criminal offending trajectories. Youth Violence and Juvenile Justice, 11, 230-248. doi:10.1177/1541204012469414

16.

Grove

W. M.

Zald

D. H.

Lebow

B. S.

Snitz

B. E.

Nelson

(2000). Clinical versus mechanical prediction: A meta-analysis. Psychological Assessment, 12, 19-30. doi:10.1037/1040-3590.12.1.19

17.

Hanley

J. A.

McNeil

B. J.

(1983). A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology, 148, 839-843. doi:10.1148/radiol.2471071321

18.

Hanson

R. K.

Bourgon

Helmus

Hodgson

(2009). The principles of effective correctional treatment also apply to sexual offenders: A meta-analysis. Criminal Justice and Behavior, 36, 865-891. doi:10.1177/0093854809338545

19.

Hanson

R. K.

Morton-Bourgon

K. E.

(2009). The accuracy of recidivism risk assessments for sexual offenders: A meta-analysis of 118 prediction studies. Psychological Assessment, 21, 1-21. doi:10.1037/a0014421

20.

Hilterman

E. L. B.

Nicholls

T. L.

Nieuwenhuizen

C. N.

(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. doi:10.1177/1073191113498113

21.

Hoge

R. D.

Andrews

D. A.

(2006). Youth Level of Service/Case Management Inventory (YLS/CMI) user’s manual. Toronto, Ontario, Canada: Multi-Health Systems.

22.

Hoge

R. D.

Andrews

D. A.

(2011). Youth Level of Service/Case Management Inventory 2.0 (YLS/CMI 2.0) user’s manual. Toronto, Ontario, Canada: Multi-Health Systems.

23.

Jung

Rawana

E. P.

(1999). Risk and need assessment of juvenile offenders. Criminal Justice and Behavior, 26, 69-89. doi:10.1177/0093854899026001004

24.

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

25.

Marczyk

G. R.

Heilbrun

Lander

DeMatteo

(2003). Predicting juvenile recidivism with the PCL:YV, MAYSI, and YLS/CMI. International Journal of Forensic Mental Health, 2, 7-18. doi:10.1080/14999013.2003.10471175

26.

Marshall

Egan

English

Jones

R. M.

(2006). The relative validity of psychopathy versus risk/needs-based assessments in the prediction of adolescent offending behaviour. Legal and Criminological Psychology, 11, 197-210. doi:10.1348/135532505X68719

27.

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, 331-344. doi:10.1037/h0093927

28.

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

29.

Olver

M. E.

Stockdale

K. C.

Wormith

J. S.

(2014). Thirty years of research on the Level of Service Scales: A meta-analytic examination of predictive accuracy and sources of variability. Psychological Assessment, 26, 156-176. doi:10.1037/a0035080

30.

Onifade

Davidson

Campbell

Turke

Malinowski

Turner

(2008). Predicting recidivism in probationers with the Youth Level of Service/Case Management Inventory (YLS/CMI). Criminal Justice and Behavior, 35, 474-483. doi:10.1177/0093854807313427

31.

Rice

M. E.

Harris

G. T.

(1995). Violent recidivism: Assessing predictive validity. Journal of Consulting and Clinical Psychology, 63, 737-748. doi:10.1037/0022-006X.63.5.737

32.

Rice

M. E.

Harris

G. T.

(2005). Comparing effect sizes in follow-up studies: ROC area, Cohen’s d, and r. Law and Human Behavior, 29, 615-620. doi:10.1007/s10979-005-6832-7

33.

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

34.

Storey

J. E.

Watt

K. A.

Jackson

K. J.

Hart

S. D.

(2012). Utilization and implications of the STATIC-99 in practice. Sexual Abuse: A Journal of Research and Treatment, 24, 289-302. doi:10.1177/1079063211423943

35.

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

36.

Thompson

A. P.

Pope

(2005). Assessing juvenile offenders: Preliminary data for the Australian adaptation of the Youth Level of Service/Case Management Inventory. Australian Psychologist, 40, 207-214. doi:10.1080/00050060500243491

37.

Thompson

A. P.

Putnins

A. L.

(2003). Risk-need assessment inventories for juvenile offenders. Psychiatry, Psychology and Law, 10, 324-333. doi:10.1375/132187103322742158

38.

Upperton

R. A.

Thompson

A. P.

(2007). Predicting juvenile offender recidivism: Risk-need assessment and juvenile justice officers. Psychiatry, Psychology and Law, 14, 138-146. doi:10.1375/pplt.14.1.138

39.

Vieira

T. A.

Skilling

T. A.

Peterson-Badali

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

40.

Viljoen

J. L.

Mordell

Beneteau

J. L.

(2012). Prediction of adolescent sexual re-offending: A meta-analysis of the J-SOAP-II, ERASOR, J-SORRAT-II, and STATIC-99. Law and Human Behavior, 36, 423-438. doi:10.1037/h0093938

41.

Worling

J. R.

Curwen

(2001). Estimate of risk of adolescent sexual offense recidivism (Version 2.0). Toronto, Canada: Ontario Ministry of Community and Social Services.

42.

Wormith

J. S.

Hogg

Guzzo

(2012). The predictive validity of a general risk/needs assessment inventory on sexual offender recidivism and an exploration of the override. Criminal Justice and Behavior, 39, 1511-1538. doi:10.1177/009385481245574