Abstract
The Juvenile Sex Offender Assessment Protocol–Revised (J-SOAP-II) is the most commonly used measure in the assessment of recidivism risk among juveniles who have committed sexual offenses (JSOs), but mixed support exists for its predictive validity. This study compared the predictive validity of the J-SOAP-II across two offender characteristics, age and sexual drive, in a sample of 156 JSOs who had been discharged from a correctional facility or a residential treatment program. The J-SOAP-II appeared to be a better predictor of sexual recidivism for younger JSOs (14-16 years old) than for older ones (17-19 years old), with significant differences found for the Dynamic Summary Scale and Scale III (Intervention). In addition, several of the measure’s scales significantly predicted sexual recidivism for JSOs with a clear pattern of sexualized behavior but not for those without such a pattern, indicating that the J-SOAP-II may have greater clinical utility for JSOs with heightened sexual drive. The implications of these findings are discussed.
Juveniles who have committed sexual offenses (JSOs) pose a risk to community safety, accounting for 14.3% of arrests for forcible rape and 17.9% of arrests for other sexual offenses (Office of Juvenile Justice and Delinquency Prevention, 2013). Although meta-analytic findings suggest that JSOs sexually reoffend at a relatively low rate (approximately 7%; Caldwell, 2010), public concern about JSOs has intensified over the past two decades, and the legal consequences for them have become more severe (Caldwell, 2007). In many states, JSOs may be involuntarily committed or required to register in public sex offender databases (Caldwell, Ziemke, & Vitacco, 2008). Given the gravity of these consequences, clinicians often are called upon to assess JSOs’ risk of recidivism before placement and treatment decisions are made (Hempel, Buck, Cima, & van Marle, 2013).
Several measures have been developed to assist in the assessment of JSO recidivism risk (Caldwell et al., 2008), the most widely used of which is the Juvenile Sex Offender Assessment Protocol–Revised (J-SOAP-II; Murrie, 2012; Prentky & Righthand, 2003). The J-SOAP-II is designed for use with male JSOs who have a history of a sex offense conviction or sexually coercive behavior. The instrument is comprised of 28 static and dynamic risk factors for recidivism, each of which is scored on a 0- to 2-point scale. These risk factors are grouped into four scales, Sexual Drive/Preoccupation (Scale I), Impulsive/Antisocial Behavior (Scale II), Intervention (Scale III), and Community Stability/Adjustment (Scale IV), with higher scores typically reflecting greater risk. These scales are grouped into two summary scales, Static (Scales I and II) and Dynamic (Scales III and IV), and a total score encompassing all 28 items. Although the J-SOAP-II yields total, scale, and summary scale scores, no cutoff scores have been established for classifying youth according to risk level (e.g., low, moderate, high). Rather, clinicians typically use J-SOAP-II scores to gauge risk of recidivism and inform decisions regarding treatment and placement.
To date, 15 studies have examined the predictive validity of the J-SOAP-II, yielding mixed results (Aebi, Plattner, Steinhausen, & Bessler, 2011; Caldwell & Dickinson, 2009; Caldwell et al., 2008; Chu, Ng, Fong, & Teoh, 2012; Faniff & Letourneau, 2012; Martinez, Flores, & Rosenfeld, 2007; Martinez, Rosenfeld, Cruise, & Martin, 2015; McCoy, 2007; Parks & Bard, 2006; Petersen, 2011; Powers-Sawyer & Miner, 2009; Prentky et al., 2010; Ralston & Epperson, 2013; Rajlic & Gretton, 2010; Viljoen et al., 2008). Some have found the total score to be a significant predictor of both sexual and non-sexual recidivism. In a sample of male JSOs who had been enrolled in a community sex offender treatment program, Martinez and colleagues (2007) found the total score to be a significant, albeit moderate, predictor of both sexual and non-sexual recidivism, area under the curve (AUC) = .78 and .76, respectively. Other studies utilizing community samples of male JSOs have generated similar results (Prentky et al., 2010; Rajlic & Gretton, 2010). However, some have found that the total score predicts only non-sexual recidivism. In a sample of male JSOs who had been on probation or placed in secure juvenile correctional facilities, Chu et al. (2012) found that the total score was a significant predictor of non-sexual recidivism, AUC = .79, but performed little better than chance in predicting sexual recidivism, AUC = .51. Using a similarly mixed sample of male JSOs, Aebi et al. (2011) also found that the total score significantly predicted only non-sexual recidivism. However, they found a considerably weaker association between the total score and non-sexual recidivism (AUC = .61) than did Chu et al. (2012), suggesting that the J-SOAP-II may not have clinical utility for the prediction of non-sexual recidivism risk. Indeed, other studies have found the total score to be a poor predictor of both sexual and non-sexual recidivism (Caldwell et al., 2008; Faniff & Letourneau, 2012; McCoy, 2007; Parks & Bard, 2006; Viljoen et al., 2008). Support for the J-SOAP-II scales is similarly equivocal. Some studies have found only Scale I (Sexual Drive/Preoccupation) to predict sexual recidivism (Chu et al., 2012; Viljoen et al., 2008), whereas others have found Scales II (Impulsive/Antisocial Behavior) and IV (Community Stability/Adjustment) to do so as well (Aebi et al., 2011; Petersen, 2011; Prentky et al., 2010). In short, whether the J-SOAP-II accurately predicts recidivism among JSOs remains unclear.
This inconsistency in the JSO risk assessment literature is not limited to the J-SOAP-II. A recent meta-analysis examined the predictive validity of the J-SOAP-II and two other commonly used JSO-specific risk instruments, the Estimate of Risk of Adolescent Sexual Offense Recidivism (ERASOR; Worling & Curwen, 2001) and the Juvenile Sexual Offense Recidivism Risk Assessment Tool-II (J-SORRAT-II; Epperson, Ralston, Fowers, DeWitt, & Gore, 2006; Viljoen, Mordell, & Beneteau, 2012). The authors found no significant differences between the predictive validity of total scores on these instruments, with AUC estimates ranging from .64 to .70 for sexual recidivism. This finding suggests that the J-SOAP-II, like other JSO-specific risk assessment instruments, has moderate predictive validity for sexual recidivism overall. However, the authors also found significant heterogeneity in AUC estimates for all three instruments across studies. Although they examined whether a number of variables (i.e., study setting, treatment status, publication bias, author allegiance, geographic location, sample size, sexual recidivism rate, and instrument interrater reliability) accounted for this heterogeneity, they found that none of these variables significantly moderated the predictive validity of the instruments studied. Thus, the cause of the inconsistency in the J-SOAP-II literature, and in the JSO risk assessment literature more generally, remains unclear.
The heterogeneity of JSOs themselves may account for these inconsistent findings (Rajlic & Gretton, 2010). One offender characteristic that could impact the predictive validity of the J-SOAP-II and other JSO-specific risk assessment instruments is age. Given the rapid cognitive, psychosocial, and sexual development that occurs during adolescence, a number of authors have suggested that there may be developmental differences in risk factors for recidivism among juvenile offenders (Caldwell, 2002; Viljoen, Elkovitch, Scalora, & Ullman, 2009; Vincent, Perrault, Guy, & Gershenson, 2012). Consequently, it is possible that a JSO’s age at the time of an evaluation may impact the predictive validity of the risk assessment instrument used, particularly if that instrument includes dynamic risk factors. Dynamic risk factors, by definition, change over time (Vincent et al., 2012). However, they may be more stable during certain periods of adolescence than others, and thus, may serve as more accurate predictors of recidivism for some age groups of JSOs than others (Ralston & Epperson, 2013). Yet very few studies have examined whether age moderates the predictive validity of JSO risk assessment instruments. Viljoen et al. (2008) examined whether age at assessment moderated the predictive validity of the J-SOAP-II in a sample of male JSOs who had completed a residential sex offender treatment program. They found that the total score significantly predicted sexual and non-sexual aggression during treatment for JSOs 16 to 18 years old at admission, but not for those 12 to 15 years old. They also found that the total score significantly predicted violent non-sexual recidivism for youth 16 to 18 years old at discharge, but not for those 12 to 15 years old. These findings are consistent with those of a subsequent study utilizing the same sample, which found that the ERASOR total score was a stronger predictor of non-sexual recidivism among JSOs 16 to 18 years old at discharge (the time of assessment) than among those 12 to 15 years old (Viljoen et al., 2009). These results suggest that the J-SOAP-II and other JSO risk assessment instruments may have greater clinical utility for older JSOs than for younger JSOs.
Ralston and Epperson (2013) also examined whether age moderated the predictive validity of the J-SOAP-II, using a sample of male JSOs who had been adjudicated delinquent and received a variety of juvenile justice system services (e.g., secure placement, treatment, probation). Rather than examining age at assessment, they examined age at reoffense, finding that Scales I and II significantly predicted sexual recidivism occurring before the age of 18 but not after 18. This finding indicates that the J-SOAP-II may have greater clinical utility for younger JSOs. However, when they examined the moderating effect of age at index offense, they found no age differences in the predictive accuracy of the J-SOAP-II for recidivism occurring before age 18, suggesting that the predictive validity of the J-SOAP-II may not vary with offender age. As the authors noted, however, their failure to account for time at risk may have confounded their findings, as some youth in the sample may have had fewer opportunities to recidivate by virtue of being in a secure placement during the follow-up period. Moreover, as they only analyzed J-SOAP-II scales that measure static risk factors for recidivism, it is unclear whether the same pattern would have emerged had they analyzed the J-SOAP-II scales that measure dynamic risk factors. It is also unclear whether age differences would have emerged had they measured age at assessment rather than at index offense and reoffense. In short, the relationship between offender age at the time of assessment and the predictive validity of the J-SOAP-II remains uncertain.
A second offender characteristic that may impact the predictive validity of JSO risk assessment instruments is sexual drive. Some scholars suggest that heightened sexual drive and preoccupation may serve as the link between social, emotional, and neuropsychological problems and sexual offending. For example, Ward and Beech (2005) posit that emotion dysregulation may lead to sexual offending for youth with high levels of sexual drive, as these youth are more likely to engage in sexual behavior to cope with negative emotional states. Similarly, Cortoni and Marshall (2001) suggest that relationship difficulties during adolescence may lead to sexual offending because youth who fear rejection from peer-aged partners may become preoccupied with impersonal, non-affectionate sexual activity (e.g., use of pornography, masturbation), eventually using sex as a coping mechanism. Moreover, research indicates that sexual drive and preoccupation are associated with sexual recidivism among JSOs (Worling, Bookalam, & Litteljohn, 2012). Consequently, it is possible that sexual drive may moderate the predictive validity of JSO risk assessment instruments. Youth who score highly on these instruments because they have a number of social and emotional risk factors for recidivism may be more likely to reoffend if they also have heightened sexual drive. However, no research to date has examined this potential moderating effect.
In sum, despite the inconsistent findings regarding the predictive validity of the J-SOAP-II, a growing body of research suggests that the instrument may have greater clinical utility for some groups or types of JSOs than for others. In particular, there is evidence suggesting that age moderates the predictive validity of the J-SOAP-II, although it remains unclear whether the instrument has greater clinical utility for older or younger JSOs. Moreover, it is possible that the J-SOAP-II may have greater clinical utility for JSOs with high levels of sexual drive. However, very few studies have examined whether these and other offender characteristics impact the predictive validity of the J-SOAP-II.
The present study sought to extend the JSO risk assessment literature by examining the moderating effect of age and sexual drive on the predictive validity of the J-SOAP-II for sexual recidivism. We expected age to moderate the predictive validity of the J-SOAP-II, but given the mixed findings of prior research, we had no a priori hypothesis regarding the age group for whom the instrument would have better predictive validity. We also expected that the instrument would have better predictive validity for youth with heightened sexual drive than for those without heightened sexual drive.
Method
Participants
The sample was comprised of 156 male JSOs who were discharged from one of two New Jersey Juvenile Justice Commission (NJJJC) facilities, the Pinelands Residential Community Home or the New Jersey Training School, between 1998 and 2009. At discharge, participants’ ages ranged from 14 to 19 years, with a mean age of 17.43 years (SD = 1.10 years). The majority of participants were African American (63.46%, n = 99), 19.87% (n = 31) were Caucasian, and 14.74% (n = 23) Latino. Participants spent an average of 14.69 months in NJJJC custody (SD = 7.22 months). More than half were discharged to live with a family member or in a foster home (58.97%, n = 92), whereas 21.79% (n = 34) were discharged to a community-based residential treatment program, and 12.17% (n = 19) to another juvenile justice facility; discharge location for the remaining 7.05% (n = 11) of participants was unavailable.
Participants’ index offenses varied. Although the majority of the sample was incarcerated for a sexual offense (69.87%, n = 109), 30.13% (n = 47) committed a non-sexual index offense (i.e., non-sexual violent assault, drug or property offense, violation of parole) but were receiving JSO-specific treatment because they had also committed a prior sexual offense. The most common sexual offense charge (47.44%, n = 74) was aggravated sexual assault (e.g., sexual assault involving penetration and a child under the age of 13, the use of a weapon, or the use of physical force or coercion and consequent severe personal injury), followed by sexual assault (28.85%, n = 45), sexual contact (13.46%, n = 21), and non-contact offenses (i.e., lewdness, harassment, endangering the welfare of a minor; 10.26%, n = 16).
Measures
Intended for use with male JSOs 12 to 18 years old, the J-SOAP-II is an empirically informed 28-item checklist of static and dynamic risk factors for non-sexual and sexual recidivism (Prentky & Righthand, 2003). Each item is scored on a 0- to 2-point scale, with a 0-point score reflecting the absence of a risk factor, a 1-point score reflecting some evidence of a risk factor, and a 2-point score reflecting the clear presence of a risk factor. The J-SOAP-II generates seven scores: a total score, four scale scores (Scale I: Sexual Drive/Preoccupation, Scale II: Impulsive/Antisocial Behavior, Scale III: Intervention, and Scale IV: Community Stability/Adjustment), and two summary scale scores (Static, Dynamic). As all risk factors are weighed equally, these scores are generated by summing the relevant items. No cutoff scores for risk-level classifications have been established (Prentky & Righthand, 2003).
Studies in which J-SOAP-II ratings were based on a file review generally have found the instrument to have good interrater reliability and moderate to strong internal consistency. For instance, Parks and Bard (2006) found good interrater reliability for the total and scale scores, ranging from r = .95 for the total score to r = .81 for Scale I, and strong internal consistency, ranging from α = .77 for Scale I to α = .90 for Scale III. Martinez et al. (2007) and Rajlic & Gretton (2010) had similar findings.
In the current study, internal consistency was moderately strong for the J-SOAP-II total score (α = .79), and stronger for the Dynamic Summary Scale (α = .89) than the Static Summary Scale (α = .67). Internal consistency was variable for the subscales, but generally adequate (α = .79 for Scale I, α = .75 for Scale II, α = .92 for Scale III, α = .71 for Scale IV; Martinez et al., 2015). After omitting Item 7, which was used to group youth in the sample according to their level of sexual drive, internal consistency increased slightly for the total score (α = .80) and decreased slightly for Scale I (α = .73) and the Static Summary Scale (α = .65).
Two-way mixed model intraclass correlation coefficients (ICC) were used to estimate interrater reliability for a subset of cases that were rated by a second clinician (n = 25). Interrater reliability was strong for the total score, Dynamic Summary Scale score, and Static Summary Scale score (ICC = .82, .82, and .83). Interrater reliability for the subscales varied (ICC = .74 for Scale I, ICC = .87 for Scale II, ICC = .77 for Scale III, ICC = .65 for Scale IV; Martinez et al., 2015).
Procedures
The procedures used to generate the data set on which the present study was based are detailed elsewhere (Martinez et al., 2015). Therefore, only a brief description is provided here.
J-SOAP-II protocols
J-SOAP-II ratings were based on a retrospective review of all clinical and legal system records available at the time of discharge from NJJJC placement. A subset of these cases (n = 25) were reviewed by a second clinician to assess interrater reliability. The extent of records available differed across cases but typically included demographic information, legal documents (e.g., victim and witness statements, police reports, pleadings), and treatment information (e.g., incident reports, progress notes, quarterly treatment summaries). Ratings were completed in accordance with the J-SOAP-II manual (Prentky & Righthand, 2003), with the exception of Scale IV (Community Stability/Adjustment), which typically is omitted for youth who are in a secure setting. To examine the potential utility of this scale, and the Dynamic Summary Scale (which cannot be generated without inclusion of Scale IV), Scale IV was scored on the basis of participants’ behavior and the support they received from guardians during the last 6 months of treatment.
Outcome data
Outcome data were collected via review of official criminal justice records, the New Jersey Judiciary’s Family Automated Case Tracking System (FACTS) and PROMIS/GAVEL, to determine whether participants had sexually reoffended after discharge. Recidivism was determined on the basis of re-arrest rather than conviction because sexual offenses can result in non-sexual offense convictions through plea bargaining. Participants were considered to have reoffended if they were arrested for a charge that was sexual in nature, including those for non-contact offenses (e.g., lewdness). Any time that participants spent in NJJJC custody or under NJJJC supervision as a result of having committed a non-sexual reoffense during the follow-up period was not documented. In addition, although some participants became adults during the follow-up period, sexual offenses committed during adolescence were not distinguished from those committed during adulthood.
Statistical analysis
Receiver operating characteristic (ROC) curve analyses were used to estimate the predictive validity of the J-SOAP-II in identifying youth who committed a sexual reoffense. ROC curve analysis provides a measure of discrimination, and is common in risk assessment research because it is relatively resistant to low base rates (Mossman, 1994; Singh, 2013). This approach generates an AUC estimate by plotting the true positive rate of an instrument against its false positive rate along all possible cutoff scores. AUC estimates range from 0 to 1, with those exceeding .70 and .80 reflecting “adequate” and “good” predictive validity, respectively (Mossman, 1994).
The predictive validity of the J-SOAP-II was compared across age and the presence/absence of heightened sexual drive using the z-score formula for comparing ROC curves described by Hanley and McNeil (1982), and by comparing 95% confidence intervals (CI), a more conservative method of estimating group differences (Gardner & Altman, 1986). With respect to age, the sample was divided into two groups, JSOs 14 to 16 years old at discharge (n = 59) and those 17 to 19 years old at discharge (n = 97), given research suggesting that psychosocial maturity develops more rapidly after age 16 (Cauffman & Steinberg, 2000; Steinberg, Cauffman, Woolard, Graham, & Banich, 2009).
The effect of sexual drive on the predictive accuracy of the J-SOAP-II was examined by grouping JSOs in the sample according to their scores on J-SOAP-II Item 7, which assesses excessive sexual activity or preoccupation with sexual gratification, “relative to what might be considered normative for youths of [the same] age” (Prentky & Righthand, 2003, p. 15). Evidence considered in scoring this item includes “paraphilias . . . , compulsive masturbation, chronic and compulsive use of pornography, frequent highly sexualized language and gestures, and indiscriminant sexual activity with different partners out of the context of any relationship” (Prentky & Righthand, 2003, p. 15). JSOs with a 2-point score on this item, which reflects a high level of sexual drive and preoccupation and is assigned on the basis of six or more instances of sexualized behavior (n = 70; “higher sex drive group”), were compared with those with a 0-point item score, which reflects normative/minimal sexual drive and/or preoccupation and assigned on the basis of one or two instances of sexualized behavior (n = 53; “lower sex drive group”; Prentky & Righthand, 2003). JSOs with a 1-point score on this item were omitted from this analysis to permit comparison of youth with a clear pattern of sexualized behavior with those without evidence of this pattern. For all analyses involving sexual drive, the total, Scale I, and Static Summary Scores were recalculated without Item 7 to account for its use as a grouping variable.
Finally, the utility of the J-SOAP-II in predicting time to sexual reoffense was estimated using Cox regression, a type of survival analysis that controls for variation in time at risk (Cox, 1972). Sexual reoffense was entered as the event, and separate analyses were performed for J-SOAP-II total and scale scores as predictors. To examine whether age and sexual drive moderated the association between the J-SOAP-II and time to sexual reoffense, the main effects of the J-SOAP-II score and the moderator variable were entered in Block 1 of the model, and the relevant interaction term (Age Group × J-SOAP-II Score; Sex Drive Group × J-SOAP-II Score) was entered in Block 2 (see Holmbeck, 1997, 2002). To facilitate the interpretation of coefficients in models with significant interaction effects, the reference category was inverted to generate group-specific effects and Wald statistics. Because race, offense history (i.e., top sexual offense charge), index offense (i.e., sexual vs. non-sexual), and discharge location did not significantly predict time to sexual reoffense, these variables were not included as covariates in the Cox regression models.
Results
Thirteen youth (8.33% of the sample) sexually reoffended during the follow-up period (M = 63.70 months, SD = 30.44 months, range = 8.88-130.92). These offenses occurred an average of 12.31 months (SD = 21.99 months, range = 0-82) after discharge from NJJJC custody. Discharge location was not significantly associated with sexual recidivism, as the rate of sexual recidivism for participants discharged to their families or foster homes (9.52%; 8 of 84) was not significantly different from that for participants who were discharged to another correctional facility or residential treatment (7.55%; 4 of 53), χ2(1, N = 145) = 0.06, p = .81. The rate of sexual recidivism for the younger age group (8.47%; 5 of 59) did not differ significantly from that for the older age group (8.25%; 8 of 97), χ2(1, N = 156) = 0.002, p = .96. Likewise, the rate of sexual recidivism did not differ significantly between the higher sex drive (10.00%; 7 of 70) and lower sex drive (11.32%; 6 of 53) groups, χ2(1, N = 123) = 0.06, p = .81. Sexual drive was not significantly associated with age, χ2(1, N = 123) = 0.18, p = .67. Older participants comprised the majority of both the higher sex drive (64.29%; 45 of 70) and lower sex drive (67.92%; 36 of 53) groups.
Impact of Age on J-SOAP-II Predictive Validity
ROC curve analyses
The total score was an adequate, significant predictor of sexual recidivism for JSOs 16 years old and younger at discharge, AUC = .75, p = .07, 95% CI = [.58, .92], but not for those 17 years old and older, AUC = .58, p = .45, 95% CI = [.37, .79], although the difference between these estimates was not significant (see Figure 1; z = 0.10, p = .32, two-tailed test). Notably, the 95% CI for the AUC for the younger age group does not include .50, yet the p value for this estimate exceeds .05. Given the modest sample size used in this analysis, the CI is considered to more accurately represent the estimate’s statistical significance than the p value (Gardner & Altman, 1986).

Comparison of the predictive validity of the J-SOAP-II for sexual recidivism across age.
The Static Summary Scale was not a significant predictor of sexual recidivism for either age group, generating AUC estimates of .44, p = .63, 95% CI = [.14, .73], and .55, p = .62, 95% CI = [.38, .73], for the younger (14-16 years old) and older (17-19 years old) groups, respectively. Also, Scale I did not perform better than chance for both groups (see Table 1). Scale II was an adequate, significant predictor for the younger group, AUC = .73, p = .09, 95% CI = [.61, .86], but not for the older group, AUC = .52, p = .83, 95% CI = [.36, .68], although neither comparison of their 95% CIs (see Figure 1) nor the z-score method (z = 1.22, p = .22, two-tailed test) reflected a significant difference between the AUCs.
ROC Curve Analyses: Age at Discharge and Sexual Recidivism.
Note. ROC = receiver operating characteristic; AUC = area under the curve; CI = confidence interval; Scale I = Sexual Drive/Preoccupation; Scale II = Impulsive/Antisocial Behavior; Scale III = Intervention; Scale IV = Community Stability/Adjustment.
The Dynamic Summary Scale and its component subscales (Scales III and IV) were also significant predictors for JSOs 14 to 16 years old but not for JSOs 17 to 19 years old. The Dynamic Summary Scale was a good predictor for the younger group, generating an AUC estimate of .87, p = .006, 95% CI = [.78, .97], but was a poor predictor for the older group, AUC = .58, p = .44, 95% CI = [.35, .82]. As can be seen in Figure 1, the 95% CIs for these estimates overlap, indicating that the difference between them was not significant. However, the z-score method of comparison indicates that this difference approached significance, z = 1.91, p = .056 (two-tailed test). Likewise, Scale III was a good predictor of sexual recidivism for the younger group, AUC = .89, p = .004, 95% CI = [.79, .98], but not for the older group, AUC = .55 (p = .61, 95% CI = [.33, .78]. Comparison of the 95% CIs indicated a significant difference between these estimates, as did the z-score comparison, z = 2.32, p = .02 (two-tailed test). Scale IV displayed adequate predictive accuracy for the younger group, AUC = .72, p = .10, 95% CI = [.56, .88], and poor predictive accuracy for the older group, AUC = .59, p = .39, 95% CI = [.39, .79], but the difference between these estimates was not significant, z = 0.78, p = .44.
Cox regression
Interaction terms for the J-SOAP-II total score, Static and Dynamic Summary Scale scores, and Scales I, II, and IV scores (i.e., Age Group × Score) were not significant, indicating that age did not moderate the association between these scores and time to sexual reoffense. However, the interaction term for Scale III was significant, β = .55, SE = .28, Wald = 3.95, p = .047), indicating that age moderated the accuracy of this score in predicting time to reoffense. Scale III was a better predictor for the younger group (14-16 years old), β = .61, SE = .27, Wald = 5.25, p = .022, exp(β) = 1.83, than for the older group (17-19 years old), β = .057, SE = .09, Wald = 0.44, p = .51, exp(β) = 1.06.
Impact of Sexual Drive on J-SOAP-II Predictive Accuracy
ROC curve analyses
The total score displayed adequate predictive accuracy for the higher sex drive group, AUC = .70, p = .08, 95% CI = [.50, .91], and poor predictive accuracy for the lower sex drive group, AUC = .64, p = .26, 95% CI = [.40, .88], but neither of these estimates, nor the difference between them, was significant (see Figure 2). The Static Summary Scale was an equally poor predictor for both groups, as was Scale I (see Table 2). Scale II was a significant, albeit poor, predictor of sexual recidivism for the higher sex drive group, AUC = .67, p = .15, 95% CI = [.53, .81], but not for the lower sex drive group, AUC = .55, p = .68, 95% CI = [.36, .75], although this difference was not significant.

Comparison of the predictive validity of the J-SOAP-II for sexual recidivism between higher sexual drive and lower sexual drive JSOs.
ROC Curve Analyses: Level of Sexual Drive and Sexual Recidivism.
Note. The total, Static Summary Scale, and Scale I scores were recalculated without Item 7 to account for its use as a grouping variable. Sexual drive was assessed using J-SOAP-II Item 7. The higher sexual drive group was comprised of youth in the sample with a 2-point score on this item, and the lower sexual drive group was comprised of youth with a 0-point score. ROC = receiver operating characteristic; AUC = area under the curve; CI = confidence interval; Scale I = Sexual Drive/Preoccupation; Scale II = Impulsive/Antisocial Behavior; Scale III = Intervention; Scale IV = Community Stability/Adjustment; J-SOAP-II = Juvenile Sex Offender Assessment Protocol–Revised.
The Dynamic Summary Scale was an adequate predictor of sexual recidivism for the higher sex drive group but not for the lower sex drive group, generating AUC estimates of .73, p = .044, 95% CI = [.49, .97], and .69, p = .13, 95% CI = [.44, .94], for the higher sex drive and lower sex drive groups, respectively. Scale III was also an adequate, albeit non-significant, predictor for the higher sex drive group, AUC = .72 p = .053, 95% CI = [.47, .98], but not for the lower sex drive group, AUC = .63, p = .31, 95% CI = [.37, .89], but the difference between these AUC estimates was not significant. There was also no significant difference in the predictive validity of Scale IV between the two groups, although this scale was a significant predictor of sexual recidivism for the higher sex drive group, AUC = .73, p = .046, 95% CI = [.54, .92], but not for the lower sex drive group, AUC = .61, p = .39, 95% CI = [.39, .83].
Cox regression
None of the interaction terms (i.e., Sex Drive Group × J-SOAP-II Score) was significant, indicating that sexual drive (as assessed by J-SOAP-II Item 7) did not moderate the accuracy of the J-SOAP-II in predicting time to sexual reoffense.
Discussion
Research has repeatedly shown that relatively few JSOs sexually reoffend (Caldwell, 2010), and among those who do, few go on to commit sex offenses as adults (Lussier & Blokland, 2014; Lussier, Van Den Berg, Bijleveld, & Hendriks, 2012; McCuish, Lussier, & Corrado, 2016). However, because of public misperception about JSOs’ sexual recidivism risk, JSOs can face severe and highly restrictive sanctions. Accurate risk assessment is thus essential to ensure that JSOs at low risk for reoffending do not face overly restrictive penalties, and those at heightened risk receive the supervision and treatment they need. The J-SOAP-II is the most commonly used instrument in the assessment of sexual recidivism risk among JSOs (Murrie, 2012), but support for its predictive validity is mixed. Some studies have found the predictive validity of the J-SOAP-II to vary across offender subgroups, suggesting that the instrument may better predict recidivism for some offenders than others. Yet, few studies have compared its predictive validity across offender subgroups. This study sought to redress this gap in the literature by comparing the predictive validity of the J-SOAP-II across two offender characteristics, age and sexual drive.
Overall, the J-SOAP-II was a better predictor of sexual recidivism for younger JSOs (those 14-16 years old at assessment) than for older JSOs (those 17-19 years old at assessment). All J-SOAP-II scores but Scale I and the Static Summary Scale were significant predictors (based on 95% CIs for the AUC estimates) for the younger group, whereas none of the scores significantly predicted sexual reoffense for the older group. In addition, significant differences were found in the predictive validity of the Dynamic Summary Scale and Scale III (Intervention) scores across the two groups. Moreover, when time to reoffense was taken into account, Scale III significantly predicted time to sexual reoffense for the younger group but not the older one. These findings complement those of Ralston and Epperson (2013), who found that the J-SOAP-II predicted recidivism occurring before the age of 18 but not after. However, Viljoen et al. (2008) found the opposite—that the J-SOAP-II was a better predictor of recidivism for older JSOs than younger ones. This inconsistency may have resulted from differences in study methodology and sample characteristics. Whereas Viljoen and colleagues included JSOs as young as 12 years old in their sample and used age 15 as the cutoff for the younger group, the present study did not include JSOs younger than 14 years of age and used age 16 as the cutoff for the younger group.
These findings suggest that risk factors for sexual recidivism may change over the course of adolescence. In particular, it appears that risk factors related to amenability to intervention (e.g., empathy, remorse, motivation for change), as measured at discharge by Scales III and IV, and the Dynamic Summary Scale, may better predict recidivism in younger JSOs than in older ones. Currently, little is known regarding developmental changes in risk factors for sexual recidivism during adolescence and how these changes impact assessment.
Some have suggested, contrary to the findings of the current study, that risk assessment instruments are poorer predictors of sexual recidivism for younger adolescents because scores on an instrument such as the J-SOAP-II may reflect transient characteristics that will change with development (e.g., lack of empathy, impulsivity) rather than stable personality traits (Viljoen et al., 2008; Vincent et al., 2012). However, research suggests that the development of psychosocial maturity, comprised of traits that relate to dynamic items included in JSO risk assessment instruments, continues throughout late adolescence and into early adulthood (Cauffman & Steinberg, 2000; Cauffman & Steinberg, 2012; Steinberg et al., 2009). Steinberg and Cauffman (1996) posited that psychosocial maturity is comprised of temperance, or the ability to resist impulses and control one’s behavior; perspective, or the ability to consider the consequences of one’s actions and others’ perspectives; and responsibility, or autonomy and identity formation. They found that these traits continue to develop during late adolescence, with the steepest increase occurring at some point between 16 and 19 years of age (Cauffman & Steinberg, 2000). Similarly, Steinberg and colleagues (2009), who operationalized psychosocial maturity in terms of impulsivity, susceptibility to peer influence, sensation seeking, risk perception, and future orientation, found that psychosocial maturity increased between mid-adolescence and early adulthood, with significant increases occurring between the ages of 17 and 22.
Many of these aspects of psychosocial maturity relate to dynamic risk factors included in the J-SOAP-II, such as temperance and impulsivity, which relate to Scale IV items that assess a youth’s ability to manage his sexual urges and anger, and perspective, which relates to the Scale III item that assesses empathy. Consequently, it may be more difficult to assess these dynamic risk factors accurately in older adolescents because they are experiencing more rapid change in these domains than younger adolescents are. However, no study to date has compared these or other risk factors for sexual recidivism across different age groups of JSOs, and the few studies that have examined the moderating effect of age on the predictive validity of risk assessment instruments have yielded mixed results (Vincent et al., 2012). Moreover, as the current study examined only age at discharge, it is possible that a different pattern would have emerged had the moderating effect of age at index offense or reoffense been analyzed. Additional research in this area is warranted to determine whether different risk assessment approaches should be used for adolescents of different ages.
A second potential explanation for the age-related findings of this study lies in the developmental course of sexual offending among JSOs. Whereas some JSOs commit one sexual offense during adolescence and desist thereafter, others commit multiple sexual offenses, and some persist in sexually offending into adulthood (Lussier et al., 2012; Smallbone, 2006). Some have suggested that these different offending trajectories may represent distinct developmental pathways to sexual aggression (Lussier, 2005; Lussier et al., 2012; Nisbit, Smallbone, & Wortley, 2010; Smallbone, 2006), and thus, that risk factors for sexual recidivism differ across offending trajectories (Knight, Ronis, & Zakireh, 2009; Lussier et al., 2012). However, very little research has examined this hypothesis. Knight and colleagues (2009) compared risk factors for sexual recidivism between JSOs in a residential treatment program and incarcerated adult sex offenders who had committed sex offenses during adolescence. They found that adult sex offenders who had committed sex offenses during adolescence were more likely than JSOs to have heightened sexual drive/preoccupation, hostility toward women, exposure to pornography, and interpersonal/affective psychopathic traits. Conversely, they found that JSOs were more likely than adult sex offenders with a history of adolescent sex offenses to have school problems, general delinquency, and childhood sexual abuse, suggesting that risk factors may differ between JSOs who desist and those who continue to offend into adulthood. Subsequently, Lussier and colleagues (2012) examined the sexual offending trajectories of JSOs during adolescence and adulthood, finding that youth in their sample fell into two trajectories. The majority followed an “adolescent-limited” (AL) trajectory, with sexual offending peaking, on average, at age 14 and terminating by the end of adolescence. However, a small portion of their sample followed a “high-rate, slow desister” (HRSD) trajectory, with sexual offending peaking at an average age of 12 and declining thereafter, but persisting into adulthood (p. 1569). Given these findings and those of Knight et al. (2009), Lussier et al. (2012) suggested that transient features of adolescence, such as increased risk-taking, may increase sexual recidivism risk for the AL group, whereas more stable personality traits, such as callousness and hypersexuality, may increase risk for the HRSD group.
Consequently, it is possible that J-SOAP-II Scale III and the Dynamic Summary Scale were stronger predictors of sexual recidivism for the younger group (14-16 years old at assessment) than the older group (17-19 years old at assessment), not because these groups differed in age but rather because they differed in offending trajectory. In particular, it is possible that youth in the younger group may have been more likely to have committed a sex offense at an earlier age than those in the older group, and thus, may have been more likely to fall within the HRSD trajectory than youth in the older group. As a result, Scale III and the Dynamic Summary Scale may have been significant predictors of sexual recidivism for the younger group because these scales assessed stable personality traits that placed this group at higher risk. However, absent data about the offending trajectories of the youth in this sample, whether this is true cannot be determined. Indeed, Lussier and colleagues (2012) suggested that heterogeneity in offending trajectories among JSOs may confound the results of research examining the predictive accuracy of JSO risk assessment tools, as few studies examine within-individual change in offending patterns. Future research is necessary to examine whether the predictive accuracy of the J-SOAP-II and other JSO-specific risk assessment tools varies with offending trajectory.
The J-SOAP-II was also a somewhat better predictor of sexual recidivism for JSOs with heightened sexual drive. For JSOs with a clear pattern of sexualized behavior, as assessed by J-SOAP-II Item 7, Scale II (Impulsive/Antisocial Behavior) and Scale IV (Community Stability/Adjustment) scores significantly predicted sexual recidivism, and the total and Scale III (Intervention) scores approached significance. Of note, the authors of the J-SOAP-II have suggested that in addition to Scale II, the total score and Scales III and IV also tap antisocial personality characteristics, given the significant correlations between these scales and subscales and the Youth Level of Service/Case Management Inventory, a measure of risk for general delinquency (Righthand et al., 2005). In contrast, none of the J-SOAP-II scales or subscales significantly predicted sexual recidivism among JSOs without a clear pattern of sexualized behavior. However, no significant differences in the predictive validity of the measure were found between these two groups.
The lack of significant differences between ROC curves notwithstanding, these results suggest that J-SOAP-II items assessing antisocial personality characteristics may have greater clinical utility for JSOs with high levels of sexual drive. However, no studies to date have examined the relationship between sexual drive and antisociality in the prediction of sexual recidivism among JSOs. Additional research in this area is necessary to maximize the predictive accuracy of JSO risk assessments. Moreover, heightened sexual drive and preoccupation may provide early evidence of sexual deviance (Seto & Barbaree, 1997), and some suggest that there may be a subgroup of JSOs for whom antisocial personality traits and sexual deviance jointly contribute to recidivism risk (Awad & Saunders, 1989; Butler & Seto, 2002; Seto & Lalumiere, 2010; Worling, 2001). Given the debate as to whether and how sexual deviance should be assessed in youth (Worling, 2012), sexual drive may provide a more reliable measure of risk, particularly for antisocial youth. Heightened sexual drive may also have the potential to serve as a marker for JSOs who are more likely to commit sex offenses as adults, as Knight and colleagues (2009) found that sexual drive, among other indices of hypersexuality (e.g., sexual compulsivity, paraphilias, voyeurism, pornography use) differentiated adult sex offenders who had committed sex offenses as adolescents from desisting JSOs. However, whether JSOs with both heightened sexual drive and antisociality are more likely to commit sex offenses as adults remains unclear, as they did not find that measures of antisociality differentiated the two groups. Consequently, further study of the association between sexual deviance, sexual drive, and antisociality in JSOs is warranted.
The results of this study should be considered in light of several methodological limitations. First, the failure to find more significant differences between groups likely resulted from the modest sample size. The overall sample size (N = 156) and the rate of sexual recidivism in this sample (8.33%; n = 13) were comparable with those found in other studies (Caldwell, 2010) and reflect the infrequency of sexual recidivism among JSOs. However, analyzing the predictive validity of the J-SOAP-II across subgroups of these offenders resulted in comparisons between relatively small groups, each with only a handful of JSOs who sexually reoffended. Thus, as the large CIs for a number of the AUC estimates indicate, many analyses had limited power to detect between-group differences. Second, very few of the significant AUC estimates for the J-SOAP-II fell within the “good” range, suggesting that even when the instrument significantly predicted recidivism, it had only modest accuracy at best. Third, the use of the J-SOAP-II in this study departs from the way in which the instrument is typically used in clinical practice. Whereas this study treated the J-SOAP-II as an actuarial instrument, interpreting scores rather than clinical judgments about risk, clinicians are typically advised to use the J-SOAP-II as a tool to guide treatment and placement decisions, rather than as a set of scores. Although all published studies of the J-SOAP-II have used the measure in a similar manner, future research should examine its utility in guiding clinical decision making. Fourth, having rated the J-SOAP-II on the basis of a retrospective file review, the accuracy of these ratings may have been impacted by variability in file content across participants. Despite strong interrater reliability, even for dynamic risk factors that require more subjective judgments, future research utilizing a prospective design would be helpful to better understand the influence of development on J-SOAP-II scores and predictive accuracy. Fifth, the results of the Cox regression analyses should be interpreted with caution, given the possibility that some youth in the sample may have been incarcerated or under NJJJC supervision during the follow-up period because they committed a non-sexual reoffense. Consequently, true time at risk was not accounted for in this study. Finally, findings regarding Scale IV should also be interpreted cautiously. Although the J-SOAP-II manual recommends omission of this scale for incarcerated JSOs, we chose to rate this scale based on the 6 months preceding discharge. The findings regarding predictive accuracy of this subscale support this decision, as Scale IV appeared to have a significant association with recidivism for some JSOs.
These limitations aside, this study provides an important contribution to the literature on the predictive validity of the J-SOAP-II. These results suggest that the J-SOAP-II may have greater predictive validity for JSOs between 14 and 16 years of age at the time of evaluation than for older youth, and that development during adolescence or the developmental course of sexual offending in youth may impact the accuracy of risk assessment. The results also indicate that J-SOAP-II scales that assess antisocial personality characteristics may be better predictors for JSOs with heightened sexual drive, and that these risk factors may jointly contribute to recidivism risk. These findings may have particular utility for clinicians who are responsible for assessing or managing JSOs’ risk of recidivism after treatment or discharge from a correctional facility, as this study is one of only a handful of studies to analyze the predictive validity of the J-SOAP-II when administered at discharge. Still, given the dearth of research on the psychometric properties of the J-SOAP-II, and on risk factors for sexual recidivism in adolescent offenders, additional study is necessary to identify the types or groups of offenders for whom the J-SOAP-II has the strongest predictive validity.
Footnotes
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
