Predicting violent recidivism

Research summary
Vol. 12 No. 3
May 2007


How well can risk assessment instruments predict violent recidivism?


Predicting future violence is one of the most important tasks of the criminal justice system. It is important to the police, the courts, prisons and community corrections for the purpose of identifying those offenders who require enhanced security, supervision and treatment. From the offender's perspective, accurate prediction is important to ensure that liberties are not unnecessarily restricted and, for the general public, personal safety is a goal of violence risk prediction.

In the field of corrections, various risk assessment instruments have been developed to forecast which offenders are likely to reoffend. Some of these instruments were specifically developed to predict violent recidivism, while others were originally developed to predict general reoffending. Some of these instruments depend upon interviewing offenders, others rely only on file information or a combination of both.

An important development in risk instruments is the inclusion of dynamic risk factors. These are risk factors that are changeable (e.g., substance abuse, employment) and therefore, can serve as targets for treatment intervention. Not all risk instruments have the potential for guiding treatment. Some risk instruments consist almost exclusively of static variables (e.g., criminal history).

The goal of the present investigation was to determine what type of risk instruments are most useful for predicting violent recidivism.


A quantitative review of studies evaluating various risk measures was conducted yielding 185 estimates of the ability of these measures to predict violence. Although over 70 different risk measures were identified the review focused on the more frequently used measures. The most frequently reported measures were the HCR-20 (a structured clinical judgment instrument for dangerousness), the PCL-R (for the diagnosis of psychopathy), the SIR and VRAG (static risk scales) and the LSI-R (a risk/need instrument designed to guide treatment).


Slightly more than half of the risk instruments were based upon file information (52.2%), while 17.4% depended on self-report, 11.2% on interviews and 16.5% of the risk instruments were based on a combination of file review and offender interview. In total, the studies reviewed included over 273,000 offenders. All four methods performed equally well in the prediction of violence.

Risk instruments that included dynamic risk factors showed a small advantage to risk instruments that were comprised mainly of static risk factors. Risk instruments that also provide targets of intervention for case management were particularly promising.

The five most commonly studied risk instruments (HCR-20, PCL-R, SIR, VRAG and LSI-R) performed similarly, although the LSI-R, PCL-R and SIR scales showed a slight advantage over the HCR-20 and VRAG to predicting violent reoffending.

Policy implications

  1. A number of commonly used risk scales predict violent recidivism and it does not appear to matter whether the instrument was developed solely for the purpose of predicting violent behaviour. However, instruments based on theoretically meaningful criminal behaviour constructs were stronger predictors of violence than instruments based on irrelevant content.
  2. Although risk instruments that relied on file extraction methods or self-report predict satisfactorily, instruments that include interview methods may gather important information not adequately captured by other methods.
  3. The risk instruments that provide information on treatment targets are the most useful as they assist in the delivery of services that reduce violent recidivism.


For further information

James Bonta, Ph.D.
Corrections Research
Public Safety Canada
269 Laurier Avenue West
Ottawa, Ontario K1A 0P8
Phone: 613-991-2831
Fax: 613-990-8295

Date modified: