Incorporating Change Information Into Sexual Offender Risk Assessments Using the Violence Risk Scale

Abstract

We examined the use of risk-change information in sexual offender risk assessments featuring the Violence Risk Scale–Sexual Offender version (VRS-SO), a sex offender risk assessment and treatment planning tool. The study featured a combined international sample of 539 sex offenders followed up an average of 15.5 years post-release. Pre- and posttreatment VRS-SO ratings were amalgamated from two treated samples of sex offenders from Canada and New Zealand. Analyses focused on examinations and applications of change data and its relationship to sexual and violent recidivism. VRS-SO change scores were significantly associated with decreases in these outcome criteria with, and without, controlling for indicators of pretreatment risk (e.g., Static-99R score) and individual differences in follow-up time. Applications of logistic regression using fixed 5-year follow-ups generated estimated rates of sexual and violent recidivism at different VRS-SO score thresholds. The use of logistic regression demonstrated a clinically useful and systematic means of combining risk and change information into posttreatment risk appraisals. Implications for the use of change information in the assessment and management of sexual offender risk are discussed.

Keywords

sexual recidivism VRS-SO dynamic change risk assessment

The assessment of sexual violence risk, prediction of recidivism, and treatment of sexual offenders have important implications for public and forensic mental health, the criminal justice system, research, and clinical practice. Accurate identification of high-risk sex offenders and their successful treatment and management can serve to reduce the costs of reincarceration, sexual victimization, and promote safer communities. Although risk assessment is a key component of managing and preventing sexual violence, arguably the purpose of sexual violence risk assessment should be preventing, as opposed to merely predicting, future sexual offending (Douglas & Kropp, 2002). For instance, when an individual is identified as high risk for sexual violence steps should be taken to reduce or manage risk to prevent a new incident of sexual violence.

Assessing Changes in Sexual Offender Risk

Central to the risk, need, responsivity model is that criminogenic needs are dynamic and changeable and that risk is dynamic (Andrews & Bonta, 2010). As such, effecting positive changes in these need areas (e.g., via treatment) should be linked to reductions in recidivism. That treatment works by reducing an individual’s risk is the presumed mechanism that underpins the relationship of treatment completion and recidivism; however, empirical support for the relationship between risk reduction and recidivism reduction is scant. Much of the research instead has focused on single time-point risk estimates and predictions using putatively dynamic variables (Douglas & Skeem, 2005; Kraemer et al., 1997).

Increasing lines of research have documented significant pre–posttreatment changes on dynamic domains of sexual offender risk and need such as cognitive distortions, antisocial tendencies, loneliness, and socioaffective functioning among other domains (Beggs & Grace, 2011; Hudson, Wales, Bakker, & Ward, 2002; Jung & Guyalets, 2011; Nunes, Babchishin, & Cortoni, 2011; Nunes & Cortoni, 2007; Olver, Wong, Nicholaichuk, & Gordon, 2007; Wakeling, Beech, & Freemantle, 2013), with changes ranging from small (d < .20) to large (d > .80) in magnitude of effect. Although such changes may indicate improvements in psychological health, the men making such changes still may not experience normal or healthy levels of psychological functioning in these domains. Moreover, such changes may not necessarily be risk-relevant, that is, linked to reductions in an external behavioral criterion, such as recidivism. In accordance with the risk and need principles, it would also stand to reason that risk-related treatment gains (or lack thereof) should have greater bearing on posttreatment outcome with higher risk offenders, as opposed to lower risk offenders, who are likely to remain low risk irrespective of any changes they make; this consideration would underscore the importance of controlling for baseline risk when evaluating risk changes (Olver & Wong, 2011). Some lines of support have found risk-related changes in dynamic areas of risk and need to be associated with reductions in sexual and violent recidivism in treated sexual offenders (Beggs & Grace, 2011; Olver et al., 2007; Wakeling et al., 2013).

Violence Risk Scale–Sexual Offender Version (VRS-SO)

The VRS-SO (Wong, Olver, Nicholaichuk, & Gordon, 2003) is sex offender risk assessment and treatment planning tool. Comprising 7 static and 17 dynamic items, the VRS-SO is designed to assess sexual violence risk, identify targets for treatment, and assess changes in risk through treatment or other change agents. Research on two data sets demonstrated support for the predictive validity of the VRS-SO, as well as the important premise, described hereinbefore, that changes in risk as measured by the VRS-SO are meaningfully associated with changes in recidivism outcomes. In a study of 321 male sexual offenders treated in a high-intensity sex offender program, Olver et al. (2007) found the VRS-SO static and dynamic components predicted sexual recidivism. Although the change scores had a small univariate relationship to sexual recidivism (r = −.10), after controlling for risk and follow-up time, change achieved in treatment was significantly related to reductions in sexual recidivism. Changes scores were also significantly associated with reductions among offenders scoring high on the Static-99 but not for those scoring low. In a sample of 218 treated child molesters who completed a sex offender treatment program in New Zealand, Beggs and Grace (2010, 2011) found the VRS-SO predicted sexual recidivism and that VRS-SO change scores were significantly associated with reductions in sexual recidivism, even after controlling for risk.

Applications of Risk Assessment Information

There have been important developments in applied sex offender risk assessment research with implications for the interpretation and reporting of risk information in clinical practice. These include the application of nonarbitrary metrics (e.g., measures of relative risk such as percentiles or hazard ratios; Hanson, Babchishin, Helmus, & Thornton, 2013; Hanson, Lloyd, Helmus, & Thornton, 2012), generating sample-specific norms based on differences in risk level and recidivism base rate (e.g., as with the Static-99R; static99.org), and use of statistical procedures to generate recidivism estimates based on the sample base rate and magnitude of the predictor–criterion relationship (Hanson, Helmus, & Thornton, 2010). Hanson et al. (2010), for instance, used logistic regression at fixed 5- and 10-year follow-up periods to estimate sexual recidivism base rates at all possible Static-2002 scores from the magnitude of the instrument’s relationship to outcome (unstandardized B) and the observed sexual recidivism rates of the sample. The use of logistic regression has certain advantages. Recidivism rates can be poorly estimated for extreme scores with a small number of cases (e.g., score of 13 on the Static-2002), and the use of logistic regression with fixed follow-up periods can smooth out the trend line of recidivism values at different scores while eliminating large fluctuations due to sampling variability (Hanson et al., 2010).

Thornton (2011) applied this methodology to estimate recidivism rates while combining multiple (i.e., static and dynamic) risk measures. Using the Static-99/R to provide a measure of static risk, Thornton found that the odds of sexual recidivism increased linearly with each increment in static and dynamic score, across a range of dynamic tools using unstandardized B values and base rate information. The information was gathered from studies examining the STABLE 2007 in a Canadian sample of probationers (Hanson, Harris, Scott, & Helmus, 2007), the Structured Risk Assessment: Forensic Version in a Massachusetts sexual offender inpatient sample (Thornton, 2011), and the VRS-SO in a Canadian sample of treated federal sex offenders (Olver et al., 2007). The use of logistic regression to generate recidivism estimates at specific scores (or clusters of scores) arguably increases the utility and practicality of risk assessment information communicated for decision-making purposes while reducing bias. Conceivably, this methodology could be extended to predictor variables with inverse relationships to outcome, such as treatment change, to obtain recidivism estimates at different risk and change score thresholds. This logistic regression application would enable an estimation of the recidivism rates and amount of reduction of risk associated with different amounts of change.

Rationale for the Present Study

Two earlier investigations (Beggs & Grace, 2011; Olver et al., 2007) described hereinbefore have provided support for the dynamic validity of the VRS-SO and research featuring this tool is ongoing. The present study sought to build on this body of work through a more comprehensive and nuanced examination of change data from combining these two samples. Taken together, the two groups represent a multinational sample of sex offenders, heterogeneous with respect to risk and victim types, who attended specialized comprehensive sex offender treatment programs with long-term posttreatment follow-ups in the community. Combining the samples, in turn, has important advantages of providing a larger N and increasing the variance of VRS-SO scores, which add statistical power to detect possible effects that may be overlooked in a smaller sample.

A primary objective of the present study was a novel application of analytic approaches to the change data that in our view increase the practical applications of change information with the VRS-SO; specifically, the application of logistic regression modeling as in Thornton (2011) and Hanson et al. (2010) to estimate recidivism rates as a function of VRS-SO risk and change scores. The findings are not only of theoretical interest, but also underscore the value added of incorporating change information into posttreatment assessments. A second objective was to test relationships between change and outcome not examined in previous VRS-SO work, including the relationship of changes on all dynamic VRS-SO scale components to outcome, and extending our change analyses to include sexual and general violent recidivism. The inclusion of general violence is an important consideration, given that nonsexual violent convictions have been found not uncommonly to have sexual motivations in sex offender samples (Rice, Harris, Lang, & Cormier, 2006); not to mention that the reduction of all violence, sexual and nonsexual, is a laudable goal of sex offender treatment, risk reduction, and community reintegration efforts. Finally, in order for the present investigation and its findings to have even greater currency with the sex offender risk assessment field, we computed the Static-99R (see Helmus, Thornton, Hanson, & Babchishin, 2012) and used this in place of the Static-99 in our analyses.

Method

Samples

Clearwater Sex Offender Program

This sample¹ included 321 male federal sexual offenders who had participated in a high-intensity sex offender treatment program (the Clearwater Program; Olver et al., 2007) in a maximum-security forensic mental health facility in Canada between 1983 and 1997. The program is approximately 8 months in duration, provides group and individual therapies, and is cognitive behavioral in orientation. Approximately half (52.6%) of the sample were rapists, with the remainder being fairly evenly divided among extrafamilial child molesters (17.4%), mixed offenders (14.0%), and incest offenders (15.9%). In this sample, the VRS-SO dynamic items were coded pretreatment and posttreatment from detailed and comprehensive treatment file information by trained research assistants who were blind to recidivism outcome. No interviews were conducted. The file information was highly comprehensive, often spanning several hundred or more pages, and included nursing notes, assessment reports (psychological, psychiatric), treatment progress summaries, social histories, criminal histories, correctional plans, narrative reports of phallometric and psychometric testing among other information sources. If insufficient information was available to score one of the dynamic items, the item was omitted and stepwise multiple regression procedures were used to estimate missing values. Ninety-one percent of cases had no more than two missing items with the remainder missing three or four items (see Olver et al., 2007). The VRS-SO static items, in turn, were constructed on a randomly selected half of the sample and validated on the other half. The Static-99 was also rated for comparison purposes. As reported in Olver et al. (2007), acceptable interrater reliabilities were obtained on 35 randomly selected cases with intraclass correlation coefficient (ICC) values of .74 and .79 obtained on pre- and posttreatment dynamic ratings, respectively.

Outcome data were updated May 15, 2010 through the Canadian Police Information Centre (CPIC), a nationwide electronic database of the offenders’ officially recorded criminal charges and convictions, to extend the mean follow-up period to 17.7 years (SD = 4.3). Sexual recidivism was defined as any criminal conviction for a sexually motivated offense, including offenses that were adjudicated as nonsexual crimes (e.g., nonsexual assault) when additional documentation (e.g., Criminal Profile Report) was available to determine if the offense was sexually motivated; although this additional verification could only be performed if the conviction involved a return to federal custody. Violent recidivism was defined as any new criminal code conviction for an offense against the person (e.g., assault, robbery), including sexual offenses. General recidivism included any criminal code conviction. The reader is referred to Olver et al. (2007) for further details regarding the Clearwater sample and data collection procedures.

Kia Marama Sex Offender Treatment Program

This sample included 218 intrafamilial (56.4%) and extrafamilial (43.6%) child molesters who participated in the Kia Marama Special Treatment Unit in Rolleston, New Zealand, Department of Corrections, between 1993 and 2000 (Beggs & Grace, 2010, 2011). The Kia Marama treatment reports were conducive to completing VRS:SO ratings and included an extensive pretreatment assessment describing the offender and his risk factors, followed by a thorough description of the individual’s treatment progress and his overall attitude toward treatment (Beggs & Grace, 2010). Training in VRS-SO scoring was provided over the phone and via email and Internet by the instrument’s developers. The VRS-SO static and dynamic items were then coded pre- and posttreatment from the treatment reports by one of the study authors (Beggs) and a trained research assistant while blind to outcome. The Static-99 was also rated from these information sources. On 23 randomly selected cases, high interrater reliabilities were observed for VRS-SO pretreatment (ICC = .90) and posttreatment (ICC = .92) dynamic scores.

Offenders were followed up a mean 12.2 years (SD = 1.8) postrelease and officially recorded convictions were captured July 1, 2008, from a nationwide database. Sexual offense recidivism was defined consistent with Olver et al. (2007). In addition, for the purposes of the present study, general violence (including sexual) and general recidivism (i.e., any new conviction) outcome variables were created from the data using the operational definitions hereinbefore. Readers are directed to Beggs and Grace (2010) for further details regarding the Kia Marama sample and data collection procedures.

Measures

VRS-SO

The VRS-SO (Wong et al., 2003) is a 24-item sex offender risk assessment and treatment planning tool designed to assess risk, identify targets for treatment, and identify changes in risk as a result of treatment or for other reasons. The instrument comprises 7 static items (i.e., criminal history, victim, and offender demographics) and 17 dynamic items empirically, theoretically, or conceptually related to increased risk for sexual recidivism (e.g., cognitive distortions, interpersonal aggression, deviant sexual preference, intimacy deficits; see Olver et al., 2007, for a list and brief description of dynamic items). Each item is rated on a 4-point (0, 1, 2, 3) scale, with higher scores, in general, indicating increased risk for sexual recidivism. Dynamic items rated 2 or 3 are considered to be criminogenic (i.e., linked to sexual offending) and, therefore, appropriate targets for treatment whereas those rated 0 or 1 are not. Results of a factor analysis of the VRS-SO dynamic variables (see Olver et al., 2007) generated three oblique factors labeled Sexual Deviance (e.g., sexually deviant lifestyle, deviant sexual preference), Criminality (e.g., interpersonal aggression, substance abuse, impulsivity), and Treatment Responsivity (e.g., cognitive distortions, poor treatment compliance).

The VRS-SO assesses change through a modified application of Prochaska, DiClemente, and Norcross’s (1992) transtheoretical model of change, in which individuals navigate a series of stages associated with cognitive, behavioral, and experiential changes as they remediate problem areas. Each of the five stages of change (Precontemplation, Contemplation, Preparation, Action, and Maintenance) is operationalized for each dynamic item. All treatment targets (dynamic items rated 2 or 3) are given a stages of change baseline or pretreatment rating to assess the individual’s motivation and readiness for change. Dynamic items rated 0 or 1, as they are not treatment targets, generally require no stages of change rating. All treatment targets are re-rated on the stage of change at posttreatment to assess their treatment change which is the difference in the stage of change from pre- to posttreatment. Change scores are calculated such that positive scores indicate positive change. Progression from one stage to the next stage, an indication of positive change and hence risk reduction, is scored as a 0.5-point reduction in the pretreatment rating of the item; progression of two stages, 1.0-point reduction, and so on. This is carried out for each dynamic item identified as a treatment target to arrive at a posttreatment score for each dynamic item; the exception is for progress from Precontemplation to Contemplation wherein no risk reduction is registered as there is no relevant behavioral improvement. Deterioration is marked by a 0.5-point addition for each stage that the individual has worsened. If an individual becomes substantively worse in an area previously assessed as not problematic, or more information comes to light about a problem area suggesting that it is actually worse than initially assessed (more common from our experience), the item rating can be revised using the scoring instructions. The sum of all posttreatment dynamic item ratings plus the total static items ratings is the individual’s risk rating posttreatment.

Static-99R

As outlined above, the Static 99 (Hanson & Thornton, 1999) was originally coded on the two samples. Given the revision of the Static-99 through developing an item with new age weights, we computed the Static-99R (Helmus, Thornton, et al., 2012) through applying the scoring criteria to recode the age at release item and then recomputing the total score. The Static-99R, a 10-item static actuarial sex offender risk assessment measure, has total scores that range from −3 to 12 points grouped as low (−3 to 1), medium–low (2 to 3), medium–high (4 to 5), and high (6 to 12). Results from a recent meta-analysis (k = 22, n = 8,055) supported the predictive accuracy of the Static-99R for sexual recidivism (area under the curve [AUC] = .69; Helmus, Hanson, Thornton, Babchishin, & Harris, 2012).

Planned Analyses

The analyses proceeded in several stages. First, prior to amalgamating the samples for all remaining analyses, the t-test comparisons were conducted between the two samples on their VRS-SO and Static-99R scores. Second, the predictive accuracy of the VRS-SO static, dynamic, total, and factor scores and the Static-99R (for comparison purposes) were examined with respect to sexual, violent, and general recidivism though computing point biserial correlations (r_pb) and AUC values for receiver operator characteristic (ROC) analyses. Using the guidelines provided in Rice and Harris (2005), we interpret r_pb, d, and AUC values in terms of effect size magnitudes using the language of small/low (AUC = .556-.637, Cohen’s d = .20, r_pb = .10), medium (AUC = .639-.712. d = .50, r_pb = .243) and large/high (AUC = .714 and up, d = .80, r_pb = .371).

Third, univariate change analyses were conducted through examining the effect size (Cohen’s d) magnitude of pre–post differences on the dynamic and factor score totals, as well as their bivariate associations (r_pb) with sexual, violent, and general recidivism. We present d-values in particular given their practical interpretation, representing the magnitude of difference between two groups in standard deviation units. These analyses are useful in quantifying the amount of change occurring in treatment as well as the strength and direction of relationship to outcome. In addition, Kaplan–Meier survival analyses were conducted for three broad change groups (> 1 SD, ±1 SD, < 1 SD relative to the mean change score) to illustrate sexual and violent recidivism failure rates over time as a function of different amounts of change.

The last set of analyses examined the relationship of VRS-SO change scores to sexual and violent recidivism after controlling for risk through a series of Cox and logistic regressions. The analyses are intended to answer different research questions and have potential applications. Cox regression survival analyses were first conducted to examine the relationship of VRS-SO measured treatment change while controlling for pretreatment risk and individual differences in length of follow-up. These analyses address the question as to whether VRS-SO change scores add incrementally to the prediction of recidivism after controlling for important covariates. Given that important risk-related differences emerged between the two samples (i.e., with the Clearwater sample being the higher risk of the two), we decided to treat sample as a covariate in the first block of analyses. Pretreatment risk was controlled through entering a static risk measure (either the Static-99R or the VRS-SO static factor score) and the pretreatment dynamic item total in the second block of analyses, followed by the VRS-SO change score in the third and final block of analyses. Static and dynamic measures are each incorporated as covariates as past research on these two samples demonstrate both to be incremental, hence improving predictive accuracy and translating into a more robust control for risk.

Although the information generated from Cox regression is valuable, worthwhile, and important to identify the strength and direction of effects, applying the findings in a clinically practical manner with this technique is more difficult. As such, we use logistic regression analyses using a 5-year fixed follow-up to estimate rates of sexual and violent recidivism at different amounts of treatment change (using the same change score bands used in the Kaplan–Meier analyses) at different VRS-SO score thresholds. These analyses are intended to be practical and speak to the issue as to how much risk reduction is associated with different amounts of change based on the individual’s baseline risk level. Application of the following formula from Tabachnick and Fidell (2001) enables one to generate recidivism rates at specific scores on the predictor variables: e^{B0 + B1 × Score} / (1 + e^{B0 + B1 × Score}). We apply this procedure as used elsewhere (Hanson et al., 2010; Thornton, 2011) to estimate sexual and violent recidivism base rates on the VRS-SO at different score thresholds.

Results

Group Comparisons

Group comparisons were conducted between the Clearwater (Olver et al., 2007) and Kia Marama (Beggs & Grace, 2010) samples, using independent-sample t-tests, on the VRS-SO and Static-99R. Figure 1 shows the mean scores on the measures for these samples. All group differences were significant at p < .001 with t-test values as follows (Welch’s unequal variance t-test and df reported when equal variances not assumed): Static-99R (537) = 13.43, VRS-SO static (415.54) = 6.28, dynamic (524.63) = 5.92, static + dynamic total (537) = 6.90, Sexual Deviance (536.99) = −9.63, Criminality (493.51) = 7.08, and Treatment Responsivity (511.34) = 17.88. As such, the Clearwater sample was higher risk on all scales and domains, except for Sexual Deviance, in which the Kia Marama sample was higher risk.

Figure 1.

Clearwater (Olver et al., 2007) and Kia Marama (Beggs & Grace, 2010) samples compared on the VRS-SO and Static-99 risk measures (mean values).

Predictive Accuracy

Individuals across the two samples were released between 1983 and 2000 with a median year of release of 1994 and an average follow-up of 15.5 years (SD = 4.4). Overall, 22.4% of the combined sample (n = 121) was convicted for a new sexual offense, 42.3% (n = 228) for a new violent (including sexual) offense, and 59.4% (n = 320) for any new offense. Broken down by sample, offenders released from the Clearwater program had significantly higher rates of sexual (28.7% vs. 13.3%), χ²(df = 1, N = 539) = 17.59, p < .001; violent (55.5 vs. 22.9%), χ²(df = 1, N = 539) = 56.24, p < .001; and general (71.7 vs. 41.3%), χ²(df = 1, N = 539) = 49.63, p < .001, reconviction.

Consistent with the two prior investigations, most of the VRS-SO scale components and Static-99R demonstrated moderate-to-high predictive accuracy for sexual and violent recidivism, with correlations and AUC values frequently exceeding .30 and .70, respectively (see Table 1). Although the Sexual Deviance factor significantly predicted sexual recidivism, the predictive accuracy, particularly for pretreatment scores, was notably lower, as well as negligibly related to violent recidivism, and inversely to general recidivism. Most measures also significantly predicted general recidivism, albeit at a somewhat smaller magnitude compared with other criteria, with some of the larger effects observed with the static tools and the Criminality factor. VRS-SO posttreatment ratings demonstrated predictive accuracies that were generally greater in magnitude than the pretreatment ratings to varying degrees across all recidivism outcomes. The examination of change scores formed the basis for the remaining analyses.

Table 1.

Predictive Accuracy Statistics of the VRS-SO and Static-99R for Sexual, Violent, and General Recidivism (N = 539).

	Sexual recidivism			Violent recidivism			General recidivism
Measure	r	AUC	95% CI	r	AUC	95% CI	r	AUC	95% CI
Static-99R	.31	.71	[.66, .76]	.42	.74	[.70, .79]	.41	.74	[.70, .78]
VRS-SO static	.34	.73	[.69, .78]	.34	.70	[.65, .74]	.34	.70	[.65, .74]
Dynamic (pre)	.33	.73	[.68, .78]	.31	.68	[.64, .73]	.19	.61	[.56, .66]
Dynamic (post)	.37	.75	[.70, .80]	.36	.72	[.67, .76]	.22	.63	[.59, .68]
Total (pre)	.38	.76	[.71, .81]	.37	.72	[.67, .76]	.28	.67	[.62, .72]
Total (post)	.41	.77	[.73, .82]	.41	.74	[.70, .78]	.31	.68	[.64, .73]
Sexual deviance (pre)	.11^a	.57^a	[.51, .63]	−.08 ns	.45^a	[.40, .50]	−.20	.38	[.33, .43]
Sexual deviance (post)	.20	.62	[.56, .68]	.03 ns	.50	[.45, .55]	−.11^b	.42^a	[.38, .47]
Criminality (pre)	.28	.69	[.64, .74]	.40	.73	[.69, .78]	.35	.71	[.66, .75]
Criminality (post)	.31	.71	[.65, .76]	.42	.75	[.71, .79]	.35	.71	[.66, .75]
Treatment responsivity (pre)	.27	.69	[.64, .74]	.32	.69	[.65, .74]	.23	.64	[.59, .68]
Treatment responsivity (post)	.28	.70	[.65, .75]	.34	.70	[.65, .74]	.23	.64	[.59, .69]

Note. Predictive validity correlations and AUC values are significant at p < .001 except for ^ap < .01, ^bp < .05. VRS-SO = Violence Risk Scale–Sexual Offender version; AUC = area under the curve; CI = confidence interval.

Univariate Change Analyses

Correlation and Cohen’s d analyses

VRS-SO change scores are reported in Table 2, along with the pre- and posttreatment scores on all dynamic components of the tool. All pre–post differences were statistically significant with the magnitude of change ranging from roughly one fifth to nearly one half of a standard deviation (Cohen’s d = .22-.46). The change scores of the three broad factors were also positively correlated (all ps < .001): Sexual Deviance and Criminality, r = .20; Sexual Deviance and Treatment Responsivity, r = .44; and Treatment Responsivity and Criminality, r = .48. The positive correlations suggest some shared variance in change scores on the three factors. The change scores, in turn, were differentially related to the recidivism outcomes. The dynamic total change score, representing the sum total of pre–post differences, was significantly negatively correlated with all recidivism outcomes. Cohen’s d, representing the magnitude of difference in change score in standard deviation units between recidivists and nonrecidivists, ranged from the low to medium −.30s for sexual and general recidivism, and to −.49 for the difference between violent recidivists and nonrecidivists. When examined at the factor level, it became apparent that change scores on the Sexual Deviance factor accounted for much of the relationship of change to outcome, with significant inverse relationships to sexual, violent, and general recidivism ranging from r = −.22 to −.33 and d = −.44 to −.69. Change scores on the Criminality factor did not significantly predict decreases in any recidivism criteria, while modest significant relationships were observed between Treatment Responsivity change scores and sexual and violent recidivism.

Table 2.

Magnitudes of VRS-SO Measured Therapeutic Change and its Relationship to Sexual, Violent, and General Recidivism (N = 539).

					Recidivism
	Pre–posttreatment differences				Sexual		Violent		General
VRS-SO measure	Pre	Post M (SD)	Change	d pre–post change	r	d	r	d	r	d
Dynamic total	23.55 (7.08)	20.21 (7.54)	3.34 (2.22)	.46	−.18***	−.37	−.24***	−.49	−.16***	−.32
Sexual deviance	7.47 (3.80)	6.22 (3.39)	1.25 (1.22)	.35	−.22***	−.44	−.33***	−.69	−.31***	−.64
Criminality	7.90 (3.59)	7.14 (3.40)	.76 (.77)	.22	−.06	−.12	.00	.00	.10*	.22
Treatment responsivity	5.56 (2.48)	4.58 (2.62)	.98 (.66)	.38	−.11*	−.22	−.12**	−.24	−.06	−.11

Note. For pre–post differences, the positive d-values indicate lower posttreatment scores than at pretreatment. All pre–posttreatment differences are significant at p < .001. For recidivism effect sizes, negative d-values indicate an association of positive change to lower recidivism, while positive d-values indicate an association of positive change with increased recidivism. The p values for r apply to the equivalent d for a given relationship tested. VRS-SO = Violence Risk Scale–Sexual Offender version.

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

Survival analyses

Kaplan–Meier survival analyses were conducted to examine rates of sexual and violent recidivism over time among three groups based on a mean change score of 3.34 (SD = 2.22): Change score totals that fell one or more standard deviations below the mean (low change, n = 119), one standard deviation within (above or below) the mean (medium change, n = 313), and one or more standard deviations above the mean (high change, n = 107). The two figures illustrate marked differences in recidivism trajectories among the three change groups. As illustrated in Figure 2, the low-change group had significantly higher and faster rates of sexual recidivism than the medium-change group, log rank χ²(1, N = 432) = 5.36, p = .021, and high-change group, log rank χ²(1, N = 226) = 16.51, p < .001. Significant differences in sexual recidivism rates were also observed between the high-change and medium-change groups, log rank χ²(1, N = 420) = 7.79, p = .005. For violent failure (Figure 3), again the low-change group had significantly higher and faster rates of violent recidivism than the medium-change group, log rank χ²(1, N = 432) = 11.21, p = .001, and high-change group, log rank χ²(1, N = 226) = 22.10, p < .001. The medium change group also had higher rates of violent failure than the high-change group, log rank χ²(1, N = 420) = 6.90, p = .009.

Figure 2.

Survival analysis: Cumulative sexual recidivism failure rates as a function of dynamic risk change.

Figure 3.

Survival analysis: Cumulative violent recidivism failure rates as a function of dynamic risk change.

Change, Sexual Violence Risk, and Recidivism

The preceding analyses examined the relationship of risk change to outcome without controlling for individual differences in risk level; that is, whether higher risk offenders were simply more likely to reoffend sexually and violently, irrespective of the possible changes they may have made, while lower risk offenders demonstrated lower rates of recidivism, and if change scores added incrementally to the prediction of outcome over and above risk. As such, we next examined the relationship of VRS-SO measured risk change to subsequent sexual and violent recidivism through Cox regression survival analysis. The sample site was entered as a dichotomously coded covariate in the first block, followed by the pretreatment risk variables in the second block, and the VRS-SO change score (total or one of the three factor scores) in the third block. This was done while alternately controlling for the VRS-SO static item total or the Static-99R plus the pretreatment dynamic item total. The results are presented in Tables 3 and 4, respectively. For the sake of space considerations, we present Blocks 1 and 2 of the analyses only once at the top of each table with Block 3 of each analysis presented in detail. Several themes were evident.

Table 3.

Cox Regression Survival Analyses: Prediction of VRS-SO Measured Therapeutic Change to Sexual and Violent Recidivism Controlling for Pretreatment (Static and Dynamic) Risk (N = 539).

Cox regression model (1-5)		Sexual recidivism							Violent recidivism
							95% CI for e^B							95% CI for e^B
		B	SE	Wald	p	e^B	Lower	Upper	B	SE	Wald	p	e^B	Lower	Upper
1	Block 1 (all analyses)
	Sample	.776	.214	13.116	.001	2.172	1.427	3.305	1.014	.161	39.701	.001	2.757	2.011	3.780
	Block 2 (all analyses)
	Sample	.328	.221	2.192	.139	1.388	0.899	2.142	0.709	.166	18.298	.001	2.031	1.468	2.811
	VRS-SO static	.129	.026	24.135	.001	1.137	1.081	1.197	0.089	.018	24.900	.001	1.093	1.056	1.132
	Dynamic total (pre)	.052	.013	14.966	.001	1.054	1.026	1.082	0.038	.010	14.129	.001	1.038	1.018	1.059
	Block 3
	Sample	.096	.230	0.172	.678	1.100	0.701	1.728	.477	.174	7.488	.006	1.611	1.145	2.267
	VRS-SO static	.133	.025	27.080	.001	1.142	1.086	1.200	.096	.017	29.990	.001	1.101	1.063	1.139
	Dynamic total (pre)	.052	.013	16.138	.001	1.053	1.027	1.080	.041	.010	17.847	.001	1.042	1.022	1.062
	Dynamic total change	−.135	.042	10.335	.001	0.873	0.804	0.949	−.124	.032	15.584	.001	0.883	0.830	0.939
2	Block 3
	Sample	−.236	.253	0.873	.350	0.790	0.481	1.296	.109	.194	.317	.573	1.115	.763	1.631
	VRS-SO static	.133	.026	26.901	.001	1.142	1.086	1.201	.098	.018	30.760	.001	1.103	1.065	1.142
	Dynamic total (pre)	.054	.013	17.572	.001	1.056	1.029	1.083	.044	.010	20.707	.001	1.045	1.026	1.066
	Sexual deviance change	−.412	.108	14.474	.001	0.662	0.536	0.819	−.412	.081	25.985	.001	0.663	0.566	0.776
3	Block 3
	Sample	.330	.221	2.234	.135	1.391	.902	2.146	.708	.166	18.315	.001	2.031	1.468	2.809
	VRS-SO static	.130	.026	25.224	.001	1.139	1.082	1.198	.090	.018	25.575	.001	1.094	1.057	1.133
	Dynamic total (pre)	.054	.013	16.7111	.001	1.055	1.028	1.083	.039	.010	15.395	.001	1.040	1.020	1.061
	Criminality change	−.234	.114	4.530	.033	0.784	.627	.981	−.098	.078	1.569	.210	0.907	0.778	1.057
4	Block 3
	Sample	.280	.222	1.591	.207	1.323	.856	2.046	.664	.167	15.876	.001	1.942	1.401	2.692
	VRS-SO static	.129	.026	25.259	.001	1.138	1.082	1.197	.091	.018	26.702	.001	1.095	1.058	1.133
	Dynamic total (pre)	.051	.013	15.304	.001	1.052	1.026	1.080	.038	.010	15.401	.001	1.039	1.019	1.059
	Treatment responsivity change	−.252	.127	3.921	.048	0.777	.605	.997	−.237	.095	6.230	.013	0.789	0.655	0.950
5	Block 3
	Sample	−.248	.266	.874	.350	0.780	.463	1.313	.064	.202	.099	.752	1.066	0.718	1.583
	VRS-SO static	.134	.026	27.033	.001	1.143	1.087	1.203	.098	.018	30.697	.001	1.103	1.066	1.142
	Dynamic total (pre)	.056	.013	18.191	.001	1.058	1.031	1.086	.045	.010	20.348	.001	1.046	1.026	1.066
	Sexual deviance change	−.447	.133	11.237	.001	0.640	.493	0.831	−.448	.096	21.702	.001	.639	0.529	0.771
	Criminality change	−.147	.141	1.090	.296	0.863	.654	1.138	.048	.088	0.303	.582	1.050	0.894	1.247
	Treatment responsivity change	.188	.186	1.024	.312	1.207	.839	1.737	.040	.128	0.097	.755	1.041	0.810	1.336

Note. Significant p values in bold. VRS-SO = Violence Risk Scale–Sexual Offender version; CI = confidence interval.

Table 4.

Cox Regression Survival Analyses: Prediction of VRS-SO Measured Therapeutic Change to Sexual and Violent Recidivism Controlling for Static-99R and Dynamic Factor Score (N = 539).

		Sexual recidivism							Violent recidivism
							95% CI for e^B							95% CI for e^B
Cox regression model (1-5)		B	SE	Wald	p	e^B	Lower	Upper	B	SE	Wald	p	e^B	Lower	Upper
1	Block 1 (all analyses)
	Sample	.776	.214	13.116	.001	2.172	1.427	3.305	1.014	.161	39.701	.001	2.757	2.011	3.780
	Block 2 (all analyses)
	Sample	.018	.234	0.006	.937	1.019	0.644	1.610	.366	.177	4.278	.039	1.442	1.019	2.041
	Static-99R	.194	.043	20.293	.001	1.215	1.116	1.322	.194	.031	38.934	.001	1.215	1.143	1.291
	Dynamic total (pre)	.067	.013	27.421	.001	1.069	1.043	1.096	.042	.010	19.217	.001	1.043	1.043	1.062
	Block 3
	Sample	−.164	.239	.473	.492	0.849	0.532	1.355	.174	.183	0.901	.343	1.190	.831	1.704
	Static-99R	.190	.042	20.796	.001	1.209	1.115	1.312	.194	.030	40.808	.001	1.214	1.144	1.289
	Dynamic total (pre)	.067	.012	28.712	.001	1.069	1.043	1.095	.045	.009	22.817	.001	1.046	1.027	1.065
	Dynamic total change	−.119	.042	8.106	.004	0.888	.818	0.964	−.111	.031	12.692	.001	0.895	.842	0.951
2	Block 3
	Sample	−.443	.258	2.938	.087	.642	.387	1.066	−.138	.201	0.473	.886	0.871	.587	1.292
	Static-99R	.183	.041	19.556	.001	1.200	1.107	1.302	.189	.030	38.784	.001	1.208	1.138	1.282
	Dynamic total (pre)	.070	.012	31.831	.001	1.073	1.047	1.099	.049	.009	27.070	.001	1.050	1.031	1.070
	Sexual deviance change	−.363	.108	11.204	.001	0.696	.562	0.860	−.361	.080	20.448	.001	0.697	0.596	0.815
3	Block 3
	Sample	.026	.232	.013	.910	1.027	.651	1.618	.369	.176	4.372	.037	1.446	1.023	2.043
	Static-99R	.196	.042	21.252	.001	1.216	1.119	1.322	.196	.031	39.984	.001	1.216	1.145	1.292
	Dynamic total (pre)	.068	.013	29.079	.001	1.070	1.044	1.097	.043	.010	20.448	.001	1.044	1.025	1.064
	Criminality change	−.234	.111	4.439	.035	0.792	.637	.984	−.104	.076	1.905	.168	0.901	.777	1.045
4	Block 3
	Sample	−.014	.234	.004	.952	0.986	.623	1.559	.334	.178	3.526	.060	1.396	.985	1.979
	Static-99R	.191	.042	20.304	.001	1.211	1.114	1.316	.192	.031	38.908	.001	1.212	1.141	1.287
	Dynamic total (pre)	.066	.013	27.270	.001	1.068	1.042	1.095	.042	.009	20.106	.001	1.043	1.024	1.063
	Treatment responsivity change	−.215	.126	2.890	.089	0.807	.630	1.033	−.201	.094	4.575	.032	0.818	.680	0.983
5	Block 3
	Sample	−.453	.269	2.830	.093	0.636	.375	1.078	−.172	.208	0.681	.409	0.842	.560	1.266
	Static-99R	.188	.042	20.482	.001	1.207	1.112	1.309	.189	.030	38.662	.001	1.208	1.138	1.283
	Dynamic total (pre)	.073	.013	32.249	.001	1.075	1.049	1.103	.050	.010	27.171	.001	1.052	1.032	1.072
	Sexual deviance change	−.398	.133	8.991	.003	0.672	.518	.871	−.394	.095	17.288	.001	0.675	.560	.812
	Criminality change	−.182	.137	1.764	.184	0.833	.637	1.091	−.006	.084	0.006	.941	0.994	.842	1.173
	Treatment responsivity change	.213	.185	1.326	.249	1.238	.861	1.780	.078	.127	0.381	.537	1.082	.843	1.387

Note. Significant p values in bold. VRS-SO = Violence Risk Scale–Sexual Offender version.

The results reinforced the incremental validity of static (VRS-SO static or Static-99R) and dynamic factors for sexual recidivism (as demonstrated in past analyses on the individual samples) and violent recidivism in the second and third blocks of all analyses after controlling for program site. Program site was significant in the first block of analyses, reflecting the higher recidivism rates observed with the Clearwater Program. Program site was no longer significant in the prediction of sexual recidivism after controlling for risk and change in subsequent blocks of analyses; however, it remained significant in the prediction of violence, indicating that additional factors unique to the Clearwater Program (e.g., more than half the sample comprising rapists and mixed offenders) accounted for its higher rates of violent recidivism.²

Block 3 of each of the analyses seemed to closely parallel the earlier results of the univariate change analyses. Significant e^B values for change effects ranged from .662 to .888 for sexual recidivism and .639 to .895 for violent recidivism; each e^B value representing the predicted decrease in the hazard (i.e., 10.5%-36.1%) for one unit increase in change after controlling for risk. Controlling for the VRS-SO static and dynamic factors (Table 3), changes on the total dynamic score (Model 1) and the Sexual Deviance factor (Model 2) were each significantly associated with reductions in sexual and violent recidivism. This was also observed while controlling for the Static-99R and pretreatment dynamic items (Table 4). Changes on the Criminality factor (Model 3) were associated with reductions in sexual but not violent recidivism. Changes on the Treatment Responsivity factor (Model 4) uniquely predicted decreased sexual and violent recidivism while controlling for the VRS-SO static and dynamic items, as well as reductions in violence while controlling for the Static-99R and VRS-SO dynamic item total. Finally, when change scores on the three broad factors were entered simultaneously in the third block of analyses (Model 5), again, only change scores on the Sexual Deviance factor significantly predicted decreases in sexual and violent recidivism. In short, the analyses demonstrated that VRS-SO measured risk and treatment change, particularly dynamic total and Sexual Deviance factor change scores, were significantly associated with decreased sexual and violent recidivism post-release after accounting for individual differences in risk level.

Logistic Regression Generated Estimates of Sexual and Violent Recidivism

The next set of analyses used logistic regression at fixed 5-year follow-ups to estimate projected rates of sexual and violent recidivism as a function of the individual’s pretreatment risk level as well as the amount of change made. The Hosmer–Lemeshow goodness-of-fit test, conducted for all logistic regression analyses, was not significant in the prediction of either outcome, indicating that the logistic distribution provided an acceptable approximation of sexual and violent recidivism rates. The first of these logistic regression derived estimates are reported in Table 5, which reports actual and estimated recidivism rates for the pretreatment dynamic total risk categories overall and at each of the three levels of change (low, medium, high) used in survival analyses. The VRS-SO pretreatment total score predicted sexual (B¹ = .109, p < .001, constant = −5.736) and violent (B¹ = .090, p < .001, constant = −4.137) recidivism. These values were used to generate the sexual and violent recidivism estimates in the “overall category” of Table 5 which represents recidivism estimates without consideration of change information; that is, for a pretreatment or Time 1 assessment.

Table 5.

Actual and Logistic Regression Estimated Rates of Sexual and Violent Recidivism (Fixed 5-Year Follow-Up) at VRS-SO Pretreatment (Static + Dynamic) and Change Categories.

				Change category
	Overall sample			low			medium			high
Pretreatment risk band and criterion	Actual	Estimate	n	Actual	Estimate	n	Actual	Estimate	n	Actual	Estimate	n
Sexual recidivism
Low	0.0	2.2	63	0.0	3.1	10	0.0	2.1	36	0.0	1.3	17
Moderate–low	5.8	5.6	190	5.6	7.5	36	7.6	5.1	119	0.0	3.3	35
Moderate–high	12.6	13.7	167	25.0	18.0	36	10.7	12.7	103	3.6	8.5	28
High	37.3	36.0	118	40.5	43.6	37	42.6	33.8	54	22.2	24.6	27
Violent recidivism
Low	9.5	7.0	63	10.0	10.3	10	11.1	6.9	36	5.9	4.4	17
Moderate–low	14.2	14.4	190	13.9	20.4	36	16.0	14.2	119	8.6	9.3	35
Moderate–high	27.5	27.8	167	33.3	36.9	36	29.1	27.3	103	14.3	18.9	28
High	53.4	52.3	118	64.9	62.6	37	55.6	51.8	54	33.3	40.0	27

Note. Pretreatment (time 1) risk category: Low 0-20 (M = 17.28, SD = 2.29), Moderate–low 21-30 (M = 26.20, SD = 2.66), Moderate–high 31-40 (M = 35.37, SD = 2.88), High 4172(M = 47.04, SD = 5.24). Change categories: low < 1.5 (M = .41.SD = .68), medium 1.5 to 5 (M = 3.36, SD = 1.08), high 5.5 + (M = 6.58, SD = .95). VRS-SO = Violence Risk Scale–Sexual Offender version.

In subsequent analyses, both sets of predictors uniquely predicted sexual (pretreatment total B¹ = .108, p < .001, change score B¹ = −.140, p = .018, constant = −5.279) and violent (pretreatment total B¹ = .090, p < .001, change score B¹ = −.149, p = .002, constant = −3.661) recidivism. The mean scores for the pretreatment and change categories were then applied to the B¹ and B⁰ values to generate the estimated recidivism rates reported in Table 5 as a function of change across three levels; that is, for a posttreatment or Time 2 assessment. There were projected decreases in the estimated (and actual) rates of sexual and violent recidivism with each increase in change category across each pretreatment risk threshold, with the most pronounced differences occurring among the high-risk group of offenders (VRS-SO pretreatment total M = 47.04). For instance, the 5-year estimate of sexual recidivism is approximately 36% for the high-risk group of offenders; however, for those who attended treatment and made low change, the projected sexual recidivism rate was 43.6%, for medium change, 33.8% (somewhat lower than the estimate for the overall sample), and a high amount of change, 24.6%. Note that the mean score for the high-change group of 6.58 amounts to approximately one full standard deviation of change in the dynamic score.

These procedures were subsequently repeated with the VRS-SO static and dynamic scores entered separately along with the VRS-SO change score. These analyses are intended to provide an illustration of the incremental value added of risk-change data over and above the individual contributions of the static and dynamic components of the VRS-SO (as opposed to a different clinical use of the VRS-SO by separately calibrating static and dynamic components in this manner). Figures 4 and 5 are a pictorial logistic regression-based counterpart to the Cox regression survival analyses reported in Tables 3 and 4. As in the preceding analyses, all sets of predictors uniquely predicted sexual (static B¹ = .179, p < .001, dynamic B¹ = .077, p < .001, change score B¹ = −.147, p = .014, constant = −5.225) and violent (static B¹ = .133, p < .001, dynamic B¹ = .069, p < .001, change score B¹ = −.152, p = .002, constant = −3.552) recidivism. Three groups were created based on the mean score ±1 standard deviation for the VRS-SO static (M = 9.01, SD = 4.46) and dynamic (M = 23.55, SD = 7.08) scores, as well as the three change groups created using the same rationale from the preceding analyses.

Figure 4.

Logistic regression generated estimates of sexual recidivism (fixed 5-year follow-up) as a function of VRS-SO static and dynamic risk levels and therapeutic change.

Figure 5.

Logistic regression generated estimates of violent recidivism (fixed 5-year follow-up) as a function of VRS-SO static and dynamic risk levels and therapeutic change.

Following Thornton (2011) as described hereinbefore, we then generated violent and sexual recidivism estimates at low, medium, and high scores on the three predictors. When stratified by static and dynamic score similar trends were observed as in Table 5, with rates of recidivism decreasing as values on the static and dynamic predictors decreased in magnitude while change increased. For instance, with the high-risk dynamic band (30.7), individuals scoring high (13.5) on the static factors and demonstrating little change had a projected rate of sexual recidivism of approximately 37.3%, while those demonstrating moderate change, approximately 27.9%, and high change, 19.4%. By contrast, those scoring low on the static factors (4.5), even if they were dynamically high risk, had markedly lower predicted rates of sexual recidivism of approximately 10.5%, 7.2%, and 4.6% for individuals demonstrating low, medium, and high amounts of changes, respectively.

Discussion

We conducted an extended analysis of VRS-SO change data on a large combined sample of sex offenders from Canada and New Zealand. The larger sample provided sufficient statistical power to examine novel applications of change information to quantify reductions in recidivism as a function of risk level. There were important differences between the two samples. Specifically, the Canadian sample (Olver et al., 2007) consisting of rapists and child molesters, was higher risk than the New Zealand sample of child molesters only (Beggs & Grace, 2010, 2011), as evidenced by higher recidivism base rates and higher scores on the VRS-SO and Static-99R. This risk-related heterogeneity of the combined samples increased the range and variance in VRS-SO scores, also serving to enhance power.

Risk, Treatment Change, and Recidivism

Positive pre–posttreatment changes on the VRS-SO were associated with lower rates of sexual and violent recidivism. This general finding remained significant irrespective of program site and when pretreatment risk level and length of follow-up were statistically controlled. Although the Sexual Deviance factor did not appear to demonstrate the same magnitude of prediction compared with the other two factors, interestingly, changes in this domain bore the most consistent relationship to decreased recidivism for all outcomes. Our interpretation of this is that many of the interventions geared toward changing the domains of this factor (e.g., developing and understanding a sexual offending cycle, managing deviant arousal, developing a prosocial and sexually healthy lifestyle) may generate changes on this factor that have broader implications for other forms of criminal behavior, sexual and nonsexual.

We used logistic regression to estimate rates of sexual and violent recidivism across different VRS-SO score thresholds as a function of different amount of change over a fixed 5-year follow-up. This methodology has important implications for the interpretation and reporting of risk information. Specifically, the use of logistic regression provides a useful projection of estimated recidivism probability at clearly defined follow-up periods. The results demonstrated systematic decreases in predicted rates of sexual and violent recidivism as a function of increasing change, at different levels of risk.

The amount of change recorded in treatment and the magnitude of its relationship to outcome was not trivial. The set of analyses demonstrated the value added of systematically incorporating change information into posttreatment risk estimates. The disparities in projected rates of sexual and violent recidivism as a function of change were most pronounced at the upper extremes, which documented a relative 50% decrease in recidivism rates between individuals registering little or no change, versus those who demonstrated a large amount of change (e.g., about 1 SD). As the creation of risk and change subgroups reduces the sample sizes within the individual cells for producing reliable recidivism estimates, this underscores the additional value of using logistic regression modeling with continuous predictors to estimate recidivism rates at specific risk and change score thresholds. A set of norms using logistic regression generated estimates of recidivism at pre- (Time 1) and posttreatment (Time 2) incorporating additional samples will be forthcoming; those interested may visit www.psynergy.ca to access these norms.

Given the numerous pressing decisions made in the criminal justice system that rely on accurate assessments of risk, including posttreatment, the important implications of these results are clear. While the dynamic nature of risk and the relationship between changes in risk across treatment and recidivism outcomes are inherent assumptions in the widely applied risk–need–responsivity model of offender rehabilitation, to date, empirical research testing these assumptions has been scant. The present study provides robust support for the premise that changes in dynamic risk factors as a result of treatment among sexual offenders can show practical links to decreases in observed sexual and violent recidivism.

Some Conditions Under Which Evaluators May Revise Their Assessments

How should risk-mitigating information (e.g., completion of sex offender treatment) be taken into consideration? First, we argue that risk-related change requires a stable and credible change agent such as the completion of a verified and established risk-reduction program. Not all treatment programs or putative risk-reduction regimes are necessarily equal, and as Hanson, Bourgon, Helmus, and Hodgson (2009) demonstrated in their sex offender treatment outcome meta-analysis, greater adherence to the risk–need–responsivity principles by sex offender treatment programs was associated with larger reductions in recidivism. Part of this also entails examining what behaviors constitute change. Gordon and Wong (2010) for instance, make the distinction between offense analogue behaviors (OABs) and offense replacement behaviors (ORBs). OABs refer to the manifestations of criminogenic needs within institutional settings, where there may otherwise be limited access to potential sexual offense victims (e.g., stalking or quasi-stalking behaviors of staff, lusty or inappropriate talk, deviant or inappropriate fantasies). ORBs by contrast refer to prosocial positive behaviors that should take the place of OABs (e.g., use of healthy coping, increase in appropriate fantasy, an increase in prosocial thinking styles). In principle, treatment improvement should be indicated by an increase in ORBs and reduction in OABs, which in turn, would signify relevant risk-related changes within a closed institutional setting. In short, a reduction in OABs and increase in ORBs, represent the observable behavioral changes that are quantified by the VRS-SO. High-change scorers, for instance, would have more ORBs and fewer OABs; low-change scorers the opposite. This would also be relative to risk level, as higher risk offenders would also be expected to exhibit more OABs.

Second, some risk assessment approaches, such as the VRS-SO, have built-in mechanisms for structuring their appraisals of change. Ultimately, a comprehensive appraisal of risk can be systematically formulated incorporating relevant change-related information, although how this is done may vary depending on the instruments or approaches used (e.g., approaches that involve re-rating items on a given tool versus applying explicit change guidelines). Here, we have reported actual and logistic regression generated estimates of sexual and violent recidivism at different levels of change as a function of the individual’s overall risk level. In addition to this approach, evaluators may also wish to take into consideration the individual’s percentile rank on a given tool, pre- and posttreatment, if such normative information is available for the tool. For instance, it would no doubt be helpful to know if an individual scored at the 95th percentile relative to other sexual offenders on a given tool, and then scored in the 15th percentile on the amount of change they made over treatment; or by contrast, if a hypothetical offender scored in the 90th percentile in the amount of change made. Percentile ranks will also be forthcoming in the VRS-SO norms including its static, dynamic, and change components.

Finally, the evidence to adjust static actuarial appraisals of risk based on risk-relevant information (e.g., downgrading the risk level of a static tool based on treatment information), although tempting in some circumstances, is limited and possibly ill-advised without a structured and empirically informed means of doing so. As an alternative, we (Olver & Wong, 2011) have suggested,

Instead of adjusting risk levels assessed using static tools, it is probably clinically more parsimonious, objective and, ultimately, more defensible to assess risk using tools with dynamic predictors and to re-assess risk at appropriate intervals. Static risk tools, if used in treatment settings, should be supplemented with dynamic tools that can assess change especially in post-treatment risk evaluations. In a post-treatment assessment context, the two sets of measures can be used in tandem and possible discrepancies in risk estimates can be reconciled using professional discretion. Part of this would entail evaluating the reliability of the change agent and the credibility of the client’s change. (p. 124)

Strengths, Limitations, and Future Directions

The present study has several strengths and limitations, with implications for future research. Important strengths include the large and heterogeneous nature of this combined sample of treated sexual offenders, with a long postrelease follow-up. The raters were carefully trained and complete pre- and posttreatment information was available. These conditions, in turn, facilitated the performance of several high-powered change analyses to illustrate the value added of incorporating change information (if available) into posttreatment assessments. The use of logistic regression methodology, as previously illustrated by Hanson et al. (2010) and Thornton (2011) provided a user friendly and interpretable metric for incorporating change information into sexual and violent recidivism estimates.

These strengths notwithstanding, the present study involves the incorporation of two archival samples and there are limitations associated with this. The absence of an interview to inform VRS-SO ratings, particularly for the dynamic variables and change, invariably removes an important source of information. This shortcoming is offset, however, by numerous studies elsewhere demonstrating detailed archival records to have ample information to generate valid and reliable scores on dynamic tools (e.g., Gossner & Wormith, 2007; Harris, Rice, & Quinsey, 1993; Viljoen et al., 2008; Wong & Gordon, 2006). A second and related shortcoming of the present study is its retrospective nature. A more ecologically valid design entails ratings made with file plus interview and the individuals followed up prospectively post-release, although such a design is also invariably more time-consuming, and does not entail the immediate access to long-term follow-ups as with retrospective designs. A prospective investigation of the VRS-SO has been ongoing and the results will be reported in a future paper. Finally, the present study included measurements at two time points, which although permitting a useful examination of change, is arguably less dynamic than examining changes over multiple (e.g., three or more) time points, such as conducting reassessments when the individual has been released to the community. This in turn is also fertile ground for further research and would permit an examination of whether continued changes in risk (increases or decreases) are associated with corresponding changes in recidivism patterns over different time periods.

Conclusion

The present study demonstrated the potential value of having a structured operationalization of change to inform sexual offender risk appraisals. We used a range of procedures to demonstrate not only the informational value of change data, but potentially clinically useful and systematic applications of this information to quantify changes in risk. Risk is a dynamic construct, and sexual offender treatment programs, specialized legislation, and community management and supervision approaches are all ways to attempt to contain it and reduce it to prevent future sexual violence. We believe the applications of change data described here offer one means of contributing toward these efforts.

Footnotes

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Mark E. Olver and Stephen C. P. Wong are also authors of the VRS-SO.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Andrews

D. A.

Bonta

(2010). The psychology of criminal conduct, fifth edition. New Providence, NJ: LexisNexis.

Beggs

S. M.

Grace

R. C.

(2010). Assessment of dynamic risk factors: An independent validation study of the Violence Risk Scale: Sexual Offender Version. Sexual Abuse: A Journal of Research and Treatment, 22, 234-251. doi:10.1177/1079063210369014

Beggs

S. M.

Grace

R. C.

(2011).Treatment gains for sexual offenders against children predicts reduced recidivism: A comparative validity study. Journal of Consulting and Clinical Psychology, 79, 182-192. doi:10.1037/a0022900

Douglas

K. S.

Kropp

P. R.

(2002). A prevention-based paradigm for violence risk assessment: Clinical and research applications. Criminal Justice and Behavior, 29, 617-658. doi:10.1177/009385402236735

Douglas

K. S.

Skeem

(2005). Violence risk assessment: Getting specific about being dynamic. Psychology, Public Policy, and Law, 11, 347-383. doi:10.1037/1076-8971.11.3.347

Gordon

Wong

S. C. P.

(2010). Offense analogue behaviours as indicator of criminogenic need and treatment progress. In Daffern

Jones

Shine

(Eds.), Offence paralleling behaviour: An individualised approach to offender assessment and treatment (pp. 171-183). London, England: Wiley-Blackwell.

Gossner

Wormith

J. S.

(2007). The prediction of recidivism among young offenders in Saskatchewan. Canadian Journal of Police & Security Services, 5, 70-82.

Hanson

R. K.

Babchishin

K. M.

Helmus

Thornton

(2013). Quantifying the relative risk ratios for Static-99R. Sexual Abuse: A Journal of Research and Treatment, 25(5), 482-515. doi:10.1177/1079063212469060

Hanson

R. K.

Bourgon

Helmus

Hodgson

(2009). The principles of effective correctional treatment also apply to sexual offenders. Criminal Justice and Behavior, 36, 865-891. doi:10.1177/0093854809338545

10.

Hanson

R. K.

Harris

A. J. R.

Scott

Helmus

(2007). Assessing the risk of sex offenders on community supervision (User Report No. 2007-05). Ottawa, Ontario, Canada: Public Safety and Emergency Preparedness Canada.

11.

Hanson

R. K.

Helmus

Thornton

(2010). Predicting recidivism among sexual offenders: A multi-site study of Static-2002. Law and Human Behavior, 34, 198-211. doi:10.1007/s10979-009-9180-1

12.

Hanson

R. K.

Lloyd

C. D.

Helmus

Thornton

(2012). Developing non-arbitrary metrics for risk communication: Percentile ranks for the Static-99/R and Static-2002/R sexual offender risk tools. International Journal of Forensic Mental Health, 11, 9-23. doi:10.1080/14999013.2012.667511

13.

Hanson

R. K.

Thornton

(1999). Static 99: Improving actuarial risk assessments for sex offenders (User Report 99-02). Ottawa, Ontario: Department of the Solicitor General of Canada.

14.

Harris

G. T.

Rice

M. E.

Quinsey

V. L.

(1993). Violent recidivism of mentally disordered offenders: The development of a statistical prediction instrument. Criminal Justice and Behavior, 20, 315-335. doi:10.1177/0093854893020004001

15.

Helmus

Hanson

R. K.

Thornton

Babchishin

K. M.

Harris

A. J. R.

(2012). Absolute recidivism rates predicted by Static-99R and Static-2002R sex offender risk assessment tools vary across samples: A meta-analysis. Criminal Justice and Behavior, 33, 1148-1171. doi:10.1177/0093854812443648

16.

Helmus

Thornton

Hanson

R. K.

Babchishin

K. M.

(2012). Improving the predictive accuracy of Static-99 and Static-2002 with older sex offenders: Revised age weights. Sexual Abuse: A Journal of Research and Treatment, 24, 64-101. doi:10.1177/1079063211409951

17.

Hudson

S. M.

Wales

D. S.

Bakker

Ward

(2002). Dynamic risk factors: The Kia Marama evaluation. Sexual Abuse: A Journal of Research and Treatment, 14, 103-119. doi:10.1177/107906320201400203

18.

Jung

Guyalets

(2011). Using clinical variables to evaluate treatment effectiveness in programmes for sexual offenders. Journal of Sexual Aggression 17, 166-180. doi:10.1080/13552601003802238

19.

Kraemer

H. C.

Kazdin

A. E.

Offord

D. R.

Kessler

R. C.

Jensen

P. S.

Kupfer

D. J.

(1997). Coming to terms with the terms of risk. Archives of General Psychiatry, 54, 337-343. doi:10.1001/archpsyc.1997.01830160065009

20.

Nunes

K. L.

Babchishin

K. M.

Cortoni

(2011). Measuring treatment changes in sex offenders: Clinical and statistical significance. Criminal Justice and Behavior, 38, 157-173. doi:10.1177/0093854810391054

21.

Nunes

K. L.

Cortoni

(2007). Assessing treatment change in sexual offenders (Research Report 2008 No R-184). Ottawa, Ontario: Correctional Service of Canada.

22.

Olver

M. E.

Wong

S. C. P.

(2006). Psychopathy, sexual deviance, and recidivism among sex offenders. Sexual Abuse: A Journal of Research and Treatment, 18, 65-82. doi:10.1007/s11194-006-9006-3

23.

Olver

M. E.

Wong

S. C. P.

(2011). A comparison of static and dynamic assessment of sexual offender risk and need in a treatment context. Criminal Justice and Behavior, 38, 113-126. doi:10.1177/0093854810389534

24.

Olver

M. E.

Wong

S. C. P.

Nicholaichuk

Gordon

(2007). The validity and reliability of the Violence Risk Scale–Sexual Offender version: Assessing sex offender risk and evaluating therapeutic change. Psychological Assessment, 19, 318-329. doi:10.1037/1040-3590.19.3.318

25.

Prochaska

J. O.

DiClemente

C. C.

Norcross

J. C.

(1992). In search of how people change: Applications to the addictive behaviors. American Psychologist, 47, 1102-1114.

26.

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. doi:10.1007/s10979-005-6832-7

27.

Rice

M. E.

Harris

G. T.

Lang

Cormier

C. A.

(2006). Violent sex offenses: How are they best measured from official records? Law and Human Behavior, 30, 525-541. doi:10.1007/s10979-006-9022-3

28.

Tabachnick

B. G.

Fidell

L. S.

(2001). Using multivariate statistics. Boston, MA: Allyn & Bacon.

29.

Thornton

(2011, November). Interpreting SRA-FV total need scores. Paper presented at the 30th annual meeting of the Association for the Treatment of Sexual Abusers, Toronto, Ontario, Canada.

30.

Viljoen

J. L.

Scalora

Cuadra

Bader

Chavez

Ullman

Lawrence

(2008). Assessing risk for violence in adolescents who have sexually offended: A comparison of the J-SOAP-II, J-SORRAT-II, and SAVRY. Criminal Justice and Behavior, 35, 5-23. doi:10.1177/0093854809340991

31.

Wakeling

Beech

A. R.

Freemantle

(2013). Treatment change and its relationship to recidivism in a sample of 3773 sex offenders in the UK. Psychology, Crime, & Law, 19, 233-252. doi:10.1080/1068316X.2011.626413

32.

Wong

S. C. P.

Gordon

(2006). The validity and reliability of the Violence Risk Scale: A treatment-friendly violence risk assessment tool. Psychology, Public Policy, and Law, 12, 279-309. doi:10.1037/1076-8971.12.3.279

33.

Wong

S. C. P.

Olver

M. E.

Nicholaichuk

T. P.

Gordon

(2003). The Violence Risk Scale: Sexual Offender version (VRS-SO). Saskatoon: Regional Psychiatric Centre and University of Saskatchewan, Saskatoon, Saskatchewan, Canada.

Incorporating Change Information Into Sexual Offender Risk Assessments Using the Violence Risk Scale–Sexual Offender Version

Abstract

Keywords

Assessing Changes in Sexual Offender Risk

Violence Risk Scale–Sexual Offender Version (VRS-SO)

Applications of Risk Assessment Information

Rationale for the Present Study

Method

Samples

Clearwater Sex Offender Program

Kia Marama Sex Offender Treatment Program

Measures

VRS-SO

Static-99R

Planned Analyses

Results

Group Comparisons

Predictive Accuracy

Univariate Change Analyses

Correlation and Cohen’s d analyses

Survival analyses

Change, Sexual Violence Risk, and Recidivism

Logistic Regression Generated Estimates of Sexual and Violent Recidivism

Discussion

Risk, Treatment Change, and Recidivism

Some Conditions Under Which Evaluators May Revise Their Assessments

Strengths, Limitations, and Future Directions

Conclusion

Footnotes

Declaration of Conflicting Interests

Funding

Notes

References