Name of Tool Violence Risk Appraisal Guide-Revised (VRAG-R)
Category Violence Risk (Validated)
Author / Publisher Quinsey, Harris, Rice and Cormier
Year 2013
Description
•The Violence Risk Appraisal Guide-Revised (VRAG-R) is a 12-item actuarial risk assessment
instrument for the prediction of violent recidivism among male forensic psychiatric patients.
•The instrument was revised in 2013 in order to make it easier to score. Four of the original VRAG
items were dropped for using outdated diagnostic criteria or because they have been shown not to
be fully applicable to individuals who committed sexual offences (Hertz et al., 2019).
•The instrument utilises the clinical records as a basis for scoring rather than structured interviews
or questionnaires (Harris et al., 2015).
•The VRAG-R provides a numerical estimate of the risk of violent recidivism. It is suitable for males
aged 18 years and older who have committed serious, violent or sexual offences.
Age Appropriateness
18+
Assessor Qualifications
Professional expertise and training on instrument.
Strengths
•The VRAG-R has a large literature base.
•This tool can be used in combination with historical notes and criminal records (Thomson et al.,
2008).
Empirical Grounding
The VRAG was developed from file reviews of 618 male criminal offenders and forensic patients
who were initially being assessed for criminal responsibility, fitness to stand trial and/ or being
treated in a secure setting; this sample was followed for 7 years. Subsequently, the tool was
recalibrated with an extended sample of 800 individuals and followed over a period of 10 years
(Quinsey et al., 2006). It was revised in 2013 to make it easier to score. The results of the
developmental sample of the VRAG-R showed good predictive accuracy with an AUC of .76 (Harris,
Rice and Quinsey, 2016).
RATED page updated: July 2019
© Risk Management Authority 2019
Inter-Rater Reliability
a) UK Research •Gray et al. (2007) – the VRAG obtained a high ICC value
of .95
•Doyle and Dolan (2006) found an inter-rater reliability
value of .99 between three raters based on seven cases.
b) International Research •Rossegger et al. (2011) - the VRAG obtained an ICC
value of .95.
•Endrass et al. (2008) utilised Krippendorff’s alpha to
determine the VRAG’s inter-rater reliability. The VRAG
attained an excellent inter-rater reliability coefficient of
.89.
•Mills et al. (2007) found an ICC value of .95 for the VRAG
in a sample of incarcerated Canadian offenders.
•Olver and Sewall (2018) found the VRAG-R displayed
excellent inter-rater reliability across 35 randomly
selected double-coded cases, with an ICC value of .97.
Validation History
General Predictive Accuracy
•The VRAG has been devised for use in forensic psychiatric settings (see section ‘IV. Mentally
Disordered Offenders’). As previously mentioned, other studies have also tested its validity in
offenders without psychiatric diagnoses (Langton et al., 2007; Loza and Dhaliwal, 1997).
a) UK Research None available at present.
b) International Research •Rice, Harris and Lang (2013) developed a revised
version of the VRAG (VRAG-R), making it easier-to-score.
Both the revised version and the original VRAG yielded
high predictive accuracy with an approximate ROC of .75.
•The VRAG was administered to 52 violent offenders in a
Swiss prison to test its ability to predict misconduct. The
VRAG displayed good predictive validity for physically
violent misconduct and any misconduct (AUCs of 0.83 and
0.81 respectively); fair predictive validity was shown for
other misconduct (AUC=0.73) (Abbiati et al., 2014).
•Using a sample of 296 sex offenders followed-up over
17.6 years, Olver and Sewall (2018) found the VRAG-R
scores demonstrated moderate to large predictive
RATED page updated: July 2019
© Risk Management Authority 2019
accuracy for sexual (AUC range=60-.67) and violent (AUC
range=.70-.78) recidivism respectively.
•In the first European cross-validation study of the VRAG-
R, 534 individuals convicted of a sexual offence were
followed up for an average of 7.62 years. The VRAG-R
showed moderate to large predictive accuracy for violent,
general and sexual recidivism (AUCs of .75, .78 and .63
respectively). It was found that predictive accuracy for
sexual recidivism was only significant for those convicted
of child sexual abuse offences but not for that who
committed them against adult victims (Hertz et al., 2019).
•When applied to a sample of 597 male juvenile sexual
offenders, the VRAG-R showed potential strength in
predicting non-sexual violent recidivism. It was found,
however, that elevated offence severity and adverse
childhood experiences encumbered the predictive
accuracy of the tool, particularly in the cases of sexual
recidivism (Barra et al., 2018).
Validation History
Applicability: Females
a) UK Research •Coid et al. (2009) – the VRAG generated moderate
predictive accuracy of recidivism in a sample of female
offenders with ROC values ranging between .65 to .66.
b) International Research •Eisenbarth et al. (2012) - the VRAG demonstrated good
accuracy in predicting general recidivism in a sample of
80 German female offenders (AUC = .72).
•Hastings et al. (2011) - the VRAG was unable to predict
institutional misconduct and post-release recidivism in
female offenders.
Validation History
Applicability: Ethnic Minorities
a) UK Research •Snowden, Gray and Taylor (2010) - in a two-year follow-
up, the VRAG obtained an AUC value of .74 in predicting
violent reconvictions for offenders of Black ethnic origin.
b) International Research None available at present.
RATED page updated: July 2019
© Risk Management Authority 2019
Validation History
Applicability: Mental Disorders
a) UK Research •Doyle et al. (2012) - the VRAG moderately predicted
post-discharge violence in a sample of patients
discharged from acute mental health units (AUC = .65).
•Coid et al. (2009) - the VRAG appeared to outperform
the HCR-20 and the PCL:R, demonstrating a ROC area of
.70 for violent recidivism; the VRAG scores also predicted
acquisitive reconviction in males with a ROC of .71.
•Ho et al. (2009) - AUC analysis revealed that the VRAG
had moderate to high accuracy in predicting minor
violence (.70), serious violence (.74) and any violent
incidents (.68).
•Snowden et al. (2009) – the VRAG obtained a ROC value
of .77 in a sample of male psychiatric patients.
b) International Research •A study within an Australian clinical forensic practice
found that the revised version of the VRAG (VRAG-R)
demonstrated predictive validity for recidivism
(Brookstein, Daffern and Ogloff, 2016).
•Glover and colleagues (2017) tested the VRAG-R on a
sample of 120 male correctional individuals. Results
indicated that the VRAG-R gave moderate levels of
predictive validity for general and violent recidivism that
was able to be sustained over time.
•Camilleri and Quinsey (2011) report the VRAG has "good
predictive accuracy with psychiatric patients of lower
intelligence".
•Verbrugge et al. (2011) - the VRAG total score attained
AUC values of .79 and .92 for violent and general
recidivism respectively in a sample of 59 community-
based offenders with intellectual disabilities.
•In their retrospective study, Rice et al. (2008) showed
that the VRAG had the ability to discriminate risk between
non-intellectually disabled individuals (control) and
intellectually disabled sex offenders.
•Kröner et al. (2007) - the tool demonstrated moderate
accuracy (AUC) in predicting general (.70) and violent
recidivism (.70) in a sample of German male offenders
undergoing clinical evaluation for criminal responsibility.
RATED page updated: July 2019
© Risk Management Authority 2019
•Pouls and Jeandarme (2018) collected VRAG scores for
52 offenders with intellectual disabilities (OIDs). AUCs
were non-significant; although a trend towards
significance was evident for physical aggression
(AUC=0.74). The results show that the VRAG
overestimated risk of OIDs and was only accurate in
identifying low-risk individuals.
Contribution to Risk Practice
•The VRAG-R has the ability to create awareness of static risk factors and can prompt further
assessment of the risk of reoffending.
•The VRAG-R shows some consideration for responsivity issues (e.g. psychopathy).
Other Considerations
•Since the VRAG is composed solely of static factors, the tool does not have the capacity to inform
treatment protocol or monitor offender progress or motivation for intervention (Daffern, 2007).
•In a review of cases decided in United States federal courts, it was found that the VRAG was mainly
introduced by the prosecution as a measure of violence risk and was rarely challenged (Cox et al.,
2018).
•The VRAG was found to have high concurrent validity with SAQ total scores (Andreau-Rodriguez,
Peña-Fernández, and Loza, 2016).
•The tool also relies on PCL:R rating scores as part of the predictive measurement. A study by Doyle,
Dolan and McGovern (2002) found that the PCL:SV was a significant contributor to the predictive
validity of the VRAG.
•For more information on the VRAG-R, please visit the following website: [Link]
RATED page updated: July 2019
© Risk Management Authority 2019