The Predictive Validity of a General Risk/Needs Assessment Inventory on Sexual Offender Recidivism and an Exploration of the Professional Override

Abstract

This study examines the predictive validity of the Level of Service/Case Management Inventory (LS/CMI) on a sample of sexual offenders extracted from a large cohort of offenders and compares predictive validities with nonsexual offenders from the same cohort. The LS/CMI predicted sex offenders’ general recidivism, which occurred at a rate of 44.1%, with about the same accuracy as less frequently occurring violent (12.34%) and sexual recidivism (3.73%; AUC = .77, .74, and .74, respectively) and with nonsexual offenders. The study revealed that allowing assessors to override the numerically derived risk level with their professional judgment, a practice that increased risk level much more often than it decreased it, reduced the predictive validity of the scale and did so particularly for sex offenders by increasing risk excessively. An exploration of factors related to these adjustments revealed that non-risk-related characteristics were used in judgments to modify risk ratings. Implications for policy and practice are considered.

Keywords

risk assessment recidivism prediction sexual offenders LS/CMI structured professional judgment override

The past decade has witnessed tremendous progress in the assessment of sexual offenders’ risk to the community. This includes the development and validation of specialized sexual offender risk assessment instruments (e.g., Rapid Risk Assessment for Sexual Offense Recidivism [RRASOR; Hanson, 1997], STATIC-99 [Hanson & Thornton, 2000], Sex Offender Need Assessment Rating [Hanson & Harris, 2000a], STABLE-2007 [Hanson, Harris, Scott, & Helmus, 2007], ACUTE [Hanson et al., 2007], Risk Matrix 2000–Sexual [Thornton et al., 2003], Minnesota Sex Offender Screening Tool–Revised [MnSOST-R; Epperson, Kaul, & Huot, 1995], Sex Offender Risk Appraisal Guide [SORAG; Quinsey, Harris, Rice, & Cormier, 2006], Sexual Violence Risk–20 [SVR–20; Boer, Hart, Kropp, & Webster, 1997], and the Violence Risk Scale: Sex Offender version [Olver, Wong, Nicholaichuk, & Gordon, 2007]) and investigations that have compared multiple measures (Barbaree, Langton, & Peacock, 2006; Craig, Browne, & Stringer, 2004; Dempster, 1998; Gentry, Dulmus, & Theriot, 2005; Hanson & Thornton, 2000; Harris et al., 2003; Yang, Wong, & Coid, 2010). Although the emerging picture from individual investigations and meta-analyses on sexual offender risk assessment is instructive for specific risk instruments, as well as general risk tools, there is some debate as to whether general risk assessment tools have any utility in the assessment of sexual offenders.

Some comparative studies of multiple instruments have included general risk assessment tools even though the primary interest was in the predictive validity of sexual recidivism by sexual offenders (e.g., Gretton, McBride, Hare, O’Shaughnessy, & Kumka, 2001; Hanson & Bussière, 1998; Hanson & Harris, 2000b; Hanson & Morton-Bourgon, 2009; Hanson & Thornton, 2000; Parent, Guay, & Knight, 2011). This strategy seems to have been partly driven by findings that sexual offenders recidivate more frequently for nonsexual offenses, including violent and property offenses, than they do for sexual offenses (e.g., Hanson & Morton-Bourgon, 2005; Harris & Hanson, 2004; Prentky, Lee, Knight, & Cerce, 1997). However, these comparisons of instruments may be misleading as they are based on different samples and follow-up times. Therefore, further examinations of the predictive validity of general risk/need assessment schemes with sexual offenders are warranted.

The Debate about the Application of General Risk/Need Assessment Instruments to Sexual Offenders

It is commonly argued that general risk and risk/need assessment tools, such as the Level of Service (LS; Andrews, Bonta, & Wormith, 2006) scales, are not helpful in the assessment of sexual offenders or in the prediction of their further antisocial behavior, particularly their sexual reoffending. Some professionals, as well as correctional agencies (e.g., Ministry of Public Safety and Solicitor General, 2004; Probation & Welfare Service, undated), have gone so far as to assert explicitly that such tools should not be employed with sexual offenders because the assessment results are at best misleading and at worst categorically wrong. Moreover, Doren (2002) has suggested that two criteria for the selection of sex offender risk assessment tools are that they be designed specifically for use with sexual offenders and that they be designed to assess sexual recidivism. Three interrelated kinds of reasons, espoused primarily by parole and probation officers and sex offender specialists in the field, are offered for rejecting the use of general risk and risk/need assessment tools in the assessment of sexual offenders.

Much of the rationale comes from the position that general risk or risk/need assessment instruments do not tap into the element of sexual deviance, which is an important predictor of sexual recidivism among sexual offenders (e.g., Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2005, 2009), perhaps the best for at least some types of sexual offenders, such as child molesters (Looman & Abracen, 2010). The second argument is that general risk and risk/need assessment instruments attend to risk factors that are essentially superfluous for the assessment of sexual offenders. Rather, it is specifically asserted that sex offenders tend not to exhibit the traditional criminogenic needs, such as an impoverished education and employment, a strong mix of criminal as opposed to noncriminal companions, and criminal attitudes, all of which are found in high-risk nonsexual offenders. However, others have reported that many sexual offenders have histories of nonsexual offenses (e.g., Maletzky, 1991), suggesting that nonsexual criminogenic variables may be more important in predicting recidivism than otherwise thought. Third, it is often contended that general risk and risk/need tools underestimate the risk level of sexual offenders. This position is supported by the low average scores that groups of sexual offenders typically produce on general risk and risk/need tools compared to the general offender population, as found by Simourd and Malcolm (1998) among familial child molesters in the Correctional Service of Canada (CSC) with the Level of Service Inventory–Revised (LSI-R; Andrews & Bonta, 1995). However, recidivism rates are also lower for sex offenders than nonsexual offenders, and hence risk assessment scores are collectively in sync with outcome. In our view, these arguments are well intentioned but largely unfounded, with considerable evidence to the contrary, some of which is reviewed below. Possibly, they come from forensic and correctional workers’ failure to differentiate sexual offenders from sexual reoffending. This is illustrated when practitioners express dismay and incredulity about an incest offender who otherwise has led a prosocial lifestyle and is rated low risk by a general risk assessment scale.

Although some jurisdictions have declared that their general risk assessment measure is not to be administered to their sexual offender population and that staff should rely exclusively on a battery of specialized sex offender risk assessment tools (e.g., Ministry of Public Safety and Solicitor General, 2004; Probation & Welfare Service, undated), others have endorsed the use of a combination of both general and specialized risk assessment instruments. The Interstate Commission for Adult Offender Supervision (2007) reported in a survey of U.S. state practices that all but one of the 47 responding states employed some kind of general risk/need assessment tool in their assessment of sexual offenders, and about one-half of these states (i.e., 23) employed some version of LS. Some researchers, such as Gentry et al. (2005), have suggested that both kinds of instruments should be used with sexual offenders to get a more complete picture of their multiple dimensions of risk. Simourd and Malcolm (1998) even warned that it may be shortsighted to focus exclusively on specialized sexual offender assessment instruments because, with the exception of familial child molesters, their sexual offender sample displayed the same criminogenic needs as their nonsexual offenders. In the absence of a general risk/need assessment, these general criminogenic needs of sexual offenders may go undetected and untreated. It is interesting that they also found a modest correlation between the LSI-R total score and their phallometric-based measure of deviant sexual arousal.

Use of the Level of Service Instruments with Sexual Offenders

The LS scales comprise one of the most commonly used general risk assessment tools in criminal justice internationally with more that one million registered uses in 2011 alone (Wormith, 2011). Two kinds of studies have examined its predictive validity with sexual offenders. One involves the use of the instrument with a cohort of offenders in which predictive validities are broken down for sexual and nonsexual offenders. These studies have found a considerable range of correlations with various types of recidivism. Following the introduction of the LSI-R in the state of Washington in 1999, an examination of its predictive validity with sexual offenders was undertaken (Washington State Institute for Public Policy, 2006). Receiver operating characteristic (ROC) analyses (Hanley & McNeil, 1982) of the Washington study produced an area under the curve (AUC) of .650. Girard and Wormith (2004) reported on the predictive validity of the Level of Service Inventory-Ontario Revision (LSI-OR; Andrews, Bonta, & Wormith, 1995) on a small subsample of 44 sexual offenders. The total score correlated .44 with any reconviction and .31 with violent (including sexual) reconvictions. Moreover, these coefficients were quite comparable to those of the remaining 586 nonsexual offenders (.39 and .28, respectively). Vrana, Sroga, and Guzzo (2008) examined the relationship between the LSI-OR and recidivism in a random sample of 198 offenders who were convicted of sexual assault in Ontario. The general, violent, and sexual recidivism rates over at least a 2-year follow-up were 26.3%, 12.6%, and 3.0%, respectively. Since the sexual recidivism rates were so low, its correlation with risk was not computed. However, the LSI-OR correlated .41 and .32 with general and violent recidivism, respectively. Using the Youth Level of Service/Case Management Inventory (YLS/CMI; Hoge & Andrews, 2002), Caldwell and Dickinson (2009) reported AUCs of .66 and .62 for predicting any recidivism and sexual recidivism, respectively, on a sample of juvenile sex offenders

The second kind of study involves the comparison of both general and specialized risk assessment instruments on a common group of sexual offenders. In predicting violent and general reoffending on a sample of 202 high-risk Canadian sexual offenders over 41 months, Bonta and Yessine (2005) found the Level of Service: Screening Version (LSI:SV; Andrews & Bonta, 1998) had higher AUCs with recidivism than the STATIC-99, but not as high as those of the Statistical Information on Recidivism (SIR; Nuffield, 1982) scale, the Violence Risk Appraisal Guide (VRAG; Quinsey, Harris, Rice, & Cormier, 1998), or the VRAG-Proxy (VRAG without the Psychopathy Checklist–Revised [PCL-R]; Bonta & Yessine, 2005). However, only the AUCs for the SIR and STATIC-99 with general recidivism were significantly different from each other. Rossegger et al. (2011) compared several tools in a sample of violent and sex offenders released from Swiss prisons. The PCL-R (Hare, 1991) had the highest AUC, followed by the FOTRES (Urbaniok, 2007), the Historical, Clinical, Risk Management–20 (HCR-20; Webster, Douglas, Eaves, & Hart, 1997), the LSI-R, and finally the VRAG, which had the lowest AUC.

In a meta-analytic review of sex offender risk items and composite scales, Hanson and Morton-Bourgon (2009) reported on four studies that examined the predictive validity of the LS instruments and compared these results to those of other general risk assessment instruments. They concluded that the LS along with the VRAG, SORAG, and SIR were better predictors of general recidivism for sexual offenders and that the LS along with the VRAG, SORAG, the Risk Matrix–Combined, and the SIR were better predictors for violent recidivism.

Drawing on four different sets of studies, Hanson (2009) compared the predictive validity of five general and violence specific risk assessment tools to four sexual offense risk tools on three kinds of outcome, general violent, and sexual offense recidivism. The specialized sex offense prediction tools clearly predicted sexual recidivism of sexual offenders better than the general and violence-specific risk assessment tools with the average correlation across four sexual offense specific tools (STATIC-99, SVR-20, MnSOST-R, and SORAG) being .32 and the average correlation among five general and violence specific tools (HCR-20, LS, PCL-R, SIR scale, and VRAG) being .21.

Finally, two studies have examined general and specialized risk assessment instruments with adolescent sexual offenders. Morton (2003) found a significant correlation for violent recidivism, but not for general recidivism or sexual recidivism using the YLS/CMI (Hoge & Andrews, 2002). However, low correlations were found for the RRASOR (Hanson, 1997), the STATIC-99 (Hanson & Thornton, 1999), and the Estimates of Risk of Adolescent Sexual Offense Recidivism (ERASOR; Worling & Curwen, 2001) for violent, general, and sexual recidivism, respectively. The YLS/CMI was correlated particularly well with the ERASOR, a tool that was designed specifically for the prediction of sexual recidivism among adolescent sexual offenders and comprised many sexually specific items. In a larger study of 220 young offenders, comparing instruments over a follow-up period of up to 14 years, Skowron (2004) found that the ERASOR (Worling & Curwen, 2001) was a better predictor of sexual recidivism but the YLS/CMI was a better predictor of general recidivism, whereas the two were comparable for violent recidivism. These studies on youth must be regarded separately from the previous studies of adult sexual offenders as adolescent sexual offenders differ from their adult counterparts, including their low rates of recidivism (U.S. Department of Justice, Office of Juvenile Justice and Delinquency Prevention, 2001). However, the potential applicability of the YLS/CMI to adolescent sexual offenders should not be overlooked.

Augmenting Statistically Based Assessment with the Professional Judgment

Another potentially fruitful discussion in the field of offender risk assessment pertains to the integration of statistically based assessments with professional judgment. Structured professional judgment (SPJ) is the use of empirically derived guidelines to serve as an aide-mémoire in a structured, but flexible and nonlinear, manner (Douglas, Cox, & Webster, 1999; Hart, 2008). SPJ has an extensive history in forensic risk assessment and is considered an important advancement over clinical or unstructured professional judgment (UPJ). It has proven to be effective in the prediction of general and domestic violence (e.g., HCR-20 [Webster et al., 1997]; Spousal Assault Risk Assessment Guide [SARA; Kropp, Hart, Webster, & Eaves, 1999; Kropp & Hart, 2000]; Brief Spousal Assault Form for the Evaluation of Risk [Kropp & Hart, 2004]). It has also been proposed for the assessment of sexual offender risk assessment (e.g., SVR-20 [Boer et al., 1997] and Risk for Sexual Violence Protocol [RSVP; Hart, Kropp, Laws, Klaver, Logan, & Watt, 2003]).

A second tempting approach, although largely untested in a systematic, empirical fashion, is to use individual or unique circumstances and patterns of the offender to augment the statistically based prognostications in the hope of boosting the predictive validity of the final assessment. The use of this “professional override” approach (Andrews, Bonta, & Hoge, 1990) is given further impetus from norms studies that have generated variable recidivism rates with a common instrument.¹ In fact, in their analysis of differing recidivism rates on the STATIC-99, Helmus, Hanson, and Thornton (2009) suggested that the assessor must make a determination as to where along a range of recidivism rates that a specific offender is likely to fall. Some interpret this to mean that the assessor should look for “red flags” to specify an offender’s degree of risk more accurately and to criticize the exclusive use of risk tools because they fail to meet the standards of scientific rigor (Sreenivasan, Weinberger, Frances, & Cusworth-Walker, 2010). Others make no such inference; rather, they believe that the assessor should use the norm group to which the offender is the best fit and use these norms without making any personal or professional judgments about placement of an offender along a range of projected recidivism rates (Abbott, 2011).

Therefore, researchers are often perceived as giving mixed messages to practitioners. For example, Hanson and Morton-Bourgon (2009) pointed out that the predictive validity of UPJ was weak in the prediction of sexual offender recidivism, whereas the predictive accuracy of SPJ fell between UPJ and actuarial measures. Yet Helmus et al. (2009) may be perceived as giving tacit approval to professional judgment by noting “until further research is conducted, however, this professional judgment may be unavoidable” (p. 42). To this end some supporting evidence may be derived from Eher and colleagues, who demonstrated that a psychiatric diagnosis of narcissistic personality disorder supplemented the predictive validity of the STATIC-99 and STABLE-2007 (Eher, Retternberger, Matthes, & Schilling, 2010).

Rather than pitting the two traditions against one another, arguments have been made for combining the statistical and clinical tradition, presumably to capitalize on the best of two approaches (Abbott, 2011). For example, how should the assessor augment a general risk/need assessment with sex offender specific information? One possible strategy is to employ a traditional risk/need assessment tool to obtain a baseline or “ballpark” risk level for the offender and then, depending on what idiosyncratic risk information might be available, augment, or override, the initial assessment by raising or lowering risk with the presence of additional risk and protective factors. In our view, the professional override, as employed in the LS/CMI, is a version of SPJ as it is consistent with the aforementioned description. However, it differs from the traditional use of SPJ, with such instruments as the HCR-20 and the SARA, in that it is applied to supplement the initial actuarial-based assessment rather than to serve as a stand-alone assessment.

The Current Investigation

This study was conducted to examine the predictive validity of a widely used general risk/need instrument (LS/CMI; Andrews, Bonta, & Wormith, 2004) on a large sample of sexual offenders and to investigate the use of the professional override to augment the risk/need assessment by examining sexual offenders taken from a large cohort of Canadian provincial offenders (probationers and prisoners).² The predictive validity of the LS/CMI was examined both before and after assessors were given the opportunity to override their initial assessment. The extent and circumstances of the override were examined, as was its impact on the predictive validity of the LS/CMI. The examination of predictive validity extended beyond traditionally assessed general risk/needs to other sections for two reasons: First, they may contain items that are particularly relevant to sexual offenders (e.g., specific risk/needs); and second, these analyses may be instructive in understanding under what conditions assessors elected to use the override (e.g., strengths). These analyses were also conducted on the remainder of the cohort, the nonsexual offenders, for comparison purposes.

Method

Participants

The sample was derived from a cohort of offenders who were under the responsibility of the Ministry of Community Safety and Correctional Services (MCSCS) in the province of Ontario, Canada. The original cohort included all male and female offenders who, during one calendar year (2004), were released from Ontario provincial correctional facilities after serving a sentence of at least one month, were sentenced to a “conditional sentence” (traditionally a custodial sentence but allowed to be served in the community, as per the Criminal Code of Canada), or began a term of probation with MCSCS.³ The sample consisted of all offenders in the cohort who had been administered an LS/CMI in conjunction with their sentence. The sexual offense sample was made up of 1,905 sex offenders, of whom 1,849 (97.1%) were male. They included 733 (38.5%) inmates who were released from a prison sentence, 349 (18.3%) offenders who were given a conditional sentence to be served in the community, and 823 (43.2%) offenders who were given a term of probation. Their mean age at the end of the follow-up period (i.e., the date at which evidence of recidivism was sought) was 41.91 (SD = 12.66) years. Most were Caucasian (61.1%), with the balance being Aboriginal (12.1%), Black (8.6%), and other or unknown (18.2%). Prisoners were sentenced to an average of 224.20 (SD = 152.52) days in custody and 184.11 (SD = 254.95) days under community supervision (i.e., probation following incarceration). Those on a conditional sentence were sentenced to an average sentence of 333.52 (SD = 202.21) days. Probationers were sentenced to an average of 573.53 (SD = 325.28) days on probation.

The nonsexual offender sample consisted of 24,545 offenders, of whom 19,767 (80.5%) were male. Broken down by type of disposition, they included 4,217 (17.2%) inmates who were released from a custodial sentence, 2,876 (11.7%) offenders who were given a conditional sentence to be served in the community, and 17,452 (71.1%) offenders who were given a term of probation. Their mean age at the end of the follow-up period was 37.63 (SD = 11.57). Most were Caucasian (59.0%), with the balance being Black (7.2%), Aboriginal (6.0%), and other or unknown (27.9%). Prisoners were sentenced to an average of 191.24 (SD = 134.54) days in custody and 148.12 (SD = 148.12) days under community supervision. Those on a conditional sentence were sentenced to an average of 270.40 (SD = 192.34) days. Probationers were sentenced to an average of 462.35 (SD = 229.04) days on probation.

Measures

LS/CMI

The Level of Service/Case Management Inventory (LS/CMI; Andrews et al., 2004) is described by some as a “fourth generation” risk assessment tool (following traditional [unstructured] clinical assessment, static [actuarial] risk, and dynamic [criminogenic] risk/need), in that it goes beyond traditional risk and needs by including other clinically relevant factors and incorporating a case management portion (Andrews et al., 2006), thus extending beyond its predecessor the LSI-R (Andrews & Bonta, 1995). However, the general risk/need score of the LS/CMI correlates very highly with the 54-item LSI-R (r = .96; Rowe, 1999; Andrews et al., 2004).⁴ The LS/CMI includes general risk/needs (Section 1) consisting of 43 items, each of which is scored in a dichotomous fashion (0 = not present, 1 = present). The items are organized into the central eight (Andrews & Bonta, 2010) subscales: Criminal History (8 items), Education/Employment (9 items), Family/Marital (4 items), Leisure/Recreation (2 items), Companions (4 items), Procriminal Attitude/Orientation (4 items), Substance Abuse (8 items), and Antisocial Pattern (4 items). The LS/CMI also includes the concept of strength. Any of the central eight subscales that are problem free or serve to “protect” the offender from other sources of risk may be declared a “strength” by the assessor. A total strength score is derived from the simple summation of strengths across the central eight domains (Andrews et al., 2004).

These items are totaled to create eight domain scores and a total general risk/need score, which is then used to determine the offender’s initial risk level on a 5-point ordinal scale ranging from very low risk to very high risk. The initial risk level may be “overridden” in either direction (i.e., from a lower to higher risk level or from a higher to a lower risk level) to create a final risk level. The two risk level variables were coded from 1 to 5, and an “override score” was calculated by subtracting the initial risk level score from the final risk level score. The scoring manual encourages assessors to exercise the override function sparingly. They are also instructed to consider other sections of the LS/CMI for aggravating (described below) and mitigating factors (e.g., strengths) that might suggest an adjustment to the score-based risk level using SPJ.

Specific risk/needs (Section 2) includes two subscales: Personal Problems with Criminogenic Potential (14 items) and History of Perpetration (9 items), also scored dichotomously. These items are intended to identify additional risk factors and criminogenic needs, as well as guide the assessors in deciding whether the risk level should be adjusted or overridden. The LS/CMI consists of three additional sections intended to guide case management: Institutional Factors (Section 3; 10 items), which records problems and management issues during previous incarcerations; Other Client Issues (Section 4; 18 items), which includes social, health, and mental health issues that are likely to deserve particular attention; and Special Responsivity Considerations (Section 5; 8 items), which includes characteristics such as ethnicity, cognitive disabilities, and personality features that are relevant to how one works with an offender. Although the items in these sections were meant to be used individually for clinical and case management purposes, they were summed to give an indication of the extent to which these issues (difficulties in prison; social, health, and mental health issues; and personal characteristics that are relevant to one’s approach with the offender) were identified by the assessor.

Recidivism

For the purpose of the current study, recidivism was defined as any criminal offense for which an offender was returned to MCSCS. These offenses are recorded in the Offender Tracking and Information System (OTIS), which is operated by MCSCS and documents all criminal offenses that occur in Ontario. The total follow-up period ran from offenders’ release from custodial sentence for prisoners or their admission to community supervision for probationers and those on conditional sentences in 2004 to the data extraction date in January 2009. The mean follow-up time was virtually the same for sex offenders and nonsexual offenders (M = 1658.72, SD = 106.68 days and M = 1658.62, SD = 106.41 days, respectively).

Five measures of recidivism were constructed from offender file information. First, a dichotomous variable (yes = 1, no = 0) was created to identify those who did and did not recidivate during the follow-up period. Second, severity of reoffense was based on the Offense Severity Scale (OSS; Stasiuk, Winter, & Nixon, 1996), which consists of 26 categories of offenses that were rank ordered in accordance with the mean sentence length for each offense category (Ontario, 1983). This scale was originally developed by MCSCS from an analysis of 60,000 sentences given to offenders in Ontario over a period of one year, where the average sentence length determined offense severity (Stasiuk et al., 1996). Offense categories ranged from 0 (no reconviction) and 1 (municipal bylaw offences) to 25 (homicide). Third, a dichotomous violent recidivism measure (yes = 1, no = 0) combined six categories of violence from the OSS (Assault and Related; Miscellaneous Offenses Against the Person; Weapons Offenses; Non-violent Sexual Offenses; Serious Violent Offenses). Fourth, a dichotomous sexual recidivism measure combined two categories (yes = 1, no = 0) from the OSS (Non-violent Sexual Offenses and Violent Sexual Offenses). The fifth recidivism variable was the time to recidivate. For the custodial sample, the time to recidivate was represented by the number of days from the release date to the date of reoffense or reentry into custody. In the community sample, this was the time from the LS/CMI assessment date to the date of reoffense or entry into custody.

Procedure

Offenders who were released from a custodial sentence or who were admitted to a conditional sentence or to probation in 2004 were identified electronically from the ministry’s OTIS. Descriptive information was obtained from OTIS including age, gender, and ethnicity (Aboriginal and non-Aboriginal) status. An automated version of the LS/CMI was introduced into the organization in 1997, allowing field staff to enter all details of their assessment into an electronic record for scoring and record keeping. The LS/CMI is administered to all adult inmates who are sentenced to at least one month in custody and to all probationers in Ontario. Therefore, a computer search on the LS/CMI database was conducted to identify all inmates of the cohort who had been administered an LS/CMI during their period of incarceration in 2004 and all community offenders in the cohort who had been administered an LS/CMI at the outset of their community supervision, also in 2004.

The LS/CMI was administered by probation officers in the community and by classification officers, and occasionally by psychologists and social workers, in custodial settings. All assessors participate in a comprehensive training prior to their being authorized to administer an LS/CMI. In Ontario, training is primarily obtained in conjunction with a mandatory training course offered at the Ontario Correctional Services College. This course also includes material on risk, need, and responsivity, lessons on effective offender supervision, and an evaluation of trainees’ ability to perform the LS/CMI. On their return to the field, supervisors perform a quality assurance function by conducting periodic audits of offender files. Versions of the LS had been used in the organization for about 20 years at the time of data collection.

Sex offender and nonsexual offender groups were created based on electronically coded offense data. The sex offender group included all offenders who had any convictions for a previous or current (the index offense) sexual offense (i.e., prior to the start of the follow-up period in 2004). The sexual offender sample included 1,905 offenders. Of these, 759 (40%) had at least one conviction for a current sexual offense.

Any evidence of recidivism, as indicated by a reconviction, was then recorded for each offender. General, violent, and sexual recidivism were coded according to the offense categories described previously. The data from the two data files were then merged by offender identification number into a single file for data analyses. The final data file included descriptive legal and demographic information about the offender, including variables to identify sexual and nonsexual offenders and the type of sentence that was being served (prison, conditional sentence, or probation), the LS/CMI total and item scores, and the five measures of recidivism.

Data Analysis

Descriptive statistics were obtained on the sample of sex offenders and comparisons made to the nonsexual offenders as well as comparisons within the sexual offenders by type of sentence. Reliability analysis was limited to assessments of internal consistency using Cronbach’s (1951) coefficient alpha. Predictive validity was assessed with correlations and ROC curves (Hanley & McNeil, 1983). The impact of SPJ (the override) on predictive validity was assessed by comparing predictive validity estimates of the risk levels before and after the override feature was exercised (i.e., initial and final risk level). Correlation and multiple regression analyses were used to identify offender characteristics that were related to the use of the override feature. Item analyses of all LS/CMI items were also conducted, correlating item scores with recidivism and the use of the override. Although the prime focus of this investigation was on sexual offenders, analyses were conducted on both sexual offenders and nonsexual offenders to compare the performance of the LS/CMI across samples.

Results

Comparisons of Sex Offenders and Nonsexual Offenders

As reported in Table 1, sexual offenders were significantly older than nonsexual offenders and were more likely to be male and Aboriginal. They also had a higher offense severity score on their index offense, which is expected given the severity ratings of the sexual offense categories. Sexual offenders scored significantly higher on all LS/CMI summary measures, except strengths, on which they scored significantly lower. These measures included the LS/CMI general risk/need total score and corresponding risk level, both before and after the uses of the override function, and the specific risk/need score. They also scored higher on the measure of risk level change, indicating that assessors used the override feature to increase their risk level significantly more than they did for nonsexual offenders. It is noted, however, that a number of these differences, although significant, are relatively small. Sexual offenders had a higher rate of general reoffending than nonsexual offenders and did so more quickly. However, there was no difference in violent and sexual reoffending.

Table 1:

Comparisons of Sex Offenders and Nonsexual Offenders on Demographic Characteristics, Level of Service/Case Management Inventory (LS/CMI), and Recidivism

	Sexual Offender	Nonsexual Offenders
	M (SD) or Percentage (SD) (n = 1,905)	M (SD) or Percentage (SD) (n = 24,545)	t Test or Chi-Square	Effect Size (Cohen’s d)
Age	41.91 (12.66)	37.63 (11.57)	t(2157.92) = −14.30***	.37
Offense severity	18.34 (4.52)	15.42 (3.88)	t(2126.31) = −27.43***	.74
Male (n [%])	1849 (97.06)	19767 (80.53)	χ²(1) = 323.28***	.22
Female (n [%])	56 (2.94)	4778 (19.47)	χ²(1) = 323.28***
Aboriginal (n [%])	231 (12.15)	1461 (5.95)	χ²(1) = 112.53***
Non-Aboriginal (n [%])	1674 (87.85)	23084 (94.05)	χ²(1) = 112.53***	.13
General risk/needs	16.53 (10.03)	12.09 (8.58)	t(2125.59) = −18.80***	.51
Specific risk/needs	4.56 (3.12)	2.33 (2.19)	t(2051.87) = −30.59***	.98
Strength	0.71 (1.50)	0.90 (1.65)	t(2277.66) = 5.29***	−.12
Initial risk level	3.06 (1.20)	2.52 (1.11)	t(26448) = −20.11***	.48
Final risk level	3.60 (.98)	2.68 (1.06)	t(2264.65) = −39.00***	.87
Risk level change	0.54 (.96)	0.16 (0.54)	t(1996.69) = −16.98***	.65
General reoffense	44.41 (0.50)	33.86 (0.47)	t(2180.55) = −8.96***	.22
Violent reoffense (%)	12.34 (0.33)	12.63 (0.33)	t(26338) = 0.37, ns	.03
Sexual reoffense (%)	3.73 (0.19)	3.17 (0.18)	t(2164.00) = −1.25, ns	.03
Time to recidivate (days)	1247.64 (703.76)	1407.95 (636.36)	t(2152.64) = 9.64***	.25
Recidivism severity	14.36 (5.14)	13.79 (4.47)	t(9151) = −3.47***	.13

Note. t test, in most cases, equal variance not assumed. Standardized mean difference for t tests, with standard deviations and weighted sample sizes, is converted to Cohen’s d and chi-square is converted to Cohen’s d with the Effect Size Calculator at http://www.psychsystems.net/manuals/StatsCalculators/Effect_Size_Calculator%2017.xls.

***

p < .001. ns = nonsignificant.

Internal Consistency

Internal consistency of the LS/CMI was examined using Cronbach’s alpha. Since three of the LS/CMI items are calculated in part based on offender’s score on previous items, the alpha calculation was calculated without these three items with α = .91 on the full sample. The alpha coefficient was lower and quite varied for the eight domains of the general risk/need section. Small coefficients were systematically related to domains having few items (e.g., Leisure/Recreation with two items; α = .43), but large coefficients with longer domains (e.g., Criminal History with 8 items; α = .87).

Predictive Validity of the LS/CMI

As seen in Table 2, the general risk/needs score was highly correlated with general recidivism for both sexual offenders and nonsexual offenders (r = .47, p < .001, and r = .43, p < .001, respectively; large effect sizes). Correlations for the prediction of violent recidivism were significantly lower (both ps < .001) for both groups (r = .28, p < .001, and r = .29, p < .001, respectively; medium effect sizes) and were significantly lower (both ps < .001) still for sexual recidivism (r = .17, p < .001, and r = .19, p < .001, respectively; small effect sizes). These findings may be expected as the base rates of violent and sexual recidivism are increasingly different from 50%. Sources of the LS/CMI’s predictive validity for general recidivism are reflected in the coefficients from the central eight domains. The correlations with general recidivism were higher for sexual offenders than nonsexual offenders on all domains, with differences being significant on criminal history leisure/recreation, companions, and substance abuse. However, this pattern did not carry over to the prediction of violent or sexual recidivism.

Table 2:

Correlations for Level of Service/Case Management Inventory (LS/CMI) Total and Section Scores With General, Violent, and Sexual Recidivism on All Offenders, Sexual Offenders, and Nonsexual Offenders

	General Recidivism			Violent Recidivism			Sexual Recidivism
LS/CMI Section	Sex Offenders (n = 1,905)	Nonsexual Offenders (n = 24,545)	z (differ.)	Sex Offenders (n = 1,905)	Nonsexual Offenders (n = 24,545)	z (differ.)	Sex Offenders (n = 1,905)	Nonsexual Offenders (n = 24,545)	z (differ.)
General Risk/Needs	.47***	.43***	2.11*	.28***	.29***	−0.46	.17***	.19***	−0.87
Initial Risk Level	.45***	.42***	1.55	.27***	.27***	0.00	.16***	.17***	−0.43
Final Risk Level	.26***	.37***	−5.14***	.18***	.23***	−2.10*	.11***	.15***	−1.71
Criminal History	.48***	.41***	3.67***	.29***	.28***	0.46	.16***	.18***	−0.87
Education/Employment	.34***	.31***	1.41	.24***	.22***	0.89	.14***	.14***	0.00
Family/Marital	.21***	.17***	1.74	.11***	.10***	0.42	.08***	.06***	0.84
Leisure/Recreation	.30***	.24***	2.72**	.16***	.17***	−0.43	.11***	.11***	0.00
Companions	.37***	.31***	2.85**	.24***	.22***	0.89	.15***	.14***	0.43
Procriminal Attitudes	.28***	.24***	1.80	.15***	.16***	−0.43	.11***	.11***	0.00
Substance Abuse	.34***	.29***	2.33*	.18***	.16***	0.87	.09***	.12***	−1.27
Antisocial Patterns	.37***	.33***	1.92	.23***	.23***	0.00	.16***	.16***	0.00
Total Strengths	−.12***	−.12***	0.00	−.05*	−.08***	1.27	−.05*	−.04***	−0.42
Specific Risk/Needs	.37***	.32***	2.39*	.15***	.16***	−0.43	.11***	.13***	−0.85
Personal Problems	.32***	.31***	0.47	.12***	.18***	−2.58***	.08***	.13***	−2.12*
Perpetration History	.34***	.23***	5.04***	.15***	.09***	2.56*	.12***	.09***	1.27
Prison Experience	.30***	.28***	0.92	.21***	.22***	−0.44	.12***	.17***	−2.15*
Social, Health, Mental Health	.23***	.18***	2.19*	.12***	.13***	−0.43	.09***	.06***	1.27
Responsivity	.18***	.18***	0.00	.07***	.10***	−1.27	.12***	.07***	2.12*

Note. z (differ.) = Fisher r to z transformation, two-tailed. r = .10 corresponds to d = .20 as per Rosenthal’s (1994) “r to d formula” (p. 239), which is described as a “small” effect size by Cohen (1988); r = .24 corresponds to d = .50, which is described as a “medium” effect size; and r = .37 corresponds to d = .80, which is described as a “large” effect size (pp. 281, 284, 285).

p < .05, two-tailed. **p < .01, two-tailed. ***p < .001, two-tailed.

Concerning the less frequently examined section of the LS/CMI, the specific risk/need section, and both of its subscales were highly correlated with general recidivism for sexual offenders (r = .37, p < .001). However, their correlations with violent and sexual recidivism for both sexual and nonsexual offenders, although significant, were substantially lower. It is interesting that sexual offenders’ noncriminogenic needs (social, health, and mental health) were also correlated with general recidivism (r = .23, p < .001), but less so for violent and sexual recidivism. Similarly, responsivity displayed modest correlation with general recidivism among sexual offenders (r = .18, p < .001), but less so with violent and sexual recidivism. As expected, strengths were negatively correlated with recidivism. All of these correlations compare quite favorably to those for nonsexual offenders. Moreover, analyses of various subgroups by gender, ethnicity (Aboriginal), and sentence type revealed a similar pattern to the preceding analyses, correlations being high for the prediction of general recidivism but low for violent and sexual recidivism, with one exception. General risk/need correlations with general (r = .36, p < .01), violent (r = .10, ns) and sexual recidivism (r = .12, ns) were consistently lower for female sexual offenders but not significantly different from their male counterparts because of the small sample size (n = 56).⁵

A series of correlation analyses were conducted to determine the LS/CMI risk levels with general, violent, and sexual recidivism and then to examine the possible decrease in predictive validity, first, when one collapses from 43 items to five risk levels and, second, when practitioners are allowed to override the score based on other pieces of information and their professional judgment. The initial risk level is the score-derived level, and the final risk level is the risk level after the override option has been applied. The correlations between the initial LS/CMI risk level for general, violent, and sexual recidivism for the entire sample of sexual offenders and the various subgroups mirror the correlations derived from the total score, although, as one would expect, they consistently show a slight decrease in predictive validity. The initial risk level correlated r = .45, p < .001 with general recidivism, followed by r = .27, p < .001 for violent recidivism and r = .16, p < .001 for sexual recidivism in the total sample. The same pattern was found for sexual offender subgroups defined by race, gender, and type of sentence.

When these analyses were repeated using the final risk level, a consistent decrease in the correlations emerged across the three measures of recidivism. The final risk level correlated with general recidivism r = .26, p < .001, violent recidivism r = .18, p < .001, and sexual recidivism r = .11, p < .001, on the complete sample of sexual offenders. These correlations were significantly lower than the corresponding initial risk level correlations with recidivism at .001, .001, and .02, respectively. The same pattern was found for sexual offender subgroups defined by race, gender, and type of sentence. It is also noted that the final risk level correlations were significantly lower for sex offenders than nonsexual offenders in the prediction of any and violent recidivism (Table 2). Possible reasons for this emerging difference are explored later.

ROC Analyses

AUC renders a better estimate of predictive validity for sexual recidivism, given its very low base rate, and affords a more equitable comparison of the predictive validity of the LS/CMI across outcome variables that have very different base rates (Rice & Harris, 2005). Therefore, a series of ROC analyses were conducted to examine the LS/CMI total and section scores with general and violent recidivism. As seen in Table 3, the general risk/need total score produced AUCs of .77 and .75 (ps < .001) for sexual offenders and nonsexual offenders, respectively, on sexual recidivism. These coefficients are comparable to those found using specialized sexual offender risk assessment tools. Moreover, they are also similar to AUCs found for the prediction of general recidivism for sexual offenders (AUC = .77, p < .001) and for nonsexual offenders (AUC = .76, p < .001) and the prediction of violent recidivism for sexual offenders (AUC = .74, p < .001) and for nonsexual offenders (AUC = .73, p < .001). Inspection of AUCs in Table 3 reveals a slight decrease going from risk score to risk level, but a larger, consistent, and significant decrease (illustrated by nonoverlapping confidence intervals, where p < .05) when going from initial to final risk level (e.g., from .75 to .64 for sexual offenders’ general recidivism). Although many AUCs for nonsexual offender less than .60 were highly significant because of the very large sample, they would be of little or no practical value.

Table 3:

Receiver Operating Characteristic Coefficients and Confidence Intervals for Level of Service/Case Management Inventory (LS/CMI) Total and Section Scores for General and Violent Recidivism on All Offenders, Sexual Offenders, and Nonsexual Offenders

	General Recidivism		Violent Recidivism		Sexual Recidivism
LS/CMI Section	Sex Offenders (n = 1,905)	Nonsexual Offenders (n = 24,545)	Sex Offenders (n = 1,905)	Nonsexual Offenders (n = 24,545)	Sex Offenders (n = 1,905)	Nonsexual Offenders (n = 24,545)
General Risk/Needs	.77 (.75 to .79)	.76 (.75 to .76)	.74 (.71 to .77)	.73 (.72 to .74)	.74 (.69 to .80)	.77 (.75 to .78)
Initial Risk Level	.75 (.73 to .77)	.74 (.74 to .75)	.73 (.70 to .76)	.72 (.71 to .73)	.73 (.68 to .79)	.76 (.74 to .77)
Final Risk Level	.64 (.61 to .66)	.71 (.71 to .72)	.65 (.62 to .69)	.69 (.68 to .70)	.66 (.59 to . 72)	.73 (.72 to .75)
Criminal History	.78 (.76 to .80)	.73 (.72 to .74)	.75 (.72 to .78)	.72 (.71 to .73)	.74 (.69 to .80)	.76 (.74 to .77)
Education/Employment	.69 (.67 to .72)	.68 (67 to .69)	.70 (.67 to .74)	.68 (.67 to .69)	.71 (.65 to .77)	.71 (.69 to .73)
Family/Marital	.62 (.59 to .64)	.60 (.59 to .60)	.59 (.55 to .63)	.58 (.57 to .59)	.61 (.54 to .68)	.60 (.58 to .62)
Leisure/Recreation	.66 (.64 to .69)	.64 (.63 to .65)	.63 (.59 to .67)	.64 (.63 to .65)	.65 (.59 to .71)	.66 (.64 to .68)
Companions	.71 (.68 to .73)	.68 (.67 to .68)	.70 (.67 to .74)	.68 (.67 to .69)	.71 (.66 to .77)	.70 (.68 to .72)
Procriminal Attitudes	.66 (.63 to .68)	.63 (.62 to .64)	.63 (.59 to .66)	.62 (.61 to .63)	.66 (.59 to .72)	.64 (.62 to .66)
Substance Abuse	.70 (.67 to .72)	.67 (.66 to .68)	.65 (.62 to .69)	.63 (.62 to .64)	.64 (.58 to .70)	.68 (.66 to .70)
Antisocial Patterns	.70 (.68 to .72)	.67 (.66 to .67)	.69 (.65 to .72)	.66 (.65 to .67)	.71 (.65 to .77)	.70 (.68 to .72)
Total Strengths^a	.57 (.55 to .60)	.56 (.55 to .57)	.56** (.52 to .60)	.56 (.55 to .57)	.58* (.51 to .64)	.55 (.54 to .57)
Specific Risk/Needs	.71 (.69 to .74)	.68 (.67 to .68)	.64 (.60 to .68)	.63 (.62 to .64)	.66 (.60 to .72)	.68 (.66 to .70)
Personal Problems	.68 (.65 to .70)	.67 (.66 to .68)	.61 (.57 to .65)	.64 (.63 to .65)	.62 (.56 to .68)	.68 (.66 to .70)
Perpetration History	.68 (.66 to .71)	.61 (.60 to .62)	.63 (.59 to .67)	.55 (.54 to .56)	.66 (.60 to .73)	.61 (.59 to .63)
Prison Experience	.66 (.63 to .68)	.62 (.61 to .63)	.68 (.64 to .71)	.63 (.62 to .64)	.68 (.61 to .74)	.67 (.65 to .70)
Social, Health, Mental Health	.62 (.60 to .65)	.61 (.60 to .62)	.60 (.56 to .64)	.62 (.61 to .63)	.62 (.56 to .70)	.60 (.58 to .62)
Responsivity	.60 (.57 to .62)	.60 (.60 to .61)	.56 (.52 to .60)	.58 (.57 to .59)	.64 (.58 to .71)	.60 (.58 to .62)

Note. 95% confidence intervals are in parentheses.

The coding of recidivism was reversed for Total Strengths to predict success as opposed to recidivism.

All ps < .001, except *p < .05 and **p < .01.

Use of the Override

Findings in Tables 2 and 3 related to risk levels raise questions about the value of the override feature and the apparent loss of predictive validity when this kind of SPJ is afforded to practitioners (e.g., the prediction of sexual offenders’ general recidivism fell from r = .45, p < .001, to r = .26, p < .001). Moreover, when the predictive validity of LS/CMI risk level on general recidivism was examined only for the 669 sex offenders whose risk level was changed, the correlation fell from .33, p < .001, for the initial risk level to a nonsignificant .02 for the final risk level. For the 3,694 nonsexual offenders whose risk level was changed, the correlation fell from .36, p < .001, to .14, p < .001. It was also noted that the override option was exercised for the complete sample much more frequently to increase risk than to decrease risk (14.9% vs. 1.6%). This difference was even more pronounced in the adjustment to sexual offender risk level (33.5% vs. 1.6%) compared to nonsexual offenders (13.5% vs. 1.6%), χ²(2) = 560.53, p < .001.

In an effort to determine what may have contributed to the decrease in predictive validity with the use of the override function, two additional analyses were performed. First, the sexual offender sample was assigned to an initial-by-final risk level matrix and recidivism rates within each cell were examined. Inspection of the recidivism rates across the initial-final risk level cells was consistent with a decrease in predictive validity of the assessment process. For example, the 263 sexual offenders who were initially in the low-risk category but were overridden to medium, high, and very high risk actually recidivated at lower rates (24.2%, 19.0%, and 5.9%, respectively) that the low-risk sexual offenders who were not overridden (31.7%). On the other hand, the relatively few sexual offenders (n = 31) who were overridden to a lower risk level appeared to have been adjusted appropriately as their recidivism rates were lower than their unadjusted counterparts. For example, the recidivism rate of the 19 very high-risk sexual offenders whose risk levels were reduced was considerably lower (63.2%) than their unadjusted counterparts (84.4%).

A second strategy was to examine the relationship between the risk category change score and a number of demographic and LS/CMI variables. However, the direction and magnitude of changes in risk level caused by use of the override were highly related to the LS/CMI total score because of the asymmetry of the override process (i.e., high-risk offenders are already high risk and are more likely to be overridden downward, whereas low-risk offenders have much more room available to be overridden upward). Therefore, partial correlations controlling for LS/CMI general risk/needs total score were computed (Table 4). For sexual offenders, controlling for risk, increases in risk level by means of the override were not correlated with age or race (Aboriginal), although they were negatively related to being female. Among LS/CMI scales, overriding risk level to a higher level was correlated with Total Specific Risk/Needs and its subscales Personal Problems with Criminogenic Potential and History of Perpetration, Prison Experience, Social, Health, and Mental Health Problems, and Responsivity Considerations. Increased risk level was also related to three general risk/need subscales: Low Criminal History, High Procriminal Attitudes, and High Antisocial Pattern. A similar pattern was found for nonsexual offenders, with one particularly notable exception—increased risk level was unrelated to criminal history.

Table 4:

Partial Correlation Matrix With Level of Service/Case Management Inventory (LS/CMI) Section Scores and Override Score Controlling for Total General Risk/Needs Score (Section A) on the Complete Sample, Sexual Offenders, and Nonsexual Offenders

LS/CMI Section	Sex Offenders (n = 1,904)	Nonsexual Offenders (n = 24,543)
Control Measure
Total Section A^a	−.57***	−.31***
Demographic Measures
Ethnicity (Aboriginal)	.03	−.01
Age	.04	.08***
Gender (Female)	−.05*	−.08***
LS/CMI Scales
Criminal History	−.11***	−.00
Education/Employment	−.00	−.07***
Family/Marital	.01	.06***
Leisure/Recreation	−.01	.01
Companions	−.01	−.06***
Procriminal Attitudes	.11***	.09***
Substance Abuse	−.03	−.01
Antisocial Patterns	.16***	.08***
Total Strengths	.02	−.03***
Total Section B	.15***	.18***
Personal Problems	.14***	.17***
Perpetration History	.08***	.11***
Prison Experience	.05*	.05***
Social, Health, Mental Health	.10***	.03***
Special Responsivity	.12***	.13***

Zero-order correlation for the control variable (Total Section A) with outcome (override score).

p < .05. **p < .01. ***p < .001.

The demographic and LS/CMI variables that were correlated with sexual offenders’ change (increase) in risk level by use of the override in Table 4 were then submitted to two multiple regression analyses, first on change in risk (Table 5) and second on general recidivism (Table 6). These analyses were performed only on sexual offenders. After entering LS/CMI total score in Block 1 and the remaining demographic and LS/CMI measures in Block 2, the regression analysis improved in a significant but minimal way, F_change(7, 1896) = 9.17, p < .01. Measures that were related to change in risk level, independent of risk score, included Personal Problems, History of Perpetration, Social, Health, and Mental Health Problems, and Responsivity.

Table 5:

Multiple Regression of Level of Service/Case Management Inventory (LS/CMI) Sections on Risk Level Change Score (final risk level minus initial risk level) for Sexual Offenders (n = 1,905)

	Unstandardized Coefficients		Standardized Coefficients		95.0% Confidence Interval for B
Model	B	SE	β	t	Lower Bound	Upper Bound
Step 1
(Constant)	1.454	0.035		41.638***	1.385	1.522
Total LS/CMI score	−0.055	0.002	−.573	−30.517***	−0.059	−0.052
Step 2
(Constant)	1.475	0.132		11.167***	1.216	1.734
Total LS/CMI score	−0.069	0.003	−.718	−23.444***	−0.075	−0.063
Personal Problems with Criminogenic Potential Total (B1 of LSI)	0.045	0.013	.096	3.363***	0.019	0.071
History of Perpetration Total (B2 of LSI)	0.030	0.016	.047	1.849^†	−0.002	0.062
Prison Experience: Institutional Factors Total (C1 of LSI)	−0.002	0.021	−.002	−0.076	−0.043	0.039
Gender (female)	−0.171	0.109	−.030	−1.572	−0.385	0.042
Social Health and Mental Health Total (F1 of LSI)	0.022	0.009	.059	2.556*	0.005	0.040
Special Responsivity considerations (G1 of LSI)	0.055	0.019	.064	2.839**	0.017	0.093
Age at Data Extraction	0.002	0.001	.023	1.188	−0.001	0.005

†

p < .10. *p < .05. **p < .01. ***p < .001.

Table 6:

Multiple Regression of Level of Service/Case Management Inventory (LS/CMI) Sections on General Recidivism for Sexual Offenders (n = 1,905)

	Unstandardized Coefficients		Standardized Coefficients		95.0% Confidence Interval for B
Model	B	SE	β	t	Lower Bound	Upper Bound
Step 1
(Constant)	0.058	0.019		3.000**	0.020	0.096
Total LS/CMI score	0.023	0.001	.471	23.315***	0.021	0.025
Step 2
(Constant)	0.361	0.073		4.951***	0.218	0.504
Total LS/CMI score	0.019	0.002	.386	11.762***	0.016	0.022
Personal Problems with Criminogenic Potential Total (B1 of LSI)	0.002	0.007	.007	0.213	−0.013	0.016
History of Perpetration Total (B2 of LSI)	0.032	0.009	.095	3.497***	0.014	0.049
Prison Experience: Institutional Factors Total (C1 of LSI)	0.012	0.012	.028	1.067	−0.010	0.035
Gender (female)	0.019	0.060	.007	0.319	−0.099	0.137
Social Health and Mental Health Total (F1 of LSI)	−0.006	0.005	−.031	−1.231	−0.015	0.004
Special Responsivity considerations (G1 of LSI)	−0.019	0.011	−.043	−1.770^†	−0.040	0.002
Age at Data Extraction	−0.007	0.001	−.170	−8.167***	−0.008	−0.005

†

p < .10. *p < .05. **p < .01. ***p < .001.

To assess the wisdom of using the above noted demographic and LS/CMI variables in exercising the override function, the predictor variables from the preceding multiple regressions were then applied to general recidivism as the dependent variable. As was the case in the previous analyses, the general risk/need score was applied in the first block, followed by the remaining demographic and other LS/CMI measures and the analyses were performed separately on the sexual offender and nonsexual offender samples. By comparing the results of this multiple regression (Table 6) to the previous one (Table 5), it becomes apparent that variables that contributed incrementally, beyond the general risk/need score, to increases in risk by means of the override were frequently not the same as those that contributed incrementally, beyond the general risk/need score, to the prediction of general recidivism, and vice versa. For example, responsivity, which was associated with an increase risk rating, was associated marginally (p < .10) with a decrease in recidivism, whereas age, which was associated with a decrease in recidivism, was not associated with any change in rated risk. Conversely, history of perpetration was associated with recidivism and marginally (p < .10) with an increase in rated risk.

Finally, relationships among individual items, change in risk level, and recidivism were examined. Particular attention was paid to items in Sections 2, 4, and 5 because these sections were related to the use of the override and specific items in these sections might have been responsible for these decisions but may not be justified empirically. Among Specific Risk/Need (Section 2) items, Inappropriate Sexual Activity (2.1.8), was correlated with increases in risk level for sexual offenders, but not nonsexual offenders (.19 and .04, respectively) when controlling for total risk score. However, this item was negatively correlated with general, violent, and sexual recidivism (–.19, –.12, and –.06, respectively) for sexual offenders, but not nonsexual offenders, suggesting it was inappropriately being used to influence assessors’ use of the override for sexual offenders. Second, Sexual Assault–Extrafamilial (2.2.1) was correlated with increases in the risk level while controlling for total risk score (.10), for sexual offenders, but not nonsexual offenders. However, it was unrelated to general, violent, and sexual recidivism for the sexual offender sample (.03, .04, and .03, respectively).

Among Other Client Issues (Section 4) items, Shy/Withdrawn (4.1.4) and Diagnosis of Serious Mental Disorder (4.14) were both correlated (.06 and .06) with increases in risk level for the sexual offender sample while controlling for total risk score but were not for the nonsexual offender sample (.00 and .03). However, for the sexual offender sample, neither was related to general or sexual recidivism and Shy/Withdrawn was negatively related to violent recidivism (–.05). For nonsexual offenders, these items correlated .02 or less with the three outcome measures, suggesting they had an inappropriate influence on the override with both sexual and nonsexual offenders.

Among Responsivity (Section 5) items, Interpersonally Anxious (5.3) was the most strongly correlated item with increases in risk level among sexual offenders (.11) while controlling for the total risk/needs score. It was also correlated with general recidivism (.08), although not for violent or sexual recidivism, suggesting the item may be appropriately contributing to the use of the override in predicting sexual offenders’ general recidivism.

Discussion

This study examined the applicability of the LS/CMI to sexual offenders by comparing its predictive validity on a large sample of provincial sexual offenders in Ontario to their nonsexual offender counterparts in the original cohort. The rate of sexual reoffending after 4.54 years was particularly low (3.73%), as has been reported elsewhere (e.g., 3.8% by the Washington State Institute for Public Policy, 2006), whereas the rate of general recidivism (44.41%) was substantial, which is also consistent with other sex offender samples (e.g., 43.0% by Lanagan, Schmitt, & Durose, 2003). Sexual offenders had a higher rate of general reoffending and reoffended more quickly than nonsexual offenders. However, there were no differences in violent and sexual recidivism between the sexual and nonsexual offenders. Finally, assessors used the override feature of LS/CMI to increase sex offenders’ risk level significantly more frequently than they did for nonsexual offenders.

Results from this study have strengthened the argument that general risk/need assessment instruments, such as the LS/CMI, may play an important role in the assessment of sexual offenders. The high rate of general recidivism found for the sex offender sample as well as the similar correlations and ROCs between the sex offender and nonsexual offender samples in examining the LS/CMI and recidivism support the relevance of the LS/CMI in sex offender risk assessment. Correlations with the LS/CMI were strongest for general recidivism, followed by violent recidivism and sexual recidivism for both sexual and nonsexual offenders, whereas ROCs remained quite consistent across outcome measures. Second, an ROC of .74 for the prediction of sexual offenders’ sexual recidivism is not out of line with findings with specialized sex offender risk scales (e.g., .76 for the RRASOR, .70 for the SORAG, .67 for the Structured Anchored Clinical Judgment Scale (SACJ-Min; Thornton, 1997), and .65 for the MnSOST-R (Craig et al., 2004)]. However, one is reminded that these findings come from different samples, and more compelling evidence would be derived from comparisons of multiple instruments on the same sample.

Sources of the LS/CMI’s predictive validity for general recidivism on both sexual and nonsexual offenders are also reflected in the coefficients from the central eight domains. It is interesting that the correlations with general recidivism were higher for sexual offenders than nonsexual offenders on all domains except substance abuse. However, this pattern did not carry over to the prediction of violent or sexual recidivism. Analyses of subgroups of sexual offenders by gender and ethnicity revealed similar patterns.⁶

Use of the Professional Override with the LS/CMI

The effect of the override was examined by comparing the predictive validity of the initial and final risk levels. Two relatively small but consistent findings were observed. First, there was a decrease in predictive validity across all comparisons after the override provision was exercised. This included sex offenders and nonsexual offenders over three outcome measures (general recidivism, violent recidivism, and sexual recidivism) and in both the correlation and ROC analyses. Second, the decrease caused by the use of the override was larger for the sexual offender sample on all three measures of recidivism. Their predictive validities were very similar to those for nonsexual offenders initially but were consistently lower than those for nonsexual offenders on the final risk level. Moreover, as evidenced by the nonoverlapping 95% confidence intervals, the decrease in validity was significant for general and violent recidivism, although all predictive validities remained highly significant. However, when the validity of the LS/CMI was examined for sex offenders on whom the override provision was applied, the predictive relationship was eliminated.

This suggests that extraneous factors completely negated the utility of the instrument when the override was introduced to sexual offenders as a group. Although this overall finding was particularly discouraging for efforts to increase the validity of the scale for sex offenders using the professional override, it does not mean that such an approach cannot work. In fact, our findings suggest that the use of the override function to lower risk level, as rare as that was, may indeed be an appropriate procedure. This makes some intuitive sense in that one can only imagine the kind of confidence and conviction that an LS/CMI assessor must have when making such a decision, knowing the potential consequences of doing so erroneously, and resisting any pressure to be excessively conservative. On the other hand, using the override to increase an LS/CMI based risk level among a large portion of the sexual offender population (i.e., 33.5%) without more explicit directions as to how to do so is likely only to deteriorate the process and the predictive validity of the instrument.

Retrospective analyses attempted to determine possible reasons for assessors’ using the override. Because there was a strong correlation between the LS/CMI total score and use of the override (i.e., low scores were associated with increases in risk level), statistical measures were invoked to control for the LS/CMI total score. This procedure generated some intriguing findings that may give some glimpse into how practitioners accommodate their own particular “theories” about sex offender risk. While controlling for overall risk, criminal history was associated with SPJ-based decreases in risk, whereas procriminal attitudes and antisocial pattern were associated with SPJ-based increases in risk. Quite possibly, assessors suspect that the LS/CMI underestimates the risk presented by sexual offenders who do not have an extensive criminal history. It is important that the same kind of adjustment to risk was not evidenced among nonsexual offenders. Conversely, it appears that assessors may believe that the presence of antisocial attitudes and an antisocial pattern are not given enough weight in the LS/CMI scoring scheme and hence tend to increase their risk level if these two domains are high, a practice that was also found in the assessment of nonsexual offenders.

Both subsections of Special Risk/Needs, Personal Problems and History of Perpetration, were also correlated with overriding to a higher risk level, as they were with nonsexual offenders. This finding is quite expected as the LS/CMI manual indicates that these factors are possible reasons for exercising the override feature. However, it was unexpected that the Special Responsivity section and the Social, Health, and Mental Health section would be positively correlated with the override function, although they were for nonsexual offenders as well. This finding, which suggests assessors may be influenced by a range of demographic and other clinical features, perhaps unknowingly, to elevate offender risk, is disconcerting, reminiscent of findings in other clinical domains (e.g., Garb, 1997), and deserves further investigation. Conversely, strengths were not correlated with a reduction of risk for sexual offenders but were, slightly so, for nonsexual offenders. These findings were augmented by a multiple regression analysis of LS/CMI section risk scores on the override change variable indicating that, after the total risk/need score is taken into consideration, personal problems, history of perpetration, social health and mental health issues, and responsivity all contributed positively to assessors’ SPJ to increase offender risk level.

Item analysis in relation to use of the override, although exploratory, shed some light on what may prove to be an ill advised use of the override, particularly for sexual offenders. A number of items were correlated with the use of the override when controlling for the total risk/need score, but not with outcome. For example, being shy and withdrawn or having a diagnosis of psychosis was correlated with override-based increases in risk level among sexual offenders (i.e., override to higher risk), while controlling for total risk. This decision was not supported by the analysis of these items’ incremental validity beyond the total risk score. On the other hand, being interpersonally anxious, although a responsivity item, appears to be used appropriately to increase risk level of sexual offenders. However, the effects of these items on the use of the override must be described as small but very reliable given the sample size. Nonetheless, the contribution of individual nonrisk items to SPJ must be considered preliminary and merit further investigation prior to making pronouncements to users in the field.

With respect to demographic characteristics, age and ethnicity (Aboriginal) were unrelated to use of the override on sexual offenders, whereas being female was correlated with lowering the risk level. Similar effects were found for nonsexual offenders, although an increase in risk level was also correlated with age. The gender finding is not surprising in light of the popular notion that risk assessment tools generally do not treat women offenders fairly (e.g., Andrews, et al., 2012; Blanchette & Brown, 2006; Van Voorhis, Wright, Salisbury, & Bauman, 2010). In sum, it appears that other complicating factors in the sexual offenders’ lives may be related to augmenting the risk level of sexual offenders. It is unknown whether this is a systematic conscious decision by the assessors, or if these factors contribute to an unspecified sense of uneasiness that leads assessors to increase sexual offenders’ risk level.

The current findings were revealing in that variables that contributed incrementally, beyond the general risk/need score, and independent of each other, to the use of the override were not always the same as those that contributed incrementally to the prediction of recidivism. Since their zero-order correlations with recidivism were significant and substantial and their scores are not used to determine a sex offender’s general degree of general recidivism, it is understandable how assessors might use personal problems and social, health, and mental health problems to make an adjustment to the final assessment, typically by raising risk level. However, based on the finding that these measures did not add any incremental validity to the prediction of recidivism, although others such as perpetration history and responsivity did, it becomes apparent how the final risk level correlates less well with recidivism than the initial risk level. In our view, these findings illustrate the potential shortcomings of using SPJ to augment a statistically based risk/need assessment scheme.

The use of the professional override to augment the results of a statistically based risk assessment is particularly relevant in the assessment of sexual offender risk. Increasingly, decisions to detain sexual offenders are being made at the judicial level through special provisions in law, such as the involuntary commitment of sexual predators in at least 17 U.S. states (Washington State Institute for Public Policy, 2005), and the Dangerous Offender provisions, which are in effect throughout Canada and are applied primarily (77%) to sexual offenders (Public Safety Canada, 2010). Consequently, pressure is only increasing on professionals to provide accurate, expert testimony in sexual offender cases and on researchers to develop and validate mechanisms to do so.

In this context, it is not surprising to find individual, case-specific efforts to augment the predictive accuracy of established statistical tools that have been designed for this very purpose. For example, Sreenivasan et al. (2010), have encouraged clinical professionals also to consider “red flags” that are indicative of such features as deviant drives, and use their “education, skill, and professional experience” (p. 405). In our view, professionals should be very cautious about engaging in any such strategy until the validity of a particular strategy being entertained has been empirically demonstrated. Failure to do so is likely to decrease predictive validity rather than increase it, at least when applying it to elevate sexual offenders risk level as derived from the LS/CMI total score.

Notably, some sexual offender characteristics that contributed incrementally to use of the override made no incremental contribution to the prediction of recidivism (i.e., personal problems, social, health, and mental health needs, and responsivity), whereas other characteristics that did not make any contribution to use of the override contributed incrementally to the prediction of recidivism (i.e., history of perpetration and age [being young]). These findings have three potentially important implications. First, they point to a possible source of the deteriorating predictive validity when the override is used (i.e., assessor bias). Second, they offer directions for researchers to improve the instrument (e.g., age-related norms). Third, they suggest valuable instructions for assessor training (i.e., cautious use of the override).

Limitations and Further Directions

Five limitations of the current investigation merit consideration, two of which concern measurement issues and three relate to inferences which may or may not be drawn from this study. First, since the LS/CMI data were derived from an existing database of the agency, it was impossible to determine the accuracy of the LS/CMI assessments. Hundreds of probation officers and correctional staff with various years of service and familiarity with the LS/CMI instruments were responsible for administering the LS/CMI. The fact that the LS/CMI data were entered into an electronic database using specially designed LS/CMI software ensures only that no logical or arithmetic errors were made in scoring the instrument.

Second, the assessment of criminal recidivism as the outcome measure was based on internal agency recontact with offenders. This included all reconvictions in the province in which the agency was located, but not out-of-province reconvictions. Since Ontario covers a very large geographic area, it is assumed that vast majority of reconvictions were captured in the agency’s database. Moreover, since reconviction was employed as the criterion for recidivism as opposed to rearrest, recently arrested offenders who await trail and conviction would be “missed.” On the other hand, conviction captures some recidivists whom reincarceration would miss. Regardless, the net effect of these two limitations in the data is to decrease estimates of predictive validity.

Third, the cohort was limited to provincial sexual offenders meaning that sexual offenders who were sentenced to 2 years or more of custody were not included. As sentence length can be interpreted as a general measure of the severity of an offense (Quirk, Nutbrown, & Reynolds, 1991), the most serious convicted offenders (sexual and nonsexual) in the province were not included. As this is a truncated cohort study, excluding the most serious sexual and nonsexual offenders in the province, one must be cautious about generalizing these findings to jurisdictions that do not make this kind of legal distinction based on sentence length. For example, it is possible that less “serious” sexual offenders are more likely to commit nonsexual offenses on release. On the other hand, the exclusion of federal sexual offenders is likely to have reduced the heterogeneity of the sex offender cohort in terms of their degree of risk, in which case the currently reported findings would represent conservative estimates of association between the LS/CMI and sexual recidivism. Including both federal and provincial offenders would increase the variance of the LS/CMI scores (Wormith, Olver, Stevenson, & Girard, 2007).

Fourth, since the agency does not routinely administer specialized sex offender risk instruments, we were unable to compare the predictive validity of the LS/CMI to a specialized sex offender tool, which may have proven very instructive as comparisons between validities across different samples are problematic for a number of reasons. Moreover, we were unable to speculate about the potential merits of using both a general and sexual-offender-specific risk assessment tool as part of an overall sex offender assessment protocol, as has been suggested (e.g., Simourd & Malcolm, 1998). This strategy remains of interest to us, in large part because it is now apparent that a sizeable portion of the sex offender population is generally antisocial. Findings from the current investigation that demonstrate the incidence of general and nonsexual violent offending among a large heterogeneous cohort of sexual offenders and the predictive validity of the LS to identify these offenders indicates that this line of research needs to be explored in a thorough and systematic fashion.

Finally, our examination of the override function and implications for SPJ is based on only one characterization of SPJ, the professional override, which may be idiosyncratic to the LS/CMI. We cannot conclude that the current findings apply to other SPJ procedures that are designed to augment quantitative risk instruments. However, the current findings do not auger well for the professional override having overall incremental predictive validity, at least as employed with the current guidelines in the LS/CMI. Rather, they partially support Abbott’s (2011) position that professional judgment should not be used to increase actuarially based assessments of sexual offenders’ risk of recidivism.

Conclusion

This study was undertaken to assess the appropriateness and value of using a general risk/need assessment instrument on a specialized offender population. Results supported the use of the LS/CMI on sexual offenders as demonstrated by its predictive validity coefficients that were comparable to nonsexual offenders from the same cohort. Many sexual offenders share the same criminogenic risks and needs as nonsexual offender and, therefore, may be assessed on the same general risk/need assessment instruments as nonsexual offenders (Simourd & Malcolm, 1998). However, we do not suggest that the LS/CMI be administered to sex offenders in lieu of specialized sex offender risk assessment tools, such as the STATIC-99. Rather, we support the recommendation of Gentry et al. (2005) that the two instruments be used in concert, although their potential contribution as a supplement to one another remains to be explored systematically. Alternatively, one might select a sexual offender risk tool that includes most of the risk/need factors that are included in the LS/CMI, instead of a “pure” sexual offender risk instrument. The Stable-2007, SVR-20, and RSVP appear to be possible candidates.

The use of the professional override with the LS/CMI led to a slight, but systematic, deterioration in the predictive validity of the LS/CMI. These results should encourage assessors to be very cautious about applying the override or other kinds of SPJ in their assessments of sexual offenders with the LS/CMI, particularly if the intent is to increase risk level. Further research on the override and use of SPJ to augment any quantitatively based predictions about sexual offenders’ recidivism using the LS/CMI is recommended.

Footnotes

Authors’ Note:

The authors wish to express their appreciation to the Ontario Ministry of Community Safety and Correctional Services (MCSCS) for access to the data described in this study. The views expressed herein do not necessarily reflect the views of MCSCS. J. Stephen Wormith receives royalties from sales of the Level of Service/Case Management Inventory from its publisher, Multi-Health Systems.

Notes

J. Stephen Wormith received his PhD from the University of Ottawa. Previously, he was psychologist-in-chief with the then named Ministry of Correctional Services of Ontario. He is currently professor, Department of Psychology, and director, Centre for Forensic Behavioural Science and Justice Studies, both at the University of Saskatchewan.

Sarah Hogg received her master of arts in applied social psychology (forensic) from the University of Saskatchewan. She is currently working as a senior statistics officer in the Program Effectiveness, Statistics and Applied Research Unit (PESAR) of the Ontario Ministry of Community Safety and Correctional Services.

Lina Guzzo (formerly Girard) received her PhD from the University of Ottawa. She practiced clinical and forensic psychology in adult and children’s mental health prior to taking a position with the Ontario Correctional Services Division. She is currently manager of the Program Effectiveness, Statistics and Applied Research Unit with the Ontario Ministry of Community Safety and Correctional Services.

References

Abbott

B. R.

(2011). Throwing the baby out with the bath water: Is it time for clinical judgment to supplement actuarial risk assessment? Journal of the American Academy of Psychiatry and Law, 39, 222-230.

Andrews

D. A.

Bonta

(1995). LSI-R: The Level of Service Inventory–Revised. Toronto, Canada: Multi-Health Systems.

Andrews

D. A.

Bonta

(1998). The Level of Service Inventory–Revised: Screening Version. Toronto, Canada: Multi-Health Systems.

Andrews

D. A.

Bonta

(2010). The psychology of criminal conduct (5th ed.). New Providence, NJ: LexisNexis Matthew Bender.

Andrews

D. A.

Bonta

Hoge

R. G.

(1990). Classification for effective rehabilitation: Rediscovering psychology. Criminal Justice and Behavior, 17, 19-52.

Andrews

D. A.

Bonta

Wormith

J. S.

(1995). The Level of Service Inventory: Ontario Revision. Toronto, Canada: Ontario Ministry of Correctional Services.

Andrews

D. A.

Bonta

Wormith

J. S.

(2004). The Level of Service/Case Management Inventory. Toronto, Canada: Multi-Health Systems.

Andrews

D. A.

Bonta

Wormith

J. S.

(2006). The recent past and near future of risk/need assessment. Crime & Delinquency, 52, 7-27.

Andrews

D. A.

Guzzo

Raynor

Rowe

R. C.

Rettinger

L. J.

Brews

Wormith

J. S.

(2012). Are the major risk/need factors predictive of both female and male reoffending? A test with the eight domains of the Level of Service/Case Management Inventory. International Journal of Offender Therapy and Comparative Criminology, 56, 113-133.

10.

Barbaree

H. E.

Langton

Peacock

E. J.

(2006). Different actuarial risk measures produce different risk rankings for sexual offenders. Sexual Abuse: A Journal of Research and Treatment, 18, 423-440.

11.

Blanchette

Brown

S. L.

(2006). The assessment and treatment of women offenders. Chichester, UK: Wiley.

12.

Boer

D. P.

Hart

S. D.

Kropp

P. R.

Webster

C. D.

(1997). Manual for the Sexual Violence Risk–20: Professional guidelines for assessing risk of sexual violence. Vancouver, BC, Canada: Mental Health, Law and Policy Institute.

13.

Bonta

Yessine

A. K.

(2005). The National Flagging System: Identifying and responding to high-risk, violent offenders (User Report 2005-03). Ottawa: Public Safety and Emergency Preparedness Canada. Retrieved from https://www.publicsafety.gc.ca/res/cor/rep/2005-04-flg-sysm-eng.aspx

14.

British Columbia Ministry of Public Safety and Solicitor General. (2004). Assessing the risk of reoffending. Retrieved from: http://www.probation.ie/pws/websitepublishing.nsf/attachmentsbytitle/Assessing+Risk+of+Re-offending/$file/Assessing+Risk+of+Re-offending.pdf

15.

Caldwell

M. F.

Dickinson

(2009). Sex offender registration and recidivism risk in juvenile sexual offenders. Behavioral Sciences and the Law, 27, 941-956.

16.

Campbell

T. W.

(2003). Sex offenders and actuarial risk assessments: Ethical considerations. Behavioral Sciences and the Law, 21, 269-279.

17.

Campbell

T. W.

DeClue

(2010). Flying blind with naked factors: Problems and pitfalls in adjusted-actuarial sex-offender risk assessment. Open Access Journal of Forensic Psychology, 2, 75-191. Retrieved from http://www.forensicpsychologyunbound.ws/OAJFP/Volume_2__2010_files/Campbell%202010.pdf

18.

Cohen

(1988). Statistical power analysis for the behavioural sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum.

19.

Craig

L. A.

Browne

K. D.

Stringer

(2004). Comparing sex offender risk assessment measures on a UK sample. International Journal of Offender Therapy and Comparative Criminology, 48, 7-27.

20.

Cronbach

L. J.

(1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16, 297-334.

21.

Dempster

R. J.

(1998). Prediction of sexually violent recidivism: A comparison of risk assessment instruments (Unpublished master’s thesis). Simon Fraser University, Vancouver, British Columbia, Canada.

22.

Doren

D. M.

(2002). Evaluating sex offenders: A manual for civil commitments and beyond. Thousand Oaks, CA: Sage.

23.

Douglas

K. S.

Cox

D. N.

Webster

C. D.

(1999). Violence risk assessment: Science and practice. Legal and Criminological Psychology, 4, 149-184.

24.

Eher

Retternberger

Matthes

Schilling

(2010). Stable dynamic risk factors in child sexual abusers: The incremental predictive power of narcissistic personality traits beyond the STATIC-99/STABLE-2007 priority categories on sexual reoffense. Sexual Offender Treatment, 5, 1-12.

25.

Epperson

D. L.

Kaul

J. D.

Huot

S. J.

(1995, October). Predicting risk of recidivism for incarcerated sex offenders: Updated development on the Sex Offender Screening Tool (SOST). Poster session presented at the 14th Annual Research and Treatment Conference of the Association for the Treatment of Sexual Abusers, New Orleans, LA.

26.

Garb

H. N.

(1997). Race bias, social class bias, and gender bias in clinical judgment. Clinical Psychology: Science and Practice, 4, 99-120.

27.

Gentry

A. L.

Dulmus

C. N.

Theriot

M. T.

(2005). Comparing sex offender risk classification using the STATIC-99 and LSI-R assessment instruments. Research on Social Work Practice, 15, 557-563.

28.

Girard

Wormith

(2004). The predictive validity of the Level of Service Inventory–Ontario Revision on general and violent recidivism among various offender groups. Criminal Justice and Behavior, 31, 150-181.

29.

Gretton

H. M.

McBride

Hare

R. D.

O’Shaughnessy

Kumka

(2001). Psychopathy and recidivism in adolescent sex offenders. Criminal Justice and Behavior, 28, 427-449.

30.

Hanley

J. A.

McNeil

B. J.

(1982). The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology, 143, 29-36.

31.

Hanley

J. A.

McNeil

B. J.

(1983). A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology, 148, 839-843.

32.

Hanson

(1997). The Development of a Brief Actuarial risk Scale for Sexual Offense Recidivism. Ottawa: Department of the Solicitor General of Canada. Retrieved from http://www.defenseforsvp.com/Resources/Hanson_Static-99/RRASOR.pdf

33.

Hanson

R. K.

(1998). What do we know about sex offender risk assessment? Psychology, Public Policy and Law, 4, 50-72.

34.

Hanson

R. K.

(2009). The psychological assessment of risk for crime and violence. Canadian Psychology, 50, 172-180.

35.

Hanson

R. K.

Bussière

M. T.

(1998). Predicting relapse: A meta-analysis of sexual offender recidivism studies. Journal of Consulting and Clinical Psychology, 66, 348-362.

36.

Hanson

R. K.

Harris

(2000a). The Sex Offender Need Assessment Rating (SONAR): A method for measuring change in risk levels. Ottawa: Corrections Research, Department of the Solicitor General of Canada. Retrieved from http://www.publicsafety.gc.ca/res/cor/rep/sonar-eebds-eng.aspx

37.

Hanson

R. K.

Harris

A. J. R.

(2000b). Where should we intervene? Dynamic predictors of sex offense recidivism. Criminal Justice and Behavior, 27, 6-35.

38.

Hanson

R. K.

Harris

J. R.

Scott

T.-L.

Helmus

(2007). Assessing the risk of sexual offenders on community supervision: The dynamic supervision project (User Report No. 2007-05). Ottawa: Public Safety Canada. Retrieved from http://publications.gc.ca/collections/collection_2007/ps-sp/PS3-1-2007-5E.pdf

39.

Hanson

R. K.

Morton-Bourgon

K. E.

(2005). The characteristics of persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73, 1154-1163.

40.

Hanson

R. K.

Morton-Bourgon

K. E.

(2009). The accuracy of recidivism risk assessments for sexual offenders: A meta-analysis of 118 prediction studies. Psychological Assessment, 21, 1-21.

41.

Hanson

R. K.

Thornton

(1999). STATIC-99: Improving actuarial risk assessments for sex offenders (User Report No. 1999-02). Ottawa: Department of the Solicitor General of Canada. Retrieved from http://www.publicsafety.gc.ca/res/cor/rep/_fl/1999-02-st99-imp-rsk-ssmnt-eng.pdf

42.

Hanson

R. K.

Thornton

(2000). Improving risk assessment for sex offenders: A comparison of three actuarial scales. Law and Human Behavior, 24, 119-136.

43.

Hare

R. D.

(1991). The Psychopathy Checklist–Revised. Toronto, Canada: Multi-Health Systems.

44.

Harris

A. J. R.

Hanson

R. K.

(2004). Sex offender recidivism: A simple question (Corrections Users Report No. 2004-03). Ottawa: Public Safety and Emergency Preparedness Canada. Retrieved from http://www.publicsafety.gc.ca/res/cor/rep/2004-03-se-off-eng.aspx

45.

Harris

G. T.

Rice

M. E.

Quinsey

V. L.

Lalumière

M. L.

Boer

Lang

(2003). A multi-site comparison of actuarial risk instruments for sex offenders. Psychological Assessment, 15, 413-425.

46.

Hart

S. D.

(2008, July). The Structured Professional Judgment approach to violence risk assessment: Core principles of SPJ. Paper presented at the annual meeting of the International Association of Forensic Mental Health Services, Vienna, Austria.

47.

Hart

S. D.

Kropp

P. R.

Laws

D. R.

Klaver

Logan

Watt

K. A.

(2003). The Risk for Sexual Violence Protocol (RSVP). Burnaby, Canada: Simon Fraser University, Mental Health, Law, and Policy Institute.

48.

Helmus

Hanson

R. K.

Thornton

(2009). Reporting STATIC-99 in light of new research on recidivism norms. The Forum, 21, 38-45.

49.

Hoge

R. D.

Andrews

D. A.

(2002). Youth Level of Service/Case Management Inventory: User’s manual. Toronto, Canada: Multi-Health Systems.

50.

Ireland Probation & Welfare Services. (Undated). Assessing the risk of reoffending. Retrieved from: http://www.probation.ie/pws/websitepublishing.nsf/attachmentsbytitle/Assessing+Risk+of+Re-offending/$file/Assessing+Risk+of+Re-offending.pdf

51.

Interstate Commission for Adult Offender Supervision. (2007). Sex offender assessment information survey (4-2007). Retrieved from http://www.interstatecompact.org/LinkClick.aspx?fileticket=jLoQPVeviaQ%3D&tabid=105&mid=431

52.

Kropp

P. R.

Hart

S. D.

(2000). The Spousal Assault Risk Assessment (SARA) Guide: Reliability and validity in adult male offenders. Law and Human Behavior, 24, 101-118.

53.

Kropp

P. R.

Hart

S. D.

(2004). The development of the Brief Spousal Assault Form for the Evaluation of Risk (B-SAFER): A tool for criminal justice professionals. Ottawa: Research and Statistics Division, Justice Canada. Retrieved from http://www.justice.gc.ca/eng/pi/rs/rep-rap/2005/rr05_fv1-rr05_vf1/p5.html

54.

Kropp

P. R.

Hart

S. D.

Webster

C. D.

Eaves

(1999). Spousal Assault Risk Assessment Guide (SARA). Toronto, Canada: Multi-Health Systems.

55.

Lanagan

P. A.

Schmitt

E. L.

Durose

M. R.

(2003). Recidivism of sex offenders release from prison in 1994. Washington, DC: U.S. Department of Justice. Retrieved from http://bjs.ojp.usdoj.gov/content/pub/pdf/rsorp94.pdf

56.

Looman

Abracen

(2010). Comparison of measures of recidivism in sexual offenders. Journal of Interpersonal Violence, 25, 791-807.

57.

Maletzky

(1991). Treating the sexual offender. Newbury Park, CA: Sage.

58.

Morton

K. E.

(2003). Psychometric properties of four risk assessment measures with male adolescent sexual offenders. Dissertation Abstracts International, 41, 1833.

59.

Nuffield

(1982). Parole decision making in Canada. Ottawa: Solicitor General of Canada.

60.

Olver

M. E.

Wong

S. C. P.

Nicholaichuk

T. P.

Gordon

(2007). The validity and reliability of the Violence Risk Scale-Sexual Offender version: Assessing sex offender risk and evaluating therapeutic change. Psychological Assessments, 19, 318-329.

61.

Ontario Ministry of Correctional Services, Research Services Division. (1983). Offences leading to incarceration. Correctional facts (No. 83-1).

62.

Parent

Guay

J. P.

Knight

R. A.

(2011). An assessment of long-term risk of recidivism by adult sex offenders: One size doesn’t fit all. Criminal Justice and Behavior, 38, 188-209.

63.

Petrila

Otto

R. K.

(2001). Admissibility of expert testimony in sexually violent predator proceedings. In Schlank

(Ed.), The sexual predator: Legal issues, clinical issues, special populations (Vol. 2, pp.). Kingston, NJ: Civic Research Institute.

64.

Prentky

R. A.

Lee

A. F. S.

Knight

R. A.

Cerce

(1997). Recidivism rates among child molesters and rapists: A methodological analysis. Law and Human Behavior, 21, 635-659.

65.

Public Safety Canada. (2010). Corrections and conditional release statistical overview: Annual report—2010. Ottawa: Public Works and Government Services Canada. Retrieved from http://www.publicsafety.gc.ca/res/cor/rep/2010-ccrso-eng.aspx

66.

Quinsey

Harris

Rice

Cormier

(1998). Violent offenders: Appraising and managing risk. Washington, DC: American Psychological Association.

67.

Quinsey

V. L.

Harris

G. T.

Rice

M. E.

Cormier

C. A.

(2006). Violent offenders: Appraising and managing risk (2nd ed.). Washington, DC: American Psychological Association.

68.

Quirk

Nutbrown

Reynolds

(1991). A practical measure of offence seriousness: Sentence severity. Brampton, Canada: Ontario Correctional Institute. Retrieved from http://regreynolds.ca/Articles/Criminality/Offence%20Severity.pdf

69.

Rice

M. E.

Harris

G. T.

(2005). Comparing effect sizes in follow-up studies: ROC area, Cohen’s d, and r. Law and Human Behavior, 29, 615-620.

70.

Rosenthal

(1994). Parametric measures of effective size. In Cooper

H. M.

Hedges

L. V.

(Eds.), The handbook of research synthesis (pp.). Thousand Oaks, CA: Sage.

71.

Rossegger

Laubacher

Moskvitin

Villmar

Palermo

G. B.

Endrass

(2011). Risk assessment instruments in repeat offending: The usefulness of FOTRES. International Journal Offender Therapy Comparative Criminology, 55, 716-731.

72.

Rowe

R. C.

(1999, November). The utilization of an interview-based classification instrument for parole board decision-making. Paper presented at the 51st annual meeting of the American Society of Criminology, Toronto, Ontario, Canada.

73.

Simourd

D. J.

Malcolm

(1998). Reliability and validity of the Level of Service Inventory–Revised among federally incarcerated sex offenders. Journal of Interpersonal Violence, 13, 261-274.

74.

Skowron

(2004, December). Differentiation and predictive factors in adolescent sexual offending (Unpublished doctoral dissertation). Carleton University, Ottawa, Canada.

75.

Sreenivasan

Weinberger

L. E.

Frances

Cusworth-Walker

(2010). Alice in actuarial-land: Through the looking glass of changing STATIC-99 norms. Journal of the American Academy of Psychiatry and Law, 38, 400-406.

76.

Stasiuk

Winter

Nixon

(1996). Awaiting trial: Accused persons remanded to custody. Toronto, Canada: Ontario Ministry of Correctional Services.

77.

Thornton

(1997). Structured anchor clinical judgment risk assessment (SACJ): Proceedings of the NOTA Conference. Brighton, United Kingdom.

78.

Thornton

Mann

Webster

Blud

Travers

Friendship

Erikson

(2003). Distinguishing and combining risks for sexual and Washington state violent recidivism. In Prentky

R. A.

Janus

E. S.

Seto

M. C.

(Eds.), Annals of the New York Academy of Sciences: Vol. 989. Sexually coercive behavior: Understanding and management (pp. 225-235). New York, NY: New York Academy of Sciences.

79.

Urbaniok

(2007). FOTRES: Forensisches Operationalisiertes Therapie-Risiko-Evaluations-System [FOTRES: Forensic Operationalized Therapy/Risk Evaluation System]. Bern, Switzerland: Zytglogge.

80.

U.S. Department of Justice, Office of Juvenile Justice and Delinquency Prevention. (2001). Juveniles who have sexually offended: A review of the professional literature report. Retrieved from http://www.ojjdp.ncjrs.org/

81.

Van Voorhis

Wright

E. M.

Salisbury

Bauman

(2010). Women’s risk factors and their contributions to existing risk/needs assessment: The current status of a gender-responsive supplement. Criminal Justice and Behavior, 37, 261-288.

82.

Vrana

G. C.

Sroga

Guzzo

(2008). Predictive validity of the LSI–OR among a sample of adult male sexual assaulters. Unpublished manuscript, Nipissing University, North Bay, Canada.

83.

Washington State Institute for Public Policy. (2005). Involuntary commitment of sexually violent predators: Comparing state laws. Olympia: Author. Retrieved from http://www.wsipp.wa.gov/rptfiles/05-03-1101.pdf

84.

Washington State Institute for Public Policy. (2006). Sex offender sentencing in Washington State: Predicting recidivism based on the LSI-R. Olympia: Author. Retrieved from http://www.wsipp.wa.gov/rptfiles/06-02-1201.pdf

85.

Webster

C. D.

Douglas

K. S.

Eaves

Hart

S. D.

(1997). HCR-20: Assessing the risk for violence (Version 2). Vancouver, Canada: Simon Fraser University, Mental Health, Law, and Policy Institute.

86.

Worling

J. R.

Curwen

(2001). Estimates of Risk of Adolescent Sexual Offense Recidivism (ERASOR) Version 2.0. Toronto, Canada: SAFE-T Program, Thistletown Regional Centre.

87.

Wormith

J. S.

(1997). Research to practice: Applying risk/needs assessment to offender classification. Forum on Corrections Research, 9, 26-31.

88.

Wormith

J. S.

(2011). The legacy of D. A. Andrews in the field of criminal justice: How theory and research can change policy and practice. International Journal of Forensic Mental Health, 10, 78-82.

89.

Wormith

J. S.

Hogg

Guzzo

(2011, December). The predictive validity of sexual offender recidivism with a general risk/needs assessment inventory. Saskatoon, Canada: University of Saskatchewan.

90.

Wormith

J. S.

Olver

M. E.

Stevenson

H. E.

Girard

(2007). The long-term prediction of offender recidivism using diagnostic, personality, and risk/need approaches to offender assessment. Psychological Services, 4, 287-305.

91.

Yang

Wong

S. C. P.

Coid

(2010). The efficacy of violence prediction: A meta-analytic comparison of nine risk assessment tools. Psychological Bulletin, 136, 740-767.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

9.23 MB