part0029

An essential task for clinicians and other decision makers who work with sex offenders against children is to identify those who will sexually offend again. In other words, clinicians and other decision makers want to know what factors will identify those who will sexually reoffend. Risk assessors are also often concerned about how quickly new offenses occur and the seriousness of any new offenses. These are questions of persistence, having to do with the likelihood that someone already known to have committed a sexual offense will commit another sexual offense. It is not the same as the question of onset, which has to do with the likelihood that someone will commit a sexual offense against a child in the first place. The variables that help answer the question of persistence may not help researchers answer the question of onset, either in the general population or among individuals who would be at risk, for example, because they have pedophilia or hebephilia; this is addressed instead by research on the origins of sexual offending (see Chapter 4 , this volume). The assessment of risk to reoffend includes questions about cross-over, such as whether someone who has committed noncontact sexual offenses will escalate to contact sexual offenses or whether someone who has committed sexual offenses against one age or gender category might offend against individuals in other categories.

In this chapter, I focus on the assessment of risk to reoffend among identified sex offenders (persistence) because that is the most pressing question and because much more research is available. But I also review what is known about escalation from noncontact to contact sexual offending and what can be said about the risk to sexually offend among putatively at-risk individuals (onset). Most of the literature examines the ability to predict sexual recidivism among sex offenders in general, rather than among offenders against children only. Also, most of this research focuses on any new sexual offenses, not specifically sexual offenses against children; any exceptions are mentioned.

The assessment of risk to reoffend represents multiple, overlapping issues: (a) whether an offender meets legally determined criteria for dangerousness, which requires a judgment about absolute probability of reoffending (e.g., Dangerous Offender hearings in Canada that can result in indeterminate prison sentences, sex offender civil commitment proceedings in the United States that allow for the indefinite commitment of sex offenders after they have served their prison sentence); (b) the rank ordering of offenders according to risk to reoffend, which can be done by looking at relative rank without focusing on absolute probabilities, to appropriately place offenders in terms of different levels of security, supervision, and treatment; (c) case formulation, to understand what factors underlie an individual’s sexual offending; (d) determining when supervision conditions should be adjusted to safely manage offenders in the community; and (e) identifying targets for intervention to reduce the likelihood or imminence of a new sexual offense. It remains the case that more is known about the first two issues, but good progress has been made on the most important treatment targets. I begin with a review of definitions and important methodological considerations in risk assessment research.

Almost all risk assessment studies have defined recidivism as new arrests, charges, or convictions for a criminal offense, using official records to collect information about these outcomes. In part, this is because of the feasibility of obtaining official records versus relying on self-report or other sources of information about recidivism. However, official records are underestimates of true reoffending because many victims do not report sexual crimes and many sexual crimes that are reported do not result in arrests, charges, or convictions. Moreover, these criminal justice outcomes have important differences: Arrests are the broadest, representing all new offenses that meet the legal threshold. But some arrests do not lead to charges or then to convictions because of insufficient evidence, reasonable doubt, or other reasons.

The mismatch between actual and recorded recidivism affects policies and practices because the size of the difference is not precisely known. The sources of this mismatch may obscure the identification of recidivism risk factors, to the extent the mismatch is related to the risk factor. For example, a psychopathic individual who has no positive ties to others, who is highly mobile, and who is very willing to engage in deceit and manipulation might be more likely to escape detection for a new sexual offense and might have more opportunities to commit sexual offenses than someone who has many relatives and friends and who returns to the same community for many years after serving his prison sentence. At the same time, psychopathic offenders are more likely to offend, and if they reoffend, they may commit new offenses that are more frequent and/or severe. Although many of these offenses might not be reported, each new offense increases the chance of eventual detection. Consistent with this idea, Snyder (2000) found that the likelihood that a minor’s report of sexual victimization would result in an arrest was positively associated with the number of victims involved. Fortunately for our interpretation of risk factor research, meta-analyses have shown that effect sizes are not related to either source of information (e.g., self-report vs. national criminal records) or to the type of recidivism outcome examined (arrest vs. conviction; Hanson & Morton-Bourgon, 2005).

Researchers have also differed in the type of recidivism outcome they have focused on in their follow-up studies. Some researchers have focused on sexual recidivism , usually defined as a sexually motivated crime involving physical contact with a victim (e.g., criminal charges of child molestation, rape, or sexual assault). Some also include noncontact sexual recidivism, which can include crimes involving child pornography, exhibitionism, or voyeurism. In our recidivism research (Seto & Eke, 2015; Seto, Harris, Rice, & Barbaree, 2004), we have examined both contact and noncontact sexual recidivism, as well as a broader violent recidivism outcome that encompasses new nonsexually violent or sexual contact offenses. I am interested in preventing both sexual and nonsexual violence, because both are harmful. Also, some apparently nonsexually violent offenses—as identified by the type of criminal charge/conviction in the records—may in fact represent sexually motivated offenses that resulted in plea bargains or that did not have sufficient evidence for conviction (e.g., an attempted sexual assault that resulted in a conviction for assault or forcible confinement; Rice, Harris, Lang, & Cormier, 2006). Focusing only on sexual contact recidivism will miss some of these sexually motivated offenses.

Some writers have suggested that the factors that predict long-term risk to reoffend may differ from those that predict imminent reoffending (e.g., Doren, 2002; Litwack, 2001). In principle, dynamic risk measures such as the Stable-2000 or -2007 (see later section regarding dynamic risk assessment) are better able to assess imminent risk, although it is also the case that those who are actuarially at higher risk of recidivism because of static or historical risk factors are also more likely to reoffend in the short as well as long term (G. T. Harris, Rice, & Cormier, 2002). In the general forensic assessment literature, a small but growing body of evidence has examined the Short Term Assessment of Risk and Treatability (START), a dynamic measure assessing both risk and protective factors to predict shorter-term outcomes, such as inpatient aggression (e.g., Nicholls, Brink, Desmarais, Webster, & Martin, 2006).

In addition to whether a sex offender with child victims is likely to reoffend within a specified time period, risk evaluators may be interested in how quickly any new offenses will occur, and the severity of any new offenses. Quinsey, Harris, Rice, and Cormier (2006) reported that Violent Risk Appraisal Guide (VRAG) scores were negatively and moderately correlated with time to any new violent offenses, including contact sexual offenses, in a sample of 343 sex offenders. In other words, among those who violently reoffended, sex offenders with higher VRAG scores were quicker to reoffend than those with lower VRAG scores. I am not aware of any similar analyses from other sexual offending risk appraisal measures.

Knowing the base rate of recidivism is not enough without knowing the time at risk when comparing the results of different outcome studies, because longer follow-up periods are associated with higher rates of recidivism, simply because of more opportunity to reoffend. One study may report a sexual recidivism rate of 7%, whereas another study reports a sexual recidivism rate of 47%, but this is not very informative until we also know that the first study followed sex offenders for 2 years whereas the second study followed them for 20 years. Given the relatively low rate of sexual reoffending observed after 4 to 5 years of follow-up (approximately 13% in the meta-analyses reported by Hanson & Bussière, 1998, or by Hanson & Morton-Bourgon, 2004, 2005), longer follow-up periods are often required to detect statistically significant associations between putative risk factors and recidivism and to detect statistically significant differences between groups, such as offenders who have participated in treatment and those who have not. Large samples are required as well, especially as more recent follow-up studies find lower sexual recidivism rates (Caldwell, 2016; Helmus, Hanson, Thornton, Babchishin, & Harris, 2012).

Analytically, variable follow-up times may cloud results: G. T. Harris and Rice (2003) demonstrated that clearer prediction results are obtained when the offenders in a sample are followed for the same fixed follow-up time. In their meta-analysis, Hanson and Morton-Bourgon (2009) found a similar effect for studies examining the Static-99, where higher accuracies were obtained in fixed follow-up time studies, but they did not show the same effect in the overall meta-analysis of sex offender risk measures. Meta-analyses suggest that a majority of any new sexual offenses will take place within the first 5 years of opportunity; for example, Helmus et al. (2012) reported an average sexual recidivism rate of 11.1% after 5 years, and 16.6% after 10 years. Few studies have longer follow-up times.

Receiver operating characteristic (ROC) analysis is increasingly becoming the standard analysis to examine the predictive accuracy of risk measures for binary outcomes such as recidivism (yes or no; Swets, 1988). The primary metric of interest is the area under the ROC curve (AUC), which is bounded by 0 and 1: An AUC value of .5 indicates prediction at the chance level, and an AUC value of 1 indicates perfect prediction in the expected direction. In the prediction of recidivism, the AUC can be interpreted as the probability that a randomly selected recidivist will have a higher score on a risk measure than a randomly selected nonrecidivist. The AUC is an attractive measure of predictive accuracy because it is less sensitive to very low (or very high) rates of recidivism or to the selection ratio (the proportion of offenders who are predicted to reoffend) than a correlation coefficient, percentage of offenders correctly classified, or the values for sensitivity and specificity at a particular cut-score (Babchishin & Helmus, 2016), although it is still affected by the distribution of scores (Howard, 2017). These overall properties have led to a call for its use as a standard measure of diagnostic and predictive accuracy (Rice & Harris, 1995; Swets, Dawes, & Monahan, 2000).

On the other hand, the AUC has been criticized because it is possible to get a high AUC value despite ordinary performance at particular cut-scores, for example, at a high specificity for recommending civil commitment of a sex offender in the United States or at a high sensitivity when considering whether a sex offender should be released on parole. Each jurisdiction has a particular base rate of sexual recidivism, and thus each jurisdiction is expected to have its own optimal cut-off score. The optimal cut-off score will also vary according to the relative costs of false positive errors (incorrectly identifying a nonrecidivist as someone who is likely to reoffend) versus false negative errors (incorrectly identifying a recidivist as someone who is unlikely to reoffend).

Fortunately, the AUC permits the comparison of different risk measures developed on samples with different base rates of recidivism, and practitioners or researchers can determine the optimum cut-scores for their particular purpose by examining each ROC curve. Rice and Harris (1995) discussed how to determine cutoff scores for different base rates and situations in which the relative costs of false positive or false negative errors are not equal. Moreover, the AUC is one (but not the only) performance indicator regarding the discrimination of higher and lower risk offenders (see J. P. Singh, 2013, but also a critical rejoinder by Helmus & Babchishin, 2017).

Recently, researchers have been interested in the problem of calibration of risk measures, that is, how well the observed recidivism rates match the expected recidivism rates from developmental samples. Calibration can be distinguished from discrimination, which is the focus of analyses based on AUC or other metrics: How well do measures distinguish those who are more from those who are less likely to reoffend? Discrimination is very helpful for decision making focusing on prioritization or rank ordering, but some psycholegal questions are specifically about the estimated probability of recidivism. For example, sex offender civil commitment case law has homed in on the idea of a greater likelihood than not as a legally meaningful threshold (i.e., greater than 50% probability of recidivism.)

The distinction between discrimination and calibration is important because meta-analysis suggests that discrimination (relative ranking) is robust across samples, whereas calibration (recidivism estimation) is not; there are significant variations in observed versus expected recidivism rates (Helmus et al., 2012). This means cross-validations of risk measures need to be concerned about both discrimination and calibration; Hanson (2017) provided a primer on how to address these concerns.

Social views and public policies about sexual offending against children seem to be driven by the belief that most sex offenders will sexually reoffend if they have opportunities to do so. This perceived risk to sexually offend again is used to justify legal policies that greatly restrict the liberties of convicted sex offenders, including long prison sentences; the option to civilly commit offenders as they leave prison (currently an option in 20 American states: see http://soccpn.org/civil-commitment-info/ ); and other legal controls, such as (public) registration of sex offender information, including photographs, contact information, and sexual offense history; residency restrictions that limit where offenders can live in the community based on proximity to schools, day care centers, and playgrounds; and community notification of nearby residents when offenders move into a neighborhood.

Putting aside the evidence for or against these policies—which I consider in the next chapter —the assumption that most sex offenders will sexually reoffend is not correct. In a landmark meta-analysis, Hanson and Bussière (1998) quantitatively reviewed sex offender follow-up studies and found that the average sexual recidivism rate for a total of 9,603 sex offenders against children was 13%, after an average follow-up time of 5 to 6 years. Hanson and Morton-Bourgon (2004, 2005) found a similar rate of 13.7% after an average follow-up time of 5 to 6 years, with data from 84 studies, representing 20,440 sex offenders (not only offenders against children). In a more recent meta-analysis of studies following female sex offenders, the median sexual recidivism rate was 1.5 years after an average 6-year follow-up.

For juveniles, Caldwell (2010) found in a meta-analysis of 63 data sets that 7% of 11,219 juveniles who had sexually offended—the large majority of whom have offended against younger children—committed another sexual offense within a 5-year follow-up. In a more recent meta-analysis of 98 data sets, representing 33,783 cases, Caldwell (2016) reported that the average sexual recidivism rate was 4.9% over 59 months. The recidivism rates were lower for a newer cohort, 2.8% for studies between 2000 and 2015, compared with 10.3% for 1980 to 1995. However, this study did not control for risk scores, only follow-up length. This cohort difference could reflect changes over time in average risk to reoffend, how offenders are managed, or societal constraints. For example, one could speculate that more public awareness about sexual abuse and greater parent involvement with their children has reduced opportunities to sexually offend. Some have even speculated that the onset of the Internet and online gaming has reduced opportunity more generally for antisocial behavior of all kinds, which could then show up in lower sexual offending (see Mishra & Lalumière, 2009).

Some might respond that these are short follow-up periods and that more sex offenders will reoffend as follow-up continues. Focusing on studies with longer follow-up periods, Hanson, Steffy, and Gauthier (1993) followed 197 adult sex offenders against children for an average of 21 years after release from custody. Forty-two percent were convicted of a violent and/or sexual offense during this time; 23% offended 10 or more years after their release. Prentky, Knight, and Lee (1997) followed 111 men who had sexually offended against an unrelated child and who were released from a treatment center between 1959 and 1984. Forty (36%) of these men had sexually reoffended after a follow-up period of up to 24 years; some did not reoffend until many years after being released. Långström (2002) examined a sample of 117 juveniles who had sexually reoffended and found 30% had committed another sexual offense during the follow-up period (of almost 10 years). In all of these longer-term recidivism studies, it was not the case that most—or even a large majority—sexually reoffended, even with decades of opportunity. 1

Another objection that might be raised is that these studies rely on official records to study recidivism, whether it is defined as arrest, charge, or conviction. Thus, only sexual offenses that are reported to authorities and legally acted on in some way are counted; other new offenses may go unreported or are not acted upon, and thus the official recidivism rate is an underestimate. This is clear when looking at official and actual criminal histories. For example, in a meta-analysis of online (mostly child pornography offenders), we found that one in eight had a criminal record showing a prior contact sexual offense; however, in the subset of six studies with self-report information because of polygraph interviewing or treatment disclosures, half had a contact sexual offense history (Seto, Hanson, & Babchishin, 2011). In another demonstration of this effect, most self-identified persons with pedophilia or hebephilia seeking assessment or treatment through a confidential community-based clinic had committed child pornography or contact sexual offenses but had never been detected by authorities.

Researchers do not know the extent to which observed recidivism underestimates actual recidivism because they would have to be able to follow and ask sex offenders about new offenses, and it is highly unlikely that an offender would ever admit to undetected sexual offenses, even with a promise of confidentiality (e.g., using a certificate of confidentiality available from the National Institutes of Health in the United States). The extent of the underestimation is unclear because multiple factors are involved, including the likelihood that a victim will report a new sexual offense, police investigate and make an arrest, and an arrest leads to prosecution (leading to conviction). The reality that many sexual offenses are unreported might be offset by the fact that identified sex offenders are followed more closely on probation or parole postcustody, being involved in community treatment, or being subject to legal obligations such as public registration. Hanson, Morton, and Harris (2003) suggested that actual recidivism rates are 10% to 15% higher than observed rates. Even if the gap was higher, it is still correct that many sex offenders do not sexually reoffend, even after long follow-up periods. The challenge for clinicians and policymakers is to distinguish those who are likely to sexually reoffend and those who are not.

Multiple quantitative reviews of decades of sex offender recidivism research have shown two major risk dimensions that can be described as atypical sexual interests and antisociality (Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2004, 2005). 2 Atypical sexual interests is a latent dimension that reflects paraphilias, such as pedophilia, and excessive sexual preoccupation. Indicators of paraphilia include phalometrically assessed sexual arousal to children or to sexual coercion and other objective measures of sexual interest, such as viewing time; sexual offending history (particularly sexual victim characteristics); paraphilic pornography use; and self-reported sexual thoughts, fantasies, urges, and behavior about paraphilic targets or activities. Indicators of excessive sexual preoccupation include self-reported frequency and intensity of sexual thoughts, fantasies, or urges, and sexual behavior including masturbation, mainstream pornography use, and activity with others. Many of the diagnostic indicators of pedophilia discussed in Chapter 2 predict sexual recidivism among identified sex offenders: pedophilia indicators associated with risk include phallometrically assessed sexual arousal to children, having a boy victim, and having unrelated victims. In fact, these atypical sexual interest variables are among the strongest predictors of sexual recidivism that have been identified so far (Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2004, 2005).

Antisociality is a latent dimension that reflects factors associated with general antisocial and criminal behavior, including antisocial personality traits such as impulsivity or callousness, antisocial attitudes and beliefs (e.g., a disregard for the rule of law), and procriminal identification and association (e.g., gang membership). Antisociality can also be inferred from behavioral evidence of early conduct problems, juvenile delinquency, and criminal history. These same variables predict reoffending among mixed groups of offenders (Gendreau, Little, & Goggin, 1996) and among other specific offender groups, including sex offenders with adult victims, female offenders, juvenile offenders, and mentally disordered offenders (Bonta, Blais, & Wilson, 2014; Lipsey & Derzon, 1998; Rettinger & Andrews, 2010).

These risk dimensions are psychologically meaningful because atypical sexual interests can be viewed as the motivations for committing sexual offenses, whereas antisociality represents factors that facilitate acting on those motivations. Among identified sex offenders, both antisociality and atypical sexual interests predict sexual recidivism, whereas antisociality also predicts nonsexual recidivism (Clift, Rajlic, & Gretton, 2009; Gretton, McBride, Hare, O’Shaughnessy, & Kumka, 2001; Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2004, 2005). In other words, sex offenders who are high on atypical sexual interest measures are more likely to commit a sexually motivated offense, and those who are high on antisociality are more likely to commit another criminal offense, whether sexual or nonsexual. The highest risk group is high on both dimensions; pedophilic and psychopathic offenders are very likely to sexually offend again (Rice & Harris, 1997; Seto, Harris, Rice, & Barbaree, 2004).

This evidence contributed to the motivation–facilitation model of sexual offending described in Chapter 4 , which suggests that individuals high in both atypical sexual interests and antisociality are the most likely to commit sexual offenses for the first time. At the same time, having an atypical sexual interest such as pedophilia might elevate risk from the general population base rate but it is not a sufficient factor because some self-identified persons with pedophilia have never sexually offended or at least have never committed contact sexual offenses (e.g., they may have accessed child pornography). I argued in Chapter 4 that what distinguishes those who act on their atypical sexual interests from those who do not is their level of antisociality. Pedophilia is also not a necessary factor, because some highly antisocial but nonpedophilic individuals commit sexual offenses against children.

The existence of these two major dimensions of risk—atypical sexual interests and antisociality—among sex offenders against children has both practical and theoretical implications. Risk assessment measures need to include both atypical sexual interest and antisociality variables. Indeed, many of the risk measures that have been developed (discussed in the next section) contain variables reflecting atypical sexual interests (e.g., having boy victims), antisociality (e.g., prior criminal history), or both (e.g., prior sexual offense history). As discussed in Chapter 4 , theories to explain sexual offending need to incorporate both major dimensions. Finally, as I discuss in the next chapter , treatments that have been demonstrated to be effective for offenders in general are likely to be valuable templates for designing effective treatments for sex offenders. At the same time, sex offenders with atypical sexual interests may also require specific treatment targeting this risk dimension.

In the next part of this chapter, I review the empirical literature on sex offender risk assessment, including the development of valid and accurate risk measures for the prediction of sex offender recidivism. This line of work is huge and therefore my review is quite selective. Moreover, there have been major advances in risk assessment since the first edition, and I can only highlight these advances; risk assessment of sex offenders deserves its own book. Here, I summarize research on practical risk assessment concerns, including clinical adjustments of risk estimates, combining the results of multiple risk scales, and communicating risk from evaluators to decision makers. I then discuss what we know about the risk for the onset of sexual offending among men who are of concern because of noncontact sexual offending involving children (child pornography offenders, online solicitation offenders who have targeted minors) or other forms of noncontact sexual offending (exhibitionism, voyeurism). Readers who want to know more should see G. T. Harris, Rice, Quinsey, and Cormier (2015); Craig, Browne, and Beech (2008); the revised Static-99R coding manual ( http://www.static99.org ); and everything authored or coauthored by Karl Hanson. For a more general overview of violence risk assessment, see the handbook by Otto and Douglas (2011).

I end the chapter with speculations about risk to offend among at-risk individuals who are not known to have offended, as well as suggestions for future directions in risk assessment research and practice.

Given our knowledge of factors empirically associated with sexual recidivism, how should this information be combined to inform risk-related decisions regarding sentencing, institutional placement, treatment allocation, and supervision conditions? The different approaches for combining risk factor information are discussed next.

Unstructured judgment involves the subjective appraisal and weighing of putative risk factors, followed by a global statement about risk of sexual recidivism. Unstructured judgment is still a common approach in traditional clinical assessment, where the professional has discretion in terms of what information to pay attention to, how to combine it, and how to express an opinion about risk. A common expression of this approach would involve interviews with the offender, review of available file information, psychometric or other testing, and then an opinion about risk given in subjective, categorical language, for example, describing a person as “low,” “moderate,” or “high” in risk of the outcome. The same process can be carried out by probation or parole officers, judges, and other professionals who are obliged to have opinions about risk.

The early pessimism about the validity of risk assessment was justified because risk assessment was based on unstructured judgment (Monahan, 1981). Indeed, the myriad challenges for unstructured professional judgment were identified decades earlier, by Meehl (1954). The challenges are clear: Professionals can idiosyncratically pay attention to irrelevant factors that are not associated with risk of sexual recidivism, because of cognitive biases, such as the representativeness heuristic (the extent to which a factor seems to fit a prototype of a persistent sex offender) or the availability heuristic (the extent to which a factor is salient when the judgment is made; see Kahneman, 2011). Or they may ignore risk factors that are important (e.g., whether the offender had boy vs. girl victims only) or fail to take statistical covariation between risk factors into account in combining information. As an example of the first challenge, practitioners and others might consider acceptance of responsibility for the sexual offenses to be an important risk factor that needs to be targeted in sex offender treatment, even though denial or minimization of responsibility is not empirically related to sexual recidivism (Hanson & Morton-Bourgon, 2004). As an example of the second challenge, evaluators might not consider whether the offender specifically has boy victims, even though having boy rather than girl victims is a robust predictor of sexual recidivism. As an example of the third challenge, both number of prior offenses and number of prior admissions to corrections are significantly related to recidivism, but usually only the number of prior offenses uniquely contributes to the prediction of recidivism when both variables are considered in multivariate statistical analyses because they are so highly correlated with each other (Quinsey, Harris, Rice, & Cormier, 1998). Professionals can also idiosyncratically weight the risk factors they consider, giving too much weight to some and too little to others. Last, subjective labels such as “low” or “high” risk are inherently imprecise and potentially misleading; this is an issue that Hanson, Babchishin, Helmus, Thornton, and Phenix (2017) attempted to address by proposing five standard risk categories anchored by typical offender characteristics, criminal trajectories, and recidivism.

Thus, the field has turned to structured risk assessment approaches which have been repeatedly supported empirically and are becoming the dominant approaches in sex offender risk assessment (McGrath, Cumming, Burchard, Zeoli, & Ellerby, 2010). Structured judgment involves using a guide, such as a checklist, to focus one’s attention on specific risk factors (although it is not always the case that each item is empirically supported). Thus, evaluators focus on a specified set of risk factors, defeating the first problem with unstructured risk assessment. However, these checklists might still offer discretion in terms of how items are weighted and how opinions about risk are reported. A well-known example of a structured risk assessment tool for assessing sex offenders is the Sexual Violence Risk–20 (SVR-20; Boer, Hart, Kropp, & Webster, 1997), in which 20 items are grouped into the following domains: psychosocial adjustment (e.g., atypical sexual interests, psychopathy, relationship problems), sexual offending patterns (e.g., frequency and severity of offenses, victim injury, versatility), and future planning (lacks realistic plans, negative attitudes about interventions). Each item is rated as absent , possibly present or present to a limited extent , or present . Rather than add up scores, the SVR-20 developers recommended scoring the items to guide the risk assessment but then considering the totality of risk information to arrive at a relative judgment of low , moderate , or high risk , which could involve a summation of items but also the importance of a few factors or even a single critical factor. Consistent with this recommendation, Hanson and Morton-Bourgon (2009) found that a structured professional judgment produced a higher predictive accuracy in the three studies that examined the SVR-20 compared with four studies that examined the predictive accuracy of the SVR-20 items when mechanically combined.

Actuarial risk assessment differs from structured professional judgment because it specifies which factors are considered, all factors contribute to the prediction of sexual recidivism, weighting (if any) is specified, and it provides probabilistic estimates of recidivism that can be used to anchor risk opinions (e.g., reporting the proportion of sex offenders with the same score or range of scores who would be expected to sexually reoffend within a specified time period). Actuarial risk assessments are well-established in such disparate areas of practice as determining insurance coverage and predicting survival times for progressive stages of cancers (e.g., when an oncologist reports that 50% of people with Cancer X are in remission 5 years after Treatment Y ). Probabilities of making a claim are rarely reported to insurance buyers, but this information is used by insurers to determine whether potential clients are too risky or what premium should be charged in order to be profitable. In a similar vein, the original and revised versions of the Screening Scale for Pedophilic Interests described in Chapter 2 (this volume) were not developed as risk measures (although they do predict recidivism; Helmus, Ó Ciardha, & Seto, 2015; Seto, Harris, Rice, & Barbaree, 2004) but as actuarial measures for determining the likelihood that a sex offender with child victims will show a sexual preference for children over adults when assessed phallometrically. Probably the best-known and most evaluated actuarial risk assessment measure for sex offenders is the Static-99R ( http://www.static99.org ; see also Helmus, Babchishin, Hanson, & Thornton, 2009).

A large body of research demonstrates that structured or actuarial assessments are better, on average, than unstructured clinical judgment, across a wide range of assessment questions (Ægisdóttir et al., 2006; Grove, Zald, Lebow, Snitz, & Nelson, 2000), including sex offender risk assessment (Hanson, Bourgon, Helmus, & Hodgson, 2009). The greater predictive accuracy provided by structured or actuarial risk scales can translate into the prevention of many sexual offenses against children and more efficient use of limited treatment and supervision resources when decisions are made about hundreds of thousands of sex offenders. 3

Many risk assessment tools have been specifically developed for sex offenders. To illustrate, in order of frequency of use in the Safer Society survey of community programs in the United States: Static-99, Minnesota Sex Offender Screening Tool [MnSOST-R], Rapid Risk Assessment of Sexual Offence Recidivism, Static-2002, Vermont Assessment of Sex Offender Risk, and Sexual Violence Risk–20 (see Table 7.1 ). Some of these measures may be mandated by state professional bodies or by statute (e.g., in the case of civil commitment laws). Some risk measures include both static and dynamic risk factors (e.g., Violence Risk Scale—Sexual Offender version; VRS–SO) and others have been developed specifically for juveniles who have sexually offended (e.g., Estimated Risk of Adolescent Sexual Offense Recidivism [ERASOR], Juvenile Sex Offender Assessment Protocol [JSOAP]; McGrath et al., 2010). I do not go into as much detail here as in the 2008 edition on individual measures other than the Static family of measures because the field is constantly advancing in terms of research on reliability and validity in different populations and contexts. The Static family of measures—Static-99, Static-99R, Static-2002R—are the best cross-validated measures, and the Static-99R in particular shows robust predictive validity across jurisdictions and settings for predicting both sexual and nonsexually violent recidivism up to 15 years (Hanson et al., 2009). Also, the Static-99 is the most widely used, with 80% of residential programs and 71% of community programs in the United States reporting that they used this risk measure (McGrath et al., 2010).