Are Two Risk Assessments Better Than One?
Part 2: General Violence and Domestic Violence

      In Part 1, I described a recent study that tried to improve prediction of sex offender recidivism, using intuitive and statistical methods. For predicting sexual or violent recidivism, the Sex Offender Risk Appraisal Guide (SORAG) was the most accurate risk assessment. No combination of tools was better than the SORAG alone. Combining tools tended to increase both false negatives and false positives. This research shows that we should only use the best tool for assessing a particular risk.

      I mentioned in Part 1 a study in which combining the components of the HCR-20 tool improved prediction of physical violence. In that study, published in Psychiatric Services in October, 2003, Douglas, Ogloff, and Hart reported that combining the H(historical), C(clinical), and R(risk management) with an overall judgement of low/moderate/high risk predicted violence better than the H, C, or R alone. Does this research contradict Seto's research? No, because Seto assumed that only accurate risk assessment tools would be combined; in the Douglas study, the H and R scales did not predict violence by themselves, so there was room for improvement.

      Seto's study was all about assessments for men charged for sexual offences. Recently, MHCP researchers have developed the Ontario Domestic Assault Risk Assessment (ODARA), to assess the risk of wife assault recidivism by men who have assaulted their partners. The ODARA uses information easily available to front-line police officers and victim support workers. Could we improve the ODARA by combining it with in-depth clinical information that is not normally available to front-line personnel, but is available to MHCP staff or corrections staff?

      The answer from a new Research Department study is "yes." A statistical combination of the ODARA and the Hare Psychopathy Checklist (PCL-R) was more accurate than either tool by itself. And it turned out that the kind of violence being predicted doesn’t seem to matter much. The best predictors of violence in general (the VRAG and PCL-R) were also among the best predictors of wife assault in particular. Some specialized domestic violence risk assessments did not work nearly as well. These less accurate tools (the Danger Assessment, the Domestic Violence Screening Inventory, and the Spousal Assault Risk Assessment) were developed without the benefits of a follow-up study to validate them. In part, these tools were based on their designers' theories about the causes of wife assault, theories which underestimate the importance of antisociality in causing violence of all kinds.

      How does Seto's research apply to violent offenders in general? Research so far suggests that combining the most accurate risk assessment with any other tool will not work, and could result in less accurate assessments. In a series of studies, MHCP researchers have tried to improve the accuracy of the VRAG and SORAG by testing the addition of such things as psychotic symptoms, insight, institutional misbehavior, progress in therapy, attitudes, emotional responses, family support, and plans. Each time, we could find no improvement. Why do these additions not help the VRAG when the addition of the PCL-R helped the ODARA? The answer is that most of these things were already considered when the VRAG was constructed, and they were rejected because they did not predict violence (e.g., insight) or because they were weak predictors and did not add to the accuracy of the VRAG (e.g., institutional misbehavior).

      In conclusion, research shows that clinicians should use only the most accurate tool for assessing the risk of any form of violence recidivism. For wife assault recidivism, the most accurate tool is the ODARA. Soon, a more accurate in-depth tool will also be available. Readers are welcome to contact us for more details. For violent recidivism in general, the tool with the greatest empirical support remains the VRAG. The HCR-20, or parts of it, is also a significant predictor, but research to date indicates that combining the VRAG with any other information would result in less accurate risk assessment. In the same way that the PCL-R improves the ODARA, future research will eventually identify new risk factors to improve such tools as the VRAG, but those factors are most likely to come from some domain not yet tested in violence risk assessment: biological markers, perhaps.