Research Evidence

Reliability and Internal Consistency: While the SAVRY and its component domains are not intended as ”scales”, for heuristic purposes in our initial validation samples, we analyzed the internal consistency of SAVRY Risk Total and found it to be .82 for the offenders and .84 for the community sample (Bartel, Forth & Borum, 2004). The primary issue of reliability for SPJ instruments, including the SAVRY, is interrater reliability. In one such study, using trained student raters the single rater intraclass correlation coefficient (ICC) was .81 for the SAVRY Total Score and .77 for the Summary Risk Rating (Catchpole & Gretton, 2003). Comparably, McEachran (2001) found relatively high reliability (.83) for the SAVRY Total Score and moderate coefficients (.72) for the Summary Risk Rating.

Validity: The concurrent validity of the SAVRY has been examined in relation to the Youth Level of Service/Case Management Inventory (YLS/CMI; Hoge & Andrews, 2002) and the Hare Psychopathy Checklist: Youth Version (PCL:YV; Forth et al., 2003). In our initial validation study, the SAVRY Risk Total correlated significantly with both instruments among offenders and in community samples (e.g., .89 with YLS/CMI and .78 with PCL-YV among offenders). The SAVRY protective domain was negatively correlated with both of the other measures. Furthermore, in a study by Catchpole and Gretton (2003), the SAVRY Risk Judgment correlated .64 with the YLS/CMI summary classification and .68 with the PCL:YV Total score. Although the correlations are significant, indicating the SAVRY shares variance with both these measures, it also possesses independent predictive power.

With regard to criterion validity, numerous studies have found significant correlations between SAVRY scores and various measures of violence in juvenile justice and high-risk community-dwelling populations (Bartel, Borum, & Forth, 2002; Fitch, 2002; Gretton & Catchpole, in press; Lodewijks, 2002; McEachran, 2001). In our validation sample (Bartel, Forth, & Borum), SAVRY Total Risk scores were all significantly related to behavioral measures of institutional aggressive behavior (.40) and aggressive conduct disorder symptoms (.52). Protective factors, again, were negatively related. Significant correlations have been found in others studies between SAVRY Risk Total scores and measures of violence among young male offenders in Canada (.32 in one study and .25 in another) (Catchpole & Gretton, 2003; Gretton & Abramowitz, 2002) and among high-risk Native American Youth (.56 for sample, .72 for females) (Fitch, 2002). SAVRY Summary Risk Ratings have also been found to correlate with violence in studies by McEachran (2001) (.67) and by Gretton and Abramowitz (2002) (.35).

The SAVRY also has demonstrated incremental (criterion) validity (or predictive power) beyond the YLSI and the PCL-YV. Results of hierarchical regression analyses showed adding the SAVRY improved the power of the YLSI and the PCL-YV in predicting both institutional aggressive behavior and serious aggressive conduct disorder symptoms. The SAVRY also accounted for a large proportion of the explained variance in each type of violence (Bartel, Forth, & Borum ). Using Receiver Operating Characteristic (ROC) analysis, which measures predictive accuracy in terms of relative improvement over chance, Areas under the Curve (AUCs) for the total score average about .74-.80 across studies. Interestingly, as with the univariate comparisons, the examiner judgments (summary risk rating), not made on the basis of any cutting score, consistently perform as well as, and often better than, the linear combination of the scores themselves. For example, using ROC, McEachran (2001) found an AUC for the SAVRY total score of .70, but the AUC for the SAVRY Summary Risk Rating was .89. This finding has been evident in research on other structured professional judgment tools as well, and provides some of the first empirical evidence that clinical judgments – properly structured and based on sound assessments – can achieve levels of accuracy that rival that of any other known predictors while maintaining latitude for case-specific analysis.

Finally, two studies have examined the classification link between Summary Risk Ratings and actual recidivism. Catchpole and Gretton (2003) found that those classified as LOW Risk had a 6% violent recidivism rate; MODERATE Risk had a rate of 14%; and HIGH Risk a rate of 40%. Similarly, Gretton and Abramowitz (2002) found that those with Risk Rating of LOW Risk had a 5.7% violent recidivism rate; MODERATE Risk had a rate of 13.1%; and HIGH Risk a rate of 40.4%. Of those who did recidivate, 69.7% were rated HIGH Risk; 24.2% MODERATE and only 6.1 LOW Risk. Those rated as HIGH Risk also recidivated more quickly than MODERATE or LOW Risk rated youth.
Concerning generalizability, while the results of these studies support the use of the Structured Assessment of Violence Risk in Youth (SAVRY) as a tool for assessing violence risk in adolescents, more research clearly is needed to clarify its applicability across genders and different ethnic groups. Of particular importance is the dearth of data on African-American and Latino youth, who are largely over-represented in many juvenile justice systems throughout the U.S. Two SAVRY studies are currently underway focusing on Latino youth in California Youth Authority, one specifically targeting males and the other exclusively females.

Bartel, P., Forth, A., & Borum, R. (2003). Development and concurrent validation of the Structured Assessment for Violence Risk in Youth (SAVRY). Manuscript Under Review.
Method: Retrospective analysis; 104 incarcerated male delinquents with relatively serious criminal histories; Age range from 15 to 19 years; Compared SAVRY with PCL:YV and YLSI
Results: Institutional aggressive behavior rating was predicted by only the SAVRY Individual/Clinical scale (R2 =.18); Violent versatility was predicted by only the SAVRY Total Risk scale (R2= .21). In predicting institutional aggressive behavior and the overall number of aggressive conduct disorder symptoms, the SAVRY showed incremental validity beyond the PCL-YV and the YLS/CMI.  In predicting frequency of aggressive behaviors, the performance of the SAVRY and PCL-YV was comparable.

Catchpole, R. & Gretton, H. (2003). The Predictive Validity of Risk Assessment With Violent Young Offenders: A 1-Year Examination of Criminal Outcome. Criminal Justice & Behavior, 30(6), 688-708.
Method: 74 youth (63 males and 11 females) drawn from two forensic institutions in British Columbia, Canada; Age ranged from 15-19; A little over half were White; Also used PCL:YV and YLSI; About half enrolled in a violent offender treatment program.
Results: An ROC Analysis showed the following Areas Under Curve (AUCs) for GENERAL OFFENDING: PCL:YV Score = .78; SAVRY Score = .74; YLS/CMI = .71. And for VIOLENT OFFENDING: PCL:YV Score = .71; SAVRY Score = .72; YLS/CMI = .73. On the SAVRY, those classified as LOW Risk had a 6% violent recidivism rate; MODERATE Risk had a rate of 14%; and HIGH Risk a rate of 40%.

Chapman, J., Desai, R., Falzer, P. & Borum, R. (Manuscript Under Review) Violence Risk and Race in a Sample of Youth in Juvenile Detention: The Potential to Reduce Disproportionate Minority Confinement.
Method: In this study, the authors used the Structured Assessment of Violence Risk in Youth (SAVRY) to examine the risk and protective factors of 758 young people admitted to juvenile detention centers in Connecticut for a pre-hearing assessment.
Results: A chi-square analysis revealed that significantly more African American youth were rated as being at low risk for violence compared to their white counterparts. African American and Hispanic youth more frequently initiated violent behavior at an early age and more frequently lived in violent and disorganized neighborhoods, however, they were significantly more likely to have pro-social involvement, strong attachments and bonds, and resilient personality traits. We conclude by discussing how risk this analysis of risk and protective factors can inform efforts to reduce the overrepresentation of minorities and their disproportionate confinement. Ideas for targeted release and intervention with minority youth are described.

Fitch, D. (2002). Unpublished data from doctoral dissertation. University of Texas-Clear Lake.
Method: Cross-sectional study (Violence History item removed from SAVRY); 82 high-risk Native American Youth (47 male, 35 female).
Results: Correlations between SAVRY and violence were as follows: Historical scale = .48; Social/Contextual scale = .54; Individual/Clinical scale = .48; SAVRY total = .56. When analyzed by gender the results were as follows for Females Only: Historical scale = .66; Social/Contextual scale = .74; Individual/Clinical scale = .64; SAVRY total = .72 and for Males Only: Historical scale = .45; Social/Contextual scale = .47; Individual/Clinical scale = .35; SAVRY total = .50.

Gretton, H. & Abramowitz, C. (March, 2002). SAVRY: Contribution of Items and Scales to Clinical Risk Judgments and Criminal Outcomes. Paper Presented at Presented at the American Psychology and Law Society, Biennial Conference, Austin, Texas.
Method: 176 youth (94% male) drawn from three forensic institutions in British Columbia, Canada; Criterion was 12 month offence outcome data from BC Corrections; Average age at follow-up was 18; Approximately 75% were White.
Results: Average SAVRY Total was 26.2 with s.d. of 9.47.Correlations with violent outcome were as follows: SAVRY Risk Rating = .35; SAVRY Total Score = .25; H scale = .22; S/C Scale = .13; I/C Scale = .24; Protective Scale = -.18 (all significant, except S/C). SAVRY Total Score showed highest correlation with Risk Rating = .59. An ROC Analysis showed the following Areas Under Curve (AUCs) for GENERAL OFFENDING: SAVRY Risk Rating = .66; SAVRY Total Score = .68; Subscales ranged from .62-.68. And for VIOLENT OFFENDING: SAVRY Risk Rating = .74; SAVRY Total Score = .67; Subscales ranged from .58-.67. Those rated as HIGH Risk recidivated more quickly than MODERATE or LOW Risk rated youth. On the SAVRY, those with Risk Rating of LOW Risk had a 5.7% violent recidivism rate; MODERATE Risk had a rate of 13.1%; and HIGH Risk a rate of 40.4%. Of those who did recidivate, 69.7% were rated HIGH Risk; 24.2% MODERATE and only 6.1 LOW Risk. SAVRY items showing the strongest relationship to violent outcome were as follows: History of Violence = .26; Past Supervision Failure = .24; Poor Compliance = .16; Psychopathy = .29; Negative Attitudes = .23; Poor Anger Management = .21; and Strong Attachment and Bonds (Protective)= -.22.

McEachran, A. (2001). The predictive validity of the PCL-YV and the SAVRY in a population of adolescent offenders. Unpublished master’s thesis. Simon Fraser University, Burnaby, British Columbia.
Method: Retrospective follow-up; 108 young male offenders evaluated at a youth forensic service; Criterion used was official crimes committed after reaching adulthood, generally around three-year follow-up. The sample included violent recidivists, nonviolent recidivists, and nonrecidivists- 36 in each.
Results: Correlations with official violence were as follows: PCL:YV=.46 ; SAVRY Score = .32; SAVRY Summary Ratings = .67. An ROC Analysis showed the following Areas Under Curve (AUCs): PCL:YV Score = .79; SAVRY Score = .70; SAVRY Summary Ratings = .89.