The aim of this essay is to provide a historical overview of the developments on the risk assessment of sex offenders literature, focusing specifically on the advances in this area of research. Firstly, a definition of risk will be provided, followed by the aims of risk assessment and an example of the practical implications of inaccurate risk assessment. Subsequently, this essay will outline four generations of risk assessment for sex offenders, which were inspired on Bonta’s (1996) three generations of risk assessment. Particular regard will be given to the third and fourth generations, since these are the ones that reflect the recent advances in risk assessment.
Due to the heterogeneity inherent to the concept of sexual offenders, and because most of the risk assessment tools have been developed and validated on adult male sexual offenders (Craig & Beech, 2010), only instruments used with this sub-group will be discussed in this essay.
As stated by Hart, Laws and Kropp (2003), the definition of “risk” is not straightforward. For the scope of this essay, risk will be conceptualised as the probability of an offender being reconvicted for a sexual or violent offence. Risk assessment has two main aims. The first one is predicting the likelihood of reoffending of an individual, and the second one is guiding risk management, including the identification of risk factors associated with the offending behaviour and the selection of appropriate treatment programmes. Inaccurate prediction of risk can have serious consequences such as prolonged incarceration, in case of overestimation, or the release of dangerous offenders, in case of underestimation of risk (A. R. Beech, Craig, & Browne, 2009).
The first generation of risk assessment procedures will be called Clinical Judgement. This is an ideographic approach, according to which the practitioner relies on his/her professional experience to evaluate the individual’s characteristics and make a decision regarding the risk he poses. However, the processes through which a practitioner reaches a decision are rarely easily observable and are not easy to replicate (A. R. Beech, 2001).
Structured assessment tools
The subjectivity inherent to clinical judgement led to the development of more structured assessment tools. Actuarial measures are the most widely used structured instruments for risk assessment of sexual offenders.
The most commonly used risk tools with adult male sexual offenders are the RRASOR (Rapid Risk Assessment for Sexual Offense Recidivism; Hanson, 1997); SACJ (Structured Anchored Clinical Judgement; Grubin, 1998; Hanson & Thornton, 2000); Static-99 (Hanson & Thornton, 2000); RM 2000 (Risk Matrix 2000; Thornton et al., 2003); and the SORAG (Sex Offender Risk Appraisal Guide; Quinsey et al., 1998).
For the purpose of this essay, only some particularities of each tool will be discussed. The RRASOR is simple and easy to use. The fact that it was designed using data from seven different follow-up studies makes it more comprehensive than other tools. It was designed to screen offenders into levels of risk. The SACJ can be used with missing data, is brief and easy to use. However, it has been stated that it works better as a predictor of non-sexual offences (Craig, Browne, & Beech, 2008). Static-99 is a combination of RRASOR and SACJ. It is the most commonly used risk tool with adult sex offenders (Hanson & Morton-Bourgon, 2009) and it is intended to measure long-term risk potential. The RM 2000 has two counterparts: sexual and violent. It is widely used in British Prison and Probation Services (Hanson & Morton-Bourgon, 2009). The SORAG was designed to predict at least one reconviction of a sexual offence. It measures both violent and sexual recidivism.
Although the predictive accuracy of these tools commonly outperforms clinical judgement (Grove, Zald, Lebow, Snitz, & Nelson, 2000), several limitations have also been pointed. Is has been argued that the static nature of actuarial measures makes it impossible to identify the psychological factors underlying risk and hence provides no guidance for treatment and case management (Craig, Browne, & Beech, 2008). Furthermore, because actuarial tools are derived from base rates which are officially recorded as reconviction, the probabilities will underestimate true offence rates. The reason for this being, for instance, the high number of underreporting of sexual offences (Janus & Meehl, 1997) and the fact that some sexually motivated offences are recorded as violent offences (Corbett, Patel, Erikson, & Friendship, 2003). In addition, sub-groups of sexual offenders, like sexual offenders with learning disabilities, should not be assessed with the use of actuarial tools that have not been developed specifically for these populations, since different sub-groups will have different base rates (Craig & Beech, 2010).
Also, we should take into account what is called the “statistical fallacy”, according to which group data are used to make decisions on a single individual. This practice has its dangers since the practitioner might disregard unusual factors that are relevant to a particular case (A. R. Beech, Fisher, & Thornton, 2003) . Finally, actuarial instruments do not estimate long-term risk since they are based on unchangeable, historical static factors, taking no account of changeable, dynamic risk factors that might indicate imminent risk (A. R. Beech, Fisher, & Thornton, 2003).
Dynamic approaches to risk assessment
The statement presented by Hart, Laws and Kropp (2003) is illustrative of the contention around the utility of actuarial tool. The main reason for this being that these tools represent probabilities of recidivism, not a certainty (A. R. Beech, Fisher, & Thornton, 2003), which are based solely on historical factors.
This debate encouraged researchers to developed instruments which additionally integrate factors that account for the context in which offences occur, and that are amenable to change. These factors have been named dynamic risk factors or “criminogenic needs” (Andrews & Bonta, 2003).
A significant landmark in the research of dynamic risk factors was the development of a “deviancy” construct (A. R. Beech, 1998) which led to the creation of a system named STEP (Sex Offender Treatment Evaluation Project; Beech, 1998). This framework is used by Probation Services to categorise child molesters into high or low deviancy groups (A. R. Beech, Fisher, & Thornton, 2003; A. R. Beech, 1998) . Evidence for significant differences between these two groups has been found by Fisher, Beech and Brown (1999) and by Beech and Ford (2006). Moreover, a six years follow-up study suggested that deviancy contributes independently to prediction of reconviction once static risk factors had been considered (A. R. Beech, Erikson, Friendship, & Hanson, 2002). It can then be suggested, that categorisation of sexual offenders into high and low deviancy groups not only informs treatment options, but it also adds to the predictive value of static risk factors.
Another marker in risk assessment literature was the differentiation between two types of dynamic risk factors: stable and acute risk factors (Hanson & Harris, 2001; Hanson & Harris, 2000). Stable dynamic risk factors are those which represent relatively persistent characteristics of the individual that can be changed through therapy (Craig & Beech, 2010). Acute dynamic risk factors are rapidly changing factors that are useful in predicting imminent risk (Craig, Browne, & Beech, 2008). From this research, the authors created the SONAR (Sex Offender Need Assessment Ratings) which has been developed into the STABLE and ACUTE scales (Hanson & Harris, 2000). These scales, as the names suggest, rate stable and acute dynamic factors. The ACUTE scale has the particularity of including a “unique factor” at the end, which can be any factor that might trigger offending behaviour. It might be, for instance, contact with a specific person that elicits a strong emotional response in the offender. This factor illustrates the changeable, contextual nature of these scales, emphasising the importance of addressing individual differences and idiographic characteristics.
Thornton (2002) reported a further innovation in risk prediction with sexual offenders. He suggested that dynamic risk factors fall within four domains: sexual interest, distorted attitudes, socio-affective functioning, and self-management (Thornton, 2002). This categorisation has been standardised in a measure called SRA (Structured Risk Assessment; Thornton, 2002) which is a risk/need instrument. It can be operationalised through structured clinical ratings, leaving open the question of how to best score needs. The IDA (Initial Deviance Assessment) has been routinely used in HMP services as part of SARN (Structured Assessment of Risk and Need, Thornton, 2002), which is a version of the SRA.
In order to score the dimensions mentioned above, the practitioner usually gathers information from several sources such as psychometric tests, clinical interview and self-report scales. “Sexual interests” in specific, have been commonly measured by physiological measures such as the plethysmograph (PPG), response-time based measures, and the polygraph (A. R. Beech, Craig, & Browne, 2009). However, these measures have been criticised on several grounds. It has been suggested that they are intrusive, lacking construct validity and standardisation (Williams, 2003). It has also been suggested that individuals can use countermeasures to suppress arousal (Laws, 2003). In order to tackle these limitations, some authors have suggested the use of the Multiphasic Sexual Inventory (MSI; Nichols & Molinder, 1984) as a reliable way of measuring psychosexual characteristics (Craig, Browne, Beech, & Stringer, 2006; Grady, Brodersen, & Abramson, 2011). The use of MSI to measure sexual interests might be considered an advance in risk assessment research. It seems to add objectivity and standardisation to physiological measures, and it is ethically more appropriate.
Thornton (2002) tested the last three aspects of the model (distorted attitudes, socio-affective functioning, self-management) with a sample of child molesters and also replicated the results with rapists, showing that the dynamic risk factors tested in the model were predictive of sexual offending independent of static factors (Craig, Browne, & Beech, 2008). One again, this provides evidence for the importance of including dynamic factors is risk assessment tools.
At this point in the history, the static and dynamic approaches to risk assessment have never been clearly integrated into a single tool (A. R. Beech & Ward, 2004; Olver, Wong, Nicholaichuk, & Gordon, 2007). The development of a tool named VRS-SO (Violence Risk Scale – Sexual Offender version; Wong, Olver, Nicholaichuk, & Gordon, 2003) does this integration, representing an important advance in risk assessment of sex offenders. This tool incorporates static and dynamic risk factors, also comprising treatment planning and the measurement of treatment change, within a single tool for assessing risk in sexual offenders. It has been suggested that this tool has good inter-rater reliability, concurrent validity, and that total scores were predictive of sexual recidivism. It has also been suggested that the dynamic scale made significant contributions to the predictive value of the tool after controlling for static risk (Beggs & Grace, 2010).
We now move to a point where the importance of integrating, static and dynamic risk factors, alongside information for case management is recognised. However, risk assessment tools have been constructed through statistical analysis and the application of meta-analysis to identify which risk factors are mostly correlated to future recidivism. It has been suggested that this work is atheoretical and reductionist in the extent that it lacks a strong theoretical background and focuses too much on specific factors of risk (Beech, A. & Ward, T., 2007).
Efforts to integrate risk within a theoretical framework mark the fourth generation of risk assessment.
One of the most promising multi-factorial approaches to risk assessment of sexual offenders has been presented by Beech and Ward (2007) as the Etiological Model of Risk. The main aim of this model is to incorporate the conceptualisation of risk within an etiological and developmental framework based on current theories of sexual offending (A. R. Beech & Ward, 2004). The model starts by including developmental factors such as rejection and attachment problems, which can increase vulnerability to sexually abuse. This first step of the model is theoretically based on Marshall and Barbaree’s Integrated theory of sex offending (Marshall & Barbaree, 1990). It is then proposed that developmental factors might impact on vulnerability trait factors. These are composed by static factors and stable dynamic risk factors. In this model static factors are seen as historical markers for psychological dispositions (stable dynamic risk factors). As follows, triggering/contextual risk factors such as victim access behaviour, non-cooperation with supervision and hostility (previously described by Hanson and Harris (2001) as acute factors) impact on the vulnerability factors. These vulnerability factors are then expressed into transient states of high-risk, since they are reflective of an individuals’ psychological state (eg. physiological arousal and affective states). This chain of events, moving gradually from distal factors to more imminent, contextual factors, places individuals at an overall level of risk (Beech, A. & Ward, T., 2007).
This approach to risk assessment has several practical implications. As a starting point, it encourages practitioners to consider a wider range of risk factors, enabling objective case formulation, which will link closely to each domain of risk (Beech, A. & Ward, T., 2007) . Additionally, the reformulation of acute and stable dynamic factors into more psychologically meaningful terms enables the practitioner to contextualise ‘risk’ into theoretical frameworks that explain the process of offending as a gradual sequence of events (A. R. Beech & Ward, 2004).
Also, this model addresses individual differences to a greater extent than other models of risk assessment. This approach perceives the offender as a changeable organism, who is shaped by early developmental experiences, psychological traits and more transient states; and will respond to environmental stimuli accordingly.
Moreover, this approach also has heuristic value, proposing ways in which research could be conducted to develop this model further. However, for the purpose of this essay, these suggestions will not be explored (see (Beech, A. & Ward, T., 2007) for more details).
The MARA model (Multiaxial Risk Appraisal; Craig, Browne, Hogue, & Stringer, 2004) was developed taking into account a number of developments in the risk assessment literature, also representing a considerable advance in the area. This model structures risk-related information into three main domains: risk scales, psychopathology and empirically guided clinical assessment, from which an overall estimate of risk can be made. It provides a more global assessment of risk by considering both idiographic (the uniqueness of individuals) and nomothetic (general traits of personality) properties. The MARA encourages the use of actuarial methods as a way of providing estimates of risk within specified time-periods, considers the use of psychometrics to assess psychological constructs and suggests the use of empirically guided clinical assessment (functional analysis) as a method of monitoring acute changes. Furthermore, this approach encourages an efficient management of risk, taking into account all the factors cited above (Craig, Browne, & Beech, 2008) . Such as the etiological model, this approach aims to investigate risk of recidivism from different trajectories, allowing a more global assessment. However, it is not yet known whether adding idiographic factors actually enhances or undermines predictive accuracy (A. R. Beech, Craig, & Browne, 2009) . Hence, until further testing, this model of risk assessment must be used with caution.
Nevertheless, both the etiological model of risk and MARA can be described as comprehensive approaches to risk assessment, which offer a broad range of topics which are relevant for an accurate assessment of risk. Regarding the practical utility of these approaches, this essay argues that even though it might not be possible to address all the areas of risk in intervention due to effective management of resources, practitioners using these models would have a wide conceptualisation of the risk posed by individuals, and which areas would be more relevant to address at a specific point in time.
As a final remark in the advances in risk assessment, this essay would like to highlight the utility of the Risk-Needs-Responsivity (RNR) model (Andrews & Bonta, 2003) . This model emphasises the importance of accurate risk assessment, in order to provide the right degree of intervention; suggests a focus on the changeable risk factors presented by individuals and encourages practitioners to consider factors that may affect or impede an individual’s response to treatment.
As means of conclusion, and referring back to the statement in discussion (Hart, Laws, & Kropp, 2003) it can be stated that throughout time risk assessment has moved to being a more understood and certain process than it when the assessment of risk was based on intuition.
In an attempt to confer objectivity to this procedure, several researchers developed statistical-based, standardised approached such as actuarial tools. Thereafter, the limitations of these tools, and mainly their static nature led researchers to explore the dynamic, changeable factors related to risk and developing several useful tools. At this point it was also felt necessary to construct tools that informed case management. As proposed by Harkins and Beech (2007), it was crucial to account for risk level to an extent that it could also influence success of treatment. The most recent advances employed multi-factorial, broad models of risk which perceive risk assessment as an ongoing process, which needs to consider idiographic, historical and contextual factors. Furthermore, the RNR model provided useful guidelines for risk assessment of sexual offenders. Applying the principles of this model to exiting assessment tools has the potential of not only conferring more certainty to this process, but also to influence treatment outcomes.
This essay suggests that multi-factorial models, together with RNR principles, represent a step forward in moving risk assessment from a probability to a more certain science. This could potentially lead to a decrease in inaccurate prediction of risk, further decreasing the consequences that it involves, such as putting society at risk or unnecessary monetary expenses. It also goes without saying, that enhancing risk assessment tools is an everlasting effort, which is far from being finished.