The Development of the Brief Spousal Assault Form for the Evaluation of Risk (B-SAFER): A Tool for Criminal Justice Professionals

2. Overview of Risk Assessment

Intimate partner violence continues to be a serious problem in Canada, accounting for at least one quarter of all violent crimes reported to police (Canadian Centre for Justice Statistics, 2003). As a result of numerous high-profile intimate partner homicides, law enforcement officers are under increased pressure to conduct systematic assessments to determine whether people who are being charged with intimate partner violence pose a high risk of serious or life-threatening violence. However, there exist few user-friendly tools to assist the police and other criminal justice professionals with this task.

Three models, or methods, of violence risk assessment have been discussed in the literature: unstructured clinical decision-making, actuarial decision-making, and structured professional judgment. Unstructured clinical decision-making is probably still the most widely used approach to spousal violence risk assessment (Campbell et al., 2001; Dutton & Kropp, 2000). This is a method that involves no constraints or guidelines for the evaluator. Decisions are based on the exercise of professional discretion and usually are justified according to the qualifications and experience of the professional who makes them. The approach has been widely criticized in the violence literature for lacking reliability, validity, and accountability (Litwack & Schlesinger, 1999; Quinsey et al., 1998), and has been labelled "informal, subjective, [and] impressionistic" (Grove & Meehl, 1996, p. 293). One traditional advantage of unstructured clinical decision-making is that it allows for an idiographic, or individual-centred, analysis of the offender's behaviour and a context-specific tailoring of risk management and violence prevention strategies. However, because the approach relies so heavily on professional discretion, it is vulnerable to missing important factors that require intervention. Recommendations for management strategies - if they are made at all - might be based more on the training, preferences, and biases of the evaluator rather than on: (1) well-reasoned consideration of dynamic and criminogenic (i.e., crime-relevant) risk factors; and, (2) intervention strategies that are either empirically valid or well accepted in the field. Given the widespread criticism of this approach, those working with spousal assaulters and their victims are moving away from this practice (Campbell, 1995; Dutton & Kropp, 2000; Hilton, Harris, Rice, Lang, & Cormier, in press). At the very least, practitioners should only consider risk factors that have some support in the empirical or clinical literature.

The actuarial method of risk assessment is strongly associated with the prediction paradigm popular in the violence literature (see Heilbrun, 1997). Such methods are designed to predict specific behaviours within a specific time frame. The stated goal of the actuarial method is to predict violence in: (1) a relative sense, by comparing an individual to a norm-based reference group; and, (2) an absolute sense, by providing a precise, probabilistic estimate of the likelihood of future violence. Grove and Meehl (1996, p. 293) have described this approach as "mechanical and algorithmic." The key strength to this approach is that it improves upon the poor reliability and validity of unstructured clinical assessments (Grove & Meehl, 1996; Litwack, 2001; Quinsey et al., 1998). The actuarial approach can assist the evaluator to estimate, in a relative sense, the risk posed by an individual over a fixed time period, compared to a reference group. In this sense, it is a worthwhile endeavour to develop and test actuarial instruments for spousal violence risk assessments. Indeed, several attempts have shown correlations between the actuarial approach - that is, the totalling of risk factors to produce a risk "score" - and various measures of violent behaviour and construct validity (Campbell, 1995; Grann & Wedin, 2002; Hanson & Wallace-Capretta, 2000; Kropp & Hart, 2000; McFarlane, Campbell, & Watson, 2002). In Canada, this approach has been used by the Ontario Provincial Police in the development of the Ontario Domestic Assault Risk Assessment, or ODARA (Hilton et al., in press).

Actuarial approaches have been criticized for their lack of practical utility (Douglas & Kropp, 2002; Hart, 1998; Litwack, 2001). Thus, there is an unresolved schism between science and practice. Practitioners resist using methods that eliminate professional discretion. This might be because they see their role as preventing violence rather than predicting it (Douglas & Kropp, 2002; Heilbrun, 1997). From a violence prevention perspective, actuarial methods can inform us about the overall level of risk management that might be required (i.e., the greater the risk, the greater the necessary resources). However, they do little to inform us about specific violence prevention strategies. Heilbrun (1997) contrasted "prediction versus management" models of risk assessment, noting that the prediction model likely has "minimal" implications for management due, in part, to its lack of sensitivity to change. To apply the actuarial approach properly, the evaluator is forced to consider a fixed set of factors and cannot consider unique, unusual, or context-specific variables that might require intervention (Hart, 1998). Moreover, actuarial instruments may lack a "goodness of fit" with offender treatment programs: there is incongruence between violence prevention program targets such as "attitudes towards violence" or "denial and minimization"and risk assessment instruments that fail to consider such aspects. Finally, although actuarial approaches give the appearance of objectivity and precision, they often yield very modest correlations with violence (Douglas, Cox, & Webster, 1999) and are subject to limitations such as statistical shrinkage [1] and measurement error. Moreover, practitioners may feel uncomfortable considering only one "test" of risk, while ignoring legal, ethical, and professional requirements to consider all available information, from all perspectives (American Psychological Association, 2002). Law and professional practice must change considerably before professionals can abandon discretion in favour of strict actuarial methods. Unless and until such changes occur, professionals must decide how to strike the balance between scientific rigor and respect for the uniqueness of cases. Meteorology provides a suitable analogy: no matter how well climate tables and computer models predict the weather, it is still a good idea to look outside before deciding what to wear.

Structured professional judgment is an approach that attempts to bridge the gap between actuarial and unstructured clinical approaches to risk assessment (Douglas & Kropp, 2002; Hart, 1998). The term "professional" (Kropp & Hart, 2000) is used to allow for the reality that there are non-clinical professionals (i.e., among police officers, probation officers, victim services personnel) who are often required to conduct violence risk assessments. The method has also been termed the "guided clinical approach" by Hanson (1998, p. 52). Here, the evaluator must conduct the assessment according to guidelines that reflect current theoretical, professional, and empirical knowledge about violence. Such guidelines provide the minimum set of risk factors that should be considered in every case. The guidelines will also typically include recommendations for information gathering (e.g., the use of multiple sources and multiple methods), communicating opinions, and implementing violence prevention strategies. The method is certainly more prescribed than the unstructured clinical approach, but much more flexible than the actuarial method. Structured professional judgment does not impose any restrictions for the inclusion, weighting, or combining of risk factors. Typically, however, this approach is still considerably more structured than traditional clinical prediction, providing guidance in terms of which risk factors to consider, as well as operational definitions for the scoring of the factors. The flexibility is in the final step of combining risk factors, which is not done algorithmically. Structured professional judgment does not abrogate the professional responsibility and discretion of the evaluator, but it does attempt to improve the consistency and visibility of risk judgments. In Canada, this approach has been used by the British Columbia Institute Against Family Violence (BCIFV) in the development of the Spousal Assault Risk Assessment Guide, or SARA (Kropp, Hart, Webster, & Eaves, 1994, 1995, 1999).

The primary goal of the structured professional approach to risk assessment is to prevent violence (Douglas & Kropp, 2002). By systematically identifying risk factors - particularly dynamic, or changeable, risk factors - relevant to a case, management strategies can be tailored to prevent violence. This approach has been popular in the corrections field for some time, demonstrating some success in preventing general criminal recidivism (Andrews & Bonta, 2003). Indeed, the corrections literature has long recognized the importance of identifying risk and needs factors in individuals in order to effectively manage their behaviour. It should also be noted that the structured professional approach resembles clinical practice parameters quite commonly used in medicine (Kapp & Mossman, 1996). The structured professional approach allows for a logical, visible, and systematic link between risk factors and intervention, in addition to the ability to identify persons who are at higher or lower risk for violence. It is vulnerable to some of the same criticisms as the unstructured clinical approach because it still allows considerable professional discretion. There is some evidence, however, of the reliability and validity of structured professional judgment guidelines such as the SARA (Douglas & Kropp, 2002). For example, a number of studies conducted in North America and Europe indicate that interrater reliability is good to excellent for professional judgements concerning the presence of individual risk factors and overall levels of risk (e.g., Belfrage, 1997; Kropp & Hart, 2000). Furthermore, professional judgements of risk have good criterion-related validity: they correlate substantially with scores on actuarial measures (e.g., Douglas & Webster, 1999; Kropp & Hart, 2000), they discriminate well between known groups of recidivists and non-recidivists in retrospective research (e.g., Hanson & Morton-Bourgon, 2004; Grann & Wedin, 2002; Kropp & Hart, 2000), and they predict recidivism in prospective research (e.g., Belfrage, Fransson, & Strand, 2000; Watterworth et al., 2001).


[1] Statistical shrinkage refers to incomplete replication on cross-validation in new populations.