Chapter 6: Sex Offender Risk Assessment

Introduction

Although the desire to predict the risk of future violence posed by individuals is likely centuries old, risk assessment efforts until recently have been relatively unsuccessful in terms of their predictive accuracy. Notwithstanding pseudoscientific methods such as phrenology (which claimed to gauge behavior propensities based on measurements of the skull), risk assessment for many decades has primarily involved individual mental health professionals applying their accumulated experience and clinical acumen to produce a clinical judgment of the degree of risk posed by a particular individual. Scientists have repeatedly questioned the validity of such unstructured clinical judgment as the basis for risk assessments (Grove, 2005; Grove & Meehl, 1996; Meehl, 1954), but it took the publication of John Monahan's Predicting Violent Behavior: An Assessment of Clinical Techniques in 1981 to usher in a truly scientific approach to violence risk assessment. In the three decades since the publication of Monahan's book, the relative accuracy of violence risk assessments has increased substantially.

The ability to accurately assess the likelihood of future violent acts — and future criminal behavior more generally — is important to clinicians, policymakers and the public alike. In this context, risk assessment typically involves arriving at an estimate of the likelihood that an offender will recidivate (that is, revert to illegal behavior) after the individual experiences legal consequences or intervention for a prior criminal act. (For more information on "Adult Sex Offender Recidivism," see Chapter 5 in the Adult section.)

Risk assessment serves many purposes throughout the adjudication process. It is often undertaken for dispositional purposes to help determine, for example, an appropriate sentence or custody level or the conditions of community supervision. In these situations, decisions are often predicated, at least in part, on the assessed likelihood of recidivism, with resources being allocated accordingly to promote community safety (Kingston et al., 2008).

Research has suggested that offenders convicted of sexual offenses have received more attention from policymakers than any other category of offenders over the past 25 years (Ackerman et al., 2011; Hanson & Bourgon, in press; Levenson, 2009), and that there is consequently a need for methods and tools that can be used to accurately assess the risk to public safety that sexual offenders pose. Indeed, estimates of risk for sex offenders are used in various community corrections, institutional corrections and civil commitment decision-making contexts. Thus, the scientific and theoretical underpinnings of risk assessment are a critical component of the successful management of adult sexual offenders (Hanson & Bourgon, in press; Mann, Hanson & Thortnon, 2010; Tabachnick & Klein, 2011). (For more on "Sex Offender Management Strategies," see Chapter 8 in the Adult section.)

In many respects, the effectiveness of sex offender management policies relies on the ability of criminal justice professionals to accurately differentiate sexual offenders according to their risk for recidivism (Hanson & Morton-Bourgon, 2005). Arguing from a policy standpoint, Tabachnick and Klein (2011) have stated that the results of actuarial risk assessments in particular should inform decision-making at all levels regarding the supervision of adult sexual offenders in order to prevent recidivism. Given the role played by risk assessment in high-stakes decisions such as those involving potential civil commitment for those designated as sexually violent predators, as well as the possibility of lifetime community supervision, reliance on methods and procedures possessing a strong scientific evidence base is especially critical.

While much progress has been made regarding the ability of professionals in the field to accurately estimate the likelihood of future sexual reoffense, no one is presently able to estimate either the timing or the severity of such future criminal conduct (J. Levenson, personal communication, May 23, 2011). Therefore, it is critically important to establish a clear understanding of exactly what risk is being assessed and to frame expectations accordingly. Current methods at present allow, in most cases, only for an estimate of the likelihood of both future sexual and nonsexual offending over a specific timeframe. The accuracy of these estimates depends in part on the degree to which the individual offender being assessed matches a known group of sex offenders (knows as the normative sample or norm group) and the degree to which the factors included in the risk assessment accurately reflect the known universe of relevant risk factors.

Review of Research

The purposes of risk assessment span the spectrum of the adjudication process.

Sex offender risk assessments are most often employed in applied forensic settings for purposes of decision-making (Doren, 2002). The typical venues for sex offender risk assessment include —

• Sentencing and criminal adjudications, during which the results of the assessment are used to ascertain appropriate levels and periods of confinement and/or community supervision.
• Determinations of treatment needs, settings and modalities.
• Sex offender registration and notification (SORN) proceedings, during which assessment results are used to classify ("level") offenders based on their assessed risk.
• Civil commitment proceedings, during which assessment results are used to argue for and against indefinite confinement based on the assessed risk for sexual recidivism.

Methods of assessing sex offender risk can generally be categorized as follows (Hanson, 1998):

• Unguided (or unstructured) clinical judgment: The evaluator¹ reviews case material and applies personal experience to arrive at a risk estimate, without relying on a specific list of risk factors or underlying theory to prioritize or weight any of the information used.
• Guided (or structured) clinical judgment: The evaluator begins with a finite list of factors thought to be related to risk, drawn from personal experience and/or theory rather than from relevant empirical evidence.
• Research-guided clinical judgment: The evaluator begins with a finite list of factors identified in the professional literature as being related to risk. While these factors are given priority in the risk assessment, they are combined with other factors and considerations using the clinician's judgment.
• Pure actuarial approach: The evaluator employs an existing instrument composed of a finite, weighted set of factors (generally static, or relatively unchanging and historical in nature) identified in the literature as being associated with risk. The instrument is used to identify the presence or absence of each risk factor, and an estimate of risk is arrived at through a standard, prescribed means of combining the factors. This approach is the only risk assessment method that can be scored using a computerized algorithm or by minimally trained non-clinicians.
• Adjusted actuarial approach: The evaluator begins with the administration of an existing actuarial instrument and then employs a finite list of considerations that can be used to raise or lower the assessed level of risk.

Comparisons of the above-described approaches to risk assessment have a long, and at times contentious, history (Grove, 2005; Grove & Meehl, 1996; Grove et al., 2000; Meehl, 1954). While the superiority of structured approaches to unstructured approaches appears to have been settled (Grove, 2005; Hanson & Morton-Bourgon, 2009; A. Phenix, personal communication, May 10, 2011), each of the structured approaches has its merits as well as its supporters and detractors (Doren, 2002; A. Phenix, personal communication, May 10, 2011). Nonetheless, recent research (Hanson & Morton-Bourgon, 2009) suggests that pure actuarial assessments should be favored over other approaches (Hanson, 2009).

As regards the adjusted actuarial approach, a number of recent studies (Hanson, Helmus & Harris, 2015; Storey, Watt, Jackson & Hart, 2012; Wormith, Hogg & Guzzo, 2012) have demonstrated that clinical adjustment of actuarial results more often than not decreases the accuracy of the actuarial measure, and thus, this practice is not recommended.

Criminologist James Bonta (1996) has identified three generations of risk assessment methods: unstructured professional opinion (corresponding to Hanson's [1998] unstructured clinical judgment), actuarial methods using static predictors (corresponding to Hanson's actuarial approach) and methods that include both static and dynamic factors (referred to by Bonta as criminogenic needs). By including dynamic risk factors in the assessment process, third-generation risk assessments can be used to both guide and evaluate the impact of intervention efforts. Current developments in the field confirm the promise of third-generation risk assessment methods, as research tells us more about the relationship between specific dynamic factors and risk for recidivism (Hanson, 2011; Mann, Hanson & Thornton, 2010; A. Phenix, personal communication, May 10, 2011). Recent studies have in fact demonstrated that the inclusion of dynamic risk factors can contribute incrementally to the ability of static (relatively unchangeable) risk factors to accurately predict risk for sexual reoffense (Eher et al., 2012; Nunes & Babchishin, 2012; Olver et al., 2014; Thornton & Knight, 2015).

There are three generations of risk assessment methods: unstructured professional opinion, actuarial methods using static predictors and methods that include both static and dynamic factors.

For accurate risk assessment to occur, the factors associated with the type of risk being assessed must be known. Knowledge about the risk factors associated with recidivism typically is generated through research in which the recidivism rate for offenders with a particular characteristic is compared to the recidivism rate for offenders without that characteristic, or for offenders possessing other characteristics (Hanson, 2000). To date, no single characteristic (that is, "risk factor") has been found in isolation to be a robust predictor of recidivism. As a result, the assessment of risk by necessity involves the combination of a number of risk factors in a meaningful manner.

Karl Hanson and his colleagues (Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2005) have published the results of a series of meta-analyses² that together have shed considerable light on the known universe of static risk factors associated with sexual recidivism. The strongest predictors of sexual recidivism are factors related to sexual criminality, such as a demonstrated sexual interest in children, a history of prior sexual offenses, the age of onset of sexual offending behavior and having committed a variety of sexual offenses. Factors relating to a lifestyle of instability/criminality were also found to be associated with sexual offense recidivism (Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2005). Criminal lifestyle characteristics (e.g., substance abuse, history of rule violation) are also the factors most strongly related to violent and/or any recidivism among sex offenders, mentally disordered offenders and offenders in general (Hanson & Morton-Bourgon, 2009). More recent analyses of the specific factor structure of actuarial risk assessments have reinforced the stability of these two factors — sexual and general criminality (Babchishin et al., 2016; Jung et al., 2015). Recent studies of the structure of risk assessment instruments have also identified potential additional factors assessed by these measures (Brouillette-Alarie et al., 2016), who together identified a factor tapping youthful stranger aggression in their study of the underlying constructs of the Static-99R and Static-2002R.

Recently, significant focus has been directed toward a more contextual understanding and assessment of risk, conceptualizing risk as but one aspect of a larger and more holistic understanding of the individual. Included in this context should be the assessment of needs, protective factors, developmental progressions and change/treatment progress and the role of desistence (Hanson & Bourgon, in press; McGrath, Lasher & Cumming, 2012).

These developments are part of a larger move away from a singular focus on deficits and psychopathology toward a view that incorporates consideration of assets and strengths. A seminal influence in this regard has been that of Positive Psychology as described by Martin Seligman (Seligman & Csikszentmihalyi, 2000). A number of researchers have assessed the contribution of strengths, assets and protective factors to the prediction of sex offender risk (Miller, 2015; Turner et al., 2014). Available at present are at least four risk assessment measures for use with adults that include assessment of assets, strengths or protective factors (Miller, 2015). Of these, the Structured Assessment of Protective Factors for violence risk (SAPROF; de Vries Robbe, de Vogel & Bogaerts, 2015) appears the most promising regarding assessment of sex offender risk. Results of a study involving 83 contact sexual offenders show good interrater reliability and negative correlations between the SAPROF and two actuarial tools. The ability of the SAPROF to assess risk for general and sexual violence appears to be acceptably good over both short-term (one- to three-year) and long-term (15 year) follow-up periods, even after controlling for static risk (de Vries Robbe, de Vogel & Bogaerts, 2015).

Another aspect of dynamic assessment that has gained recent attention concerns the measure of change associated with participation in treatment. McGrath and colleagues (McGrath, Lasher & Cumming, 2011) have developed the Sex Offender Treatment Intervention and Progress Scale (SOTIPS) to accomplish this very purpose. In a study that included 759 adult male sex offenders under correctional supervision and enrolled in sexual offender treatment, study participants were assessed using a number of actuarial measures as well as the 16-item SOTIPS. One and three years following assessment, both the Static-99R and the SOTIPS demonstrated moderate ability to rank order risk for sexual, violent and any criminal recidivism and return to prison. When combined using a statistical procedure known as logistic regression, both the SOTIPS and Static-99R consistently performed better than either instrument did in isolation. Furthermore, study participants whose SOTIPS scores suggested that they made progress in treatment demonstrated lower rates of recidivism than participants who didn't demonstrate a treatment effect.

The emergent emphasis within risk assessment on more positive and healthy aspects of individuals is also reflected in the literature concerning desistence from crime. The study of desistence involves identifying those characteristics, features and events that lead to the cessation of criminal behavior (Laub & Sampson, 2001). Explorations of the factors that lead to desistence from sexual offending have recently begun to emerge in the sex offender risk assessment literature (Cale & Lussier, 2012; de Vries Robbe et al., 2015). De Vries Robbe and colleagues reviewed the available literature regarding protective factors and desistance from sex offending, and concluded there are three compelling reasons to include the assessment of strengths in sex offender risk assessment. First, to do so may improve the predictive validity of current risk assessment instruments. Second, focusing on risk alone can lead to over-prediction of violence risk. And third, assessments that focus only on deficits and pathology lead to the stigmatization of those assessed (de Vries Robbe et al., 2015). Rather than simply viewing the absence of a need as a strength, the desistence literature identifies certain characteristics as assets in their own right, such as positive social relationships and the presence of healthy coping mechanisms.

While consideration of the above factors has added to the ability to accurately assess sex offender risk, the literature also has identified factors that do not contribute to the accurate assessment of risk. Over the past three decades, numerous studies have examined the factors that are related to sexual offense recidivism, and not a single study has found the specific type of crime an offender is convicted of to be predictive of the likelihood of recidivism (Freeman & Sandler, 2010).

Sex offender risk assessment, while similar in many ways to the assessment of other latent constructs (psychological concepts) within psychology and mental health, differs in at least one significant aspect. The construct being assessed — the commission of a new sexual offense — is unobservable and is likely never to be observed by the assessor. Sex offender risk assessment entails a process of estimating the likelihood of a future event based entirely on secondary, indicator variables (Hanson, 2009). While actuarial risk assessment tools must meet standard criteria for psychological measures (e.g., reliability and validity), the utility of these instruments depends considerably on the selection of relevant risk factors and the methods used to combine these factors to arrive at a meaningful overall assessment of risk (Hanson, 2009). It is important to keep in mind that for purposes of risk assessment, the utility of a risk factor depends on its empirical relationship to the outcome being predicted (Helmus et al., 2012). The consideration of base rates is also critical (Thornton, Hanson & Helmus, 2011). The base rate is equal to the proportion of a group that shares a specific characteristic. For purposes of sex offender risk assessment, the relevant base rate is the proportion of convicted sex offenders who commit a subsequent sexual offense, either over a specified timeframe or over the course of their lifetime. The base rate is arrived at through reference to large meta-analyses of sex offender recidivism, such as Hanson and Bussière (1998) and Hanson and Morton-Bourgon (2005). These studies found the five-year recidivism rate to be approximately 13 percent. However, it is important to remember that this figure is an underestimate, given that not all recidivist behavior is detected.

The accurate assessment of risk involves gaining an understanding of all available, relevant factors associated with the known criterion or outcome behavior. While research findings are quite consistent regarding the historical, relatively unchangeable factors referred to as "static" risk factors (e.g., age at first offense, number of previous convictions), there is less agreement at present regarding more fluid, changeable risk factors referred to as "dynamic" risk factors (e.g., employment status, cooperation with supervision). The utility of a rather fixed set of static variables associated with sex offender risk has been established in numerous studies (Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2005), and empirically identified static risk factors are a primary component of several valid and reliable instruments used in the field today (e.g., Static 99R, Static-2002R, MnSOST-3).

A number of instruments incorporating dynamic factors have appeared in recent years, such as the Stable-2007/Acute-2007 (Hanson et al., 2007) and the Structured Risk Assessment-Forensic Version (Thornton & Knight, 2009). Neither of these instruments, however, has the research backing of the more established instruments of static risk, such as the Static-99R and Static 2002R. A recent meta-analysis (Mann, Hanson & Thornton, 2010) provides the most complete understanding to date of the relationship between a host of dynamic factors and sex offender recidivism.

The use of third-generation risk assessment instruments that incorporate both static and dynamic risk factors is becoming more prevalent (Hanson & Morton-Bourgon, 2009; A. Phenix, personal communication, May 10, 2011). These instruments have the potential added benefit of providing targets for intervention. An example of a third-generation instrument is the Level of Service/Case Management Inventory (Andrews, Bonta & Wormith, 2004), which provides a general assessment of risks and needs for criminal-justice-involved persons. The Violence Risk Scale: Sexual Offender Version (VRS:SO) is a recently developed instrument specifically designed to assess risks and needs among sex offenders. This measure contains seven static factors and 17 dynamic factors; the dynamic, treatment-change factors are based on the Transtheoretical Model of Change (Beggs & Grace, 2010). (For more on treatment, see Chapter 7, "The Effectiveness of Treatment for Adult Sex Offenders," in the Adult section.)

The use of third-generation risk assessment instruments that incorporate both static and dynamic risk factors is becoming more prevalent. These instruments have the potential added benefit of providing targets for treatment.

A variety of sex offender risk assessment tools possess acceptable, empirically supported psychometric properties (Doren, 2002, 2006; Hanson, 2009; Hanson et al., 2013; Hanson et al., 2016; Helmus et al., 2012; A. Phenix, personal communication, May 10, 2011; Nunes & Babchishin, 2012). While a complete review and analysis of these instruments is beyond the scope of this chapter, a meta-analysis conducted by Hanson and Morton-Bourgon (2009) provides important insights concerning the relative accuracy of different approaches. Their analysis consisted of 536 findings drawn from 118 distinct samples with a total sample of 45,398 sex offenders in 16 countries. The follow-up periods ranged from six months to 23 years; the average follow-up period was five years and 10 months (standard deviation = 46.6 months). The following types of risk assessment approaches were included in the analysis: empirical actuarial, mechanical (using factors chosen primarily on the basis of theory or literature reviews), adjusted actuarial, structured professional judgment and unstructured professional judgment.

Hanson and Morton-Bourgon (2009) concluded that empirically derived actuarial approaches were more accurate than unstructured professional judgment in assessing risk of all outcomes — sexual, violent and any recidivism. The accuracy of structured professional judgment methods fell in between these two methods. For the prediction of sexual recidivism, actuarial instruments designed for assessing the risk of sexual recidivism had the greatest predictive accuracy, followed by mechanical approaches designed for assessing the risk of sexual recidivism and actuarial instruments designed for assessing the risk of general recidivism. Unstructured professional judgment and actuarial instruments for assessing violent recidivism risk were less accurate in assessing the likelihood of sexual recidivism. The predictive accuracy of structured professional judgment fell in between that of actuarial instruments and unstructured professional judgment approaches. In addition, structured professional judgment exhibited a large degree of variability in the few studies that examined this method (Hanson & Morton-Bourgon, 2009).

Hanson and Morton-Bourgon (2009) also found that, for assessing the likelihood of sexual recidivism, the best-supported instruments were the following:

• Static-99 (Hanson & Thornton, 2000)
• Static-2002 (Hanson, Helmus, & Thornton, 2010)
• MnSOST-R (Epperson et al., 2000)
• Risk Matrix-2000 Sex (Kingston et al., 2008)
• SVR-20, specifically using the mechanical approach of adding up the item scores (Boer et al., 1997)

For assessing the likelihood of violent (including sexual) recidivism, the best-supported instruments were the following:

• Violence Risk Appraisal Guide (VRAG) (Webster et al., 1994)
• Sex Offender Risk Appraisal Guide (SORAG) (Quinsey et al., 2006)
• Risk Matrix-2000 Combined (Thornton, 2007)
• Statistic Index of Recidivism (SIR) (Nafekh & Motiuk, 2002)
• Level of Service Inventory-Revised (LSI-R) and its variants (Andrews, Bonta, & Wormith, 2004, 2006)

Some risk assessment experts have suggested that the accuracy of purely actuarial approaches can be increased if certain dynamic risk factors (e.g., active substance abuse, demonstrated pro-offending attitudes) are included in the assessment instrument or otherwise considered as part of the assessment process. Discussions of the relative merits of this approach can be found in Wollert and colleagues, 2010; Hanson and Morton-Bourgon, 2009; Doren, 2002; and McGrath, Cumming and Lasher, 2012. One dynamic risk factor that has received considerable attention in this context is the offender's age at the time of assessment. The inverse relationship between age and criminal offending — as age increases, offending decreases — is one of the more robust findings within criminology. This relationship has been found to hold across time and geographic locations, for different types of crimes and offenders and in both community and incarcerated offender populations (Hirschi & Gottfredson, 1983). Age as an adjusting factor in risk assessment has received considerable attention not only because of the strength and consistency of its relationship to offending, but also because some actuarial instruments (e.g., Static-99 and Static-2002) have been found to underestimate the likelihood of recidivism for younger offenders and to overestimate it for older offenders (Helmus et al., 2012; Wollert et al., 2010). As a result of these findings, the Static-99 and Static-2002 have been revised to better account for the impact of the offender's age at the time of assessment, resulting in the Static-99R and Static-2002R. (Both of these revised instruments do not need to be adjusted for age.) Using age-adjusted risk tables is especially important when assessing older offenders.

Another set of factors often considered as potential adjustments to actuarial measures are those referred to as "criminogenic needs" (Bonta, 1996) or psychologically meaningful risk factors (Mann, Hanson & Thornton, 2010; Thornton & Knight, 2015). These are dynamic (that is, changeable) risk factors that can serve as targets for intervention efforts. For a risk factor to be considered psychologically meaningful, there must be a plausible rationale that the factor is a cause of sexual offending and there must be strong empirical evidence that the factor predicts sexual recidivism. This latter requirement is best demonstrated through research associating variation between groups in the predictor (proposed predicting factor) with variation between groups in the rate of failure (Hanson, 2009). Specific measures have been found helpful in the assessment of dynamic risk factors (e.g., the STABLE-2000 and STABLE-2007; Eher et al., 2012; Nunes & Babchishin, 2012) and risk-enhancing behaviors such as substance abuse have also been found to be salient dynamic factors worthy of consideration in assessing risk (Looman & Abracen, 2011).

While it stands to reason that clinicians would want to consider dynamic factors when assessing risk, doing so via clinical adjustment of actuarial instruments has not been found to be effective. Many studies have examined the effects of clinical adjustments to the results of actuarial instruments, finding that "overrides" — a clinician's consideration of factors outside the actuarial scheme (i.e., the evaluator judges whether the predicted recidivism rate is a fair evaluation of the offender's risk) — decrease predictive accuracy (Gore, 2007; Hanson, 2007; Hanson, 2009; Hanson, Helmus & Harris, 2015; Hanson & Morton-Bourgon, 2009; Storey, Watt, Jackson & Hart, 2012; Vrana, Sroga & Guzzo, 2008). All of these studies have involved the adjustment of actuarial sex offender risk assessments currently in use, with each study concluding that the adjustments made actually decreased the predictive power of the actuarial instruments.

It is important to note that empirical research undertaken to date has yet to identify a single "best" assessment instrument. With this and the limitations of using only one risk assessment instrument (particularly in especially high-stakes situations such as civil commitment evaluations) in mind, clinicians have considered the potential benefits of using more than one instrument during the assessment process (Doren, 2002; Hanson, 2009, 2011). In fact, in a study of evaluators who conduct civil commitment evaluations, Jackson and Hess (2007) reported that 79.5 percent of the evaluators use more than one actuarial instrument in their sex offender civil commitment evaluations.

Empirical research has yet to identify a single "best" risk assessment instrument.

Two primary rationales support the notion that using more than one instrument provides potential benefits. First, classical test theory suggests that increasing the number of items in an assessment increases reliability and coverage. Second, if there are multiple driving forces behind sexual offending behavior, and individual risk assessment instruments tap these underlying dimensions or pathways to sexual offense recidivism differentially, then the use of multiple instruments would have a distinct advantage over the use of a single instrument alone. As Doren (2002, p. 138) points out, "The evidence for multiple underlying dimensions potentially driving sexual offending represents the main relative weakness to using only the 'best' single risk assessment instrument in a sex offender civil commitment evaluation." 

Indeed, the empirical evidence suggests that multiple dimensions or pathways underlie sexual offending, with a number of scholars describing a convergence between two of these dimensions: sexual criminality and general criminality. Doren (2002) describes the high sexual criminality/low general criminality pathway as typical of the generally law-abiding pedophile, and the low sexual criminality/high general criminality pathway as typical of an antisocial individual for whom sexual violence is simply one of many manifestations of a criminal behavioral pattern. Evidence for these two pathways also has been found in meta-analytic studies of the factors associated with sex offender recidivism (Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2005). Hence, an evaluation of both dimensions/pathways as part of the risk-assessment process seems beneficial and advisable, whether it is done using a single instrument that assesses both dimensions or multiple instruments that tap each dimension separately. In fact, recent studies have found that combining these factors or dimensions in sexual offender risk assessment increases predictive accuracy (Brouillette-Alarie et al., 2016; Olver et al., 2016). For more about pathways, see Chapter 3, "Sex Offender Typologies," in the Adult section.)

These two underlying dimensions of sexual offending were discussed in a recent study of the incremental validity of a number of actuarial instruments (Babchishin, Hanson & Helmus, 2011). As part of that study, the Rapid Risk Assessment for Sex Offense Recidivism (RRASOR) instrument was found to tap sexual criminality, while the Static-99 was found to assess risk along the general criminality pathway. Further, Babchishin, Hanson and Helmus (2011) found that the RRASOR (which taps the sexual criminality dimension) and the Static-99R and Static-2002R (both of which tap the general criminality dimension) all added incremental validity to one another, in spite of substantial intercorrelations and substantial item overlap across the three instruments. Recent research indicates that even measures as highly correlated as the Static-99R and Static-2002R add incrementally to each other (Babchishin et al., 2012; Lehmann et al., 2013), although the size of the gains are generally quite small. This would suggest that practically all of the current well-validated risk factors provide some degree of unique and predictively useful information.

There are other compelling reasons to use more than one instrument during the risk-assessment process, even when the instruments tap the same dimension or the same theoretical domain. Including a larger number of items that assess the same construct and having similar predictive accuracy increases reliability and adds to the overall predictive accuracy of the procedure. Recently, Babchishin and colleagues (2012) conducted a meta-analysis of 20 samples (n = 7,491) in which they compared a variety of methods of combining risk scales. They found that averaging the scores produced the most psychometrically sound results. They interpreted these findings as further support of the need to understand the underlying psychological constructs of factors of criterion-referenced measures such as the Static-99R and Static-2002R (Babchishin et al., 2012). Hanson and Bourgon (in press) make the case that rather than just blindly accepting the mechanical relationships between risk factors and the outcome of interest, it is important to pay careful attention to the constructs assessed by the measures. From the standpoint of construct validity, the factors measured in assessment are indicators for unobservable (referred to as latent) traits or constructs. From this perspective (the traditional perspective of psychological assessment), the factors that successfully predict recidivism (both static and dynamic) do so because they serve as indicators for the latent traits or constructs directly related to the risk of criminal behavior. In situations in which all of the risk factors represent a single latent construct, summing the risk factors results in a sense of where the offender lies on that particular dimension. The more risk factors, the greater the risk for future criminality. This approach will not work, however, when the risk factors represent more than a single construct. As a result, assessors must be aware of the underlying constructs represented by the risk assessment(s) they employ. Many of the current actuarial risk assessment measures necessarily assess at least two constructs or domains representing general and sexual criminality (Hanson and Bourgon, in press).

Another issue of critical importance in sex offender risk assessment is the communication of risk assessment findings (Babchishin & Hanson, 2009; Doren, 2002; Hanson, 2009; Hanson and Bourgon, in press). Currently, nominal descriptors of risk (low, moderate and high) are used most commonly (Babchishin & Hanson, 2009). While qualitative descriptions in general and these particular nominal descriptors are usually preferred over numerical formats for communicating risk, the use of qualitative labels alone has certain limitations. Perhaps the most significant limitation is that clinicians (as well as decision-makers) can have very different interpretations of what these nominal categories represent. The context in which risk assessment findings are communicated can also influence interpretation.

One way to mitigate the problems associated with the exclusive use of nominal categories is to also provide numerical indicators of risk, such as a recidivism rate probability, a percentile rank (Hanson et al., 2012), or a risk ratio (Hanson et al., 2013). There are various numerical formats commonly used to convey absolute risk, such as frequencies (e.g., the likelihood of recidivism is one out of 10) and percentages (e.g., the likelihood of recidivism is 10 percent), both of which are usually accompanied by a specific timeframe (e.g., within the next five years). Relative risk estimates, such as percentile ranks (e.g., the individual's risk for reoffense is equal to or greater than 90 percent of offenders) and risk ratios (the individual is four times more likely to sexually recidivate compared to the average offender), are useful as well.

While numerical estimates provide more information and are potentially less ambiguous than qualitative descriptors alone, they too have limitations. For example, even though the assessed risk is the same, risk frequencies reported with larger denominators (e.g., 10 out of 10,000 compared to one out of 1,000) tend to result in higher perceived risk. Interpreting numerical risk estimates properly can also be a challenge when base rates for the behavior in question are unknown or are not taken into consideration. Simply put, people tend to overestimate the likelihood of low-probability events and underestimate the likelihood of high-probability events. For instance, people are more likely to fear flying than driving, even though the likelihood of dying in a car crash is many times that of dying in a plane crash.

Evaluators can also make mistakes when communicating the results of risk assessments. Doren (2002) has identified three common errors in communicating results when using a single instrument: incorrectly describing the risk percentage associated with a particular score, neglecting to address sampling error or failing to provide confidence interval estimates and ignoring or incorrectly stating the qualifiers as to what has been assessed.

Consumers of risk assessment information typically desire more than a simple nominal or numeric indicator of risk. Frequently, decision-makers want the risk-assessment process to provide them with information on the likelihood of recidivism, the potential consequences associated with recidivism and what might be done to mitigate the assessed risk (Hanson, 2009). Doren's (2002) recommendations for communicating the results of sex offender risk assessment, especially in cases involving civil commitment, include the following:

• Nominal risk categories should be accompanied by numerical risk estimates. When used in tandem, nominal and numerical means of conveying risk are more accurate and informative than either one in isolation.
• Nominal categories should be explicitly defined so as to limit the degree to which readers define for themselves the meaning of the specific nominal descriptors. Two examples would be stating that "low risk" means that the risk of sexual recidivism is similar to what would be expected from a group of non-sex offenders, and "high risk" means that an offender is more likely than not to sexually recidivate over the course of his lifetime.

To date, there exists no agreed-upon, much less universal means of either describing risk or communicating the findings of risk assessments. Different risk assessments have different categories, different rules for assigning to categories and different proportions of offenders in their categories. Therefore, even when different scales share one or more category descriptors, they do not describe the same thing. Put simply, a single offender can be assigned to different risk categories by different measures, which understandably leads to confusion on the part of consumers of risk information (Hanson and Bourgon, in press). What is agreed upon, however, is that offenders can be rank ordered from low to high with respect to their risk, and that the current crop of measures can do so quite accurately. There also exists to a certain degree of agreement as to the characteristics of the offenders that make up either end of the risk continuum. Ideally, communication of risk would include factors relevant not only to risk for reoffense but also regarding dynamic factors (criminogenic needs) to be targeted for change, as well as information regarding strengths and protective factors that can be tapped with an eye toward promoting desistence. The ideal categorical risk levels would not be tied to a particular risk instrument but rather would apply across the range of risk measures. Hanson and Bourgon (in press) describe an ongoing process by which the Council of State Governments (CSG) Justice Center is working to define risk levels for the prediction of general criminality. Similarly, The STATIC Development Group has likewise proposed updated risk categories to be used with the Static-99R and Static-2002R (Hanson et al., 2016). In a manner similar to the standardized risk assessment levels proposed by CSG's Justice Center, the new categories for the Static measures have been designed to be applied across a wide range of sex offender risk-assessment measures.

Future Directions

The pace of development in the field of sexual offender risk assessment is rapid and reflects a great deal of innovative forward thinking regarding how best to assess for and communicate about sexual offender risk. The following represent emerging trends and future directions in this field.

• Increasingly greater attention will be paid to the conceptual nature of risk factors and the interrelationships between them. This will necessarily involve moving beyond describing purely mechanical and empirical relationships to elucidating conceptual relationships among and between individual risk factors as well as the underlying (latent) constructs they represent.
• A great deal of empirical support has demonstrated the utility of actuarial risk assessments, and the best of these measures are all approximately equal in their ability to assess risk for reoffense. Research into the norms and reference tables for these measures, as well as concerning the practical applications of their use, will continue.
• It may very well be that the reason that so many of the actuarial static risk assessments possess roughly equivalent predictive ability is because they have reached a limit (asymptotic) in terms of the amount of variability in risk than can be predicted with reference to static factors alone. This paves the way for inclusion of other factors and characteristics related to risk, such as dynamic factors, treatment effects, inclusion of strengths and protective factors, and factors predictive of desistence. This more inclusive and holistic approach will more clearly reflect the complicated nature of human behavior.
• These measures will continue to be applied to new populations and settings (e.g., clergy sex offenders; offenders within youth serving organizations; different ethnic, cultural and linguistic groups), as well as find new uses (e.g., the use of a subscale and item from the Static-2002R being used to predict general criminal recidivism).
• The growing emergence and availability of large datasets, combined with ever-increasing computing power and the proliferation of emerging sophisticated statistical techniques (e.g., decision trees, classification and regression tree analysis), will result in an eventual move toward fully automated actuarial risk assessments that rely exclusively on large institutional datasets. This has the potential to eliminate the coding and mathematical errors that can occur even with the current crop of purely actuarial risk-assessment measures.

Significant growth has occurred in recent years in both the development of sex-offender-specific risk-assessment instruments and their use in the field. While significant advances have been made regarding the reliability and predictive validity of instruments, a number of limitations remain. As noted above, there is currently no single "best" risk assessment for all offenders in all situations. In fact, there are certain populations for whom there is no validated risk-assessment instrument (e.g., child pornography offenders and female offenders). (For more on child pornography offenders, see see Chapter 4, "Internet-Facilitated Sexual Offending," in the Adult section.) In addition, while development and testing of third-generation instruments continues, some experts are skeptical that a single actuarial scale containing all relevant risk factors could ever be developed (Hanson, 2000). Therefore, contemporary risk assessment involves a bit of paradox: Even though research on risk assessment has largely eliminated subjective judgment from within the risk assessment process itself, clinical judgment on the part of the evaluator is still needed to make valid, research-informed decisions about the appropriate risk assessment instrument(s) to apply in any particular setting. To that end, Hanson (2009) has provided the following set of qualities to guide the future of sex offender risk assessment:

• Assess risk factors whose nature, origins and effects can be understood.
• Enable reliable and valid assessment of clinically useful causal factors.
• Provide precise estimates of recidivism risk.
• Allow all relevant factors to be considered.
• Inform the development of treatment targets and risk management strategies.
• Allow the assessment of both long- and short-term changes in risk.
• Incorporate protective factors as well as risk factors.
• Facilitate the engagement of the patient/offender in the assessment process.
• Use risk assessment methods that are easy to implement in a broad range of settings.

Summary

Significant advancements in the science and practice of sex offender risk assessment have occurred over the past two decades. A number of reliable, valid approaches for assessing sex offender risk are now available. Rigorous scientific research has demonstrated that respectable levels of predictive accuracy have been obtained with purely actuarial risk assessment approaches, approaches using structured professional judgment and the mechanical combination of items from structured risk schemes. While research evidence to date has not indicated which of these approaches are best suited to specific testing circumstances and contexts (Hanson, 2009), recent meta-analyses (Hanson & Morton-Bourgon, 2009) suggest that purely actuarial assessment approaches should be favored over other approaches for the assessment of risk for sexual reoffense (Hanson, 2009). Ultimately, however, decisions about the best approach or instrument to use should be made in the context of the assessment setting, the characteristics of the individual being assessed and the specific purpose of the risk assessment.

Many of the purely actuarial tools in wide use today can be completed quickly and easily by a variety of trained personnel (Klima & Lieb, 2008). The advent of automated actuarial tools conceptually allows even clerical workers to compute risk scores using these instruments. It is nonetheless important to provide ongoing training and monitoring of evaluators to ensure that risk-assessment procedures and instruments are always used appropriately and with integrity. The need for training and technical assistance in the context of risk assessment was identified by the 2012 SOMAPI forum participants as well as in recent literature (Hanson et al., 2015; Storey et al., 2012).

Training and monitoring of evaluators is needed to ensure that risk assessment procedures and instruments are used appropriately and with integrity.

One of the primary challenges for the field in the future will be to identify more comprehensively the risk factors (both static and dynamic) that are related to sexual offending. Identifying these factors and incorporating them into the risk-assessment process will help clinicians and decision makers better match risk levels to treatment and management efforts, thereby fulfilling the promise of third-generation risk-assessment instruments (Bonta, 1996). The need for tailored rather than uniform interventions, and the need to match sex offender treatment and management efforts to the risk levels and criminogenic needs of sex offenders, were acknowledged by the experts— both researchers and practitioners— who participated in the SOMAPI forum. 

Given the lack of a single best risk-assessment instrument, evaluators will continue to have to rely on their professional judgment to select and employ the best risk-assessment approach for the circumstances and setting. Incorporating dynamic risk factors (e.g., assessment of treatment change) can be accomplished through the use of newly developed measures such as the SOTIPS (treatment change; McGrath et al., 2012) and the SAPROF (protective factors; de Vogel et al., 2009). Additional research concerning the use of dynamic risk factors is clearly needed, is further work to explore the factors that lead to desistence from sexual offending (Griffin et al., 2008; K. Hanson, personal communication, April 8 and June 7, 2011; Maruna & LeBel, 2003). It is recommended however that risk assessment conceptually and practically be incorporated into the larger concept of psychological assessment, encompassing a more comprehensive and holistic view that incorporates not only deficits but strengths, protective factors and consideration of desistence.

Research on the best ways to revise assigned risk based on post-index behavior or qualities also is needed. In effect, this entails identifying treatment targets and assessing the impact of treatment on risk and other factors, such as institutional misconduct or the amount of time that has elapsed without a new conviction (K. Hanson, personal communication, April 8 and June 7, 2011). The ability to detect meaningful changes in risk, especially for high-risk offenders, is particularly important (Hanson, 2011; Olver et al., 2007). The VRS:SO is a promising development in this area (Beggs & Grace, 2010; Thornton, Hanson & Helmus, 2011), as is the SOTIPS (McGrath et al., 2012). Other instruments to consider for gauging changes in risk over time include the STABLE-2007 and the SRA—Forensic Version (Thornton & Knight, 2009). Olver and colleagues (2014) describe their application of logistic regression to combine risk and change information into clinically meaningful post-treatment risk assessments (Olver et al., 2014). As noted previously, the Static-99 and Static-2002 have been revised to incorporate the impact of aging on risk, resulting in the inclusion of new age weights and the publication of the Static-99R and Static-2002R (Helmus et al., 2012).

There also is a need to devise more effective and intuitive means of communicating risk-assessment findings. Communication of risk should be tailored to the purpose and setting of the assessment, and both qualitative descriptors and numerical estimates that consumers of risk-assessment information can use to guide sex offender management decision-making should be provided. Ultimately, the development of a standardized language to communicate risk would obviate many sources of potential confusion among consumers if risk information (Hanson and Bourgon, in press). Furnishing decision makers with both an accurate, contextual understanding of risk, and also with recommendations for mitigating and managing risk, is likely to be most beneficial.

Based on current knowledge, using science-based, actuarial methods to assess sex offender risk is advisable.

In conclusion, based on current knowledge, using science-based, actuarial methods to assess sex offender risk is highly advisable (Doren, 2002; Hanson & Morton-Bourgon, 2009; Tabachnik & Klein, 2011). As Hanson and Morton-Bourgon (2009, p. 10) aptly state, "Given its genesis in data, the empirical actuarial approach will ultimately provide the best estimates of absolute risk." In fact, such instruments should not be ignored in assessing the risk for sex offender reoffense unless there is clear and justifiable reason to do so, such as in cases for which no applicable risk instrument exists (Hanson & Morton-Bourgon, 2009).

For assessing the likelihood of sexual recidivism, the best-supported instruments are the Static-99R, Static-2002R, MnSOST-3 (Duwe & Freske, 2013; potential issues with limited generalizability outside of Minnesota), Risk Matrix-2000 Sex and adding the item scores from the SVR-20 (Hanson & Morton-Bourgon, 2009). The Static family of measures are by far the most often used sex offender risk assessments in the United States and Canada (Storey et al., 2012; Swinburne Romine et al., 2012; Turner et al., 2014). These measures have been found to be effective for a variety of different populations, including clergy offenders (Montana et al., 2012), those who work with children (Turner et al., 2014) and Latinos (Leguizamo et al., 2016), and with aboriginal and non-aboriginal Australians (Smallbone et al., 2013).

For assessing the likelihood of violent (including sexual) recidivism, the best supported instruments are the VRAG, the SORAG, the Risk Matrix-2000 Combined, The Psychopathy Checklist – Revised (Hare, 2003; assessed by Parent et al., 2011) and the LSI-R and its variants (Hanson & Morton-Bourgon, 2009). For assessing the risk of general (nonsexual) recidivism among sexual offenders, Hanson and Bourgon (in press) recommend using the Brief Assessment for Recidivism Risk (BARR-2002R; Babchishin et al., 2015), which is comprised of the age at time of assessment and the general criminality factor from Static-2002R.