Psychological Assessment 2007, Vol. 19, No. 3, 318 –329 Copyright 2007 by the American Psychological Association 1040-3590/07/$12.00 DOI: 10.1037/1040-3590.19.3.318 The Validity and Reliability of the Violence Risk Scale—Sexual Offender Version: Assessing Sex Offender Risk and Evaluating Therapeutic Change Mark E. Olver Stephen C. P. Wong University of Saskatchewan Regional Psychiatric Centre of the Correctional Service of Canada and the University of Saskatchewan Terry Nicholaichuk and Audrey Gordon This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Regional Psychiatric Centre of the Correctional Service of Canada The Violence Risk Scale—Sexual Offender version (VRS–SO) is a rating scale designed to assess risk and predict sexual recidivism, to measure and link treatment changes to sexual recidivism, and to inform the delivery of sexual offender treatment. The VRS–SO comprises 7 static and 17 dynamic items empirically or conceptually linked to sexual recidivism. Dynamic items with higher ratings identify treatment targets linked to sexual offending. A modified stages of change model assesses the offender’s treatment readiness and change. File-based VRS–SO ratings were completed on 321 sex offenders followed up an average of 10 years postrelease. VRS–SO scores predicted sexual and nonsexual violent recidivism postrelease and demonstrated acceptable interrater reliability and concurrent validity. A factor analysis of the dynamic items generated 3 factors labeled Sexual Deviance, Criminality, and Treatment Responsivity, all of which predicted sexual recidivism and were differentially associated with different sex offender types. The dynamic items together made incremental contributions to sexual recidivism prediction after static risk was controlled for. Positive changes in the dynamic items, measured at pre- and posttreatment, were significantly related to reductions in sexual recidivism after risk and follow-up time were controlled for, suggesting that dynamic items are indeed dynamic or changeable in nature. Keywords: Violence Risk Scale—Sexual Offender version (VRS–SO), static and dynamic variables, sexual recidivism, prediction, change The sexual abuse of adults and children is an egregious societal concern. Despite gross underreporting (Lisak & Miller, 2002), as many as 40% of Canadian women report having been sexually assaulted since age 16 (Johnson & Sacco, 1995). Sexual abuse also has profound long-term sequelae including emotional, relationship, and mental health concerns (Browne & Finkelhor, 1986). As such, the assessment, treatment, and management of individuals at high risk for sexual recidivism are paramount in the reduction of sexual violence. Static and Dynamic Predictors of Sexual Recidivism and Current Risk Instruments Although predicting sexual violence is important, reducing and preventing such violence should be the ultimate goal of risk assessment (Douglas & Kropp, 2002; Wong & Gordon, 2006). Recent meta-analytic findings have confirmed that sex offender treatment using cognitive-behavioral relapse prevention approaches has the potential to reduce sexual recidivism (Hanson et al., 2002). It follows that sexual recidivism risk is arguably dynamic and can potentially change through intervention or experience. Large-scale meta-analytic reviews (see Hanson & Bussière, 1998; Hanson & Morton-Bourgon, 2004) have identified static risk predictors (e.g., offense history, offender and victim demographics) that have been combined to develop sexual offender recidivism prediction tools such as the Static 99 (Hanson & Thornton, 1999), the Sex Offender Risk Appraisal Guide (Quinsey, Rice, & Harris, 1995), and the Minnesota Sex Offender Screening Tool— Revised (Epperson, Kaul, & Hesselton, 1998). Although these static tools can predict sexual recidivism quite well, given their static nature they do little to inform sex offender treatment, which requires assessing dynamic, or changeable, risk variables to identify treatment targets and measure change. Dynamic risk predictors or variables, also known as criminogenic needs (Andrews & Bonta, 2003) or variable risk factors (Kraemer et al., 1997), have been defined as “changeable or potentially changeable variables that could be influenced or Mark E. Olver, Department of Psychology, University of Saskatchewan, Saskatoon, Saskatchewan, Canada; Stephen C. P. Wong, Regional Psychiatric Centre of the Correctional Service of Canada, Saskatoon, Saskatchewan, Canada, and the Department of Psychology, University of Saskatchewan; Terry Nicholaichuk and Audrey Gordon, Regional Psychiatric Centre of the Correctional Service of Canada. This study was supported by Social Sciences and Humanities Research Council Grant 752-2001-2164 awarded to Mark E. Olver and was based on a doctoral dissertation. We thank the Correctional Service of Canada for providing access to data and resources for the current study and Brenda Maire and Linda Flahr for their assistance in data collection. The views expressed in this article do not necessarily represent the views of the Correctional Service of Canada. Correspondence concerning this article should be addressed to Mark E. Olver, Department of Psychology, University of Saskatchewan, Arts Building Room 154, 9 Campus Drive, Saskatoon, Saskatchewan, Canada S7N 5A5. E-mail: mark.olver@usask.ca 318 This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. SEX OFFENDER RISK ASSESSMENT changed by psychological, social or physiological means such as treatment interventions” (Wong & Gordon, 2006, p. 283). Changes occurring in dynamic risk variables should be linked to changes in recidivism. Although both static and dynamic variables can assess sexual recidivism risk, dynamic variables can ostensibly inform treatment planning, identify treatment targets, and measure changes in risk (Beech, Friendship, Erikson, & Hanson, 2002; Hanson & Harris, 2000; Hudson, Wales, Bakker, & Ward, 2002). Some have argued that dynamic variables are useful only if they have incremental predictive validity and thus add to the predictive accuracy of static variables (Harris et al., 2003). However, others have argued, and we concur, that dynamic variables do not necessarily have to trump the predictive efficacy of static variables to be useful (Wong & Gordon, 2006). Simply put, there is more to dynamic variables than recidivism prediction; their utility in treatment and risk reduction is equally important. One of the challenges in sex offender research is to identify dynamic variables likely related to sexual recidivism as well as to validate those variables as predictors of sexual recidivism. Recent research has identified several putatively dynamic sexual recidivism predictors (Beech et al., 2002; Hanson & Harris, 2000; Hudson et al., 2002; Thornton, 2002), and meta-analytic reviews (Hanson & Morton-Bourgon, 2004) have further identified two broad domains with strong links to sexual recidivism—sexually deviant interests/behaviors and an antisocial orientation/ criminality. Static and/or dynamic variables have also been combined to develop sex offender risk appraisal guides such as the Sex Offender Need Assessment Rating (Hanson & Harris, 2001), Sexual Violence Risk–20 (Boer, Hart, Kropp, & Webster, 1997), Risk for Sexual Violence Protocol (Hart, Kropp, & Laws, 2003) and the Vermont Assessment of Sex Offender Risk (McGrath & Hoke, 2002). However, much of the research examining dynamic predictors of sexual recidivism has simply assumed that dynamic risk variables are indeed dynamic while providing little evidence to support such assumptions. Although this line of research has shown that dynamic variables can predict sexual recidivism, the dynamic variables in these studies have typically been assessed at only one time point and have not been demonstrated to change. Some exceptions include the work of Hudson et al. (2002), who found some negative correlations between pre- and posttreatment change scores on sex offender self-report measures (e.g., cognitive distortions, empathy) and sexual recidivism, and the work of Hanson (2005), who investigated the predictive validity of stable and acute dynamic variables over multiple time points. For risk predictors or variables to be deemed dynamic in nature, changes in these variables (measured at two points in time) should be linked to changes in sexual recidivism (see Douglas & Skeem, 2005; Kraemer et al., 1997). Despite more recognition of the importance of using dynamic variables to assess sex offender risk as well as an increase in corresponding research, there remains a need to integrate dynamic risk variables into a sex offender risk assessment tool to appraise risk, complement treatment, and assess treatment change. The Violence Risk Scale—Sexual Offender Version (VRS–SO) The VRS–SO (Wong, Olver, Nicholaichuk, & Gordon, 2003) was designed to integrate sex offender risk assessment and treat- 319 ment planning, including the assessment of change, within a single instrument. The VRS–SO uses both static and dynamic variables to assess sexual recidivism risk and uses dynamic variables to identify treatment targets and to measure changes in risk as a result of treatment or other change agents. The VRS–SO assesses and measures change by using a modified application of the transtheoretical model of change (TTM; Prochaska, DiClemente, & Norcross, 1992). The TTM posits that individuals progress through a series of stages (precontemplation, contemplation, preparation, action, and maintenance), characterized by different cognitive, experiential, and behavioral changes as they attempt to remediate problem areas. Although the TTM conceptualization of treatment change has been applied to other forensic populations (Levesque, Gelles, & Velicer, 2000; Willoughby & Perry, 2002; Wong & Gordon, 2006), including sexual offenders (Tierney & McCabe, 2005), it has yet to be incorporated into a risk/need assessment tool for sexual offenders. The metric for assessing treatment change and risk reduction on the VRS–SO is presented in the Method section below. Scope of the Present Study The reliability and validity of the VRS–SO are evaluated in two sets of analyses. First, the factor structure, descriptive statistics, interrater reliability, internal consistency, concurrent validity, and predictive validity of the VRS–SO are examined. Second, the relationship between changes in the VRS–SO dynamic variables at two time points (pre- and posttreatment) and changes in sexual recidivism are examined to address whether changes in these variables predict sexual recidivism and, as such, whether the purported dynamic variables are indeed dynamic. Method Participants Participants included 321 male federal offenders who had participated in a high intensity sex offender treatment program (the Clearwater Program) in a maximum-security forensic mental health facility in Canada between 1983 and 1997. The program is 6 – 8 months in duration, provides group and individual therapies, and is cognitive-behavioral in orientation.1 All offenders had one or more index convictions for sexual offenses (96.3%), or had prior sexual charge(s) or conviction(s), or had exhibited sexually inappropriate behaviors (3.7%). Offenders were serving a mean sentence length of 5.8 years (SD ⫽ 3.9). The mean ages of the sample at the time of index offense and data collection were 30.5 years (SD ⫽ 9.8) and 42.7 years (SD ⫽ 9.3), respectively. Approximately 62.6% of the offenders were Caucasian, 33.6% Aboriginal, and 2.8% other ethnic decent. Most offenders (49.3%) were single/never married, 23.7% common law/married, 25.5% separated/divorced, and ⬍ 1% widowed. Demographic information was incomplete for 3 (1%) of the cases. The offender sample 1 See Nicholaichuk, Gordon, Gu, and Wong (2000) for a detailed description of the Clearwater Program. 320 OLVER, WONG, NICHOLAICHUK, AND GORDON consisted of 169 rapists, 56 child molesters, 45 mixed offenders, and 51 incest offenders.2 This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Materials VRS–SO. The VRS–SO is modeled closely after the Violence Risk Scale (Wong & Gordon, 1999, 2006), which is designed to assess nonsexual violence risk and need. The VRS–SO is a 24-item clinician-rated scale comprising 7 static and 17 dynamic items. All items are rated on a 4-point Likert-type scale ranging from 0 to 3 on the basis of a thorough file review and a semistructured interview;3 higher ratings indicate a closer link to inappropriate sexual or nonsexual behaviors, indicating increased sexual recidivism risk. The static and dynamic items, including a brief description of each item, are presented in the Appendix. The static and dynamic components of the VRS–SO were each developed through a different set of procedures. The static component was developed with statistical-actuarial procedures on approximately one randomly selected half of the sample (n ⫽ 152) and cross-validated on the remaining portion of the sample (n ⫽ 169). A pool of 24 static variables identified from the literature was initially coded and correlated with sexual recidivism. Variables with the strongest univariate relationships to outcome were retained and rescaled to a 4-point format. All 7 static items were developed in this manner (see the Appendix) and can be summed to arrive at a static item total. The dynamic component was developed through a detailed review of the sex offender prediction and treatment literature so as to select a wide spectrum of theoretically, empirically, and treatment-relevant dynamic risk variables to maximize the content validity of the VRS–SO. The review included the meta-analytic work of Hanson and Bussière (1998), research identifying dynamic predictors (e.g., Hanson & Harris, 2000; Proulx et al., 1997), the relapse prevention work of Pithers (1990) and Ward and Hudson (1998), and the theoretical traditions of The Psychology of Criminal Conduct (Andrews & Bonta, 2003). A VRS–SO scoring manual (Wong et al., 2003) that includes detailed rating criteria for each static and dynamic item has been developed. Each item has an “objective” section that briefly describes the underlying construct (e.g., deviant sexual preference) together with detailed descriptions for 3-point and 0-point ratings for the item. If the presenting problem behaviors with respect to a given item are “less serious” than those described by a 3 rating, a rating of 2 is given. Similarly if the presenting behaviors are “less positive” than those described by a 0 rating, a rating of 1 is given. Items rated a 2 or 3 are problem areas with close links to sexual offending (i.e., are criminogenic) and should be considered as treatment targets. The different dynamic items are potentially changeable by varying degrees. Changes of the dynamic items are assessed and quantified through the application of a modified TTM. Each of the five stages of change has been operationalized for each of the 17 dynamic items. The progression in the stages of change (e.g., from contemplation at pretreatment to action at posttreatment) demonstrates the extent to which the offender has improved by developing positive coping skills and strategies that are stable, sustainable, and generalizable with respect to each dynamic item. All treatment targets, that is, dynamic items rated 2 or 3, are given a stages of change baseline rating at pretreatment to assess the individual’s motivation and readiness for change. Dynamic items that are not treatment targets, that is, those rated 0 or 1, generally require no stages of change rating. The stages of change are then rerated at posttreatment on all dynamic items identified as treatment targets. Change is quantified by comparing the stages of change rating for each dynamic item at pretreatment with that at posttreatment. Advancing from one stage of change to the next on a given item is an indication of positive change and, hence, risk reduction.4 Progression from one stage to the next stage is scored as a 0.5-point reduction in the pretreatment rating of the item, progression in two stages as a 1.0-point reduction, and so on. This is repeated for each dynamic item identified as a treatment target.5 The total point deductions for each dynamic item at posttreatment are summed across all 17 dynamic items to arrive at a total change score reflecting the total amount of change. The total change score is subtracted from the total pretreatment dynamic ratings to obtain the total posttreatment dynamic ratings. For instance, a child molester entering sex offender treatment may receive a 3-point rating on the item deviant sexual preference and may be assessed as being in the contemplation stage of change for this item. Following treatment during which the offender learns arousal modification and fantasy monitoring strategies, major gains are observed and progress is made to the action stage of change. The individual has progressed two stages (i.e., contemplation to preparation to action), resulting in 2 ⫻ 0.5 ⫽ 1.0 points of change, the change score. The change score is deducted from the pretreatment rating of this item, resulting in a posttreatment rating of 3 – 1 ⫽ 2. The total change score is the sum of the change scores for all the dynamic items. The total VRS–SO pre- and posttreatment scores represent the offender’s overall risk level for sexual recidivism at two points in time. Higher scores indicate higher risk. As the VRS–SO was designed with an emphasis on assessing the relationship between therapeutic change and corresponding changes in risk, we argue that using the stages of change model to measure therapeutic change has important advantages over measuring change by simply rerating the dynamic items at posttreatment. Some rating criteria for the VRS–SO dynamic items and in other risk-rating tools are based on behaviors observable only in 2 Rapists were defined as perpetrators with victim(s) who were at least 14 years of age and child molesters as those with boy or girl victim(s) under the age of 14. Mixed offenders were defined as having both child and adult victims, and incest offenders were defined as those with victims who were family members or relatives. Relatives are individuals sufficiently closely related to the offender such that marriage would normally be prohibited (Hanson & Thornton, 1999). 3 File information may be used to rate the items for research purposes. 4 An exception to this scoring rule is for progress from precontemplation to contemplation, wherein no risk reduction is registered as there is no relevant behavioral improvement. 5 Pretreatment: static total score ⫹ dynamic total score ⫽ VRS–SO total pretreatment scores. Posttreatment: static total score ⫹ dynamic total score – total change score ⫽ VRS–SO total posttreatment score. Total change scores ⫽ sum of differences between pre- and posttreatment Stage of Change ratings for all dynamic variables. Should the individual worsen during treatment, a score of 0.5 is added to the individual’s item score for every stage of deterioration. As such, it is possible for an individual’s posttreatment score to increase (i.e., get worse) rather than decrease (i.e., improve). This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. SEX OFFENDER RISK ASSESSMENT the community but not in prisons or forensic treatment settings, which are restrictive environments. For instance, incarcerated child molesters do not have access to potential child victims. Similarly, it is problematic to infer sexually compulsive behaviors when individuals do not have the freedom to engage in promiscuity, the frequent use of prostitutes, or multiple one night stands. The stages of change descriptions of the dynamic items (the measurements of treatment change) were developed specifically to capture offense-linked proxy behaviors that are relevant to treatment and are observable in forensic and/or treatment settings.6 Therapists can use these observations as a metric to guide their decisions as to how much change has occurred. Similarly, in most forensic inpatient treatment, therapists can assess changes only by observing in-treatment behaviors, not behaviors that can only be observable in the community. A further advantage in using the TTM to guide the measurement of change is that the model also describes how offenders change in therapy in addition to assessing how much change has occurred. Therapists can also use the stages of change ratings to inform their conceptualization of treatment and treatment delivery. Static 99. The Static 99 (Hanson & Thorton, 1999) is a static actuarial sex offender risk assessment measure composed of 10 static items pertaining to sexual and nonsexual offense history, victim characteristics, and offender demographics. Total scores range from 0 to 12 points and can be organized into four risk groups: low (0 –1), medium-low (2–3), medium-high (4 –5), and high (6 –12). The Static 99 is a widely used sexual recidivism risk assessment tool (Hanson, 2006) and is included to assess the VRS–SO’s concurrent and incremental validity. Sexual offense variables. All convictions in the present study were violations of the federal Canadian Criminal Code verified by fingerprinting and available from the nationwide Canadian Police Information Centre database. The index offense was defined as the most recent sexual offense prior to program admission; prior offenses are those committed before program admission. Recidivism was defined as any conviction for a new sexual or nonsexual violent offense following first release to the community after program participation. A sexual offense is any conviction for an offense that was clearly sexual in nature (e.g., sexual assault, sexual interference) or was sexually motivated as determined by reviewing police reports.7 A nonsexual violent offense was an offense against a person that was not sexually motivated (e.g., nonsexual assault, robbery, uttering threats, murder). The binary (yes–no recidivism, coded 1– 0) sexual and nonsexual violent recidivism variables are reported and used in all analyses. Procedure Data coding. The VRS–SO dynamic items and Static 99 were scored by two trained and experienced research assistants by using archival records extracted from the participant’s institutional paper files and a computerized Offender Management System.8 The research assistants made both pre- and posttreatment ratings on the VRS–SO dynamic items. To guard against potential rater biases, care was taken to ensure that all pretreatment ratings were blind to posttreatment and recidivism information. Pretreatment ratings were generally made up to and including the initial and early program evaluations, whereas posttreatment ratings were made incorporating information from the final program evaluation and 321 other available documents such as community prerelease assessments. No interviews were conducted. One of the authors served as the second rater to assess interrater reliability and only collected recidivism data after all ratings were completed. Missing data. Only dynamic items had missing data. No more than two items were missing from the vast majority of the sample (91%); the remaining sample had three to four missing items. The item community support had the most missing data owing to limited information concerning the release possibility for some offenders, although it still had complete ratings for 75.4% of the sample. All missing values were estimated with a regression procedure that can yield accurate estimates of missing values.9 A potential shortcoming of the procedure is that the resulting mathematically derived estimate may be more accurate than the actual item score itself (Tabachnick & Fidell, 2001). We also used the regression procedure to estimate the missing change scores. Rather than entering a stage of change score for the missing values, we regressed a missing change score item on the remaining change score items. Psychometric analyses. Psychometric analyses of the VRS–SO included (a) factor analysis of the dynamic items to identify latent constructs for sexual recidivism risk, (b) descriptive statistics and comparisons between different sex offender groups, (c) measurements of interrater reliability and internal consistency, (d) concurrent validity of the VRS–SO with the Static 99, (e) predictive validity of the VRS–SO for sexual and nonsexual violent recidivism, and (f) evaluating whether computed change scores are associated with reduced sexual recidivism. Four different sets of predictive validity analyses were conducted. First, correlation coefficients (r) and receiver-operating characteristic analyses were used to investigate the relationships of 6 For instance, a part of the description of the maintenance stage for deviant sexual preference (D16) is “the individual is now able to effectively suppress or control preferences for deviant sexual stimuli. Evidence of this includes the identification of external and internal risk factors and high-risk situations that contribute to sexually deviant preferences and associated behaviours, as well as adherence to relapse prevention strategies. Other examples may include the reliable use of libido inhibiting medication . . . , compliance with phallometric sessions, or other reliable observations” (Wong et al., 2003, p. 61). 7 Official information sources, such as police reports or criminal profile report in the Offender Management System, were reviewed to ascertain sexual motivation of an offense primarily for the index conviction(s) for which the offender was sentenced. In some cases, a recidivist offender received a federal sentence for a nonsexual violent conviction, and it was possible to access information concerning the motivation of the recidivistic offense from Offender Management System as a result. An example would be an individual who recidivated by committing second degree murder, although a review of Correctional Service of Canada documents revealed this to be a sexual homicide. 8 The records included nursing notes, psychological reports, treatment program summaries and performance evaluations, criminal records, social histories, community assessments, collateral reports, results of psychological testing, and phallometric assessments. 9 With stepwise multiple regression, the variable with missing data was regressed on the remaining dynamic items to create a linear combination of weighted variables. The intercept and unstandardized regression beta weights were then summed to estimate the value of the missing item for a specific case. This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 322 OLVER, WONG, NICHOLAICHUK, AND GORDON the predictors with sexual and nonsexual violent recidivism. The area under the curve (AUC) statistic in receiver-operating characteristic analyses, unlike r, is independent of changes in base rate (Rice & Harris, 1995). The AUC (e.g., .70) can be interpreted as the probability (70% chance) that a randomly selected recidivist will have a higher score on a given risk measure than will a randomly selected nonrecidivist (Hanson, 1997). Second, Cox regression analyses were used to examine the incremental contributions of dynamic variables in predicting sexual recidivism over and above that made by static variables, including a wellestablished and validated static measure such as the Static 99. Third, predictive validity was examined through life tables survival analysis (Hanson & Thornton, 1999), which involved measuring the relative cumulative sexual recidivism failure rates of four VRS–SO risk groups (described below) over a 10-year follow-up. Predictive validity was also evaluated through examining the relationship between risk groups, or “bins,” derived from VRS–SO total scores with percent sexual recidivism for each of the four groups. Based, in part, on an examination of the distribution of VRS–SO total scores and attempts to retain a sufficient number of offenders in each category, four risk groups were developed: low (0 –20), moderate-low (21–30), moderate-high (31– 40), and high (41–72). The percentages of offenders sexually recidivating within each risk group were computed with fixed 3-, 5-, and 10-year follow-up windows to test the predictive validity of the VRS–SO over short-, medium-, and longer term follow-ups, respectively. Within the duration of a certain fixed follow-up window, offenders who received a new sexual conviction and those who were offense free were included in the analyses. Offenders who were at risk for a shorter duration than the specific follow-up window were not included. The change analyses involved computing change scores on the dynamic items (as outlined in the Method section) and examining their relationship to sexual reconviction. An inverse relationship between change and recidivism is predicted (i.e., as positive change increases, sexual recidivism decreases). Although simple correlational analyses are an obvious means to test for such a relationship, a problem with this method is that it does not take into account that individuals of different risk levels are being compared on the changes they make. For instance, an offender scoring quite high on the VRS–SO at pretreatment (such as 50) and registering a certain amount of change during treatment, would still be at higher risk for recidivism than would an offender receiving a lower VRS–SO score (such as 30) and evidencing the same amount of change. Ideally, a test of the relationship between change and recidivism should be done separately for each possible score on the VRS–SO, thereby controlling for level of risk. However, given pragmatic limitations in sample size, this would not be possible. Arguably, the best alternative would be to control for risk statistically and thereby examine the relationship of change to sexual recidivism independent of risk. As such, Cox regression survival analysis is used to examine the relationship of change to sexual recidivism while risk as measured by the VRS–SO static scores and VRS–SO total score (sum of static and dynamic items) is controlled for. Important advantages of Cox regression survival analysis are that it takes into account and adjusts for differences in follow-up time and is less influenced by variations in recidivism base rate (Hanson, 2006). Results Factor Analysis An exploratory principle components analysis with varimax rotation was conducted on the pretreatment-rated dynamic items for the entire sample to identify underlying factors informative to treatment providers. The scree plot and eigenvalues criteria suggested a three-factor solution which was then extracted with principle axis factoring with Promax oblique rotation (see Tabachnick & Fidell, 2001, p. 615) to determine whether the derived factors were correlated. A cutoff loading criterion of .32 was used (Tabachnick & Fidell, 2001). The three factors were labeled Sexual Deviance (␣ ⫽ .87), Criminality (␣ ⫽ .79), and Treatment Responsivity (␣ ⫽ .72). Their intercorrelations were moderate and significant: Sexual Deviance ⫻ Criminality (r ⫽ .18, p ⬍ .05), Criminality ⫻ Treatment Responsivity (r ⫽ .55, p ⬍ .001), and Sexual Deviance ⫻ Treatment Responsivity (r ⫽ .38, p ⬍ .001). Table 1 presents the pattern matrix where the factor loadings represent the unique relationship between the different variables that compose the three factors. The intimacy deficits item did not load above the cutoff on any of the three factors, and the emotional control item had an idiosyncratic loading pattern (i.e., loading negatively on Treatment Responsivity and positively on Criminality). As such, these items were not included in the computation of factor scores, that is, summing the items loading on their respective factors. Descriptive Statistics and Between-Groups Comparisons Table 2 presents descriptive statistics for the total sample and the different sex offender subtypes for all VRS–SO scale components (including change scores) and the Static 99. Overall group differences were tested with one-way analyses of variance, and individual differences with post hoc Tukey beta comparisons (at Table 1 VRS–SO Dynamic Items Pattern Matrix Dynamic item Sexually deviant lifestyle Deviant sexual preference Offense planning Sexual offending cycle Sexual compulsivity Impulsivity Interpersonal aggression Substance abuse Compliance with community supervision Criminal personality Community support Intimacy deficits Insight Treatment compliance Cognitive distortions Released to high-risk situations Emotional control Sexual Deviance Criminality Treatment Responsivity .852 .788 .786 .750 .661 ⫺.037 ⫺.182 ⫺.251 ⫺.027 ⫺.059 ⫺.243 .014 .232 .841 .697 .502 .017 .067 ⫺.007 ⫺.064 ⫺.103 ⫺.114 .150 ⫺.056 .057 .124 .230 .142 ⫺.114 ⫺.125 .135 .462 .431 .334 .208 ⫺.137 .046 ⫺.099 .216 .190 .239 ⫺.113 .808 .694 .651 .300 .170 .242 .363 .411 ⫺.372 Note. Variables loading on a given factor are highlighted in bold. VRS– SO ⫽ Violence Risk Scale—Sexual Offender version. SEX OFFENDER RISK ASSESSMENT 323 Table 2 Means and Standard Deviations of the VRS–SO (Pre-, Posttreatment, and Change Scores) and Static 99 Among Sexual Offender Subtypes This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Rapist (n ⫽ 169) Child molester (n ⫽ 56) Mixed offender (n ⫽ 45) Incest offender (n ⫽ 51) Total sample (N ⫽ 321) Measure M SD M SD M SD M SD M SD F (3, 317) p Static 99 VRS–SO static Dynamic (pre) Dynamic (post) Total dynamic change VRS–SO total (pre) VRS–SO total (post) Sexual deviance (pre) Sexual deviance (post) Sexual deviance change Criminality (pre) Criminality (post) Criminality change Treatment responsivity (pre) Treatment responsivity (post) Treatment responsivity change 4.7 10.0 24.2 21.5 2.67 34.2 31.5 4.8 4.4 0.42 9.5 8.5 0.97 6.6 5.7 0.93 1.7a 2.9a,b 7.0a,b 6.8 2.17 8.8a,b 8.6 3.4a,b,d 3.1a,b,d 0.69b 3.0 2.8 1.0 2.0 2.1 0.72 4.8 11.6 27.5 24.9 2.59 39.1 36.5 10.1 9.2 0.95 6.9 6.5 0.37 7.1 6.2 0.93 2.0a 3.3a 7.9 7.9 2.01 10.0 9.8 3.8 3.5 0.89 3.8a,c 3.6a,c 0.55a,c 2.2 2.3 0.70 5.5 14.4 28.2 25.6 2.62 42.6 40.0 7.3 6.7 0.62 10.5 9.6 0.90 7.3 6.5 0.84 2.0 2.3 8.6 8.6 2.32 9.6 9.6 3.8b 3.5b 0.73b 3.7 3.5 0.99 2.4 2.4 0.74 2.1 4.4 21.8 19.7 2.13 26.1 24.0 6.4 5.8 0.61 6.8 6.2 0.56 6.5 5.7 0.80 1.6a,b,c 2.2a,b,c 5.6a,b 6.0 1.67 6.4a,b,c 6.6 3.0b 2.9b 0.63b 3.2a,c 3.0a,c 0.72c 1.9 2.1 0.67 4.4 10.0 24.9 22.4 2.56 34.9 32.4 6.3 5.8 0.57 8.8 8.0 0.79 6.8 5.9 0.90 2.0 4.0 7.5 7.4 2.09 10.0 9.9 4.0 3.7 0.75 3.6 3.3 0.93 2.1 2.2 0.71 35.7 110.6 9.3 8.6 0.88 32.8 31.7 34.1 32.6 7.8 19.0 15.7 7.6 2.2 2.2 0.61 ⬍.001 ⬍.001 ⬍.001 ⬍.001 ns ⬍.001 ⬍.001 ⬍.001 ⬍.001 ⬍.001 ⬍.001 ⬍.001 ⬍.001 ns ns ns Note. All multiple comparisons were computed with Tukey beta post hoc comparisons at p ⬍ .05. Lower score than mixed offenders. b Lower score than child molesters. c Lower score than rapists. d Lower score than incest offenders. a p ⬍ .05) where significant between-groups differences were detected. The most prominent results emerging from the sex-offendertype comparisons were the differences observed in the Sexual Deviance and Criminality mean factor scores for the four groups. Sexual Deviance scores were significantly higher for child molesters than the other three groups. Criminality scores, however, were significantly higher for rapists and mixed offenders than for the child molesters and incest offenders. These results are consistent with what is known about the sexual offending and victim selection characteristics of rapists and child molesters (Blanchard et al., 2006). In contrast, Static 99 scores were the same for the child molesters and rapists. The four groups did not differ on the Treatment Responsivity factor, suggesting that differences in the Sexual Deviance and Criminality factors did not necessarily lead to differences in orientation to treatment. The size of the change scores also appeared to be a function of the size of pretreatment scores, that is, the larger the pretreatment score, the larger the change score. Whereas child molesters had the largest pretreatment Sexual Deviance score and the largest change score for Sexual Deviance, the rapists and mixed offenders had the largest Criminality pretreatment and change scores. There were no differences in total dynamic change scores across the four groups, given that the differential change scores on Sexual Deviance and Criminality would have served to cancel each other out when summed to arrive at a total change score. Overall, the change scores might seem fairly modest at about 2.5 points (approximately a third of a standard deviation), which is not an unexpected finding given the short duration of the program, thus limiting the amount of change. The sample can be classified as a medium-high-risk group by using established Static 99 criteria (Static 99 scores of 4 to 5; Hanson & Thornton, 1999). Among the four groups, mixed offenders had the highest VRS–SO total and Static 99 scores, and incest offenders, the lowest. Interrater Reliability and Internal Consistency The interrater reliability of the VRS–SO dynamic items and the Static 99 was assessed on 35 randomly selected cases and evaluated with intraclass correlation coefficients (ICCs). All reported ICCs (single measure) were significant at p ⬍ .001: Static 99 ICC ⫽ .82; pretreatment dynamic item total ICC ⫽ .74; and posttreatment dynamic item total ICC ⫽ .79. The interrater reliability of the factor-analytically derived factors were as follows: Sexual Deviance (pre) ICC ⫽ .72, (post) ICC ⫽ .73; Criminality (pre) ICC ⫽ .77, (post) ICC ⫽ .80; Treatment Responsivity (pre) ICC ⫽ .66, (post) ICC ⫽ .73.10 Finally, the reliability of the stages of change ratings were evaluated by correlating the dynamic item change scores between the two sets of ratings (r ⫽ .68). The VRS–SO also demonstrated acceptable internal consistency: dynamic items (pretreatment; ␣ ⫽ .81), static items (␣ ⫽ .67), and combined scale total (pretreatment; ␣ ⫽ .84). 10 An interrater reliability coefficient could not be calculated for the VRS–SO static item total, as static scores were computed from numeric data by using SPSS and were not rated by a human rater. As a result, there were no initial static ratings made by an independent coder to compare with those made by a second rater. OLVER, WONG, NICHOLAICHUK, AND GORDON 324 Concurrent Validity The VRS–SO was positively correlated (all ps ⬍ .001) with the Static 99, supporting its concurrent validity: Static total and Static 99 (r ⫽ .70), dynamic total (pre) and (post) with Static 99 (r ⫽ .37 and .35, respectively), and VRS–SO total (pre) and (post) with Static 99 (r ⫽ .55 and .54, respectively). VRS–SO static and dynamic item totals were also significantly correlated (pre-r ⫽ .48, post-r ⫽ .45). This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. Prediction of Sexual Recidivism Recidivism base rates. The sample was followed up for a mean of 10.0 years (SD ⫽ 4.0; range ⫽ 2.0 to 19.0) following release. Overall, 79 offenders (24.6%) were convicted for a new sexual offense, and 115 (35.8%) were convicted for a new nonsexual violent offense. The mean time to first sexual conviction was 3.8 years (SD ⫽ 3.3; range ⫽ 0.12 to 15.7) and to first nonsexual conviction, 2.6 years (SD ⫽ 2.7; range ⫽ 0.01 to 13.7). Predictive accuracy of the VRS–SO and selected risk measures. The VRS–SO static, dynamic, and factor scores, and the Static 99 (for comparison) were significantly correlated with sexual recidivism and had significant AUC values (see Table 3), confirmed by none of the 95% confidence intervals for the AUCs falling below .50. The three factors of the VRS–SO showed interesting differential relationships with sexual and nonsexual violence. Whereas the three VRS–SO factors correlated significantly with sexual recidivism, nonsexual violence was positively correlated with Criminality and negatively correlated with Sexual Deviance. This pattern of strongly opposing correlations of the two factors within the overall dynamic domain of the VRS–SO likely accounts for the weak correlations observed between the dynamic total score and nonsexual violence. The Treatment Responsivity factor predicted both recidivism criteria. Finally, the VRS–SO static items and the Static 99 each significantly predicted sexual and nonsexual violent re- cidivism but demonstrated stronger prediction for sexual recidivism. Incremental contributions of dynamic items to sexual recidivism prediction. The extent to which the dynamic items made unique contributions to the prediction of sexual recidivism after the Static 99 and the static items of the VRS–SO were controlled for was examined through Cox regression survival analysis. The VRS–SO static item total and the dynamic item total were each entered in separate steps, with the criterion variable defined as any new sexual conviction and the time variable defined as time to first sexual conviction or total follow-up time for nonrecidivists. Significant independent contributions were made by the VRS–SO static item total, Wald(1) ⫽ 22.64, p ⬍ .001; and dynamic item total, Wald(1) ⫽ 6.27, p ⫽ .012. Independent contributions were also observed when these analyses were repeated for the Static 99, Wald(1) ⫽ 7.76, p ⬍ .01; and the dynamic items, Wald(1) ⫽ 18.67, p ⬍ .001. Survival analysis. Life tables survival analyses were performed to track the relative sexual recidivism failure rates of four VRS–SO risk groups: low (L) score 0 –20, n ⫽ 39; moderate-low (ML) score 21–30, n ⫽ 110; moderate-high (MH) score 31– 40, n ⫽ 107; and high (H) score 41–72, n ⫽ 65, after offenders had been released into the community and followed up for 10 years (see Figure 1). Significant differences in failure rate were observed among the four risk groups overall, ␹2(3) ⫽ 36.51, p ⬍ .001, and in pairwise comparisons. The H group failed at a higher and faster rate than did the MH, ␹2(1) ⫽ 5.25, p ⬍ .05; ML, ␹2(1) ⫽ 25.06, p ⬍ .001; and L, ␹2(1) ⫽ 18.87, p ⬍ .001, groups. The MH group had higher failure rates compared with the ML, ␹2(1) ⫽ 9.70, p ⬍ .01; and L, ␹2(1) ⫽ 10.24, p ⬍ .01, groups, and no significant difference was observed between the ML and L groups, ␹2(1) ⫽ 2.60, p ⫽ .107. Despite having VRS–SO scores of up to 20, the L group failed at an extremely low rate even up to 10 years. Percent sexual recidivism and risk level. The percentage of individuals who sexually recidivated was computed for each of the four VRS–SO risk groups or “bins” over fixed 3-, 5-, and 10-years Table 3 Predictive Accuracy of the VRS–SO Scale Components (Pre- and Posttreatment) and Static 99 for Sexual and Nonsexual Violent Recidivism Sexual recidivism Measure Static 99 VRS–SO static Dynamic (pre) Dynamic (post) Pretreatment total Posttreatment total Sexual deviance (pre) Sexual deviance (post) Criminality (pre) Criminality (post) Treatment responsivity (pre) Treatment responsivity (post) r *** .21 .36*** .23*** .26*** .32*** .34*** .14* .16** .17** .21*** .12* .15** AUC ** .63 .74*** .66*** .67*** .71*** .72*** .59* .61** .63*** .65*** .58* .59** Nonsexual violent recidivism 95% Cl .56, .68, .59, .60, .64, .66, .52, .54, .56, .58, .51, .52, .70 .80 .73 .74 .77 .78 .66 .68 .70 .72 .65 .66 r AUC * .13 .17** .03 .06 .09 .11* ⫺.26*** ⫺.25*** .26*** .28*** .15* .15** * .57 .60** .53 .55 .56 .57* .35*** .35*** .65*** .67*** .59** .60** Note. N ⫽ 321. VRS–SO ⫽ Violence Risk Scale—Sexual Offender version; AUC ⫽ area under the curve statistic; CI ⫽ confidence interval. p ⬍ .05. ** p ⬍ .01. *** p ⬍ .001. * 95% Cl .51, .53, .46, .48, .50, .51, .29, .29, .59, .61, .53, .53, .64 .66 .59 .61 .62 .64 .41 .41 .71 .73 .65 .65 SEX OFFENDER RISK ASSESSMENT 325 1.1 Cumulative proportion surviving This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 1.0 .9 .8 .7 .6 .5 VRS–SO Risk Level .4 High (41+) .3 Mod-High (31–40) .2 Mod-Low (21–30) .1 Low (0–20) 0 2 4 6 8 10 12 Survival time (years) to first sex offense Figure 1. Survival analysis. Cumulative sexual recidivism failure rates as a function of Violence Risk Scale—Sexual Offender version (VRS–SO) risk level over a 10-year follow-up. Mod ⫽ moderate. follow-up (Figure 2).11 VRS–SO risk groups (L, ML, MH, and H) were found to be significantly associated with sexual recidivism at each follow-up window: 3 years, ␹2(3) ⫽ 33.85, ␸ ⫽ .33, p ⬍ .001; 5 years, ␹2(3) ⫽ 34.31, ␸ ⫽ .34, p ⬍ .001; and 10 years, ␹2(3) ⫽ 36.31, ␸ ⫽ .44, p ⬍ .001. Evaluating the Relationship of Change to Reductions in Sexual Recidivism Aggregate change scores for each participant were computed by summing the individual item change scores for all 17 dynamic items. Total change scores ranged from –2.50 to ⫹10.13 points of change (M ⫽ 2.56, SD ⫽ 2.09), and for Sexual Deviance, – 0.50 to 3.50 (M ⫽ 0.57, SD ⫽ 0.75), Criminality, –1.0 to 6.50 (M ⫽ 0.78, SD ⫽ 0.93), and Treatment Responsivity, –1.50 to 3.03 (M ⫽ 0.90, SD ⫽ 0.71).12 Cox regression survival analyses were conducted to examine the relationship of therapeutic change to sexual recidivism while risk and follow-up time were controlled for. In the first analysis, the VRS–SO static score and total dynamic change score were entered as covariates, and both variables demonstrated significant unique relationships to sexual recidivism: VRS–SO static score, Wald(1) ⫽ 41.33, p ⬍ .001, exp(B) ⫽ 1.230; total dynamic change, Wald(1) ⫽ 3.94, p ⬍ .05, exp(B) ⫽ 0.896. In the second analysis, the VRS–SO pretreatment total score (i.e., combined static and dynamic to further control for risk) and total dynamic change score were entered as covariates. Both variables also demonstrated significant relationships to sexual recidivism and in the expected direction: VRS–SO pretreatment total, Wald(1) ⫽ 42.99, p ⬍ .001, exp(B) ⫽ 1.070; total dynamic change, Wald(1) ⫽ 4.42, p ⬍ .05, exp(B) ⫽ 0.900. Exp(B) is an odds ratio statistic representing the predicted change in hazard (e.g., risk for recidivism) per unit increase in the predictor (see Fox, 2002; Tabachnick & Fidell, 2001). Exp(B) values exceeding 1 indicate that higher scores on a measure are associated with increased recidivism, and values less than 1, with decreased recidivism. An exp(B) of 0.900, for instance, would be interpreted to mean that for every 1-point increase in change score (i.e., a unit of positive change), there would be a predicted 10% decrease in sexual recidivism after accounting for risk. In sum, whereas scores on the risk measures maintained a significant association with increased recidivism, therapeutic change scores made significant and unique contributions to sexual recidivism prediction. In other words, reductions in VRS–SO dynamic scores, an indication of improvement during treatment, were found to be related to reductions in sexual recidivism after measurements of risk and follow-up time were controlled for. In the present sample of sex offenders, a seemingly modest average change score of approximately 2.5 (see Table 2) could be extrapolated to mean a predicted 25% reduction in sexual recidivism with treatment after accounting for risk. We further examined the relationship of therapeutic change to sexual recidivism as a function of risk by dividing the offender sample into broad high- (n ⫽ 204) and low- (n ⫽ 117) risk groups 11 See Figure 2 note for ns within each bin. As a result of using the regression procedures (described in the Procedure subsection) to estimate, or prorate, missing data, some of the reported values are not whole numbers or in denominations of 0.5. Negative change scores indicate the offender deteriorated in treatment. 12 OLVER, WONG, NICHOLAICHUK, AND GORDON 326 Low (0–20) Mod-Low (21–30) Mod-High (31–40) High (41–72) 80 70 Percent sexual reconviction This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 70 60 50 50 41.9 40 34.3 30 24.2 20 17 20 8.1 10 2.6 4.6 6.7 3 0 3 years 5 years 10 years Figure 2. Base rates of sexual recidivism at derived Violence Risk Scale—Sexual Offender version (VRS–SO) risk categories over fixed 3-, 5-, and 10-year follow-up periods. Group bin sizes for each follow-up period were as follows: 3 years, N ⫽ 317, low n ⫽ 38, moderate (mod)-low n ⫽ 109, moderate-high n ⫽ 106, high n ⫽ 64; 5 years, N ⫽ 293, low n ⫽ 33, moderate-low n ⫽ 99, moderate-high n ⫽ 99, high n ⫽ 62; 10 years, N ⫽ 187, low n ⫽ 15, moderate-low n ⫽ 70, moderate-high n ⫽ 62, high n ⫽ 40. on the basis of their Static 99 scores (MH- and H-risk groups were collapsed into a single high-risk group, and ML- and L-risk groups were collapsed into a single low-risk group). The distributions of therapeutic change scores for each group were broadly normal and fell within acceptable limits of skewness and kurtosis. Cox regression analyses were performed separately for each group. Within the high-risk group, significant independent relationships were observed for total dynamic change, Wald(1) ⫽ 4.28, p ⬍ .05, exp(B) ⫽ 0.882; and the VRS–SO static items, Wald(1) ⫽ 21.23, p ⬍ .001, exp(B) ⫽ 1.215. However, within the low-risk group, no relationship was observed for total dynamic change, Wald(1) ⫽ 0.005, ns, exp(B) ⫽ 1.01; although the static items continued to predict sexual recidivism, Wald(1) ⫽ 11.11, p ⫽ .001, exp(B) ⫽ 1.31. As dichotomizing the groups by risk level is a crude means of controlling for risk, it is worth noting that consistent with the above Cox regression analyses of the high- and low-risk groups, change scores were significantly negatively correlated with sexual recidivism in the high-risk group (r ⫽ –.15, p ⬍ .05) but not the low-risk group (r ⫽ .01, ns) nor with the entire group (r ⫽ –.09, p ⫽ .10). The results demonstrate therapeutic change to be more predictive (and hence, likely more informative) among higher risk sex offenders. Discussion The VRS–SO was designed to integrate within a single instrument the assessment of sex offender risk, the identification of dynamic variables linked to sex offending as treatment targets, and the evaluation an offender’s treatment readiness and change; in short, to develop a sex offender risk assessment tool that can also inform and guide sex offender treatment. The present study investigated the psychometric properties of the VRS–SO in a sample of treated sex offenders, including its reliability, concurrent and predictive validity, and the relationship of changes in the dynamic variables to changes in sexual recidivism. Psychometric Properties of the VRS–SO A factor analysis of the dynamic items suggested that the dynamic component of the scale (17 of 24 variables) can be summarized by three factors labeled Sexual Deviance, Criminality, and Treatment Responsivity. Two of the three factors extracted are consistent with the two major sex offender risk factor domains (i.e., sexual deviance, antisocial orientation) that have been demonstrated to predict sexual recidivism in the empirical literature (e.g., Doren, 2004; Hanson & Morton-Bourgon, 2004). Specifically, Sexual Deviance taps a constellation of variables that include deviant sexual interests, lifestyle, and preoccupations. Criminality comprises variables that reflect a generalized antisocial lifestyle. The Treatment Responsivity factor reflects distorted attitudes and beliefs supportive of sexual offending and resistance to change and would likely entail a lack of compliance with change agents such as sex offender treatment. Of note is that scores on the Sexual Deviance and Criminality factors indicated differences among sex offender types. Consistent with empirical findings (e.g., Pithers, 1993), rapists and mixed offenders tended to be a more criminalized group of offenders, scoring highest on the Criminality factor, whereas child molesters This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. SEX OFFENDER RISK ASSESSMENT were more likely to have deviant preferences (e.g., Blanchard et al., 2006) and scored highest of the four groups on the Sexual Deviance factor. Clinically, assessing sex offenders with the VRS–SO can provide a criminogenic profile based on scores on the three dynamic factor domains that can provide differential information to inform sex offender treatment and case conceptualization. As a result, treatment resources, usually in short supply, can be better focused onto the most salient problem areas as indicated by the broad factor domains and then onto specific treatment targets, as indicated by the individual dynamic items, to increase treatment efficacy. The VRS–SO has acceptable interrater reliability (.66 to .80), even when raters were required to rate fairly difficult dynamic items from file information. Acceptable internal consistency was also obtained for the dynamic item total (␣ ⫽ .81) and the three dynamic factor domains (␣s ranging from .72 to .87). Concurrent and Predictive Validity VRS–SO static, dynamic, total, and factor scores each significantly predicted sexual recidivism. VRS–SO total scores also predicted sexual recidivism over short-, medium-, and long-term follow-ups. The relationship of the different VRS–SO factors to nonsexual violent recidivism was also consistent with what one would expect from a perusal of the literature (Hanson & MortonBourgon, 2004). Although Criminality was significantly positively correlated with nonsexual violent recidivism, Sexual Deviance was significantly negatively correlated with this outcome. The concurrent validity of the VRS–SO was examined with respect to the Static 99, a widely used static actuarial risk measure developed and validated on a large international sample of sex offenders (Hanson & Thornton, 1999). The VRS–SO static items had a substantially higher correlation (r ⫽ .70) with the Static 99 than did the VRS–SO dynamic items (r ⫽ .37), suggesting that there is limited overlap between the dynamic items and the Static 99 and raising the possibility that the dynamic items may capture sexual recidivism variance not explained by the Static 99. This possibility was tested with Cox regression, and the results demonstrated that the total dynamic score made significant incremental contributions to predicting sexual recidivism over and above that of the Static 99. The dynamic items also showed incremental validity for sexual recidivism prediction beyond the VRS–SO static items. Are the VRS–SO Dynamic Variables Indeed Dynamic? To evaluate whether the dynamic items are indeed dynamic, we examined the relationship of changes in the VRS–SO dynamic items to changes in sexual recidivism. The results supported the contention that positive changes in the dynamic items, indicating a reduction in risk, are related to reductions in sexual recidivism. Change scores suggested a significant unique inverse relationship with sexual recidivism after statistically controlling for actuarial risk by using Cox regression survival analysis. The results can be interpreted in practical terms to mean for every 1-point increase in change score (i.e., one unit of positive change), there would be a predicted 10% decrease in sexual recidivism after accounting for risk. It follows that, in the present sample, with an average total change score of approximately 2.5, there would be a predicted 327 25% overall reduction in the probability of sexual recidivism with treatment after accounting for risk. This is not a trivial effect in terms of reducing sexual revictimization, and the magnitude of this anticipated reduction is broadly consistent with meta-analytic sex offender treatment outcome findings (Hanson et al., 2002) as well as prior treatment outcome research on a similar Clearwater sample (Nicholaichuk, Gordon, Gu, & Wong, 2000). The present results are consistent with the notion that the VRS–SO dynamic items are dynamic and that the dynamic items and their change scores are potentially informative of treatment change and risk reduction. However, a cross-validation of the present findings on a different sample of treated sex offenders is necessary to evaluate the generalizability of these results. Not all dynamic items, however, are equally dynamic. Some items, such as sexually deviant lifestyle, may be more resistant to change than is an item such as cognitive distortions. Given the substantial correlation (r ⫽ .48) between the static and dynamic items, it could be argued that the VRS–SO dynamic and static items share considerable common variance, and therefore, some dynamic items may be more like static items. We would argue that the static and dynamic items are more like two sides of the same coin. For instance, an individual with many dynamic items endorsed is expected to be more criminalized and sexually deviant and, therefore, more likely to transgress social and legal boundaries and run afoul of the law, the consequences of which (e.g., sexual assault) are captured by the static items. As such, the static and dynamic items would be expected to be moderately or even quite highly correlated. Whether one focuses on the functional characteristics of the individual (the dynamic items) or the lifetime outcomes of his or her antisocial behaviors (static items) is the choice one makes in looking at one or the other side of the same coin. Treatment providers would be well advised to pay attention to dynamic items to capture clinically relevant information about the client presenting for treatment. The results also demonstrated that dynamic item change scores are more closely linked to sexual recidivism in higher risk than lower risk offenders, as found when change analyses were conducted separately in the two risk groups through Cox regression. The results are not unexpected and suggest that not only is there more room to change for higher risk offenders, but changes made in treatment (or lack thereof) are potentially more informative in sexual recidivism risk assessment for this group of offenders. In conclusion, the results of the present study provide preliminary empirical support for the use of the VRS–SO as a sex offender risk assessment and prediction tool that can also be useful for treatment purposes. The tool uses dynamic items to inform treatment planning and delivery and to measure changes in risk that, as the data suggest, are linked to changes in sexual recidivism. It is our hope that continued efforts to validate the use of dynamic variables to assess risk and evaluate therapeutic change will lead to improvements in the assessment and treatment of sex offenders and, ultimately, the reduction of sexual victimization. Limitations and Future Directions Although the present study has generated some promising findings to support the validity and reliability of the VRS–SO, there are limitations that may be redressed through future research. First, the static items were developed and cross-validated on separate OLVER, WONG, NICHOLAICHUK, AND GORDON This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. 328 halves of the sample, but the entire sample was used in all prediction analyses, which could result in inflating the predictive validity correlations of the static items. Yet, on the other hand, a more conservative estimate of the relationship of change to sexual recidivism may have been obtained as a result of statistically controlling for a static measure with potentially inflated prediction. Second, the Static 99 was the only sex offender risk assessment tool used as a comparison measure with the VRS–SO in the current study. Future research could compare the VRS–SO against other sex offender risk measures. Third, as data were collected by using only file information, a replication of the study with both file and interview is warranted. Conducting interviews, in turn, should increase the volume of clinical information available for making VRS–SO ratings and thus potentially enhance the validity and reliability of the assessment. As this is the first major study investigating the psychometric properties of the VRS–SO, future research efforts would be well served to cross-validate the tool prospectively on a different sample of sex offenders to increase the generalizability of the current findings. Of particular interest may be further research examining the relationship of changes in the dynamic items to reductions in sexual recidivism and how therapeutic movement on these variables can inform supervision, risk management, and decision making with sex offenders. References Andrews, D. A., & Bonta, J. (2003). The psychology of criminal conduct (3rd ed.). Cincinnati, OH: Anderson. Beech, A., Friendship, C., Erikson, M., & Hanson, R. K. (2002). The relationship between static and dynamic factors and reconviction in a sample of U.K. child abusers. Sexual Abuse: A Journal of Research and Treatment, 14, 155–167. Blanchard, R., Kuban, M. E., Blak, T., Cantor, J. M., Klassen, P., & Dickey, R. (2006). Phallometric comparison of pedophilic interest in nonadmitting sexual offenders against stepdaughters, biological daughters, other biologically related girls, and unrelated girls. Sexual Abuse: A Journal of Research and Treatment, 18, 1–14. 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, British Columbia, Canada: Institute against Family Violence and the Mental Health, Law, and Policy Institute, Simon Fraser University. Browne, A., & Finkelhor, D. (1986). Impact of sexual abuse: A review of the research. Psychological Bulletin, 99, 66 –77. Doren, D. M. (2004). Towards a multidimensional model for sexual recidivism risk. Journal of Interpersonal Violence, 19, 835– 856. 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. Douglas, K. S., & Skeem, J. (2005). Violence risk assessment: Getting specific about being dynamic. Psychology, Public Policy, and Law, 11, 347–383. Epperson, D. L., Kaul, J. D., & Hesselton, D. (1998). Minnesota Sex Offender Screening Tool—Revised (MnSOST–R): Development, performance, and recommended risk level cut scores. St. Paul, MN: Iowa State University and Minnesota Department of Corrections. Fox, J. (2002). Cox proportional-hazards regression for survival data. In J. Fox (Ed.), An R and S-PLUS companion to applied regression (pp. 1–18). Thousand Oaks, CA: Sage. Hanson, R. K. (1997). The development of a brief actuarial scale for sexual offense recidivism (User Report 97– 04). Ottawa, Ontario, Canada: Department of the Solicitor General of Canada. Hanson, R. K. (2005, June). The assessment of criminogenic needs of sexual offenders by community supervision officers: Reliability and validity. Symposium conducted at the 66th annual meeting of the Canadian Psychological Association, Montreal, Quebec, Canada. Hanson, R. K. (2006). Does Static 99 predict recidivism among older sexual offenders? Sexual Abuse: A Journal of Research and Treatment, 18, 343–355. Hanson, R. K., & Bussière, M. T. (1998). Predicting relapse: A metaanalysis of sexual offender recidivism studies. Journal of Consulting and Clinical Psychology, 66, 348 –362. Hanson, R. K., Gordon, A., Harris, A. J. R., Marques, J. K., Murphy, W., Quinsey, V. L., & Seto, M. C. (2002). First report of the Collaborative Data Outcome Project on the effectiveness of psychological treatment for sexual offenders. Sexual Abuse: A Journal of Research and Treatment, 14, 169 –194. Hanson, R. K., & Harris, A. J. R. (2000). Where should we intervene? Dynamic predictors of sexual offense recidivism. Criminal Justice and Behavior, 27, 6 –35. Hanson, R. K., & Harris, A. J. R. (2001). A structured approach to evaluating change among sexual offenders. Sexual Abuse: A Journal of Research and Treatment, 13, 105–122. Hanson, R. K., & Morton-Bourgon, K. (2004). Predictors of sexual recidivism: An updated meta-analysis (User Report 2004 – 02). Ottawa, Ontario, Canada: Public Safety and Emergency Preparedness Canada. Hanson, R. K., & Thornton, D. (1999). Static 99: Improving actuarial risk assessments for sex offenders (User Report 99 – 02). Ottawa, Ontario, Canada: Department of the Solicitor General of Canada. Harris, G. T., Rice, M. E., Quinsey, V. L., Lalumière, M. L., Boer, D., & Lang, C. (2003). A multi-site comparison of actuarial risk instruments for sex offenders. Psychological Assessment, 15, 413– 425. Hart, S. D., Kropp, P. R., & Laws, D. R. (2003). The Risk for Sexual Violence Protocol (RSVP)—Structured professional guidelines for assessing risk of sexual violence. Burnaby, British Columbia, Canada: Mental Health, Law and Policy Institute, Simon Fraser University. Hudson, S. M., Wales, D. S., Bakker, L., & Ward, T. (2002). Dynamic risk factors: The Kia Marama evaluation. Sexual Abuse: A Journal of Research and Treatment, 14, 103–119. Johnson, H., & Sacco, V. F. (1995). Researching violence against women: Statistics Canada’s national survey. Canadian Journal of Criminology, 37, 281–304. 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. Levesque, D. A., Gelles, R. J., & Velicer, W. F. (2000). Development and validation of a stages of change measure for men in batterer treatment. Cognitive Therapy and Research, 24, 175–199. Lisak, D., & Miller, P. M. (2002). Repeat rape and multiple offending among undetected rapists. Violence and Victims, 17, 73– 84. McGrath, R. J., & Hoke, S. E. (2002). Vermont Assessment of Sex Offender Risk manual: Research edition 2001. Waterbury, Vermont: Author. Nicholaichuk, T. P., Gordon, A., Gu, D., & Wong, S. (2000). Outcome of an institutional sexual offender treatment program: A comparison between treated and matched untreated offenders. Sexual Abuse: A Journal of Research and Treatment, 12, 139 –153. Pithers, W. D. (1990). Relapse prevention for sexual aggressors: A method for maintaining therapeutic gain and enhancing external supervision. In W. L. Marshall, D. R. Laws, and H. E. Barbaree (Eds.), Handbook of sexual assault: Issues, theories, and treatment of the offender (pp. 343–361). New York: Plenum Press. Pithers, W. D. (1993). Treatment of rapists: Reinterpretation of early outcome data and exploratory constructs to enhance therapeutic efficacy. In G. C. N. Hall, R. Hirschman, J. R. Graham, & M. S. Zaragoza (Eds.), Sexual aggression: Issues in etiology, assessment, and treatment (pp. 167–196). Philadelphia, PA: Taylor & Francis. This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly. SEX OFFENDER RISK ASSESSMENT 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. Proulx, J., Pellerin, B., Paradis, Y., McKibben, A., Aubut, J., & Ouimet, M. (1997). Static and dynamic predictors of recidivism in sexual aggressors. Sexual Abuse: A Journal of Research and Treatment, 9, 7–27. Quinsey, V. L., Rice, M. E., & Harris, G. T. (1995). Actuarial prediction of sexual recidivism. Journal of Interpersonal Violence, 10, 85–105. Rice, M. E., & Harris, G. T. (1995). Violent recidivism: Assessing predictive validity. Journal of Consulting and Clinical Psychology, 63, 737–748. Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate statistics. Boston, MA: Allyn and Bacon. Thornton, D. (2002). Constructing and testing a framework for dynamic risk assessment. Sexual Abuse: A Journal of Research and Treatment, 14, 139 –153. Tierney, D. W., & McCabe, M. P. (2005). The utility of the trans- 329 theoretical model of behavior change in the treatment of sex offenders. Sexual Abuse: A Journal of Research and Treatment, 17, 153–170. Ward, T., & Hudson, S. M. (1998). A model of the relapse process in sex offenders. Journal of Interpersonal Violence, 13, 700 –725. Willoughby, T., & Perry, G. P. (2002). Working with violent youth: Application of the transtheoretical model of change. Canadian Journal of Counseling, 36, 312–326. Wong, S., & Gordon, A. E. (2000). The Violence Risk Scale. Unpublished manuscript. Wong, S. C. P., & Gordon, A. E. (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. Wong, S., Olver, M. E., Nicholaichuk, T. P., & Gordon, A. (2003). The Violence Risk Scale—Sexual Offender version (VRS–SO). Saskatoon, Saskatchewan, Canada: Regional Psychiatric Centre and University of Saskatchewan. Appendix VRS–SO Static and Dynamic Items and Brief Item Descriptions Static Items S1. Age at release: ⬍ age 25 years; 25–34 years; 35– 44 years; 45 years and up S2. Age at first sex offense: ⬍ age 20 years; 20 –24 years; 25–34 years; 35 years and up S3. Sex offender type: mixed offender; child molester; rapist; incest offender S4. Prior sex offenses: 4 or more prior sexual charges/ convictions; 2–3 prior; 1 prior; 0 prior S5. Unrelated victims: 4 or more unrelated victims; 2–3 unrelated; 1 unrelated; 0 unrelated (all related) S6. Victim gender: 2⫹ male victims; 1 male and 1 female/or 2⫹ female; 1 male victim only; 1 female victim only S7. Prior sentencing dates: 11⫹ prior sentencing dates; 5–10 prior; 2– 4 prior; 0 –1 prior Dynamic Items D1. Sexually deviant lifestyle: lifestyle hobbies, interests, work, or relationships involve sexually deviant behaviors D2. Sexual compulsivity: strong sex drive and high frequency of sexual behavior and cognitions D3. Offense planning: victim grooming and premeditation involved in sexual offending D4. Criminal personality: interpersonal and emotional attributes conducive to criminal behavior (e.g., lack of remorse) D5. Cognitive distortions: attitudes and distorted thinking supportive of sexual offending D6. Interpersonal aggression: physically and/or verbally aggressive behavior in interpersonal interactions D7. Emotional control: tendency to overcontrol or undercontrol emotions linked to sexual offending D8. Insight: poor understanding of causes of sexual offending and unwillingness to discuss/explore sexual offending D9. Substance abuse: substance use problems linked specifically to sexual offending D10. Community support: lack of positive support people, services, or plans in community (or unwilling to use) D11. Released to high-risk situations: offender seems likely or has shown pattern of returning to situations linked to sex offending D12. Sexual offending cycle: pattern of interpersonal, situational, and personal factors linked to sexual offending D13. Impulsivity: behavior displays tendency to “act first, think later” and lacks reflection or forethought D14. Compliance with community supervision: poor attitude and/or cooperation with community supervision D15. Treatment compliance: poor attitude and/or cooperation with sex offender treatment D16. Deviant sexual preference: interests or preferences for deviant sexual stimuli or behaviors (e.g., children, violence) D17. Intimacy deficits: incapacity to form or maintain adult romantic relationships Note. All items are rated on a 4-point (3, 2, 1, 0) scale. Item descriptions are abbreviated examples of the originals and are not intended to be used for clinical or research purposes. Please consult the VRS–SO rating manual (Wong, Olver, Nicholaichuk, & Gordon, 2003) for more detailed item descriptions, stages of change ratings, and scoring instructions. Received December 19, 2006 Revision received June 5, 2007 Accepted June 6, 2007 䡲