Violence Risk Assessment - Criminal Psychology

Violence risk assessment is a professional task that is required in numerous legal and clinical settings. It has as its most basic premise the identification of persons who pose higher versus lower levels of risk of future violent behavior. More comprehensive approaches also facilitate decisions about why persons have acted violently, why they might do so in the future, and what risk management efforts can be put into place to reduce risk. Such decisions are made daily, thousands of times, across myriad contexts, pan-globally. Common settings include parole decisions, sentencing determinations, discharge from forensic hospitalization, civil commitment, workplace health and safety determinations, immigration hearings, law enforcement case management, community risk management of patients and offenders, bail decisions, and sexually violent predator and other indeterminate sentencing proceedings. Many different professional groups are involved in violence risk assessment: clinical psychologists, psychiatrists, psychiatric nurses, social workers, occupational health and safety workers, law enforcement officers, judges, and immigration officials. In this article, the following topics are reviewed: a descriptive overview of the task of violence risk assessment; the landscape of the most common applications of violence risk assessment; the evolution of the violence risk assessment field, including landmark legal events that spurred its growth; major approaches to the task of violence risk assessment; the essential applied tasks involved in carrying out violence risk assessments; the fundamental state of the field in terms of its empirical evaluation; and core gaps to consider moving forward.

Descriptive Overview of Violence Risk Assessment

The essence of violence risk assessment is to make decisions about another person’s future behavior. To predict the future behavior of a single individual—the task required within risk assessment contexts—is a complex endeavor. There are numerous pitfalls that can detract from its accuracy and reliability. Most prominent among these is simply the fact that the future is unknown, and there are many unknown future events and circumstances that can elevate or decrease a person’s risk for violence. Nonetheless, predictive decisions must be made for legal and ethical reasons.

Credible approaches to violence risk assessment require evaluators (e.g., clinical psychologists, social workers, law enforcement officers) to consider a set of factors that are known through research to increase the risk of violence, on average, across people. These factors are called risk factors. Typically, they are assembled together in a manual or a test, with instructions to evaluators on how to rate or code them. Such manuals or tests are called violence risk assessment instruments or measures. The tests or measures also have instructions and explanations for evaluators to produce estimates of the likelihood of future violence. Some such measures produce numeric probability estimates, whereas others produce principled, narrative categorical estimates. Regardless of this, evaluators use these measures to guide their judgments about the nature and level of risk posed by a given individual. Researchers evaluate these in terms of their reliability (i.e., Do different evaluators reach the same, or nearly the same, decisions based upon them?) and validity (i.e., To what extent do the measures help to differentiate those who go on to be violent from those who do not?).

The results of a violence risk assessment are then used directly by the evaluator or reported to a different decision maker (e.g., a judge, a parole board member, a review board member in a forensic setting), to inform any one of a number of vitally important dispositions for which the issue of risk posed by future violent behavior, or threat to public safety, is critical. For example, risk assessment decisions are considered in determining whether a person is sentenced to prison or probation, is granted parole or must remain in prison, is allowed out of the forensic hospital (and whether this must be with an escort or not), and is classified as maximum, medium, or minimum security within correctional facilities; whether a child should be taken away from his or her family in a child protection case; whether an individual with mental disorder can be civilly committed against his or her will; whether a person can be fired from a job; whether a young offender should be tried as an adult; whether a person who has committed a capital offense should be sentenced to death; or whether a person applying for immigration status within a country poses a risk to the public.

Common Applications of Violence Risk Assessment

As noted earlier, there are many settings in which violence risk assessment is required. Here, just a few major ones are described. In mental health law, a person can be involuntarily hospitalized (and hence deprived of liberty) if that person is mentally ill and poses a risk of violence to others (or self). This has been one of the most contentious legal arenas in the risk assessment field. Criticism of the mental health field as state agent spurred the evaluation of the accuracy with which mental health professionals can estimate future risk of violence. Despite legal attacks, most states, provinces, and countries permit involuntary hospitalization of a person due to mental illness in combination with the risk of violence to others (or self). The strictness of requirements varies greatly across jurisdictions. In some, risk of violence must be imminent, and the violence under consideration must be serious and must be against an identifiable person. In others, the risk must not be so clear. Under U.S. law, civil commitment law has been broadened to allow post-sentence detention of sex offenders who pose a risk to the public of continued sexual violence (the so-called sexually violent predator laws). However in traditional civil commitment, the length of stay in hospital is short (often days or weeks); in the sexually violent predator context, most detainees are hospitalized for many years.

Another common application of violence risk assessment is discharge from prison or from forensic hospitalization. In most cases, a person will have been imprisoned or hospitalized for years and becomes eligible for potential release into the community. Partial releases are also possible (e.g., day parole, supervised day leave, unsupervised day leave). In such contexts, parole board members (in the prison context) or review board members (in the forensic release context) must follow guiding legislation to determine whether the release would pose an undue risk to society. For this, they often receive risk assessment reports from psychologists, psychiatrists, or related professionals. They then integrate these reports into their own risk assessments to determine whether release poses an undue risk as defined by statute.

In such contexts, if a person is released, he or she typically will be supervised in the community by a parole officer or by a forensic mental health team. These professionals then engage in an ongoing risk assessment and management role. That is, they adjust and revise the risk assessment over time to ensure that the person under supervision does not pose an undue risk. They have the authority to send a person back to the prison or to the forensic hospital if conditions of supervision are violated or if they deem that the person poses an undue risk to the public. Risk assessment approaches that are used in these contexts must facilitate the tracking of change in risk over time, which is called dynamic risk. Some risk assessment approaches do not permit this and hence are of limited utility in these settings.

Another common application is presentence reports to court. Judges often need to know whether a person is safe to serve his or her sentence in the community (e.g., probation) or must be incarcerated to protect the public. Judges have limits placed on them (e.g., first-degree murderers will not receive community sentences, regardless of future risk). However, in some cases, they have discretion about whether a person should receive a custodial or community sentence. In such cases, violence risk assessments play an important role.

Finally, workplaces have a legal duty to protect the safety of their workers. Such threats to safety might stem from asbestos exposure, electrocution risk, or heavy machinery—or they might come from coworkers. This setting has been subjected to much less empirical scrutiny, likely because companies generally are not interested in publicizing the extent of violence perpetrated by their workers. Yet, all companies are subject to occupational health and safety law.

Evolution of the Violence Risk Assessment Field and Major Approaches to Violence Risk Assessment

Johnnie Baxstrom changed the world—that is, the risk assessment world. He also changed the lives of nearly 1,000 people in the same situation he was in. He was serving a sentence for violent crimes including robbery, and he also had a mental disorder. Once his sentence expired, New York state classified him as a dangerous mentally disordered accused, an option available to it in the 1950s and 1960s. This meant that he would be indeterminately imprisoned due to his dangerousness, despite having served his sentence. In essence, he had been predicted to be violent in the future and, despite having served his sentence, was therefore detained. Ultimately, the U.S. Supreme Court deemed that such treatment was unfair and ordered that he and close to 1,000 people in the same situation be released outright or transferred to less secure facilities.

In the mid-1960s, the sociologist Henry J. (Hank) Steadman decided to follow up these so-called dangerous mentally ill offenders to see how many of the 967 went on to commit crimes of violence against others. All 967 had been deemed to be dangerous, that is, likely to commit a serious act of violence in the future if they were released into the community. Steadman and his colleagues found, and published in a series of articles and books, that only a small minority went on to be violent. Indeed, the vast majority of these predictions of violence were simply wrong.

This observation shook the academic and mental health fields. Some argued that mental health professionals were incapable of predicting violence. People had been detained arbitrarily, and rights had been violated. At times, psychiatrists and other mental health professionals had been presumed to be able to accurately predict future violence. Yet, the Baxstrom case changed everything for the concept of dangerousness, as it was commonly referred to, and led to the modern era of violence risk assessment.

Sharp criticism of the mental health field followed. Legal commentators derided mental health professionals’ abilities in such cases. Along with critical scholarly comment and debate, numerous lawsuits called into question the ability of mental health professionals to forecast future violence with any degree of accuracy. Yet, courts generally dismissed arguments that such a feat was beyond the ken of psychiatry or other mental health fields. By the early to mid-1980s, mental health professionals or their employers were being successfully sued for not taking appropriate action to prevent foreseeable violence against third parties (see the jurisprudentially watershed case, Tarasoff v. the Regents of the University of California).

The works of Steadman and a handful of other scholars led John Monahan to declare in his 1981 book Predicting Violent Behavior: An Assessment of Clinical Techniques that only “one in three predictions were accurate.” Yet, shortly after this assertion, Monahan himself in 1984 tried to stem the flow of caustic aspersion hurled upon the mental health field. He reminded the field that although only one in three predictions of future violence was accurate, nine in 10 predictions of nonviolence were accurate. That is, although positive predictions were of questionable accuracy, negative predictions fared much better. He opined that perhaps accurate shorter term predictions were possible. He argued that a focus on validated risk factors for violence might bear fruit.

What followed was an acute increase in scholarly interest and activity about how to improve violence risk assessment. Researchers focused on which risk factors were related to violence. For instance, by the mid- to late 1980s, there were dozens of studies on the role that mental illness and psychopathy might play in understanding violence. Shortly after this risk factor stage, scholars developed and evaluated the efficacy of combining risk factors into the so-called violence risk assessment instruments.

In 1993, a 12-item violence risk assessment instrument called the Violence Risk Appraisal Guide (VRAG) demonstrated strong predictive effects in a sample of forensic patients. Estimates of future violence came in the form of numeric probability estimates (e.g., that the patient X has a 72% chance of being violent in the coming 7 or 10 years). This finding spurred the modern clinical versus actuarial debate in the violence risk assessment field that had been taken up by Paul Meehl decades earlier. The argument was that formulas (and, generally, actuarial procedures) were better able to make decisions than were human decision makers. Human decision makers, it was argued, were too susceptible to biases and heuristics to make reliable decisions about future violence (or any future behavior, for that matter).

Yet, some scholars questioned whether the VRAG was the long sought-after panacea for the ills that plagued the risk assessment field. For instance, it was developed based on a certain method (i.e., file review without interviews of patients), within a certain setting (i.e., forensic patients), using a single sample, a certain follow-up period, and a particular definition of violence that excluded, for instance, robbery. It was noted that certain potentially important risk factors were not included in the VRAG—previous violence and drug use problems, for instance. Were clinicians really going to ignore past violence and drug use when conducting a violence risk assessment? Numerous other studies suggested that these risk factors were related to violence and hence should be considered in a violence risk assessment. What if the issue was violence in the coming days to weeks rather than 7 or 10 years? And, given that the VRAG was based just on Canadian men, what about women and people outside of Canada?

Shortly after the VRAG was published, another approach to violence risk assessment emerged: the Structured Professional Judgment (SPJ) model. The most commonly used exemplar of this model is called the Historical, Clinical, Risk Management-20, the first version of which was published in 1995. According to a large survey of over 2,000 clinicians published in 2014 by Jay Singh, the VRAG is the most commonly used actuarial violence risk assessment instrument across more than 40 countries, whereas the Historical, Clinical, Risk Management-20 is the most commonly used violence risk assessment instrument of any type across these 40+ countries.

SPJ instruments such as the Historical, Clinical, Risk Management-20, like actuarial instruments, include a set of risk factors and a method of determining level of risk. However, they do not limit risk factors to those found to be important in just one study, but rather draw on the larger scientific violence literature. They also do not use equations or algorithms because such equations have yet (as of 2017) to be shown to produce generalizable findings across different samples. SPJ instruments also tend to focus more than actuarial approaches on issues such as risk management and monitoring risk over time.

Interestingly, one of the major developments was the Level of Service approach to risk assessment. This approach was never quite woven into the debates and controversies cited earlier, perhaps because its genesis was within provincial corrections in Canada and because the initial aim of the Level of Service approach was to determine the level of risk for any sort of correctional violation, not violence per se. The legal precedents and controversies unfolding in the United States were mainly within the mental health context, not within the correctional context. Nonetheless, the Level of Service system represents probably the first empirically based risk assessment process with widespread adoption. However, historically, it has not focused on violence as an outcome but, rather, any type of criminality or breach of conditions.

It should be noted that both actuarial and SPJ approaches, despite their differences, are improvements in what preceded them, which was essentially unstructured approaches that depended almost entirely on the skill and experience of the evaluator. Large discrepancies across evaluators existed. It has been demonstrated that structured approaches such as the actuarial and SPJ approaches are more reliable and valid than unstructured approaches.

Carrying Out Violence Risk Assessments

Fundamentally, when carrying out violence risk assessments, evaluators need to consider and collect information from a variety of sources, including an interview with the person of interest (if not contraindicated as in some law enforcement or workplace contexts); a thorough review of any official files (e.g., criminal history, treatment history, institutional files, employment records); observation (in the case of professionals working on inpatient units where the person of interest resides); psychological testing, if available; and interviews with collateral sources (e.g., family members, coworkers, victims).

Areas of functioning that should be covered within a risk assessment include, but are not limited to, a person’s history of violence and violent ideation, plans for the future, health and mental health history, past criminal involvement, history of incarceration, substance use history, relationships, educational and employment functioning, peer network, use of leisure time, personality and attitudes, and gang involvement. With respect to a history of violence, evaluators are interested in a variety of factors, including its frequency, severity, pattern, trajectory, victims, weapon use, who the victims are, and the ostensible motivations for violence. It is also important for evaluators to ascertain how the person felt before, during, and after any past violent incidents because there are important differences between people who enjoy being violent and those who are anxious while being violent.

Then, the evaluator collates all the information, completes one or more violence risk assessment instruments, and makes recommendations to decision makers about the level of risk and, depending on the referral question, a formulation of violence, scenarios of future violence, and risk management and reduction strategies. As mentioned previously, this is not a simple task, and the responsibility for such decisions ultimately rests with the evaluator and not with the measures he or she used. Tests and measures are just aids that professionals use.

State of the Field of the Empirical Evaluation of Violence Risk Assessment

There are numerous books, chapters, and articles that review the empirical findings on violence risk assessment. Before the 1990s, there were few published studies on the reliability and validity of violence risk assessment. As of 2017, there are hundreds of such studies. Despite some variability in findings, most research indicates that risk assessments can be conducted with a reasonable degree of reliability (i.e., agreement between raters) and produce moderate to large effect sizes with future violence. That is, they are much better than chance, but far from perfect. There is also some research demonstrating that the nonnumerical summary risk ratings produced by SPJ instruments are as or more accurate than actuarial methods. Moreover, studies also indicate that instruments designed to capture change in risk factors over time are able to do so and that such changes are associated with concomitant changes in future violence.

Core Gaps to Consider Moving Forward

As with any field, despite considerable progress, there is always more to do. For instance, there remain some contexts, such as workplace violence, where the reliability and validity of traditional measures have been subjected to much less evaluation. Research tends to suggest that risk assessment instruments perform comparably for men and women, although there is much less research with women and girls relative to men and boys. Research has also shown that most contemporary risk assessment instruments perform comparably well across different countries. However, within countries, more research is needed among ethnic minority groups. Finally, the field is in need of implementation research. Despite hundreds of studies showing reasonable reliability and validity, there are many fewer evaluations of the success of agencies to implement risk assessment approaches. Although many agencies do so well, as of 2017, the evidence remains rather anecdotal.

References:

Douglas, K. S., Hart, S. D., Groscup, J. L., & Litwack, T. R. (2014). Assessing violence risk. In I. Weiner & R. K. Otto (Eds.), The handbook of forensic psychology (4th ed.), (pp. 385–441). Hoboken, NJ: Wiley.
Douglas, K. S., Hart, S. D., Webster, C. D., & Belfrage, H. (2013). HCR-20 (version 3): Assessing risk for violence, user guide. Burnaby, BC: Mental Health, Law, and Policy Institute, Simon Fraser University.
Douglas, K. S., Hart, S. D., Webster, C. D., Belfrage, H., Guy, L. S., & Wilson, C. (2014). Historical-Clinical Risk Management-20, Version 3 (HCR-20 V3): Development and overview. International Journal of Forensic Mental Health, 13, 93–108.
Hart, S. D., Sturmey, P., Logan, C., & McMurran, M. (2011). Forensic case formulation. International Journal of Forensic Mental Health, 10, 118–126. doi:10.1080/14999013.2011.577137
Heilbrun, K., Douglas, K. S., & Yasuhara, K. (2009). Violence risk assessment: Core controversies. In L. Skeem, K. S. Douglas, & S. O. Lilienfeld (Eds.), Psychological science in the courtroom: Consensus and controversy (pp. 333–357). New York, NY: Guilford.
Litwack, T. R., Zapf, P. A., Groscup, J. L., & Hart, S. D. (2006). Violence risk assessment: Research, legal, and clinical considerations. In I. B. Weiner & A. K. Hess (Eds.), The handbook of forensic psychology (3rd ed., pp. 487–533). Hoboken, NJ: Wiley.
Monahan, J. (1981). Predicting violent behavior: An assessment of clinical techniques. Beverly Hills, CA: Sage.
Monahan, J. (1984). The prediction of violence behavior: Toward a second generation of theory and policy. American Journal of Psychiatry, 141, 10–15.
Steadman, H. J., & Cocozza, J. (1974). Careers of the criminally insane. Lexington, MA: Heath.
Steadman, H. J., & Cocozza, J. (1980). The prediction of dangerousness—Baxstrom: A case study. In G. Cooke (Ed.), The role of the forensic psychologist (pp. 204–215). Springfield, IL: Charles C. Thomas.
Tarasoff v. Regents of the University of California, 17 Cal. 3d 425, 551 P. 2d 334 (1976).
Yang, M., Wong, S. C. P., & Coid, J. (2010). The efficacy of violence prediction: A meta-analytic comparison of nine risk assessment tools. Psychological Bulletin, 136, 740–767.