A model for the assessment of static and dynamic factors in sexual offenders.

The assessment, treatment, and management of sex offenders has become a preoccupation of the criminal justice system, despite low rates of recidivism among released sex offenders (Hanson & Bussiere, 1998). Public concern has heightened regarding community safety, at least in part due to a relatively small number of sensational sex crimes which have received media attention. Such attention promotes the mistaken view that all sex offenders are alike. These issues require a criminal justice response regarding both policy and resource allocation.

Given the community concern and the significant resources invested in this population, it is not surprising that views abound regarding the preferred management strategy. At the same time, competing perspectives regarding the utility of treatment (see Abracen & Looman, 2004: Marshall, Anderson, & Fernandez, 1999, for recent reviews on this topic) and the use of prolonged incarceration in the form of detention (Nugent & Zamble, 2001) make consensus problematic. Addressing the concerns of one group by definition ignores the concerns of another. We believe that an integrated assessment strategy regarding the management of sex offenders must, at the very least, account for the following issues: (a) the marked heterogeneity among sex offenders regarding needs and risk (b) the preference for a differentiated strategy such that lower risk offenders are managed differently than high risk offenders; and (c) the fact that multi-method assessment and reassessment will be more effective than attempting to develop a single, one-time, "ideal" approach for sex offenders. Based on the available evidence (Andrews & Bonta, 1998; Marshall, Anderson, & Fernandez, 1999), it seems fair to speculate that sexual offender assessment needs may overlap with other groups of offenders but that unique treatment targets (e.g., inappropriate arousal) may require assessment in the case of sexual offenders as well. Theoretical developments within the field of sex offender treatment should also be incorporated into any comprehensive assessment strategy.

The purpose of this paper is to provide guidelines regarding the necessary domains which should be addressed in any comprehensive assessment of sex offenders. We established these assessment targets based on the available empirical research, and they represent an attempt to integrate the various approaches that have been suggested for use with sex offenders. Several reviews are available regarding the assessment and treatment of sex offenders (Abracen & Looman, 2004; Blanchette, 1996; Hanson & Bussiere, 1998: Marshall et al. 1999). Blanchette's (1996) conclusion is representative:

   Sexual aggression is a complexly-determined phenomenon, with
   varied antecedents and sequelae. Perpetrators of sexual crimes
   differ in their personal and criminal histories, the circumstances
   preceding their offences, their victim age and gender preferences,
   the attitudes and beliefs that support their deviant behaviour, and
   the degree to which they have used force or brutality or caused
   physical harm to their victims. Thus, sexual offenders are a
   heterogeneous group of individuals, with diverse evaluative and
   treatment needs. (p. 4)

Multimethod assessment is thus a critical issue with reference to the management of sexual offenders (Abracen & Looman, 2004; Blanchette, 1996). We believe that the investigation of systematic, multimethod assessments completed on large numbers of sex offenders can provide clarity, such that an assessment of risk that incorporates dynamic factors as well as static variables is possible.

Static or actuarial risk consists of an appraisal of historical factors (e.g., criminal history) and, with reference to some measures, personality characteristics that are not generally susceptible to significant change over time. Static risk is typically assessed using such instruments as the Hare Psychopathy Checklist-Revised (PCL-R) as well as other indices. Dynamic risk consists of an analysis of daily variables which are susceptible to change via treatment or other interventions.

In terms of treatment, the most highly regarded approach, a cognitive-behavioral model employing relapse prevention in high risk situations, is described by Marques, Day, Nelson, and West (1994). Several authors have suggested revisions to the model based upon recent research. For example, Marshall (1993) has suggested the importance of insecure attachment patterns and intimacy deficits in the etiology of sexual offending and has offered suggestions as to how cognitive-behavioral approaches may be adapted based on these developments (See Marshall et al., 1999, for a review of this research). This research suggests that the stability of relationships and relationship difficulties are particularly salient issues to address in the assessment of sexual offenders. In addition, substance abuse and in particular alcohol abuse may be an important domain to assess with reference to sexual offending (Abracen & Looman, 2003). Inappropriate sexual arousal has also been identified as an important risk factor related to recidivism (Hanson & Bussiere, 1998).

Other authors have developed actuarial risk instruments or scales (see Andrews & Bonta, 1998: Hare, 1998; Quinsey, Harris, Rice, & Cormier, 1998: Wong, 2000, for reviews of some of these measures) as a requisite management strategy. While these results are encouraging, additional research specific to the full range of offenders seen by Correctional Service of Canada (CSC) and other jurisdictions is necessary. These scales highlight the importance of criminal history variables as well as a variety of personality and/or dynamic variables that are important to address in any comprehensive assessment of sexual offenders. Two of the measures that have been most widely supported are the PCL-R (Hare, 1991) and the Level of Supervision Inventory (LSI; Bonta & Andrews, 1993). Although these measures have proven very useful for predicting risk for future recidivism, more research is needed on dynamic or changeable risk factors (e.g., inappropriate arousal, treatment motivation).

The majority of CSC programs lack a sufficient sample size for rigorous analysis. The goal of the present research was to investigate the utility of a model that incorporates literature from each of the domains listed above regarding the management of sex offenders. It is intended to add to existing research conducted by our team (e.g., Mailloux et al., 2003; Mailloux & Serin, 2001) that is relevant to the assessment and treatment of the broad range of sexual offenders seen by a variety of correctional jurisdictions. We expected that the final model would be generic across types of sex offenders (i.e., rapists, child molesters, and incest offenders) but would incorporate information specific to these subtypes. In this way, differentiated assessment strategies would potentially be possible. To offer a preliminary test of this model, we used an assessment sample taken from the Millhaven Assessment Unit (MAU), which assesses all sex offenders admitted to the Ontario Region of CSC. The present investigation represents an initial evaluation of the model. Should this initial validation yield significant findings, we will attempt to validate the model in future investigations. It is important to emphasize, however, that this model does not represent a substitute for the actuarial assessment of sex offenders. Rather, the model incorporates elements of traditional risk assessments but also incorporates additional information (e.g., relevant to relationship history and treatment motivation).

Seto, Barbaree, Serin, and Malcohn (1997) developed a two-tier model to begin to organize the many variables that appear to be related to sexual offending. The goal was first to develop a model consisting of a relatively small number of factors which might account for the complexity of issues related to sexual offending. The first tier reflects core constructs considered central to risk factors in sexual offenders: deviant sexual interest and criminality (Andrews & Bonta, 1998; Lalumiere & Quinsey, 1996). The second tier describes moderating variables: substance abuse, social competence, and treatment readiness. Within each of these domains it is possible to conceptualize both static and dynamic variables. Similarly, the moderating variables also represent responsivity factors that are expected to have an impact on treatment. Deficiencies in these areas will tend to attenuate positive treatment effects, independent of the assessment strategy employed (i.e., self-report, behavioral observation, interview-based).

We reviewed the literature related to these domains and selected robust variables for inclusion in our analyses. The domains and the variables proposed for the assessment model are presented in Table 1. These variables were selected consensually among the authors based on empirical support (correlated with sexual recidivism) or clinical utility (impact on treatment response). This list is not unique, as many are represented in various assessment strategies either partially or fully.

Upon compiling this list of variables, we needed to determine the extent to which explicit or proxy measures for each are available in the MAU database. That is, for some variables included in the model the available data in the MAU database only partially reflected the desired variable. For example, it was difficult to obtain information regarding intimacy deficits from the MAU database. As a result, we used data regarding the absence of stable relationships and markers of relationship difficulties in place of questions directly related to intimacy deficits. In developing scoring criteria for each of the variables in Tables 1 and 2, we determined that there were two notable areas where the proxy measures might be considered marginal. Existing measures in the MAU database for criminal attitudes and associations and sexual preoccupation and compulsivity are quite limited. This exercise in itself was therefore useful, as it identified domains where more detailed assessment would better reflect the conceptual model. This is also consistent with the long term intent to use these analyses to inform MAU and possibly other sites about how to further refine their comprehensive assessment batteries.

Although this model provides a clinical assessment protocol consistent with standards of practice (CSC, 2001), the initial focus was to determine their relative contribution to the assessment of risk for different types of sex offenders. Having determined which variables are significantly related to sex offender risk, we can then turn to the question of differential assessment. For example, do different domains postdict different types of sex offenders? Although these data are preliminary, they should be viewed as a further attempt to identify a set of domains with reasonable conceptual integrity which may further aid in the assessment of sexual offenders. This is particularly important with reference to dynamic factors (e.g., responsivity issues) that have received insufficient attention in the literature.

METHOD

Participants

Participants consisted of all offenders assessed at MAU between 1993 and 1998. By definition, all participants were incarcerated in Canadian federal penitentiaries. Within the Canadian context, any offender sentenced to 2 years or more falls under federal jurisdiction. The sample consisted of 519 participants. All participants had been convicted of a sexual offence or an offence to which there was a suspected or actual sexual component. Detailed descriptions regarding the entire sample can be found in Mailloux and Serin (2001). In brief, the mean age at index offEnce for the sample was 32 (SD = 8.95). Participants in the current sample had an average of 3.94 (SD = 5.0) sexual convictions on their records. The mean PCL-R score for the sample was 18.79 (SD = 8.12; n = 470), which corresponds to the low end of the moderate risk range for general or violent recidivism.

Procedure

Detailed information regarding each participant was coded based on file information. We produced a coding manual that included information regarding demographics, juvenile history, type of sexual offender, and characteristics of the sexual offence (e.g., degree of force used, degree of planning, use of instrumental violence, etc.). We also coded information regarding nonsexual offences. Background information regarding substance abuse and motivation for treatment (e.g., prior sex offender treatment, denial, minimization) were also scored. In all, we obtained data for over 500 variables. The items listed in the coding manual were itemized according to their level of importance to the construction of each domain (critical, desirable, or speculative). We performed a frequency analysis for each item, eliminating items for which more than 25% of the data was missing. Of the remaining items, those which were considered to be a duplicate of other information were eliminated. For most of the variables in the analyses, we coded data in a binary fashion (i.e., 0 or 1) depending on the presence or absence of that characteristic. However, for a number of variables this was not possible. For example, type of sex offender, number of partners, and Blishen Index (social class) were scored either categorically (in the case of type of sex offender) or as interval data (e.g., in the case of number of sexual partners).

We grouped the items according to the domains established by the model: criminality, sexual deviance, social competence, substance abuse, and treatment readiness. Since the items included in the coding manual were scored on a variety of measurement scales, we transformed the item scores into standardized scores so that each item was on the same measurement scale. The items could then be summed without concern for the differential weighting of items. The items included in each of the domains are listed in Table 1. With reference to three domains (criminality, sexual deviance, and substance abuse) all items were coded such that higher scores reflected more criminogenic attributes. For example, higher scores on the criminality domain reflected a more significant criminal history. With reference to social competence and treatment readiness, higher scores on these domains reflected increased levels of prosocial attributes. For example, higher scores on treatment readiness reflected more motivation for treatment or less resistance on the part of the client.

RESULTS

Background Information

In all the sample consisted of 519 consecutive admissions to the Millhaven Assessment Unit (MAU). Data with reference to type of sex offender were not available or could not be coded for 7 participants. Among the remaining 512 participants, we identified three groups of sex offenders: rapists (n = 166), child molesters (n = 168), and incest offenders (n = 178). The overall ANOVA indicated that the groups did not differ significantly on age at index offense, F(2, 479) = 2.71, ns. With reference to number of sexual assault victims, the ANOVA indicated that there were significant group differences, F(2,507) = 4.11, p < .05, [[eta].sub.2] = .016. Post-hoc testing using the Scheffe procedure indicated that rapists had significantly fewer victims than either child molesters (mean difference = -1.42, p < .05) or incest offenders (mean difference = -1.39. p < .05). There were no significant differences between child molesters and incest offenders.

Alpha Reliabilities of Domains

Each of the five domains, which we computed using standardized scores, were subjected to a reliability analysis to determine whether the items reflected a cohesive unit. We further refined the domains to maximize the associated alpha reliability for that domain. The alpha reliabilities for each domain as well as the final number of items constituting that domain are listed in Table 2. The number of items in each domain varied from 5 to 16. Standardized alphas were in the moderate range with the exception of sexual deviance and treatment readiness. The low reliability for treatment readiness is perhaps not surprising given the limited number of items included in the domain. Unfortunately, given that relatively few items related to treatment readiness could be coded from file information, it was not possible to increase the number of items in the domain.

Type of Sex Offender by Domain

To investigate whether there were differences between rapists, child molesters, and incest offenders on each of the five domains, we computed a 3 (Group) x 5 (Domain) MANOVA. The overall MANOVA using the Wilks criterion was significant, F(10,1008) = 15.41, p < .001, [[eta].sub.2]= .13. Univariate tests of significance revealed that the groups differed significantly with reference to criminality, F(2,508) = 31.0, p < .001: social competence, F(2,508) = 27.82, p < .001: sexual deviance, F(2,508) = 38.61, p < .001: and substance abuse, F(2,508) = 26.52, p < .001. There were no significant differences between the groups with reference to treatment readiness. The means and standard deviations for each of the standardized domain scores are listed in Table 3. Table 4 lists the between-groups significance levels using Tamhane's test with reference to each of the four domains for which significant univariate results were obtained. As can be seen from inspection of Tables 3 and 4, rapists were significantly more criminal than either of the other groups, with incest offenders being the least criminal of the groups. The incest offenders were rated as being the most socially competent group, followed by child molesters. We found rapists to be significantly less socially competent than either of the other groups. Both groups of child molesters were found to be more deviant than the rapists. Last, rapists had significantly greater difficulty with substance abuse than either of the other groups. There were no statistically significant differences between the two groups of child molesters with reference to either sexual deviance or substance use. Table 5 lists the correlations between the five standardized domain areas.

Regression analyses

We conducted a series of regression analyses to further investigate the relevance of the standardized domains. We performed a binary logistic regression using the five domains as predictor variables, and child molesters (including incest offenders) and rapists as the outcome variables. A test of the full model with all five predictors against a constant-only model was statistically reliable, [chi square] (5, N = 511) = 116.93, p < .001, indicating that the predictors as a set reliably distinguished between child molesters and rapists. The amount of variance accounted for by the model was moderate (Nagelkerke's [R.sup.2] = .285). Overall, the model correctly classified 73.6% of the sample. The model was much better at predicting child molesters (85.8% correct classification) than rapists (48.2% correct classification). According to the Wald criterion, only sexual deviation (z = 43.66, p < .001) and substance abuse (z = 8.71, p < .01) reliably predicted group outcome.

To investigate whether the model predicted severity of offending behaviour, we conducted a regression procedure (backward) using the five domains as predictor variables and number of sexual assault victims as the outcome variable. The best fit model indicated that the selected domains significantly predicted number of sexual assault victims, F(3,510) = 33.03, p < .001. Analyses of the individual domains indicated that three of the domains were retained in the final model: criminality, t = -2.38, p < .05; sexual deviance, t = 9.56, p < .001 ; and substance abuse, t = 2.15, p < .05. The adjusted [R.sup.2] for the best fit model was .16, suggesting that the best fit model accounted for a limited degree of variance.

Last, to investigate issues associated with predictive validity, we divided participants included in the present sample into high and low with respect to the LSI (based on a median split). We performed a binary logistic regression using LSI low (2-23) and high (24-50) scores as the outcome measure and the five domain variables as predictors. A test of the full model with all five predictors against a constant-only model was statistically reliable, [chi square] (5, N = 488) = 334.832, p < .001. The amount of variance accounted for by the model was high, (Nagelkerke's [R.sup.2] = .662). Overall, correct classification was very high with 84% of the sample correctly classified. Low LSI score was correctly classified 86% of the time, and high LSI score was correctly classified 82% of the time. According to the Wald criterion, three of the domains were significantly associated with the prediction of low or high LSI score: criminality, z = 37.42, p < .001; social competence, z = 36.14, p < .001; and substance abuse, z = 16.41, p < .001.

DISCUSSION

The results of the present investigation offer mixed support for the model described above. Three of the five domains evidenced reasonable levels of internal reliability. However, internal reliabilities for both sexual deviation and treatment readiness were low. In all likelihood, the internal reliabilities associated with these two domains are related to limitations associated with the information available in the present database. There was insufficient information to code phallometric data for over 75% of the sample: therefore, information related to this variable was not included in the present analyses. This is perhaps not surprising in light of the fact that the model was developed after the data had been collected. Given that phallometric data are related to outcome (Hanson & Bussiere, 1998: but see Marshall & Fernandez, 2000, for additional comments), the fact that data related to this variable were not included in the analyses represents a potentially important loss of information. Nonetheless, it is interesting to note that deviant arousal was a significant predictor of type of sex offender and number of sexual assault victims.

Relatively little information was available in the file information regarding treatment readiness. As such, there were few items (e.g., in comparison to the criminality domain) upon which to develop the domain. These data suggest that treatment readiness is one area in which more comprehensive assessment might be useful. Other jurisdictions where comprehensive assessment of sexual offenders is undertaken may also be interested in investigating whether this domain has received adequate attention.

Treatment readiness is a potentially important responsivity factor. Motivational interviewing techniques might be used to address issues related to treatment readiness (e.g., minimization: Miller & Rollnick, 2002). The lack of information available regarding this domain is perhaps not surprising because many people view treatment responsivity as a static (i.e., unchangeable) factor. This is perhaps the reason some programs deny entry to the majority of offenders who significantly minimize or deny their history of sexual offending (e.g., Hunter & Figueredo, 1999). This domain may be particularly important in light of the fact that the majority of sex offenders either minimize or deny some aspect of their sexual offending history (Barbaree & Cortoni, 1993).

With reference to the question of whether types of sex offenders differed on each of the five domains, a number of findings are relevant. First, the fact that rapists were significantly more criminal than either the child molesters or incest offenders is certainly in keeping with clinical impression. Likewise, the finding that child molesters evidenced higher scores on the sexual deviance domain is also not surprising in light of past research on the subject. The evidence appears more robust with reference to the presence of deviant sexual arousal being a criminogenic need for child molesters than for rapists (Marshall, 1996), a finding supported by the current data.

In terms of social competence, the results of the present analyses contradict some research in the area. Our study found that rapists had significantly lower levels of social competence than both child molesters and incest offenders. Nonetheless, Segal and Marshall (1985) found that rapists functioned reasonably well in terms of social skills in comparison to other incarcerated sexual offenders. Likewise, Overholser and Beck (1986) observed a trend for child molesters to be more deficient than rapists in assertiveness. As Gordon, Marshall, Loeber, & Barbaree (1977) observed several decades ago (an observation that still applies at present), social scientists need to be clear about what we mean when we state that sex offenders have social deficits. The present study adds to the existing knowledge with reference to this domain. Of course, more research using standardized measures and comparison groups involving nonsexual offenders are also necessary to further elucidate the relationship between social skills deficits and sexual offending. In point of fact, the lack of correspondence between the current study and others in the area are likely due to differences in methods and the relative paucity of information related to the topic. For example. Overholser and Beck (1986) investigated issues associated with assertiveness in sexual offenders. The social competence domain used in our study was certainly more global in nature than the more restricted definition used by these authors.

One aspect of social skills deficits that has received attention in the literature is the fact that sex offenders as a group appear to have difficulties with intimacy and loneliness (see Marshall et al., 1999, for a review of this research). It seems reasonable to speculate that such deficits may be in keeping with our social competence domain variables. This domain incorporated a number of variables specifically related to the absence of a stable relationship or difficulties in maintaining a relationship. As such, the variables included in our model seem to have received substantial support in the literature.

To determine whether the five domains were able to predict group membership, we performed a binary logistic regression that indicated that the model was very good at predicting child molesters but much less accurate at predicting rapists. Nonetheless, the model did account for a moderate amount of variance associated with the prediction of group membership. We found sexual deviation and substance abuse to be significant predictors of group membership. It is possible that subsequent refinements to the model might result in even better prediction of group membership. It should also be kept in mind that recent research suggests that categorizing sexual offenders on the basis of index sexual offence may be less sophisticated an approach than classifying sex offenders on the basis of theoretically derived criteria such as attachment style (e.g., Ward, Hudson, & McCormack, 1997).

We conducted analyses on the five domains to determine whether they could predict the number of sexual assault victims. Regression analyses revealed that the criminality, sexual deviance, and substance abuse domains were all significant predictors. The fact that social competence did not significantly predict number of victims is surprising. It may be that future studies will have to investigate more specific elements of social competence for significant relationships to emerge (e.g., related to intimacy deficits specifically). With reference to treatment readiness, more specific measures that account for a greater range of information may need to be included before significant relationships emerge.

Last, to investigate issues associated with predictive validity, we divided subjects based on their score on the LSI into low and high groups. The domain variables did an excellent job at predicting LSI score based on median split data. The model accounted for over 65% of the variance. This suggests that across type of sexual offender, the model did an excellent job of predicting risk level. Criminality, social competence, and substance abuse were all significant predictors in the model. Nonetheless, the fact that the model predicted LSI score so well should not be viewed as surprising. The assessed domains and the LSI overlap in terms of the items included in the respective assessment strategies. Nonetheless, each assessment strategy also investigates a unique subset of items. For example, the LSI does not investigate deviant arousal or issues related to treatment readiness directly.

The only domain that emerged as a significant predictor in each of the analyses was substance abuse. This is perhaps not as surprising as one might first think. Recent research has demonstrated that substance abuse is one of the most salient issues associated with sexual offending. This finding applies whether path analytic techniques (Johnson & Knight, 2000) or balanced placebo designs (Marx, Gross, & Adams, 1999) are employed. Further, data available from a recent national sample of female college students in the U.S. revealed that alcohol plays both direct and indirect roles in the outcomes of sexual assault (Ullman, Karabatsos, & Koss, 1999). Abracen and Looman (2003) reported that sexual offenders evidenced a distinct pattern of substance abuse in comparison to violent nonsexual offenders. Further, our data suggest that the pattern of substance abuse shown by sexual offenders may act synergistically with the intimacy deficits seen in this group as well. Based on the available evidence, it appears that substance abuse may play a more substantial role in sexual offending than previously believed.

Criminality also emerged as a significant predictor in three of our analyses, supporting the inclusion of this variable in the model. Sexual deviation emerged as a significant predictor in three of the analyses. Nonetheless, given that the sexual deviation domain had a low internal reliability, it is perhaps more difficult to determine what exactly this domain measures. The low internal reliability may in part be due to the fact that phallometric data were not included.

Although treatment readiness did not emerge as a significant predictor in several analyses and evidenced low levels of internal reliability, we do not believe that the variable should be dropped from the model. On the contrary, there is a relative paucity of information regarding this domain. However, both the view of clinicians as to the importance of this variable and the results of recent research (e.g., Serin & Kennedy, 1997) attest to the potential impact of this variable.

In summary, our proposed model received some support in this study, with three of the five domains showing acceptable levels of internal reliability. Analyses of predictive validity also supported the use of the model. This model offers clinicians and researchers a list of both dynamic and static variables that may inform clinical decision making and the prediction of recidivism. It should not be viewed as a substitute for the actuarial assessment of sex offenders; rather, our study suggests that a comprehensive assessment should also include data related to other domains (e.g., inappropriate arousal, patterns of substance abuse, issues related to treatment responsivity). It may also be the case that the first tier of the model may need to incorporate issues associated with intimacy and loneliness as well as substance abuse. The use of criminality in the first tier of the model was supported by the available data.

Rather than viewing the domains in our model hierarchically, it may be more relevant to investigate the manner in which the domains interact with one another. For example, as Hanson (2000) has argued, it may be the density of risk factors present which is of primary importance rather than any particular ordering of the variables. Alternately, the manner in which the assessed domains work in concert and the impact of various protective mechanisms may be more important than the use of a hierarchical model. The present investigation offers guidelines regarding important topics to assess in any comprehensive assessment of sexual offenders. It is only when such comprehensive assessments are undertaken that important questions regarding the relative density of risk factors and the impact of protective mechanisms can be investigated with more rigor. The investigation of each of our five domains using standardized scales with demonstrated psychometric properties and the refinement of the current model are areas where additional research would clearly be desirable. At present, gains have been made in the area of actuarial assessment (e.g., with the use of the PCL-R). However, additional research on dynamic variables is clearly necessary. This investigation represents a preliminary step in that direction.

Table 1. Variable List by Domain

Domain                Variables

Criminality           Number of nonsexual (violent & nonviolent)
                        convictions
                      Young offender arrests
                      Juvenile convictions
                      Early onset of behavioural problems
                      History of fighting
                      Trouble with police
                      Predominantly procriminal peers
                      Criminal attitudes
                      Employed at time of arrest
                      Quit jobs without another arranged
                      Crime as primary means of self support
                      Dangerous offender designation

Sexual deviance       Gender of victim
                      Deviant fantasies
                      Planning/grooming
                      Number of sexual convictions
                      Stranger victim
                      Paraphilias endorsed
                      Proportion sexual/total
                      Use of deceit
                      Total number of victims
                      Early onset of sexual offending
                      Diverse sex crimes
                      Type of sexual contact

Social competence     Marital status
                      Social class
                      Education level
                      Developmental history
                      Indicators of dysfunctional childhood
                      Unstable employment
                      Relationship difficulties (a)
                      Absence of stable relationship
                      Longest job in the community
                      Number of times fired
                      Used others for support
                      Employment as primary means of support
                      Global rating of lifestyle stability

Substance abuse       Alcohol used during offense
                      Drugs used during offense
                      Alcohol used as a teen
                      Alcohol used as an adult
                      Drugs used as a teen
                      Drugs used as an adult
                      Leisure activities related to substances

Treatment readiness   Prior sex offender treatment
                      Had non-sex offender treatment
                      Poor treatment motivation
                      Denial of offense
                      Indicators of denial/minimization/rationalization

(a) Four variables were included in this subdomain.

Table 2. Domain Reliabilities

Domain                Number of items   [alpha]

Criminality                 12            .83
Sexual deviance             12            .59
Social competence           16            .75
Substance use                7            .74
Treatment readiness          5            .51

Table 3. Means (Standard Deviations) by Group for Each Domain
(standardized scores)

Domain                   Rapists     Child molesters   Incest offenders
                         n = 166         n = 168            n = 177

Criminality            2.80 (7.41)      .29 (6.96)     -2.87 (5.64)
Social competence     -2.77 (6.44)     -.38 (7.08)      2.88 (7.57)
Sexual deviance       -2.61 (4.13)     1.79 (5.84)       .82 (4.29)
Substance use          1.94 (4.04)     -.57 (4.53)     -1.15 (3.90)
Treatment readiness    -.22 (2.83)      .20 (3.04)      2.62E-02 (2.81)

Table 4. Mean Differences by Group on Each Domain

                      Rapists vs.      Rapists     Child molesters
                    child molesters   vs. incest      vs. incest
Domain                  n = 512         n = 512         n = 512

Criminality              2.51 *         5.67 **          3.15 **
Social competence       -2.40 *        -5.66 **         -3.26 **
Sexual deviance         -4.40 **       -3.43 **           .96
Substance use            2.52 **        3.10 **           .58

* p < .01. ** P < .001.

Table 5. Pearson Correlations (ns in parentheses) Between the Domains

                                          Social
                        Criminality     competence

Criminality
Social competence     -.655 ** (518)
Sexual deviance       -.115 ** (519)    .104 * (518)
Substance abuse        .542 ** (519)   -.477 ** (518)
Treatment readiness    .064 (519)      -.014 (518)

                          Sexual       Substance
                         deviance        abuse

Criminality
Social competence
Sexual deviance
Substance abuse       -.180 ** (519)
Treatment readiness    .296 ** (519)   .014 (519)

Note. Analyses computed based on standardized domain scores.

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

REFERENCES

Abracen, J., & Looman, J. (2003). Alcohol, drug abuse, and treatment of sexual and violent non-sexual offenders. Directions in addiction treatment and prevention, 7, 23-30.

Abracen, J., & Looman, J. (2004). Issues in the treatment of sexual offenders: Recent developments and directions for future research. Aggression and Violent Behavior, 9, 229-246.

Andrews, D. A., & Bonta, J. (1998). The psychology of criminal conduct (2nd ed.). Cincinnati, OH: Anderson.

Barbaree, H. E., & Cortoni, E A. (1993). Treatment of the juvenile sex offender within the criminal justice and mental health systems. In H. E. Barbaree, W. L. Marshall, & S. M. Hudson (Eds.), The juvenile sex offender (pp. 243-263). New York: The Guilford Press.

Blanchette, K. (1996). Sex offender assessment, treatment and recidivism: A literature review. Ottawa: Correctional Services Canada (R-48).

Bonta, J., & Andrews, D. A. (1993). The level of supervision inventory: An overview. IARCA Journal, 5, 6-8.

Correctional Service of Canada. (2001). Standards for the provision of assessment and treatment services to sex offenders. Ottawa, Canada: Author.

Gordon, A., Marshall, W. L., Loeber, R., & Barbaree, H. E. (1977). Toward a definition of social competence in sexual aggressors. Paper presented at the National Conference of the Evaluation and Treatment of Sexual Aggressives, Memphis, TN.

Hanson, R. K. (2000). What is so special about relapse prevention? In D. R. Laws, S. M. Hudson, & T. Ward (Eds.), Remaking relapse prevention with sex offenders: A sourcebook (pp. 27-38). London: Sage.

Hanson, R. K., & Bussiere, M. T. (1998). Predicting relapse: A meta-analysis of sexual offender recidivism studies. Journal of Consulting and Clinical Psychology, 66, 348-362.

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

Hare, R. D. (1998). Psychopaths and their nature: Implications for the mental health and criminal justice systems. In T. Millon, E. Simonsen, M. Birket-Smith, & R. D. Davis (Eds.), Psychopathy: antisocial, criminal and violent behavior (pp. 188-212). New York: The Guilford Press.

Hunter, J. A. Jr., & Figueredo, J. (1999). Factors associated with treatment compliance in a population of juvenile sexual offenders. Sexual Abuse: A Journal of Research and Treatment, 11, 49-67.

Johnson, G. M., & Knight, R. A. (2000). Developmental antecedents of sexual coercion in juvenile sexual offenders. Sexual Abuse: A Journal of Research and Treatment, 12, 165-178.

Lalumiere, M. L., & Quinsey, V. L. (1996). Sexual deviance, antisociality, mating effort, and the use of sexually coercive behaviors. Personality and Individual Differences, 21, 33-48.

Mailloux, D. L., Abracen, J., Serin, R. C., Cousineau, C., Malcolm, P. B., & Looman, J. (2003). Dosage of treatment to sexual offenders: Are we overprescribing? International Journal of Offender Therapy and Comparative Criminology, 47, 171-184.

Mailloux, D. L., & Serin, R. C. (2001). Issues impacting on programming for sex offenders. Manuscript submitted for publication.

Marques, J. K., Day, D. M., Nelson, C., & West, M. A. (1994). Effects of cognitive-behavioral treatment on sex offender recidivism: Preliminary results of a longitudinal study. Criminal Justice and Behavior, 21, 28-54.

Marshall, W. L. (1993). The role of attachment, intimacy, and loneliness in the etiology and maintenance of sexual offending. Sexual and Marital Therapy, 8, 109-121.

Marshall, W. L. (1996). Assessment, treatment, and theorizing about sex offenders: Developments during the past twenty years and future directions. Criminal Justice and Behavior, 23, 162-199.

Marshall, W. L., Anderson, D., & Fernandez, Y. (Eds.). (1999). Cognitive behavioural treatment of sexual offenders. New York: John Wiley & Sons.

Marshall, W. L., & Fernandez, Y. M. (2000). Phallometric testing with sexual offenders: Limits to its value. Clinical Psychology Review, 20, 807-822.

Marx, B. P., Gross, A. M., & Adams, H. E. (1999). The effect of alcohol on the responses of sexually coercive and noncoercive men to an experimental rape analogue. Sexual Abuse: A Journal of Research and Treatment, 11, 131-145.

Miller, W. R., & Rollnick, S. (2002). Motivational interviewing (2nd ed.). New York: The Guilford Press.

Nugent, P., & Zamble, E. (2001). Affecting detention referrals through proper selection. Forum on Corrections Research, 13, 27-30.

Overholser, C., & Beck, S. (1986). Multimethod assessment of rapists, child molesters, and three control groups on behavioral and psychological measures. Journal of Consulting and Clinical Psychology, 54, 682-687.

Quinsey, v. L., Harris, G. T., Rice, M. E., & Cormier, C. A. (1998). Violent offenders: Appraising and managing risk. Washington, DC: American Psychological Association.

Segal, Z. V., & Marshall, W. L. (1985). Heterosexual social skills in a population of rapists and child molesters. Journal of Consulting and Clinical Psychology, 53, 55-63.

Serin, R. C., & Kennedy, S. (1997). Treatment readiness and responsivity: Contributing to effective correctional programming. Ottawa: Correctional Service of Canada (R-54).

Seto, M. C., Barbaree, H. E., Serin, R. C., & Malcolm, P. B. (1997, October). Applying a differentiated assessment model to treatment outcome. Paper presented at the annual meeting of the Association for the Treatment of Sexual Abusers, Arlington, VA.

Ullman, S. E., Karabatsos, G. E., & Koss, M. P. (1999). Alcohol and sexual assault in a national sample of college women. Journal of Interpersonal Violence, 14, 603-625.

Ward, T., Hudson, S. M., & McCormack, J. (1997). Attachment style, intimacy deficits, and sexual offending. In B. K. Schwartz & H. R. Cellini (Eds.), The sex offender: New insights, treatment innovations, and legal developments (Vol. 2, pp. 2.1-2.14). Kingston, NJ: Civic Research Institute.

Wong, S. (2000). Psychopathic offenders. In S. Hodgins and R. Muller-Isberner (Eds.), Violence, crime and mentally disordered offenders: Concepts and methods for effective treatment and prevention (pp. 87-112). New York: John Wiley & Sons, Ltd.

Manuscript accepted April 6, 2004

Jeffrey Abracen

Correctional Service of Canada

Donna L. Mailloux and Ralph C. Serin

Carleton University, Ottawa, Canada

C. Cousineau, P. Bruce Malcolm, and Jan Looman

Correctional Service of Canada

Address correspondence to Jeffrey Abracen, Central District Parole. Department of Psychology, 330 Keele St., Toronto, Ontario, M6P 2K7: e-mail: [e-mail].