The Scottish Government

« Previous | Contents | Next »

Listen

Risk Assessment and Management of Serious Violent and Sexual Offenders: A review of current issues

CHAPTER FOUR: CHOICE OF RISK ASSESSMENT TOOLS

4.1 Reliable methods of assessing risk are crucial in an area in which practitioners and their agencies may be exposed to public accountability, legal liability and media scrutiny (Carson, 1996; Monahan, 1993). Effective risk assessment and risk management are crucial to public protection and to the reduction of harm to potential victims.

This chapter will consider assessment tools in respect of two types of offenders: sex offenders; and those others who present a potential or actual risk to others through violent offending. It is important to note that risk assessment tools are subject to development and adaptation in what is a rapidly changing area, and new risk tools for violent and sexual offenders are likely to be introduced. Consequently, this review can only reflect the major assessment tools available at time of writing.

4.2 In addition, space precludes a detailed review of all the individual risk factors for sexual or violent offending. These can be found in Offenders' risk of serious harm: a literature review (Powis 2002) prepared for the Home Office Research, Development and Statistics Directorate.

SEX OFFENDERS

4.3 Grubin's review of sex offending against children (1998: 30) confirmed that the most commonly accepted broad factors for the prediction of sex offence recidivism are: 'offending history, deviant sexual arousal patterns, and previous prison sentences '. Hanson and Bussiere's meta-analysis (1998) confirmed static and historical factors such as offending history and choice of stranger victims as predictive of sex offence recidivism.

4.4 Grubin (1998) also notes that various risk assessment tools have been developed to harness this range of risk factors into useful predictive tools, but that only two have been extensively studied in America and Britain:

• the 'Rapid Risk Assessment for Sex Offence Recidivism' (RRASOR) (Hanson 1997); and,
• the 'Structured Anchored Clinical Judgement' (SACJ) ⁸ (Thornton and Travers, 1991), more recently updated into MATRIX 2000.

4.5 Other tools such as the Sex Offender Risk Appraisal Guide (SORAG) (Quinsey et al, 1998) and the Minnesota Sex Offender Screening Tool-Revised (MnSOST-R) (Epperson et al, 1998) have been developed and pursued in America and Canada. The SORAG is an adaptation of the Violence Risk Appraisal Guide (VRAG) by Quinsey et al and is principally designed for use with men convicted (or committed to mental hospitals) for offences of rape or child molestation (p.119), and is informed by a desire to distinguish appropriately for prediction purposes between variations in sex offenders and their offence preferences (p.121). Briefly summarised, their findings indicate that criminal history, gender (male), relationship to previous victims, and sexual deviance are 'strongly related to sexual and violent re-offending amongst rapists and child molesters. Offenders who are both psychopathic and sexually deviant are the most likely to recidivate'. (p.137). The SORAG comprises a fourteen-item multi-variate assessment guide that includes:

• a psychopathy score;
• criminal history score for both non-violent and violent offending;
• criminal history for sex offending;
• history of sexual offending against children or adults;
• age at index offence;
• never married;
• previous response to conditional release;
• phallometrically measured sexual deviance score;
• alcohol abuse; and,
• DSM criteria III for personality disorder.
  (Quinsey et al, 1998: 157).

4.6 Whilst initially the SORAG has not out-performed the VRAG, and has a prediction (ROC adjusted ⁹) score of 0.62 ¹⁰ (Rice and Harris 1997), Quinsey et al claim that when adjusted to include more low risk offenders, a ROC score of around 0.70 will be obtained. This they claim, coupled with the grounding of the SORAG in Hanson and Bussiere's meta-analysis (1998), will increase the predictive accuracy of the SORAG . It is currently the subject of further evaluation.

4.7 The MnSOST-R was similarly developed to assess rapists and non-familial child molesters. As with the SORAG, a multi-variate approach is used. Sixteen items, based again in those predictors most validated by meta-analysis, are generated covering sexual and non-sexual offence history, victim's age and relationship to the offender, age of offender, treatment history and previous responses, substance abuse, and unstable employment history (Epperson et al, 1998). As with the SORAG, predictive accuracy is claimed by the designers (a score of 0.45), however Hanson and Thornton (2000) interpret this cautiously (p.131), and draw attention to Epperson et al's own acknowledgement that it has yet to be fully cross-validated. Hanlon et al (1999) conducted a retrospective rating of 26 sex offenders between 1993-1994 using the MnSOST and concluded that 'Although group mean score for sexual offenders was almost fourteen points higher than that for the non-sexual offenders, groups were very small and differences not statistically significant' (p.76). In addition, Epperson et al (1995) do not recommend its use with intra-familial child molesters as the baseline recidivist rates are low and consequently false-positive rates are high.

4.8 This section therefore concentrates upon the RRASOR and SACJ based upon their more extensive evaluation (Hanson and Thornton, 2000) and their likely relevance to personnel in Scotland working with high-risk offenders. The section will also review the recent comparison of these two methods, and their combination to form a new tool, STATIC 99 (Hanson and Thornton, 2000), and the transition to MATRIX 2000.

Rapid Risk Assessment for Sex Offence Recidivism (RRASOR)

4.9 This is essentially an actuarially based tool that weights a number of key variables in terms of their predictive utility. The initial seven items were based upon Hanson and Bussiere's meta-analysis (1998), and subsequently four were substantiated as having predictive accuracy for sex offence recidivism:

• the number of past sex offence convictions or charges (with additional weight given to sex offence history);
• age of the offender less than 25;
• unrelated to victim; and,
• gender of victim
  (Hanson 1997).

4.10 These variables can be scored to produce an overall risk weighting. The ability of the tool to distinguish between high and low risk has been validated with a distinction between an 80 per cent 'low' and 'middle' risk group and a 20 per cent high-risk group (Hanson, 1997; Grubin, 1998). It has been extensively tested both on the 'developmental and validation samples' achieving a ROC adjusted score of 0.71 (Hanson and Thornton, 2000).

The Structured Anchored Clinical Judgement (SACJ)

4.11 Whilst this tool is clearly rooted in empirical research on sex offence recidivism, it seeks to avoid over-dependence upon static predictors (e.g. age, gender) and archival data (e.g. previous convictions). The tool has a somewhat more dynamic component to allow for changes in risk status over time, and operates as a three-stage 'step-wise' system rather than the 'simple summation of weighted items' (Hanson and Thornton 2000:121) with:

• Stage One: initial actuarially based screening;
• Stage Two: a more in-depth analysis of aggravating factors;
• Stage Three: careful monitoring of offender performance over time to note the impact of treatment on risky dispositions.

4.12 The first stage is designed as an initial screening of 'low', 'medium' and high-risk based upon five items:

• a current sex offence
• a past conviction(s) for a sexual offence
• past convictions for non-sexual violence
• current non-sexual violent offences
• four or more previous convictions of any sort.
  (Hanson and Thornton, 2000:121).

Four or more factors mean high-risk, two to three mean medium risk, and below this means low risk. Stage Two adds key dynamic factors (Hanson and Thornton, 2000:121):

• any stranger victims;
• any male victims;
• never married;
• convictions for non-contact sex offences (e.g. obscene phone calls);
• substance abuse;
• placement in residential care as a child;
• deviant sexual arousal; and,
• psychopathy, a score of 25+ on the PCL-R.

If two or more of these factors are present then the risk category is increased by one category.

4.13 Stage Three considers in-depth clinical information on treatment response and progress, and improvement on dynamic risk factors. This stage was particularly developed to monitor progress on prison treatment programmes and has been less well evaluated than stages 1 and 2. In addition, stages 2 and 3 are heavily dependent upon the availability of clinical data and information on dynamic factors. To compensate for this difficulty, a shortened version of the SACJ using stage 1 and the first four variables of stage 2 and known as SACJ -MIN can be used (Hanson and Thornton, 2000). The SACJ-MIN has been validated on approximately 500 sex offenders released from HM Prisons in 1979 and subjected to a 16-year follow-up. In this sample, 'the SACJ-MIN correlated 0.34 with sex offence recidivism and 0.30 with any sexual or violent recidivism' although the tool has yet to be extensively tested outside the United Kingdom prison population (Hanson and Thornton, 2000:122).

4.14 SACJ-MIN is already in extensive use in police sex offender assessments in registration units, and to a more limited extent in multi-agency public protection assessments in England and Wales (Maguire et al, 2001). The Association of Chief Police Officers working party on sex offender risk assessment has recommended the adoption of the SACJ has an initial screening tool (ACPO, 1999).

Combining RRASOR and SACJ-MIN: the development of STATIC 99

4.15 RRASOR and the SACJ-Min were compared in four diverse samples from the United Kingdom and Canada, and 'showed roughly equivalent predictive accuracy' (Hanson and Thornton, 2000:119). Both scales have since been combined to produce STATIC 99 (Hanson and Thornton 1999). Data from the four same samples indicates that STATIC 99 outperformed both the RRASOR and SACJ-MIN, although Hanson and Thornton state that the 'incremental improvement of the STATIC 99 …was relatively small' (p.129), with a ROC adjusted score of 0.71 for sexual recidivism, and a ROC adjusted score of 0.69 for violent (including sexual) recidivism (p.129). In comparison to other methods, STATIC 99 has a similar predictive accuracy as the SORAG based upon one data set only, but does not outperform the MnSOST-R. The latter has not however been subjected to cross-validation.

STATIC 99 is a developing tool, and Hanson and Thornton note that 'actuarial risk scales can improve on STATIC 99 by including dynamic (changeable) risk factors as well as additional static variables' (p. 131). Three additional indicators of sexual deviance, repetitive victim choice and early onset of sexual offending, are suggested. Hanson and Harris (1998; 2000) have completed further work on dynamic risk factors, distinguishing between acute and stable factors in a tool called the Sex Offender Need Assessment Rating (SONAR) to enable targeted risk management plans. This work and MATRIX 2000 (discussed below) has been drawn on in the development of the TAY PROJECT risk assessment tool in Scotland (TAYPREP30) (Tay Project 2001). This provides a structured assessment tool drawing on both dynamic and actuarial risk factors but has yet to be subjected to the level of evaluation of MATRIX 2000. This development also highlights that there is progress on sex offender risk assessment tools in Scotland but that developments are often localised and can result in inconsistent practice, a situation lamented by the Social Work Inspectorate's report Managing the Risk (SWSI, 2000).

MATRIX 2000

4.16 Since the comparison of three actuarial scales by Hanson and Thornton (2000) and the development of STATIC 99 for use in Canada, Thornton has updated the SACJ into MATRIX 2000. The tool represents an important improvement on the SACJ as it provides for greater accuracy and refinement in the identification of high-risk offenders, and offers two versions, one for sex offenders and one for violent offenders. Whilst the tool has not yet been subject to extensive published evaluations, it has been validated retrospectively against a twenty-year follow up of reconvictions and identified a very high-risk group (comprising 13% of the sample), of whom 60 per cent were reconvicted. This type of categorisation enables more accurate targeting of high-risk offenders. Similar findings have been found for a sample of violent offenders (Grubin 2000). The tools have however, been developed and validated against male offenders and often male prisoners, and may have a limited transferability to other groups. As Cooke et al (2001) have argued risk tools must be able to predict well in the community as well in institutions. These tools are also designed to predict recidivism and not levels of harm per se, a key concern to staff tasked with decisions about release, community location, treatment interventions and victim safety.

USEFULNESS OF SEX OFFENDER TOOLS TO POLICE SERVICE RISK ASSESSMENTS

4.17 Whilst all the tools have relevance to the risk assessment of sex offenders, early evaluations suggest that MATRIX 2000 can best identify high-risk offenders. However, all the tools are limited to male offenders, and are designed to predict recidivism and not levels of harm. They are therefore unlikely to operate as 'stand alone' instruments and will need to be supported by detailed analysis of antecedents, behavioural patterns, and the individual situational factors and circumstances that have led to offending in the past.

VIOLENT OFFENDERS

4.18 Traditionally, unacceptable levels of unreliability have plagued violence prediction (Monahan 1981). The research literature on dangerousness and the development of assessment tools for violence prediction derive predominantly from the mental health field and reflect psychiatric concerns to predict dangerousness accurately. Assessment tools have therefore been developed largely for use with mental health in-patients, psychiatric assessments at point of sentence, or prisoners under consideration for parole. Research populations have been largely male and institutionalised, and transferability to other offenders is acknowledged as problematic (Hagell, 1998). As Hagell (1998: 69) states, the tools vary in their 'definitions, purposes and the quality of evaluation' and consequently the reliability of tools both in the field, and in terms of producing accurate predictions has been questioned (Menzies et al., 1994). Due to their empirical rooting in particular populations and particular violent behaviours and victim groups, assessment tools tend to be highly specific, and this remains a barrier to the development of a single all-embracing tool. In this situation comparative evaluations are limited as like cannot necessarily be compared with like. Pure actuarial scales in particular have low transferability across settings and groups (Cooke et al 2001). Tools have largely been generated from institutional groups, either prisons or mental hospitals, but 'the balance between individual determinants and situational determinants of violence may be different in prison settings than in community settings' (p. 116), and 'situational factors may be more influential in generating violence than individual factors' (Cooke 1991; Cooke 2000; Ditchfield 1991).

4.19 This section will examine those assessment tools most discussed in the research literature in America, Canada and the UK, and focus on those most likely to have relevance to those working with high-risk violent offenders in Scotland. The assessment tools will be discussed under three main headings: actuarial tools, structured clinical tools, and multi-factoral tools.

Actuarial violence assessment

4.20 The Violence Risk Assessment Guide (VRAG) (Quinsey et al, 1998) is the most widely used actuarial tool for violence offence recidivism (Cooke, 2000). The VRAG was developed in Canada, based upon patients detained in secure hospitals between 1965 and 1980, and has been the subject of extensive evaluation (Quinsey et al, 1998). The VRAG contains twelve items:

• Revised Psychopathy Checklist score
• Elementary School Maladjustment score
• Meets DSM III criteria for any personality disorder
• Age at time of index offence
• Separation from either parent (except death) under age 16
• Failure on prior conditional release
• Non-violent offence history score (using the Cormier-Lang scale)
• Never married
• Meets DSM III criteria for schizophrenia
• Most serious victim injury (from the index offence)
• Alcohol abuse score
• Female victim in the index offence
  (Quinsey et al, 1998:147).

4.21 The factors are scored using a weighting system 'that calculates the weight on the basis of how different the individual is from the base rate' (p.147). Based upon a number of evaluations (Harris et al 1993; Rice and Harris 1995; Quinsey et al 1995) the VRAG has an adjusted ROC score between 0.73 and 0.77. The VRAG score is used to assign individuals to one of nine risk categories (or 'bins' as Quinsey et al designate them) and individual's 'actual risk scores' do not differ 'by more than one 'bin' from his obtained score' (p.150).

4.22 The VRAG does, however, have recognised limits. First, the probability prediction of recidivism does not include any assessment or prediction of the nature, severity, imminence, and frequency of future violence (Cooke, 2000). Secondly, statements of probability recidivism over long time periods (for example five, seven or ten years) do not assist individual case managers in individual cases where issues of severity and imminence can be more important. Finally, the VRAG encourages assessors to ignore clinical and dynamic factors outside the 12 items even in the face of research that may show their relevance to violent behaviour (Cooke, 2000; Hart, 1999). It is difficult to see that such decisions would be defensible in the light of risk assessment failures.

4.23 The VRAG is subject to on-going evaluation (Quinsey, et al 1998) but has established a reputation for predictive accuracy (Cooke, 2000), but as with the RRASOR for sex offenders it cannot 'provide any guidance on how that risk might be managed' (Cooke et al 2001: 116). In order to bridge this gap to risk management structured tools have been preferred, with the HCR-20 for example recently adopted by the Scottish Prison Service (Cooke et al 2001).

Structured assessment tools for violence

4.24 In an important comparative study of three assessment tools: the VRAG, PCL-R and the HCR-20, Cooke et al have argued that 'risk assessment should entail more that prediction, it should entail consideration of what can be done to avert further violence in the future' (2001: 116-117). Structured assessment tools combining static actuarial factors and dynamic ones have the most efficacy in indicating treatment plans and guiding practitioner interventions, not least because they guide practitioner judgement to the 'risk factors that have received empirical support in the literature' and they engage the assessor more readily in the assessment process (p. 13). Assessments are therefore individualised but are more valid as they are rooted in 'empirically validated, structured decisions' and can take account of particular and 'idiosyncratic risk markers' (Douglas et al 1999: 156). Such tools are seen as crucial to consistency in risk practice, whilst remaining 'flexible enough to handle the diversity of human beings and the contexts in which assessments are conducted' as well as promoting 'transparency and accountability' along with the 'appropriate use of professional discretion' (Hart 1998: 125). From a number of tools on both sides of the Atlantic three tools have emerged as front runners:

• The Violence Prediction Scheme (VPS).
• The HCR-20.
• The PCL-R.

4.25 The Violence Prediction Scheme (VPS) of Webster et al, (1994) is designed for the assessment of dangerousness in high-risk men. The scheme utilises the twelve items of the VRAG (called RAG) to produce an actuarial score, combined with structured assessment of ten, largely dynamic items: antecedent history, self presentation, social and psychological adjustment, expectations and plans, symptoms, supervision, life factors, institutional management, sexual adjustment, and treatment progress (Webster et al 1994:47). The authors acknowledge that the addition of the dynamic ASSESS-LIST adds very little to the accuracy of the actuarial (V)RAG score, however they stress the importance of the structured clinical assessment for the establishment of treatability and formulation of appropriate risk management plans (Webster et al 1994:57).

4.26 The HCR-20 is a systematic model for assessing the risk of violence. The assessment combines historical factors that have a track record in predicting risk, with clinical variables such as respondent insight, attitude, motivation to change and to treatment, stability, and general symptomology. In addition, the assessment tool has the 'value-added' component of structuring the assessor's attention towards case management plans, motivation to change and individual coping mechanisms. The HCR-20 is divided into 3 sub-scales:

Historical Scale

• Previous violence
• Young age at first violent incident
• Relationship instability
• Employment problems
• Substance use problems
• Major mental illness
• Psychopathy
• Early maladjustment
• Personality disorder
• Prior supervision failure

Clinical Scale

• Lack of insight
• Negative attitudes
• Active symptoms of major mental illness
• Impulsivity
• Unresponsive to treatment

Risk Management Scale

• Plans lack feasibility
• Exposure to destabilisers
• Lack of personal support
• Non-compliance with remediation attempts
• Stress
  (Webster et al, 1997: 11)

4.27 Whilst initially formulated as an 'aide memoire' in order to make decisions transparent (pp.5, 73), the predictive validity of the HCR-20 has been evaluated (Douglas et al, 1999) with persons 'scoring above the HCR-20 median were six to thirteen times more likely to be violent than those scoring below the median' (p.917). The HCR-20 in this evaluative study was found to add incremental validity to the Psychopathy Checklist-Screening Version (PCL-SV), although the sample was restricted to civil psychiatric patients.

4.28 This research validated the importance of the History and Risk Management scales (with the Clinical scale having a limited significance to short-term risk prediction), and the dynamic factors were seen as particularly pertinent to the ongoing assessment of risk. Cooke's short review of the HCR-20 (2000) indicates that these findings have been supported by Klassen's evaluation of the Historical Scale of the HCR-20. This found a 'moderate strength correlation' to in-patient violence by civil psychiatric patients (Klassen, 1999). Further work by Strand et al (1999) revealed that the HCR-20 was related to violence, while Wintrup's study (1996) which found a moderate strength correlation to patients who committed violence after release from secure forensic settings. However, limits to this study were acknowledged. The small scale of the sample (193 patients) over a relatively short time frame (626 days) does require longer follow-up, particularly post-discharge.

4.29 In a prospective study of 41 long-term sentenced offenders in two high-security prisons, Belfrage et al (2000) found that the historical scale was of little use for high-risk men, but that there was a high predictive value for the clinical and risk management scales. These two scales can provide more sensitive discrimination for high-risk groups (p.173). Although recently adopted by the Scottish Prison Service, the tool has however been almost exclusively applied in the mental health arena. As with other methods, severity and impact of offending are less well covered (Douglas et al, 1999), the tool does however indicate key areas for treatment and intervention (Cooke et al 2001). Cooke et al have also suggested 'using variables optimised for the Scottish prison populations' (p. 115) to increase accuracy, and that:

Predictions based on variables shown to be important in previous Scottish studies were found to have substantially better predictive utility than the primary instruments where the prediction of offences against discipline were concerned. This highlights the importance of developing predictive tools that are specific to the population of concern. It suggests that actuarial scales may have low generalisability across settings and across outcomes. (p. 116).

4.30 Their report also suggests that further research is required, particularly using other sources of information on violence in order to add to the accuracy of predictions 'for the more serious individuals who were the focus of the work of the MacLean Committee' (p. 119).

4.31 Whilst not the subject of a direct comparison, the above evaluation of the HCR-20 suggests that it will out-perform the Dangerous Behaviour Rating Scheme (DBRS) (Menzies et al, 1994), a tool which includes a variety of items such as personality attributes, situational variables, triggers for violence, and inhibitors of violence. The tool was developed for the assessment of dangerousness for pre-trial forensic patients.

4.32 The DBRS is a semi-structured tool and has been subject to a rigorous six-year follow-up. The conclusion was gloomy with Menzies et al concluding that 'a standardized, reliable, generalizable set of criteria for dangerousness prediction….is still an elusive and distant objective' (Menzies et al., 1994: 25).

4.33 The identification of psychopathy and links to violence prediction has also preoccupied researchers (for instance, Hare, 1991; Hare and Hart, 1993). Hart et al, (1994) define psychopathy as a distinct personality disorder comprising interpersonal, affective and behavioural symptoms. These are expressed in terms of egocentricity, emotional coldness and manipulation of others, lack of empathy and remorse, and a tendency towards anti-social behaviour and violation of social norms. In relation to these offenders, Hare's psychopathy checklist has gained increasing currency in forensic settings as a structured interviewing tool. It has also been found effective in predicting those offenders most likely to violate parole (Hart et al, 1994), and those young male offenders most likely to re-offend (Forth et al, 1990).

4.34 The Psychopathy Check List-Revised (PCL-R) and its derivatives (the PCL:YV for adolescents and the PCL:SV 'screening version' ¹¹) is a clinical construct rating scale used in semi-structured interview rating 20 items on a 3 point scale divided into three broad categories:

Interpersonal/affective:

• Glibness/superficial charm
• Grandiose sense of self-worth
• Pathological lying
• Conning/manipulative
• Lack of remorse or guilt
• Shallow affect
• Failure to accept responsibility

Social Deviance:

• Need for stimulation/proneness to boredom
• Parasitic lifestyle
• Poor behavioural controls
• Early behavioural problems
• Lack of realistic long-term goals
• Impulsivity
• Irresponsibility
• Juvenile delinquency

Additional items:

• Promiscuous sexual behaviour
• Many short-term marital relationships
• Criminal versatility
  (Hare 1991: 1, 73-77).

4.35 Whilst initially developed from research on male forensic patients and offenders, various studies have confirmed the applicability of the PCL-R to other offender and patient populations. These include women, ethnic minorities and offenders from different cultures (Brown and Forth, 1997; Cooke, 1998; Cooke et al, 1998; Hare, 1998). Various studies have established that the PCL-R can identify psychopathy accurately amongst forensic patients (Cooke and Mitchie, 1999; Hare, 1991; McDermott et al, 2000) with interpersonal and affective items proving to be more discriminating (Cooke et al 1998). It is highly reliable when used by well-trained assessors. Meta-analyses by Salekin et al (1996) and Hemphill et al (1998) have established PCL-R as a robust risk predictor for violence recidivism, with psychopathic prisoners four times more likely to offend violently within one year of release. Harris et al (1993) found that the PCL-R was the best predictor of future violence for those released from a maximum security unit and a pre-trial psychiatric assessment centre. Subsequently the PCL-R score was integrated into the VRAG assessment criteria. Hare (2000) has stated that whilst the PCL was not designed as a measurement of violence risk, it may measure the most important factor in the risk of predatory violence, that is, psychopathy.

4.36 The PCL-R is currently the subject of an evaluation on male offenders from the Prison Service in England. The PCL screening version designed for use as a stand alone screening tool for use with forensic patients including non-criminal civil patients is currently the subject of evaluation under the MacArthur Risk Assessment study (Steadman et al, 1999). The latter forms one example of a multi-factoral approach discussed next.

Multi-factoral approaches and classification trees

4.37 The MacArthur assessment tool has also received attention, and is part of a long term study into the release of patients from acute psychiatric hospitals into the community and is used for the assessment of mentally disordered offenders (Steadman et al, 1994). The tool is extensively reviewed in Steadman et al, (1994), and its importance is the emphasis placed upon a multi-factoral approach to violence prediction. The tool subsumes risk factors to four general categories:

• dispositional factors (demographic factors such as age, gender, social class, as well as particular personality variables);
• historical factors (factors that delineate the patient's life history and would include family and employment history, as well as a history of violent behaviour by the patient);
• contextual factors (social supports and relevant social networks, access to victims and weapons); and,
• clinical factors (distinct mental or personality disorders, and factors which affect stability and personal functioning such as drug and alcohol abuse).
  (Steadman et al, 1994: 302-303).

4.38 The tool is open to criticism because the boundaries between categories may not always clear. For example, some demographic factors could be re-framed as clinical ones. Furthermore, the relative weighting of the different factors in terms of their role as predisposing factors in risk is not made clear.

4.39 The tool does, however, usefully distinguish between those risk factors which are dynamic and hence most amenable to change and intervention (e.g. contextual and clinical factors), and those which are static and unlikely to diminish (those which are demographic and historical). This can helpfully guide practitioners' interventions towards those factors most likely to change.

4.40 Classification trees are another recent example of a multi-factoral approach (Monahan et al 2000, Steadman et al 1999). In essence, the Iterative Classification Tree (ICT) takes a binary approach to risk decision making with assessors following a pre-set guide through a series of options. The questions are empirically and theoretically grounded, and each question is dependent upon the answer to the preceding one. The model starts with initial screening (for example using the PCL-R) and the classification is refined through the questioning process. The ICT is designed to assist practitioners with the use of actuarial data in clinical settings in an efficient manner (Monahan et al, 2000:312). The ICT 'partitioned 72.6 per cent of a sample of discharged psychiatric patients into one of two categories with regard to their risk of violence to others during the first 20 weeks after discharge' (p. 317). However, as Monahan et al (2000) point out, this approach can only classify individuals as either high or low risk (p.312). A number of individuals remain unclassified. As Cooke (2000) states: 'It is these individuals, whose risk level is equivocal, with whom the assessor needs most assistance' (p.154).

RELEVANCE AND SELECTION OF ASSESSMENT INSTRUMENTS

4.41 As already outlined, neither sex offenders nor violent offenders can be considered as homogeneous groups (Grubin, 1998; Walker, 1996). The range of offending, settings, and victims is diverse. This makes transferability of assessment instruments across offender groups difficult, and can also raise issues of specificity in the application of a single tool to whole categories of offenders. The area is also subject to further development and it is likely that in the course of time other tools will be introduced. Of those reviewed for violent and sexual offending:

• the VRAG is the most robust predictor of future violence per se;
• the HCR-20 provides both prediction and identification of areas pertinent to the formulation of treatment interventions and risk management strategies particularly for forensic and prison populations;
• the PCL-R has a proven track record for the identification of psychopathy and for the prediction of predatory violence across a number of offender types (including women and ethnic minorities);
• MATRIX 2000 has increased the accuracy of both the SACJ and STATIC 99 whilst retaining the dynamic character of the SACJ, and can target very high-risk offenders with acceptable levels of accuracy;
• multi-factoral approaches and classification trees are largely restricted to the forensic field and are the subject of on-going evaluation. However, their inability to deal adequately with medium risk is problematic (Cooke, 2000).

4.42 Recent research results on both sides of the Atlantic, and within Scotland (Cooke et al 2001) indicate that these assessment tools are likely to be useful to those workers undertaking tasks arising from the MacLean recommendations and recent legislative proposals in Scotland. Parallel research on recidivism of serious violent and sexual offenders (Loucks forthcoming) and the audit of risk tools (McIvor, Kemshall and Levy forthcoming) will also add to the general knowledge base about the relevance of current risk tools to this population and the actual use and performance of such tools in the field.

« Previous | Contents | Next »