1
Violence Risk Assessment Tools
Overview and Critical Analysis
KIRK HEILBURN, KENTO YASUHARA, and SANJAY SHAH
The problems presented by violence in contemporary society have been an important consideration for many decades. As the law has turned to the behavioral and medical sciences to improve the prospects for accurately appraising and managing the risk of violent behavior, the past twenty years in particular have witnessed the development of specialized tools for the prediction and management of certain kinds of serious violence and criminal offending. This book will offer a description of some of the tools that have been developed for specific purposes in legal contexts.
The present chapter will provide a context in which the specific tools described in subsequent chapters can be considered. There are six important considerations applicable to risk assessment tools that will be addressed in the first section: context, purpose, population, parameters, approach, and applicability. Subsequent sections of the chapter will focus on approaches to risk assessment, the major features of each approach, the strengths and weaknesses of each approach, and their respective states of validation. The present chapter will not provide an in-depth review of the validation data relevant to specific tools. That will be covered in the respective chapter on each of the specialized risk assessment tools included in this book.
Context
The first consideration in conducting a violence risk assessment involves the broad context in which this appraisal is being conducted. The context influences the nature of the decision to be made, who is responsible for that decision, and the consequences that can result from differing appraisals. There are four important (and different) contexts in which a risk assessment is likely to be conducted: legal, clinical, school/workplace, and threats to protectees. Each will be described briefly. Legal contexts are those in which a legal (or quasi-legal) decision-maker must render a decision in the course of litigation or administrative tribunal. Such contexts may be in the domain of criminal, civil, or child/family proceedings (see, e.g., Melton, Petrila, Poythress & Slobogin, 2007). Decisions in this context may involve initial commitment or sentencing, release from incarceration or secure hospitalization, or steps that may be associated with a planned release (such as community notification or postsentence civil commitment for sexual offenders). Clinical contexts involve circumstances under which interventions are delivered to reduce the risk of violence or offending in the broader context of a treatment relationship. This context is sufficiently broad to include interventions in secure settings (e.g., jails, prisons, forensic hospitals) as well as those delivered using some leverage (e.g., to individuals on parole, probation, diversion from prosecution, or juvenile home-based placement) and others using no leverage at all (e.g., psychotherapy delivered on a voluntary basis to individuals in the community, in a jurisdiction in which there is a settled Tarasoff-type obligation to warn or protect). School/ workplace contexts encompass threats of harm to others that are typically not yet a part of formal legal proceedings, in which both the seriousness of the threat and the nature of the indicated risk-reduction strategies are important parts of the overall appraisal. Assessment in these settings is typically conducted in response to the concerns raised by the actions of specific individuals, rather than in a broader process involving all individuals in the setting. Finally, threats to protectees also involve the dual issues of threat seriousness and risk management. However, they differ because they occur outside a specific environment and are directed toward specific individuals (those under the protection of the Secret Service or Federal Marshalls, for example) or the kind of broader targets involved in domestic or international terrorism. Included as well in this category are threats involving potential harm to a domestic partner, where there is typically a specified victim and the initial appraisal is made by police who are called to the scene.
The subsequent chapters in this book are devoted to different specialized risk assessment tools that are almost entirely applicable in legal contexts. Accordingly, that will also be the focus of discussion in this chapter. It should be noted that a number of specialized measures referenced in this chapter and described further in this book would be appropriate for use in clinical and threats to protectees contexts, as these have been described in the present section.
Purpose
One of the important influences on contemporary conceptions of risk assessment is the risk/ needs/responsivity (RNR) model described by Canadian researchers (Andrews & Bonta, 2006; Andrews, Bonta, & Hoge, 1990; Andrews, Bonta, & Wormith, 2006). This involves the appraisal of three related domains. Risk refers to the probability that the examinee will engage in a certain kind of behavior in the future, typically either violence/violent offending, or criminal offending of any kind, with higher-risk individuals receiving more intensive intervention and management services. This kind of risk classification has typically employed static risk factors, which do not change through planned intervention, although some tools (for example, the Level of Service Inventory [LS I] measures) (see Andrews & Bonta, 2001; Andrews, Bonta & Wormith, 2004) use both static risk factors and risk-relevant needs. Needs are variables describing deficits which are related to the probability of such targeted outcomes; they are composed of dynamic risk factors (called criminogenic needs in the RNR model) or protective factors that have the potential to change through such planned intervention. Responsivity refers to the extent to which an individual is likely to respond to intervention(s) designed to reduce the probability of the targeted outcome behavior.
A comparable distinction involving risk assessment in legal contexts has been made between prediction and risk management (Heilbrun, 1997), in which it was observed that some legal decisions are best informed by a prediction of whether the individual will reoffend or otherwise behave violently. Perhaps the most frequent example of such a decision involves civil commitment, in which the court or other decision-maker must decide whether the likelihood that the individual will harm others (or self) is sufficient to justify involuntary hospitalization. There is a limited risk management component to this decision, although future decisions regarding such individuals (such as release from hospitalization) must consider risk management to a much greater extent. If the risk is sufficiently high, the court will presumably grant the petition for civil commitment. If not, the petition is likely to be denied (or granted on grounds other than risk to harm others). But under neither circumstance would the court be particularly concerned with a specific approach to lowering the violence risk. By contrast, when the court maintains jurisdiction over the individual following the decisionāfor example, when the defendant is diverted to a mental health court or committed as not guilty by reason of insanityāthen both the level of risk and the nature of the risk-relevant needs can help inform the courtās decision.
Specialized risk assessment tools are designed to provide information that is either particular to the question of prediction, or that addresses both risk and needs (with associated implications for intervention). This will be discussed in greater detail subsequently in this chapter. To our knowledge, there are no specialized tools that focus only on risk reduction. However, there is a technique (anamnestic assessment) derived from applied behavior analysis that promotes the informed selection of risk-relevant needs based on the individualās history. Such an approach can also be used with risk-needs tools, in a manner to be discussed later in this chapter.
Populations
One of the important considerations in risk assessment involves the population to which the individual being assessed actually belongs. There are important differences in base rates of violence, risk factors and protective factors, and risk-relevant interventions for differing populations. In addition, specialized tools are typically derived and validated to apply to a single population (for example, juvenile offenders) or related populations (adult offenders or insanity acquittees, sex offenders), but not across widely discrepant populations.
There are four considerations in delineating a population for risk assessment purposes. These are age, gender, mental health status, and location. Age typically refers to three distinct groups: (1) preadolescent children, (2) adolescents, and (3) adults. Gender is important because a specialized tool may not be validated for females, or may use different norms. The mental health status of a population refers to whether it is selected through assessment, intervention, or legal action as having a number of individuals with mental health problems. Since offender and mental health problems are among the populations to which risk assessment is most often applied, it is particularly important to distinguish mental health populations without formal criminal involvement at the relevant time (for example, those who are civilly committed) from general offender populations in which mental health problems are not part of the selection criteria. This can become more complex when considering populations such as defendants acquitted as Not Guilty by Reason of Insanity, which by definition include both mental health and offending among the selection criteria. Finally, the variable of location refers to the setting from which the population is drawn. Offending populations can be drawn from the community (for example, those on probation or parole) or from an incarceration setting. School and work are examples of settings with populations that are typically not offenders. There can again be increased complexity when the risk appraisal seeks to assess individuals in one setting but considers their risk for violence or offending in anotherāas when individuals in prison are appraised for their risk of violence in the community following release. The consideration of these four variablesāage, gender, mental health status, and settingāallows the identification of a population with sufficient specificity to determine whether a given risk assessment tool should be used. This choice is straightforward when the examinee is part of a specific population for which a particular tool was developed (for example, a sexual violence risk tool for an individual convicted of a sexual offense).
Parameters
The next important consideration in risk assessment involves what is being predicted, with what frequency of outcome, at what probability or category of risk, in what setting, over what period of time, and the nature of the risk factors and protective factors involved in the appraisal. It is important to specify, in the beginning, what outcome(s) will be predicted. Typically the broader the class of outcomes, the higher the base rate will be; the impact of base rates is discussed elsewhere in this chapter. It is also necessary to consider how this outcome will be measured. Common approaches to measuring outcomes such as violent behavior or violent offending include: (1) self-report, (2) the report of collateral observers, and (3) official records (for example, rearrest, rehospitalization). In the context of noncriminal community violence in the United States, the most sensitive of these measures is self-report, with collateral observation second and official records a distant third (Monahan et al., 2001; Steadman et al., 1998), but this may vary widely across jurisdictions, countries, and contexts, depending on whether the individual has a strong incentive to deny or minimize self-report, whether she or he is seen regularly by a collateral observer, and the extent to which official records are likely to reflect the occurrence of such behavior.
Next we consider the specific parameters of the behavior itself. Are we targeting serious violence only, or is more minor aggression considered as well, or only a specific type of violence (e.g., sexual reoffending)? Are verbal threats included? What of behavior (such as arson) that may be directed at property, but has the potential to harm other persons? It is also important to indicate whether the appraisal would consider a single act of the behavior as a āyes,ā or whether the specified outcome encompasses possible multiple acts of the behavior. In addition, there is an outcome period of time that the appraisal must designate. For some risk assessment, the relevant outcome period might be quite shortāperhaps no longer than 24 to 48 hours. Intermediate outcome periods often used in community and correctional outcome research in the last decade tend to range from 6 to 12 months. Much longer periods, up to 5 to 10 years, have been used i...
