Violence risk assessment is now widely assumed by policy makers and the public to be a core skill of the mental health professions and plays a pivotal role in mental health law throughout the world. Dangerousness to others is a principal standard for inpatient commitment, outpatient commitment, and commitment to a forensic hospital. The imposition of tort liability on mental health professionals who negligently fail to anticipate and avert a patient’s violence to others has become commonplace. Despite the pervasiveness of violence risk assessment in mental health law, research continues to indicate that the unaided abilities of mental health professionals to perform this task are modest at best. Many have suggested that making available to clinicians statistical information on the relationships between various risk factors and subsequent violent behavior is the only way to reduce the disconnect between what the law demands and what clinicians are able to provide. The MacArthur Violence Risk Assessment Study was one attempt to generate the necessary empirical information to improve clinical practice in the area of violence risk assessment. The approach to risk assessment developed in this project appears to be highly accurate when compared with other approaches to assessing risk among people hospitalized in acute-care psychiatric facilities. But it is also much more computationally complex than other approaches. Therefore, software has been developed to ease the administration of the MacArthur procedures in clinical practice.
The MacArthur Study’s General Research Strategy
The MacArthur Violence Risk Assessment Study had two core goals: to do the best “science” on violence risk assessment possible and to produce an actuarial violence risk assessment “tool” that clinicians in the world of managed mental health services could actually use. From these initial intellectual commitments, the Study evolved in six stages over the decade it took to plan, execute, and analyze the research.
Identifying Gaps in Methodology
Almost all existing studies of violence risk assessment suffer from one or more methodological problems: They considered a constricted range of risk factors, often a few demographic variables or scores on a psychological test; they employed weak criterion measures of violence, usually relying solely on arrest; they studied a narrow segment of the patient population, typically males with a history of prior violence; and they were conducted at a single site. Based on this critical examination of existing work, the MacArthur researchers designed a piece of research that could, to the greatest extent possible, overcome the methodological obstacles that had been identified. They studied a large and diverse array of risk factors. They triangulated the outcome measurement of violence, adding patient self-report and the report of a collateral informant to data from official police and hospital records. They studied both men and women, regardless of whether they had a history of violence. And they conducted the study at several sites rather than at a single site.
Selecting Promising Risk Factors
Although the MacArthur researchers lacked any comprehensive theory of violence by people with mental disorder from which they could derive hypothesized risk factors, recent studies suggested that a number of variables might be potent risk factors for violence among people with a mental disorder. The researchers assessed personal factors (e.g., demographic and personality variables), historical factors (e.g., past violence and mental disorder), contextual factors (e.g., social support and social networks), and clinical factors (e.g., diagnosis and specific symptoms). They chose what they believed to be the best of the existing measures of these variables, and where no instrument was available to adequately measure a variable, they commissioned the development of the necessary measure.
Using Tree-Based Methods
The MacArthur researchers developed violence risk assessment models based on the “classification tree” method rather than the usual linear regression method. A classification tree approach reflects an interactive and contingent model of violence, one that allows many different combinations of risk factors to classify a person at a given level of risk. The particular questions to be asked in any assessment grounded in this approach depend on the answers given to prior questions. Factors that are relevant to the risk assessment of one person may not be relevant to the risk assessment of another person. This contrasts with a regression approach in which a common set of questions is asked of everyone being assessed, and every answer is weighted to produce a score that can be used for purposes of categorization.
Creating Different Cutoffs for High and Low Risk
Rather than relying on the standard single threshold for distinguishing among cases, the MacArthur researchers decided to employ two thresholds—one for identifying higher risk cases and one for identifying lower risk cases. They assumed that inevitably there will be cases that fall between these two thresholds, cases for which any actuarial prediction scheme is incapable of making an adequate assessment of high or low risk. The degree of risk presented by these intermediate cases cannot be statistically distinguished from the base rate of the sample as a whole (therefore, they referred to these cases as constituting an average risk group).
Repeating the Classification Tree
To increase the predictive accuracy of a classification tree, the MacArthur researchers reanalyzed the cases that had been designated as “average risk.” That is, all people not classified into groups designated either as high risk or as low risk in the standard classification tree model were pooled together and reanalyzed. The logic here was that the people who were not classified in the first iteration of the analysis might be different in some significant ways from the people who were classified and that the full set of risk factors should be available to generate a new classification tree specifically for these people who were not already classified as high risk or as low risk. They referred to the resulting classification tree model as an iterative classification tree.
Combining Multiple Risk Estimates
Finally, the MacArthur researchers estimated several different risk assessment models in an attempt to obtain multiple risk assessments for each case. That is, they chose a number of different risk factors to be the lead variable on which a classification tree was constructed. In attempting to combine these multiple risk estimates, they began to conceive of each separate risk estimate as an indicator of the underlying construct of interest— violence risk. The basic idea was that patients who scored in the high-risk category on many classification trees were more likely to be violent than patients who scored in the high-risk category on fewer classification trees. (And analogously, patients who scored in the low-risk category on many classification trees were less likely to be violent than patients who scored in the low-risk category on fewer classification trees.)
Specific Research Methods in the MacArthur Study
More than 1,000 admissions were sampled from acute civil inpatient facilities in Pittsburgh, Pennsylvania, Kansas City, Missouri, and Worcester, Massachusetts. The MacArthur researchers selected English-speaking patients between the ages of 18 and 40, who were of White, Black, or Hispanic ethnicity, and who had a chart diagnosis of thought or affective disorder, substance abuse, or personality disorder. The median length of stay was 9 days. After giving informed consent to participate in the research, the patient was interviewed in the hospital by both a research interviewer and a research clinician to assess him or her on each of the risk factors.
Three sources of information were used in the MacArthur Study to ascertain the occurrence and details of a violent incident in the community. Interviews with patients, interviews with collateral individuals (i.e., persons named by the patient as someone who would know what was going on in his or her life), and official sources of information (arrest and hospital records) were all coded and compared. Patients and collaterals were interviewed twice over the first 20 weeks—approximately 4 to 5 months— from the date of hospital discharge.
Violence to others was defined to include acts of battery that resulted in physical injury, sexual assaults, assaultive acts that involved the use of a weapon, or threats made with a weapon in hand.
Results of the MacArthur Study
At least one violent act during the first 20 weeks after discharge from the hospital was committed by 18.7% of the patients in the MacArthur Study. Of the 134 risk factors measured in the hospital, approximately half had a statistically significant bivariate relation-ship with later violence in the community (p < .05). Some examples of specific risk factors that were—or were not—significantly related to violence are as follows:
- Gender: Men were somewhat more likely than women to be violent, but the difference was not large. Violence by women was more likely than violence by men to be directed against family members and to occur at home and less likely to result in medical treatment or arrest.
- Prior violence: All measures of prior violence—self-report, arrest records, and hospital records—were strongly related to future violence.
- Childhood experiences: The seriousness and frequency of having been physically abused as a child predicted subsequent violent behavior, as did having a parent— particularly a father—who was a substance abuser or a criminal.
- Diagnosis: A diagnosis of a major mental disorder— especially a diagnosis of schizophrenia—was associated with a lower rate of violence than a diagnosis of a personality or adjustment disorder. A co-occurring diagnosis of substance abuse was strongly predictive of violence.
- Psychopathy: Psychopathy, as measured by a screening version of the Hare Psychopathy Checklist, was more strongly associated with violence than any other risk factor. The “antisocial behavior” component of psychopathy, rather than the “emotional detachment” component, accounted for most of this relationship.
- Delusions: The presence of delusions—or the type of delusions or the content of delusions—was not associated with violence. A generally “suspicious” attitude toward others was related to later violence.
- Hallucinations: Neither hallucinations in general nor “command” hallucinations per se elevated the risk of violence. If voices specifically commanded a violent act, however, the likelihood of violence was increased.
- Violent thoughts: Thinking or daydreaming about harming others was associated with violence, particularly if the thoughts or daydreams were persistent.
- Anger: The higher a patient scored on the Novaco Anger Scale in the hospital, the more likely he or she was to be violent later in the community.
These are only bivariate relationships between single risk factors measured in the hospital and violence during the first 20 weeks after discharge into the community, however. The more important question is how the risk factors performed when combined as described above. The MacArthur researchers ultimately combined the results of five prediction models generated by the iterative classification tree methodology. This combination of models produced results not only superior to those of any of its constituent models but also superior to many other actuarial violence risk assessment procedures reported in the literature. Using only those risk factors commonly available in hospital records or capable of being routinely assessed in clinical practice, the researchers were able to place all patients into one of five risk classes for which the prevalence of violence during the first 20 weeks following discharge into the community was 1%, 8%, 26%, 56%, and 76%.
Violence Risk Assessment Software
To operationalize the risk assessment procedures developed in the MacArthur Violence Risk Assessment Study, five tree-based prediction models need to be constructed, each involving the assessment of many risk factors. It would clearly be impossible for a clinician to commit the multiple models and their scoring to memory, since different risk factors are to be assessed for different patients, and using a paper-and-pencil protocol would be very unwieldy. Fortunately, however, the administration and scoring of multiple tree-based models lends itself to software. In clinical use, the risk assessment instrument developed in the MacArthur Study consists simply of a series of questions that flow one to the next on a computer screen— through the various iterations of each of the models as necessary—depending on the answer to each prior question. Under a grant from the National Institute of Mental Health, the MacArthur researchers developed such a “violence risk assessment software,” the Classification of Violence Risk™.
- Banks, S., Robbins, P., Silver, E., Vesselinov, R., Steadman, H., Monahan, J., et al. (2004). A multiple models approach to violence risk assessment among people with mental disorder. Criminal Justice and Behavior, 31, 324-340.
- Monahan, J., & Steadman, H. (Eds.). (1994). Violence and mental disorder: Developments in risk assessment. Chicago: University of Chicago Press.
- Monahan, J., Steadman, H., Silver, E., Appelbaum, P., Robbins, P., Mulvey, E., et al. (2001). Rethinking risk assessment: The MacArthur study of mental disorder and violence. New York: Oxford University Press.
- Steadman, H., Mulvey, E., Monahan, J., Robbins, P., Appelbaum, P., Grisso, T., et al. (1998). Violence by people discharged from acute psychiatric inpatient facilities and by others in the same neighborhoods. Archives of General Psychiatry, 55, 393-101.
- Classification of Violence Risk (COVR)
- HCR-20 for Violence Risk Assessment
- Violence Risk Appraisal Guide (VRAG)
- Violence Risk Assessment