Assessing Risk for Sexual Recidivism: How Inclusion of Psychologically Meaningful Risk Factors Can Improve Predictive Validity

(1)

PSYCHOLOGICALLY MEANINGFUL RISK FACTORS CAN IMPROVE PREDICTIVE VALIDITY

by

ALISON B. CONCANNON B.A., Creighton University, 2017

A thesis submitted to the Graduate Faculty of the University of Colorado Colorado Springs

in partial fulfillment of the requirements for the degree of

Master of Arts Department of Psychology

(2)

This thesis for the Master of Arts degree by Alison B. Concannon

has been approved for the Department of Psychology

by

Kelli J. Klebe, Chair

Robert L. Durham, Co-Chair

Andrew Lac

Frederick L. Coolidge

(3)

Concannon, Alison B. (M.A., Psychology)

Assessing Risk for Sexual Recidivism: How Inclusion of Psychologically Meaningful Risk Factors Can Improve Predictive Validity

Thesis directed by Professor Kelli J. Klebe and Robert L. Durham

ABSTRACT

The purpose of the present study is to develop a comprehensive actuarial scale designed to assess risk for sexual recidivism that will be appropriate for use with all sexual offender types. Although assessments for predicting risk for violence in general have been suggested to have reached a ceiling, this is not the case for assessments for predicting sexual recidivism. Recent research has suggested that these measures can be improved upon and that current measures are likely explaining unique aspects of risk for future sexual violence. Psychologically meaningful risk factors, which are factors that are more suggestive of the causes of reoffending (e.g., deviant sexual preferences or a

preoccupation with sex), could be of use in augmenting existing measures of risk for sexual recidivism. As such, the present study examined archival data for these

psychologically meaningful risk factors in order to identify if these factors are able to significantly improve upon a commonly used measure for assessing risk of sexual recidivism through logistic regression and examination of correlations with sexual recidivism. Results were mixed in their support for the utility of psychologically meaningful risk factors in predicting sexual recidivism. Although it appears that some factors meaningfully relate to sexual recidivism, others fail to improve upon more

traditional static factors of risk. Implications of these findings, and suggestions for future research regarding psychologically meaningful risk factors are discussed.

(4)

ACKNOWLEDGEMENTS

I want to thank Bob Durham, who supported me and helped me develop this work, and to Kelli Klebe, who was endlessly accommodating and gracious in helping me finish it. Thanks to Andrew Lac and Fred Coolidge for being wonderful additions to my committee and for their support throughout this process. Special thanks to Matt Huss for first introducing me to research and providing me with guidance and compassion

throughout my professional development all these years. And finally, to my family and friends for listening to me passionately talk about my research and always being so excited for me and proud of me.

(5)

TABLE OF CONTENTS

CHAPTER

I. INTRODUCTION ... 1

History and Background in Assessing Risk for Violence ... 2

Current State of the Field ... 2

Predictive Capability ... 3

Assessing Risk for Sexual Recidivism ... 5

Current State of the Field ... 5

Limitations in Assessing Sexual Recidivism ... 8

Purpose and Guiding Study: Psychologically Meaningful Risk Factors ... 10 Scale Development ... 12 II. METHOD ... 14 Participants ... 14 Measures ... 15 Static-99R ... 15

Psychologically Meaningful Risk Factors ... 15

Offense-Supportive Cognitions ... 19

Offense Supportive Attitudes ... 19

(6)

Lack of Emotionally Intimate Relationships with

Adults ... 20

Social Difficulties ... 20

Negative Social Influences ... 20

Affective Functioning ... 20

Emotion Management ... 20

Deviant Sexual Preferences ... 21

Sexualized Violence ... 21

Deviant Sexual Preferences ... 21

Sexual Preoccupation ... 21

Impulsivity and Self-Regulation ... 22

Impulsivity ... 22

Poor Problem Solving ... 22

Analysis Plan ... 22

III. RESULTS ... 24

Regression Models Predicting Recidivism Five Years Following Community Placement ... 26

Regression Models Predicting Recidivism Three Years Following Community Placement ... 28

Correlational Comparisons ... 32

Psychologically Meaningful Risk Factor Total ... 32

IV. DISCUSSION ... 35

REFERENCES ... 41

(7)

LIST OF TABLES

TABLE

1. Static-99R Items and Scoring ... 16 2. Psychologically Meaningful Items and Scoring ... 17 3. Omnibus Test Statistic Comparison of Regression Models ... 25 4. Hierarchical Logistic Regression Analysis Predicting Sexual Recidivism

at Five Years Using Individual Static Items ... 27 5. Hierarchical Logistic Regression Analysis Predicting Sexual Recidivism

at Five Years with Static-99R Total Score ... 29 6. Hierarchical Logistic Regression Analysis Predicting Sexual Recidivism

at Three Years Using Individual Static Items ... 30 7. Hierarchical Logistic Regression Analysis Predicting Sexual Recidivism

at Three Years with Static-99R Total Score ... 31 8. Correlations Between Risk Factors and Recidivism at Three- and Five-

(8)

INTRODUCTION

Assessing risk for future violent behavior and reoffending has long been dominant within the field of forensic psychology. However, there has been much less progress in assessing a particular subset of violent recidivism, namely, sexual recidivism.

Assessments designed to predict risk for violent recidivism have been suggested to have reached a ceiling in their predictive accuracy, in that most well-validated instruments seem to tap into the same four common factors of risk: an individual’s criminal history, irresponsible or impulsive lifestyle, criminal thoughts or attitudes, and issues related to substance use (Kroner et al., 2005; Skeem & Monahan, 2010). In contrast, assessments for sexual recidivism have lagged behind traditional violence risk assessment, and there is evidence that sexual recidivism measures can be improved upon. Different measures that assess risk for sexual recidivism seem to be tapping into different aspects of that risk, which suggests that there may be room for improvement in our understanding of risk for sexual recidivism in terms of instrument development (Buttars et al., 2015; Thornton, 2016; Thornton & Knight, 2015). To this end, this thesis will focus on the potential for psychologically meaningful risk factors to augment the predictive capability of an existing measure of risk for sexual recidivism (the Static-99R) by creating a more comprehensive actuarial scale that can be scored easily and simply requires a records review and brief scale administration. Inclusion of psychologically meaningful risk factors—those that are more suggestive of the cause of the re-offense—might provide

(9)

more insight for understanding sexual reoffending and be more inclusive of factors that could be changeable with the potential of targeting these factors in treatment to reduce risk. To this end, the history and background of violence risk assessment and assessment for sexual recidivism will be discussed, as well as further exploration of psychologically meaningful risk factors.

History and Background in Assessing Risk for Violence

Although sexual recidivism, which is when an individual that has been convicted of a sexual offense commits a new sexual offense following release to the community, may not always be violent (e.g. possession of child pornography), sexual recidivism is considered a subset of violent recidivism so examining the development of violence risk assessment is important. The nature of violence risk assessment has evolved significantly in the past few decades and has moved through several “generations” of theories and approaches (Monahan, 2003), with most approaches currently following the format of actuarial measures or structured interviews.

Current State of the Field

Skeem and Monahan (2011) commented on the state of the field, now nearly a decade ago, with some useful insights moving forward, which are still relevant today. Modern risk assessment approaches are currently largely distinguished by their structure (on a continuum from clinical judgement to actuarial approaches) and their content (the type of items may include static or historical factors and dynamic factors, which are changeable and theoretically amenable to treatment interventions). The empirical

literature has largely supported actuarial measures in their accuracy for predicting risk for future violence, whereas clinical judgement has had the least amount of empirical support

(10)

(Skeem & Monahan, 2011). In the realm of content, inclusion of dynamic factors of risk, rather than static ones, has been somewhat more controversial. Dynamic factors of risk are those that are changeable and are often psychological or behavioral in nature—for example, possessing offense supportive thoughts, beliefs, or attitudes a risk factor for recidivism, which can be remedied through interventions targeted to examine and adjust these maladaptive thinking patterns (Andrews & Bonta, 2010). Static factors, on the other hand, are those that do not change—for example, the number of prior convictions an individual has cannot be changed, short of being exonerated of charges and convictions or attaining more convictions over time (Mann et al., 2010). Advocates for inclusion of dynamic risk factors suggest that these factors might help present a more complete picture of risk that is not fully dependent upon the client’s criminal history, which cannot change. However, those that do not support the full inclusion of dynamic risk factors suggest that these factors are of more use to risk management and risk reduction, rather than the prediction of violence (Cording et al., 2016). Additionally, there has been research that suggests that both static and dynamic factors each possess predictive utility and one does not necessarily outweigh the other (Eisenberg et al., 2019). Skeem and Monahan (2011) further suggest that the field of violence risk assessment has likely reached a ceiling in instrument development, and that “The time is ripe to shift attention from predicting violence to understanding its causes and preventing its (re)currence” (p. 38).

Predictive Capability

In an investigation related to this ceiling effect, Kroner et al. (2005) sought to compare the predictive ability of the Psychopathy Checklist—Revised (PCL-R) (Hare et

(11)

al., 1988), Violence Risk Appraisal Guide (VRAG) (Harris et al., 1993), Level of Service Inventory Revised (LSI-R) (Andrews & Bonta, 1995), and General Statistical

Information on Recidivism (GSIR) (Bonta et al., 1996), to predict violent recidivism, specifically examining violations of probation or parole as well as criminal convictions, with the hypothesis that these four measures were all tapping into the same, rather than unique, construct of criminality. All four of these measures have been found to possess sufficient predictive validity with Cohen’s d values for each as follows: PCL-R (0.55), VRAG (0.68), LSI-R (0.51), and GSIR (0.67) (Yang et al., 2010). To test the hypothesis that all of these measures were tapping into the same construct of criminality, Kroner et al wrote each item from all four measures onto individual pieces of paper, placed them into a coffee can and mixed together, and then randomly drew slips to create four new measures. Thirteen items were drawn out of the coffee can to create a measure, then were replaced into the coffee can and then another thirteen items were drawn from the coffee can. This process was repeated until four measures were created. They found that the new random “coffee can” measures performed just as well as the original four instruments and, as a result, concluded that these measures were all tapping into the same factors of violence risk, a finding later supported by Skeem and Monahan (2011), who suggested that efforts should be redirected to understanding the causes of reoffending and on the reduction and management of risk for violence (Kroner et al., 2005). Although violence risk assessment may be at the stage to where instrumentation development is no longer as fruitful as it once was to improve predictive validity, including psychologically

meaningful risk factors in assessing risk for sexual recidivism may improve predictive accuracy and contribute to the advancement of our understanding of that risk.

(12)

Assessing Risk for Sexual Recidivism

Despite getting a later start than assessments for more general violent recidivism, risk assessment that is specifically designed to predict sexual recidivism has undergone massive changes, especially in the past few decades. The justice system, and targeted legislation in particular (e.g., sexually violent predator laws), has led to immense advancement of investigation related to the re-offending of sexual offenders. However, despite these recent developments, recent research has suggested that there is much more progress to be made in these assessments (Harris & Hanson, 2010).

Current State of the Field

Hanson and Morton-Bourgon (2009) conducted a meta-analysis of 118 studies to assess the predictive capability of risk assessment for sexual offenders and found that many measures currently in use could likely be improved upon. Overall, they found that actuarial measures designed to assess sexual recidivism were the most effective in predicting of sexual recidivism (Cohen’s d = 0.67). They also found that the Static-99R was the most commonly used measure with adult sexual offenders, with the Sexual Violence Risk-20 (SVR-20) falling in second place as a measure of structured professional judgment (Hanson & Morton-Bourgon, 2009). In an updated survey of current practices in the field conducted by Kelley et al. (2020), the Static-99R was still the most widely used instrument by far with 82.4% of practitioners surveyed reporting using it regularly, the STABLE-2007 in second (42.0%) the Static-2002R came in third (19.3%), and the SVR-20 fell in fourth (16.8%).

The Static-99R is the most commonly used instrument and includes just 10 static risk factors that are scored and then summed to indicate an overall level of risk (Helmus

(13)

et al., 2012; Phenix et al., 2016). The Static-2002R contains many of the same items as the Static-99R, however also includes a few additional items related to the rate of sexual offenses, community supervision violations, arrest for sexual offenses as both an adult and a juvenile, and years spent in the community prior to the index sexual offense (Helmus et al., 2012; Phenix et al., 2008). These additions were made to be more

inclusive of theoretical attributes of sexual offending, however, the Static-2002R (AUC = 0.68) fails to improve upon the predictive ability of the Static-99R (AUC = 0.68) in predicting sexual recidivism at five years following release to the community (Helmus et al., 2012). Despite the introduction of the Static-2002R over 15 years ago, the Static-99R remains dominant in both clinical practice and in research literature, likely a result of the measure only using information that is widely available and the fact that the measure can be scored by various types of professionals, rather than only by psychologists (Harris & Hanson, 2010; Helmus et al., 2012).

The STABLE-2007 is an actuarial measure of assessing risk for sexual recidivism, and tracking changes in risk level over time, and includes many dynamic factors and propensities identified by Mann et al. (2010). The STABLE-2007 is designed to look at those facets of risk that are changeable in order to prioritize the reduction of risk. However, despite the theoretical improvements made by STABLE-2007, the measure on its own does not appear to outperform the Static-99 in predicting sexual recidivism at 3 years, with the STABLE-2007 possessing an AUC value of 0.67 and a value of 0.74 for Static-99 (Hanson et al., 2007). While the Static-99R has been criticized for focusing too heavily on static factors, the STABLE-2007 appears to suffer from the opposite problem—placing too much of a focus on dynamic ones.

(14)

The SVR-20, in contrast to the Static-99R, Static-2002R, and STABLE-2007, is not an actuarial measure, but rather one of structured professional judgment. The SVR-20 was designed not only to assess recidivism, but also to be used by care providers to assist in the case management of individuals that sexually offend, as such, the SVR-20 includes more dynamic or changeable risk factors, such as resistance to supervision, the presence of major mental illness, and sexual deviation (Boer et al., 1997). However, the SVR-20 does not have clear guidelines as to how these factors should be weighed to indicate risk level, which has resulted in problematic variability in examiner scoring methods to compare across studies (Hanson & Morton-Bourgon, 2009) which may contribute to mixed findings regarding the measure’s ability to predict sexual recidivism at various time points, including 2, 5, and 10 years. (Holoyda & Newman, 2016).

Perhaps the most crucial finding of Hanson and Morton-Bourgon’s (2009) meta-analysis, however, was that measures designed to assess recidivism in general were nearly as effective at predicting sexual recidivism as the specialized instruments; actuarial measures of general recidivism (Cohen’s d = 0.62) were nearly as accurate in predicting sexual recidivism as actuarial measures designed to assess sexual recidivism (Cohen’s d = 0.67). Van den Berg and colleagues (2018) conducted a similar meta-analysis examining the predictive properties of dynamic assessments for sexual recidivism (Cohen’s d = 0.71) and found similar results as Hanson and Morton-Bourgon’s (2009) results on largely static, actuarial measures assessing sexual

recidivism. This suggests that there are rather notable limitations of existing instruments in predicting sexual recidivism, and approaches that have sought to assess solely static or

(15)

solely dynamic factors seem to be lacking—research suggests that both types of these factors contribute meaningfully to the prediction of risk (Thornton, 2016).

Limitations in Assessing Sexual Recidivism

A particularly enlightening study conducted by Buttars et al. (2015) on the

limitations of existing risk assessment measures for sexual recidivism utilized the “coffee can” approach developed by Kroner and colleagues (2005) and the disparities in findings between the two studies highlight the discrepant advancement of measures designed to assess violent recidivism versus sexual recidivism. While Kroner et al. (2005) found that all measures of violent risk assessment seemed to share the same common factors, which was further supported by the randomly generated measures performing just as well as the well-validated measures, Buttars et al. (2015) used the same methodology, however, the randomly generated measures outperformed the well-validated measures.

Buttars et al. (2015) utilized the “coffee can” methodology to examine if the same phenomenon was occurring with measures designed to assess risk for sexual

recidivism—if these measures seemed to have hit the same ceiling that measures assessing general violence risk had, with the hypothesis that measures for sexual

recidivism would not have done so. The recent meta-analysis conducted by Hanson and Morton-Bourgon (2009) had suggested that while actuarial sexual recidivism measures were the best at predicting sexual recidivism, measures that assess general recidivism performed nearly as well in predicting sexual recidivism (d = .67 for actuarial measures designed for sexual recidivism and d = 0.62 for actuarial measures designed for general recidivism). Given that sexual recidivism is a subset of violent recidivism, this suggests that both types of measures (those assessing sexual recidivism and those assessing violent

(16)

recidivism) might be tapping into the aspects of reoffending that are characteristic of both types, and therefore that measures of sexual recidivism might be lacking in specificity in terms of factors that uniquely predict sexual recidivism.

Buttars and colleagues (2015) chose the Iowa Sex Offender Risk Assessment (ISORA8), Level of Service Inventory-Revised (LSI-R), and the Static-99R for their original measures and utilized a similar coffee can approach to derive three new measures to predict sexual recidivism within 5 years. The LSI-R (Andrews & Bonta, 1995) was one of the measures used in the original coffee can study described above, and is similar to ISORA-8 in that both of these measures make supervision/risk recommendations for offenders upon release, however, the ISORA-8 was specifically designed for use with sexual offenders (Iowa Department of Corrections, 2010). The Static-99R is the most widely used measure for assessing risk for sexual recidivism and consists of 10 items representing static risk factors including those related to current and prior sexual and non-sexual offenses, age at release, and victim type (Kelley et al., 2020). The hypothesis that instruments assessing sexual recidivism had not reached the same ceiling as those for violent recidivism was supported in that the randomly generated measures outperformed the three existing and well-validated measures in predicting sexual recidivism. This appears to suggest that each of these measures are tapping into different predictive constructs for sexual recidivism, and that combining items from these various measures likely resulted in a more complete picture of an individual’s risk. The authors here concluded that “Current measures appear incomplete, but have a clear and empirically noted ability to improve” (Buttars et al., 2015, pg. 31). As a whole, these findings suggest that different measures currently in use each appear to be assessing distinct aspects

(17)

related to recidivism and our understanding of sexual recidivism could be improved by incorporating elements from these existing measures into a single measure in order to compile a more comprehensive and valid picture of risk.

Purpose and Guiding Study: Psychologically Meaningful Risk Factors

Given these findings, improvement upon measures that assess risk for sexual recidivism warrants exploration. The current dominant theoretical model in the literature, the Risk Needs Responsivity (RNR) model, has been found to be somewhat incomplete in its understanding and addressing risk for violence (Andrews & Bonta, 2010). This model’s emphasis on addressing dynamic factors of risk lends particularly well to the idea of risk management, rather than predictive assessment, in targeting dynamic factors of risk that are theoretically amenable to change (Polaschek, 2012). Although RNR has remained perhaps the most prevalent approach in practice, this model has also received criticism in that this model alone, which primarily focuses on dynamic risk factors as opposed to static ones, has not been found to be sufficient in explaining re-offending, predicting risk, or treatment adherence, and at the very least, that there is a need for further empirical and theoretical research as to the appropriateness of applying dynamic factors to these domains (Cording et al., 2016; Heffernan et al., 2019; Thornton, 2016). Existing measures and models of risk, including both static and dynamic factors, have been criticized for neglecting theoretical conceptualizations of risk and recidivism in favor of empirical correlation to recidivism (Ward & Beech, 2014). In the domain of predictive assessment, combining static and dynamic factors may be a better approach as findings suggest that static risk factors serve as more than simply a past record of

(18)

approach pursued in this study. This inclusion of risk factors that are more causal in nature has the potential to add more nuance to predicting recidivism, as well as create a measure that is more in line with theories of offending and criminal behavior in

considering the motivations of the offender, and is also more consistent with approaches to treatment following assessment (Marshall, 2018).

The psychologically meaningful risk factors proposed by Mann et al. (2010) are not exclusively dynamic or static, but rather include those factors that have been found to have the potential to explain the reason for the higher risk and represent what they refer to as propensities or traits that tend to be enduring, but at any given time may or may not contribute to risk. For example, a sexual attraction to children is traditionally understood to be a fairly stable trait, and a risk factor for sexual recidivism, but this is amenable to intervention focused on other strengths or to be targeted in the course of therapy to manage that risk (Mann et al., 2010). Mann et al. identify several psychologically meaningful risk factors of interest, however, it is important that these factors also be representative of the field’s theoretical understanding of the nature of sexual offending. Ward and Beech (2014) takes this position and criticized recent efforts to include

dynamic risk factors, including those of Mann et al. (2010), for not having “theoretically reflected on the nature of dynamic risk factors and their role in theory construction” (Ward & Beech, 2014, pg. 4). Given Ward and Beech’s (2014) criticism of many existing measures lacking theoretical grounding in the nature of sexual offending, in this study psychologically meaningful risk factors identified by Mann et al. (2010) will be limited to those factors reflective of deficits identified in the clinical presentation of sexual

(19)

deficits often present in individuals that sexually offend, namely cognitions supportive of offending, intimacy problems, social difficulties, problems with affective functioning, deviant sexual preferences, sexual preoccupation, and issues with impulsivity and self-regulation. Cognitions supportive of offending are those that are congruent with sexual offending and a criminal lifestyle (Ward & Beech, 2014). Intimacy problems are

concerned with the extent to which individuals are able to maintain healthy relationships with adults. Social difficulties are often combined with intimacy issues, but reflect an individual’s ability to maintain non-romantic relationships with individuals that are not involved in criminal activity and are supportive of the individual maintaining a healthy lifestyle that does not involve criminal activity (Casey, 2016). Problems with affective functioning are concerned with the ability of the individual to manage negative emotional states. Deviant sexual preferences are concerned with preferences and sexual arousal associated with sexual offending and especially with inappropriate targets of that arousal, such as children or animals (Ward & Beech, 2014). Sexual preoccupation is concerned with the extent to which the individual is focused on their sexual arousal, offending, and behaviors (Mann et al., 2010; Ward & Beech). Issues of impulsivity and self-regulation are concerned with the extent the individual is able to live a stable life and consider choices and their consequences (Casey, 2016; Ward & Beech, 2014).

Scale Development

This study will take an actuarial approach of scale development given that actuarial measures have been found to be the most accurate in predicting recidivism in general as well as for sexual recidivism (Hanson & Morton-Bourgon, 2009). Scale items were selected from the psychologically meaningful risk factors in the guiding study of

(20)

Mann et al. (2010) that matched Ward and Beech’s (2014) and Casey’s (2016) characterization of the clinical attributes of sexual offenders, as well as items derived from the Static-99R as a baseline for comparison, given that this has been identified as the most commonly used assessment for assessing risk for sexual recidivism (Kelley et al., 2020). Each of these components and how they are measured will be specified in detail in the method section and all will be investigated for inclusion in the scale.

Factors identified by Mann et al. (2010) that map onto Ward and Beech’s

identified clinical attributes of sexual offenders will be investigated for their incremental predictive utility, including sexual preoccupation/addiction, deviant sexual interests, sexualized violence, offense supportive attitudes, a lack of emotionally intimate

relationships with adults, lifestyle impulsiveness, poor problem solving, negative social influences, and an emotion management variable that captures several variables related to emotional regulation. It is hypothesized that the presence of sexual preoccupation and addiction, deviant sexual interests, sexualized violence, offense supportive attitudes, a lack of emotionally intimate relationships with adults, lifestyle impulsiveness, poor problem solving, negative social influences, and an absence of ability to manage and regulate negative emotions will be predictive of sexual recidivism at both the 5 and 3 year time points.

It is hypothesized that these psychologically meaningful risk factors will predict sexual recidivism above and beyond that of the Static-99R items. The scale will be analyzed to assess the validity of the model in predicting sexual recidivism at time points of 5 years and also to potentially eliminate items of the scale that fail to significantly contribute to predictive validity.

(21)

METHOD Participants

Archival data were accessed from the Inter-university Consortium for Political and Social Research (ICPSR) for analysis by the researchers from a study by McGrath, Lasher, and Cumming (2014). All participants had been convicted of at least one sexual offense, enrolled in a community-based sex offender treatment program between 2001 and 2007, and the time they started the treatment program to the end of the study (2010) was at least 3 years. This selection criteria resulted in a sample of 759 male offenders ranging in age from 18 to 75 years. 96.4% of the sample is Caucasian, 1.3% African American, 2.2% that identified their race as a category not represented. Offender type in this sample varied: 51.1% were categorized as female child molesters, 18.1% as adult rapists, 14.6% as incest offenders, 8.4% as male child molesters, and 7.8% as non-contact offenders. There were no child pornography offenders or statutory rapists included in this sample. Participants were followed up at fixed intervals to determine recidivism.

Recidivism in this study is operationally defined as any new charges for sexual offenses following release and was assessed at two time points: 3 years and 5 years. The dataset was assessed for issues related to missing data for required variables. Participants that did not have complete data for all variables were dropped for analyses.

(22)

Measures

Static-99R

Given its prevalence among practitioners, the Static-99R will be used as a baseline from which to compare results. The Static-99R consists of 10 items that pertain to an individual’s offense history (sexual and non-sexual), characteristics of victims reflected in their conviction history, and various demographics. Rather than using the total score for baseline comparison, all risk factors will be taken individually as variables entered into a regression. Static-99R items and their original scoring (Phenix et al., 2016), as well as adjusted scoring for analysis, are presented Table 1. Scoring for the age

variable will be treated as continuous for this study, as there are not enough participants in the sample for a sufficient sample size for all categorical groups.

Psychologically Meaningful Risk Factors

Psychologically meaningful risk factors identified by Mann et al. (2010) will be investigated for inclusion in this scale—factors will be mapped onto traits of sexual offenders identified by Ward and Beech (2014) and Casey (2016). Information related to each risk factor was collected by a treatment provider and then later coded and scored by researchers. Given that this study utilizes archival data, details on how variables that represent each risk factor are operationally defined and measured within this sample are discussed within the clinical attribute they represent: cognitions supportive of offending, lack of intimacy, social difficulties, problems with affective functioning, deviant sexual preferences, sexual preoccupation, and issues with impulsivity and self-regulation. Items, along with their original and adjusted scoring, are presented in Table 2.

(23)

Static-99R Items and Scoring

Variable Name Operational Definition Values/Levels Age Age at time of release from index sexual

offense 1 = 18 to 34.0 0 = 35 to 39.9 -1 = 40 to 59.9 -3 = 60 or older *continuous Intimate relationships Has the individual ever lived with a lover for at

least two years?

0 = yes 1 = no Index non-sex violent

convictions

Does the individual have any non-sexual violent convictions in their index offense?

0 = no 1 = yes Prior non-sex violent

convictions

Does the individual have any non-sexual violent convictions prior to their index offense?

0 = no 1 = yes Prior sex offense How many prior sexual offense charges and

convictions does the individual have excluding the index offense? (Whichever column results in the higher final score is the column that is used.) Charges None 1-2 3-5 6+ Convictions None 1 2-3 4 0 = none 1 = 1 2 = 2-3 3 = 4 *continuous 4+ prior sentencing dates Does the individual have four or more prior

sentencing dates excluding their index offense?

0 = 3 or less 1 = 4 or more Non-contact sex offenses Does the individual have any convictions for

non-contact sexual offenses?

0 = no 1 = yes Unrelated victims Does the individual have any unrelated victims

in their offense history?

0 = no 1 = yes Stranger victims Does the individual have any stranger victims in

their offense history?

0 = no 1 = yes Male victims Does the individual have any male victims in

their offense history?

0 = no 1 = yes

Note: Descriptions of items and scoring taken from Phenix et al. (2016). * Indicates

(24)

Table 2

Psychologically Meaningful Items and Scoring

Variable Name Operational Definition Values/Levels Offense supportive

attitudes +

The degree to which the individual recognizes and self-corrects his attitudes and thoughts that support or condone general criminal and rule-breaking behavior.

0 = minimal or no difficulty making corrections

1 = some difficulty

2 = considerable difficulty 3 = does not make corrections *continuous Lack of emotionally intimate relationships with adults +++

Whether the individual is involved in a committed adult love relationship, and if so, how the relationship is functioning.

0 = stable adult love relationship 1 = moderately stable adult love relationship

2 = moderately unstable adult love relationship

3 = no stable adult love relationship

*continuous Negative social

influences +

The degree to which an individual’s social influences are positive or negative.

0 = primarily positive

1 = more positive than negative 2 = more negative than positive 3= primarily negative

*continuous Emotion

management +

The degree to which an individual is able to manage negative emotional states, such as: loneliness, anger, anxiety, hostility,

depression, jealousy, and resentment.

0 = no emotion management problems

1 = minor emotion management problems

2 = moderate emotion management problems

3 = serious emotion management problems

*continuous Sexualized

violence: Presence of force ++

Indicates the presence of force used during the index sexual offense. If more than one victim, score the most force used during the index sexual offenses.

*0 = non-contact offense 1 = contact offense

2 = force greater than necessary to gain compliance or clear threats of physical harm to victim or others 3 = use of potentially deadly weapon

*continuous

(25)

Table 2 (cont.)

Variable Name Operational Definition Values/Levels Sexualized

violence: Physical harm ++

Indicates whether physical harm was caused to the index sexual offense victim. If more than one victim, score the most physical harm caused to a victim

*0 = no medical treatment required 1 = injury not requiring formal medical attention

2 = treated for injury and released 3 = hospitalized

4 = death resulting *continuous Deviant sexual

preferences +

The type of partners and behaviors the individual finds sexually arousing.

0 = all sexual interests in appropriate themes

1 = most sexual interests in appropriate themes

2 = most sexual interests in offense-related themes

3 = all sexual interests in offense-related themes

*continuous Sexual

preoccupation ++

The degree to which an individual displays offense related sexual fixation

0 = single victim and history of consenting, age appropriate sexual relationships

2 = two to four victims and history of consenting, age appropriate sexual relationships

3 = five or more victims and/or little or no history of consenting, age appropriate sexual

relationships *continuous Impulsivity:

Employment Stability +

The degree to which an individual maintains full, satisfying, and stable employment (at least 35 hours per week), if a

student, maintaining studies and a full time course schedule.

0 = full time employment or school with stability and general satisfaction or productive

1 = full time employment with moderate or greater dissatisfaction or 2 job changes or part time or seasonal employment or school 2 = 3 or more job changes or unemployed more than 50% of the time

3 = unemployed or unproductive *continuous

(26)

Offense-Supportive Cognitions

Offense Supportive Attitudes. Offense supportive attitudes were operationally

defined as possessing attitudes consistent with criminality and criminal lifestyle. This risk factor was derived from and measured consistent with the Sex Offender Treatment

Intervention and Progress Scale (SOTIPS) coding manual (McGrath, Cumming, & Lasher, 2013). This variable was treated as continuous for analysis.

Table 2 (cont.)

Variable Name Operational Definition Values/Levels Impulsivity:

Residence Stability +

The degree to which the individual’s accommodation is stable and satisfying.

0 = not more than one address change and satisfied with residence

1 = two address changes or somewhat dissatisfied with residence

2 = three or more address changes or very dissatisfied with residence 3 = no fixed address

*continuous Impulsivity:

Impulsive Behavior +

The degree to which the individual’s behavior is impulsive (e.g. disregards obligations, does not consider consequences). 0 = not impulsive 1 = occasionally impulsive 2 = frequently impulsive 3 = regularly impulsive *continuous Poor problem solving +

The degree to which an individual is able to identify and solve life problems (e.g. responding to emergencies getting a job, maintaining relationships, etc.).

0 = no problem-solving deficits 1 = some problem-solving deficits 2 = considerable problem-solving deficits

3 = serious impairment *continuous

Note:+ Indicates item was derived from SOTIPS, ++ Indicates item was derived from VASOR, +++ Indicates item was derived from SO TX Needs and Progress Scale, * Indicates adjustments made to variable coding for analysis in the present study.

(27)

Lack of Intimacy

Lack of Emotionally Intimate Relationships with Adults. A lack of

emotionally intimate relationships with adults was operationally defined as having either no stable adult intimate relationships in the past six months, or if relationships possessed serious problems or were instable. This risk factor was derived from and measured consistent with the Sex Offender Treatment (SO TX) Needs and Progress Scale— Research Version (2003) coding manual (McGrath & Cumming, 2003). This variable was treated as continuous for analysis.

Social Difficulties

Negative Social Influences. This variable was assessed by the treatment provider.

Social influences were determined to be negative if those individuals possessed an antisocial lifestyle, are not aware of sexual offending behavior, do not support efforts to live a crime-free lifestyle, or do not take risk management efforts seriously. At a contrast, positive social influences were identified as those that lead a prosocial lifestyle, are aware of the sexual offending behavior, support efforts to live a crime-free lifestyle, and take risk management efforts seriously. This risk factor was derived from and measured consistent with the SOTIPS coding manual (McGrath, Cumming, & Lasher, 2013). This variable was treated as continuous for analysis.

Affective Functioning

Emotion Management. This variable was assessed by the treatment provider and

measures the degree to which the individual is determined to be able to cope with and manage negative emotional states, such as loneliness, anger, anxiety, depression, and hostility. This risk factor was derived from and measured consistent with the SOTIPS

(28)

coding manual (McGrath, Cumming, & Lasher, 2013). This variable was treated as continuous for analysis.

Deviant Sexual Preferences

Sexualized Violence. Sexualized violence was measured by the presence of force

used during the sexual offense and/or by whether physical harm was caused to the victim. This risk factor was derived from and measured consistent with the Vermont Assessment of Sex Offender Risk-2 (VASOR) coding manual (McGrath, Hoke, & Lasher, 2013). Possessing either of these risk factors qualifies an individual for this risk factor. This variable was treated as continuous for analysis.

Deviant Sexual Preferences. Deviant sexual interests were assessed by the

treatment provider and measures the extent to which the individual engages in deviant sexual behavior, attitudes, and interests. “Deviancy” is determined by behavior, attitudes, or interests that include offense-related themes, such as coercive sex, inappropriate subjects (e.g., children), viewing others as objects for sexual pleasure, hostility toward women, or viewing oneself as emotionally congruent with children rather than adults, among others. This risk factor was derived from and measured consistent with the

SOTIPS coding manual (McGrath, Cumming, & Lasher, 2013). This variable was treated as continuous for analysis.

Sexual Preoccupation

Sexual Preoccupation. Sexual preoccupation and addiction will be assessed by

the treatment provider and measures whether the individual has five or more sexual offense victims, displays greater levels of arousal to offense-related themes rather than appropriate ones, or displays minimal to no history of consensual sexual relationships.

(29)

This risk factor was derived from and measured consistent with the VASOR coding manual (McGrath, Hoke, & Lasher, 2013). This variable was treated as continuous for analysis.

Impulsivity and Self-Regulation

Impulsivity. Lifestyle impulsiveness was measured by whether or not the

individual had demonstrated an absence of employment stability, demonstrated residence instability, or demonstrated impulsive behavior as identified by the treatment provider, including disregarding obligations, changing plans suddenly, and failure to consider consequences. Impulsivity, employment instability, and residence instability were derived from and measured consistent with the SOTIPS coding manual (McGrath, Cumming, & Lasher, 2013). These variables were treated as continuous for analysis.

Poor Problem Solving. Poor problem-solving skills were assessed by the

treatment provider and measure the extent to which the individual is able to address life problems, such as being able to find a job, build and maintain relationships, and establish community support and relationships. This risk factor was derived from and measured consistent with the SOTIPS coding manual (McGrath, Cumming, & Lasher, 2013). This variable was treated as continuous for analysis.

Analysis Plan

The relative contribution of predictors in accounting for variance in sexual recidivism will be evaluated using a two-step binary logistic regression model. The outcome of sexual recidivism was classified as a binary (yes/no) outcome for any new sexual offense charges within a 5-year time period after treatment began. The predictors estimated in each block of the model were entered as follows. The Static-99R baseline

(30)

items will be entered in Step 1. All psychologically meaningful risk factors will then be entered in Step 2. The parameter used to determine whether each factor represents a unique predictive contribution will be statistical significance, evaluated at an alpha level of .05, and the odds ratio. Odds ratios can range from 0 to +∞, with a value less than 1 indicating that the target event (in this case, sexual recidivism) is less likely to happen than not recidivating, and a value greater than 1 indicating that the target event is more likely to occur than the alternative outcome. Although it has been found that OR values of 1.68, 3.47, and 6.71 are equivalent to small (0.20), medium (0.50), and large (0.80) Cohen’s d effect sizes (Chen et al., 2010), in examining research specific to sexual reoffending, Hanson et al. (2012) found that an increase of 1 point on the Static-99R total score was approximately equivalent to an odds ratio of 1.40, so this metric, and its

inverse of 0.70, will be used to include and eliminate items from the scale. Odds ratios will be evaluated in the final model in order to control for all other predictors in the model (Lipsey & Wilson, 2001). Given that binary logistic regression does not yield a true multiple R value, the Nagelkerke R2_{, a pseudo R}2 _{that approximates the percentage}

of variance explained by the model, will be used to determine the incremental predictive validity of the second block (Norusis, 2003; Warner, 2013).

(31)

RESULTS

Descriptive information reveals that for the full sample of 759 participants, a total of 35 participants had a sexual re-offense within 3 years of community placement. The full sample also had follow-up information for sexual recidivism 5 years after community placement, of those, 43 had recidivated. In terms of Static-99R scores, 28% of individuals were assigned a low risk classification, 44% were determined to be of low to moderate risk, 22% as posing moderate to high risk, and 6.5% as high risk for sexual recidivism. For the 35 individuals that recidivated within 3 years, 11% had been assigned a low risk classification, 26% as posing low to moderate risk, 31% as moderate to high risk, and 31% as posing high risk for sexual recidivism. In terms of offender type, 17% were classified as rapists, 17% as non-contact offenders, 9% as incest offenders, 46% as female child molesters, and 11% as male child molesters. For victim type, 71% offended only against children, 20% only against adults, and 9% offended both adults and children. For the 43 individuals that recidivated within 5 years, 12% had been assigned a low risk classification, 33% as posing low to moderate risk, 28% as moderate to high risk, and 28% as posing high risk for sexual recidivism. In terms of offender type, 16% of these individuals were classified as rapists, 16% as non-contact offenders, 9% as incest offenders, 49% as female child molesters, and 9% as male child molesters. For victim type, 72% offended only against children, 19% only against adults, and 9% against both children and adults.

(32)

For all analyses, participants that did not have information for all variables were not included, which resulted in a sample of 606 individuals for all regression analyses. Statistical assumptions for logistic regression were all evaluated. The outcome variable for all logistic regression analyses is dichotomous, exhaustive and mutually exclusive, and scores on the outcome variable are statistically independent of each other. The outcome variable of recidivism fulfills all of these requirements, and frequency of the recidivism outcome variable is described with each model. Multicollinearity was assessed for each model and was not violated in any model examined, with tolerance values

ranging from .46 to .86. Table 3 contains all the omnibus tests for each of the regression models discussed below for comparison.

Table 3

Omnibus Test Statistic Comparison of Regression Models

Static-99R Sexual

Recidivism

Step χ² df Nagelkerke R²

Individual Items 5 years 1 27.72* 10 .12

2 20.13 12 .20

Full 47.85* 22

Total Score 5 years 1 22.46* 1 .10

2 19.99 12 .18

Full 42.45* 13

Individual Items 3 years 1 32.62** 10 .15

2 14.19 12 .22

Full 46.81* 22

Total Score 3 years 1 27.89** 1 .13

2 15.91 12 .20 Full 43.80** 13 Total Score + Total PMRF Score 5 years 1 22.46** 1 .10 2 3.08 1 .11 Full 25.54** 2

(33)

Regression Models Predicting Recidivism Five Years Following Community Placement

A hierarchical binary logistic regression was conducted to investigate the

relationship between Static-99R risk factors and psychologically meaningful risk factors on sexual recidivism at 5 years, and to examine incremental predictive validity of psychologically meaningful risk factors over the Static-99R baseline. The outcome variable had a binary classification, with 39 participants recidivating within 5 years, and 567 not recidivating within that timeframe. Table 4 includes the following statistics for each variable included in each step of the regression model: B, SE, OR, and the statistical significance of each predictor. In the first step of the regression model consisting of the 10 individual Static-99R baseline items, the set of predictors did significantly predict sexual recidivism, χ²(10, N = 606) = 27.72, p < .05, with prior sexual offenses uniquely predicting sexual recidivism (OR = 1.97, p < .05). Nagelkerke R² = .12. In the second step of the regression model psychologically meaningful risk factors were added. The second step did not significantly predict sexual recidivism at 5 years, χ²(12, N = 606) = 20.13, p > .05, however, the full model did significantly predict sexual recidivism, χ²(22, N = 606) = 47.85, p < .05, Nagelkerke R² = .20. Only sexual preoccupation (OR = 1.17, p < .05) was a statistically significant predictor of sexual recidivism, although does not have an odds ratio meeting the 1.40 criterion. Although not statistically significant predictors of recidivism, the following Static-99R variables possess an odds ratio of greater than 1.40: intimate relationships (OR = 1.49, p > .05), index non-sexual violent convictions (OR = 1.80, p > .05), prior non-sexual violent convictions (OR = 1.46, p > .05), prior sexual offenses (OR = 1.54, p > .05), 4+ prior sentencing dates (OR = 1.45, p > .05), and

(34)

the presence of stranger victims (OR = 1.41, p > .05); as well as two PMRF variables:, residence instability (OR = 1.45, p > .05), and poor problem solving (OR = 1.54, p > .05). Additionally, the presence of unrelated victims (OR = 0.58, p > .05), male victims (OR = 0.55, p > .05), and PMRF employment instability (OR = 0.68, p > .05) all possessed an odds ratio of less than 0.70.

Table 4

Hierarchical Logistic Regression Analysis Predicting Sexual Recidivism at Five Years Using Individual Static Items

Step 1 Step 2 Block Predictor B SE OR B SE OR Static-99R Baseline Age 0.26 0.21 1.30 0.29 0.22 1.34 Intimate relationships 0.48 0.40 1.62 0.40 0.44 1.49 Index non-sex violent convictions 0.45 0.61 1.57 0.59 0.68 1.80 Prior non-sex violent convictions 0.53 0.45 1.70 0.38 0.48 1.46 Prior sex offense 0.68* 0.26 1.97 0.43 0.31 1.54 4+ prior sentencing dates 0.23 0.46 1.26 0.37 0.49 1.45 Non-contact sex offenses 0.63 0.53 1.88 0.021 0.68 1.021 Unrelated victims -0.38 0.51 0.68 -0.54 0.55 0.58

Stranger victims 0.44 0.47 1.55 0.34 0.53 1.41

Male victims 0.00 0.53 1.00 -0.60 0.61 0.55

Nagelkerke R² = .12

PMRF Offense supportive attitudes 0.16 0.25 1.18

Lack of emotionally intimate relationships with adults

-0.01 0.21 0.99

Negative social influences -0.28 0.27 0.75

Emotion management -0.02 0.30 0.98

Sexualized violence: Presence of force -0.04 0.07 0.96 Sexualized violence: Physical harm -0.05 0.07 0.96

Deviant sexual preferences 0.22 0.25 1.25

Sexual preoccupation 0.16* 0.07 1.17

Impulsivity: Employment stability -0.39 0.22 0.68

Impulsivity: Residence stability 0.37 0.23 1.45

Impulsivity: Impulsive behavior 0.07 0.29 1.07

Poor problem solving 0.43 0.30 1.54

Nagelkerke R² = .20 Note: *p < .05 **p < .01.

(35)

This same logistic regression model was used to examine sexual recidivism at 5 years using the Static-99R total score, rather than individual predictors, in the first step of the model as a baseline from which to compare the incremental predictive validity of psychologically meaningful risk factors. The outcome variable had a binary

classification, with 39 recidivating within 5 years, and 567 not recidivating within that timeframe. Table 5 includes the following statistics for each variable included in each step of the regression model: B, SE, OR, and the statistical significance of each predictor. The first step of the regression model consisting of the Static-99R total score significantly predicted sexual recidivism, χ²(1, N = 606) = 22.46, p < .05, with the Static-99R total score possessing an odds ratio of 1.49 (p < .05), Nagelkerke R² = .10. In the second step of the regression model psychologically meaningful risk factors were added. The second block on its own did not significantly predict sexual recidivism, χ²(12, N = 606) = 19.99, p > .05, however, the full model was significantly predictive of sexual recidivism at 5 years, χ²(13, N = 606) = 42.45, p < .05, Nagelkerke R² = .18. The Static-99R total score (OR = 1.30, p < .05) and sexual preoccupation (OR = 1.14, p < .05) both significantly predicted the outcome of sexual recidivism, but do not exceed an OR of 1.40. Although not a statistically significant predictor, residence instability did have an odds ratio of over 1.40 in the final model (OR = 1.48, p > .05).

Regression Models Predicting Recidivism Three Years Following Community Placement

This model was also used to investigate the relationship between Static-99R risk factors and psychologically meaningful risk factors on sexual recidivism at 3 years. The outcome variable had a binary classification, with 33 recidivating within 3 years and 573

(36)

not recidivating within that timeframe. Table 6 includes the following statistics for each variable included in each step of the regression model: B, SE, OR, and the statistical significance of each predictor. In the first step of the regression model consisting of the Table 5

Hierarchical Logistic Regression Analysis Predicting Sexual Recidivism at Five Years with Static-99R Total Score

Step 1 Step 2

Block Predictor B SE OR B SE OR

Static-99R Baseline

Static-99R total score 0.40** 0.09 1.49 0.26* 0.11 1.30 Nagelkerke R² = .10

0.02 0.20 1.02

Emotion management 0.04 0.33 1.05

Sexualized violence: Presence of force

-0.03 0.06 0.97

Sexualized violence: Physical harm -0.03 0.06 0.97

Static-99R baseline items, the set of predictors significantly predicted sexual recidivism, χ²(10, N = 606) = 32.62, p < .001, with prior sexual offenses as the only significant predictor (OR = 1.91, p < .05), Nagelkerke R² = .15. In the second step of the regression model psychologically meaningful risk factors were added. The second block on its own did not significantly predict sexual recidivism, χ²(12, N = 606) = 14.19, p > .05, and there were no significant predictors in the full model, however, the full model as a whole did significantly predict sexual recidivism, χ²(22, N = 606) = 46.81, p < .05, Nagelkerke R² = .22. Although not statistically significant, the following Static-99R variables possessed an odds ratio over 1.40: intimate relationships (OR = 1.87, p > .05), index non-sexual violent

(37)

convictions (OR = 2.03, p > .05), prior sexual offenses (OR = 1.61, p > .05), 4+ prior sentencing dates (OR = 1.71, p > .05), the presence of stranger victims (OR = 1.66, p > .05); as well as PMRF variable poor problem solving (OR = 1.61, p > .05).

Table 6

Hierarchical Logistic Regression Analysis Predicting Sexual Recidivism at Three Years Using Individual Static Items

Step 1 Step 2 Block Predictor B SE OR B SE OR Static-99R Baseline Age 0.14 0.22 1.14 0.15 0.23 1.17 Intimate relationships 0.69 0.44 2.00 0.63 0.47 1.87 Index non-sex violent convictions 0.65 0.63 1.91 0.71 0.70 2.03 Prior non-sex violent convictions 0.41 0.50 1.51 0.28 0.53 1.33 Prior sex offense 0.65* 0.27 1.91 0.48 0.32 1.61 4+ prior sentencing dates 0.39 0.50 1.47 0.54 0.53 1.71 Non-contact sex offenses 0.69 0.54 1.99 0.03 0.73 1.03 Unrelated victims 0.13 0.67 1.14 0.00 0.69 1.00 Stranger victims 0.61 0.49 1.85 0.51 0.54 1.66

Male victims 0.27 0.55 1.31 -0.21 0.63 0.81

Nagelkerke R² = .15

0.02 0.22 1.02

-0.09 0.08 0.91

Sexualized violence: Physical harm -0.02 0.07 0.99

Sexual preoccupation 0.12 0.07 1.13

This model was also used to investigate the relationship between the Static-99R total score and psychologically meaningful risk factors on sexual recidivism at 3 years. The outcome variable had a binary classification, with 33 recidivating within 3 years and 573 not recidivating within that timeframe. Table 7 includes the following statistics for each variable included in each step of the regression model: B, SE, OR, and the statistical significance of each predictor. The first step of the regression model consisting of the

(38)

Static-99R total score significantly predicted sexual recidivism, χ²(1, N = 606) = 27.89, p < .001, with the Static-99R total score possessing an odds ratio of 1.62 (p < .001),

Nagelkerke R² = .13. In the second step of the regression model psychologically meaningful risk factors were added. The second block on its own did not significantly predict sexual recidivism, χ²(12, N = 606) = 15.91, p > .05, however, the full model as a whole did significantly predict sexual recidivism, χ²(13, N = 606) = 43.80, p < .001, with the Static-99R total score (OR = 1.42, p < .05) and sexual preoccupation (OR = 1.13, p < .05) both uniquely predicting sexual recidivism. Nagelkerke R² = .20. In addition to the Static-99R total score, poor problem solving also possessed an odds ratio greater than 1.40 (OR = 1.63, p > .05).

Table 7

Hierarchical Logistic Regression Analysis Predicting Sexual Recidivism at Three Years with Static-99R Total Score

Step 1 Step 2

Block Predictor B SE OR B SE OR

Static-99R Baseline

Static-99R total score 0.48** 0.10 1.62 0.35** 0.12 1.42 Nagelkerke R² = .13

0.04 0.21 1.04

-0.07 0.07 0.93

Sexualized violence: Physical harm

-0.01 0.06 0.99

Impulsivity: Employment stability -0.34 0.22 0.71 Impulsivity: Residence stability 0.30 0.24 1.35

(39)

Correlational Comparisons

Given the overall significance of these models, but comparative lack of significant predictors, correlations were examined for each predictor and the outcome of recidivism at 3 and 5 years. Correlations between all variables (individual Static-99R items, total Static-99R score, and all psychologically meaningful risk factors) and recidivism at both time points are presented in Table 8. For binary predictors, phi was used as the

correlation coefficient. One binary variable, the presence of unrelated victims, had at least one cell count that was less than 5, and results involving this variable should be interpreted with extreme caution, but are included for comparison to the established literature using the Static-99R risk factors. For continuous predictors, Spearman’s r was used as the correlation coefficient. The following predictors were significantly correlated with sexual recidivism at both the 3 and 5 year follow up time frame: prior sex offense, prior non-sexual violent convictions, 4+ prior sentencing dates, non-contact sexual offenses, the presence of stranger victims, offense supportive attitudes, the presence of force in the index offense, deviant sexual preferences, sexual preoccupation, impulsive behavior, and poor problem solving. The following variables were significantly

correlated with recidivism at 5 years but were not significantly correlated at the 3 year follow up: emotion management and residence instability.

Psychologically Meaningful Risk Factor Total

Examining the meaningful odds ratios of psychologically meaningful risk factors from the logistic regression models and significant correlations with sexual recidivism resulted in selection of the following risk factors for an exploratory cumulative

(40)

Table 8

Correlations Between Risk Factors and Recidivism at Three- and Five-Year Time Points

Variable Name 3 year Recidivism 5 year Recidivism

Static-99R

Age + .04 .06

Intimate relationships ++ .07 .07

Index non-sex violent convictions ++ .07 .05

Prior non-sex violent convictions ++ .07* .07*

Prior sex offense + .17** .17**

4+ prior sentencing dates ++ .10** .09*

Non-contact sex offenses ++ .14** .13**

Unrelated victims ++ .07 .04

Stranger victims ++ .14** .12*

Male victims ++ .01 -.01

Static-99R total score + .18** .16**

Psychologically Meaningful Risk Factors

Offense supportive attitudes + .09* .10*

Lack of emotionally intimate relationships with adults + .04 .04

Negative social influences + .05 .06

Emotion management + .08 .08*

Sexualized violence: Presence of force + -.10** -.09*

Sexualized violence: Physical harm + .00 -.02

Deviant sexual preferences + .08* .10*

Sexual preoccupation + .16** .17**

Impulsivity: Employment stability + .03 .03

Impulsivity: Residence stability + .07 .10*

Impulsivity: Impulsive behavior + .09* .09*

Poor problem solving + .11** .11*

Note: *Indicates p < .05, **Indicates p < .01; For correlation, + indicates Spearman’s r was used, ++ Indicates phi.

(41)

impulsive behavior, poor problem solving, emotion management, residence instability, the presence of physical force used during the index offense, and sexual preoccupation. All variables were coded as binary classifications to reflect no deficits (0) or any

impairment present (1) and recidivism was examined at 5 years. As such, the final score has a possible range of 0 to 8, with the sample possessing a mean of 5.29 and a standard deviation of 1.81. The total PMRF score was significantly correlated with sexual

recidivism at 5 years, Spearman’s r = .31, p < .001. A logistic regression model was then used to examine the incremental predictive ability of the PMRF total score to predict sexual recidivism at 5 years over the Static-99R total score and is included in Table 3 for comparison to other logistic regression models. The first step of the regression model consisting of the Static-99R total score significantly predicted sexual recidivism, χ²(1, N = 606) = 22.46, p < .001, with the Static-99R total score possessing an odds ratio of 1.49 (p < .001), Nagelkerke R² = .10. In the second step of the regression model the PMRF total score was added. The second block on its own did not significantly predict sexual recidivism, χ²(1, N = 606) = 3.08, p = .08, however, the full model as a whole did

significantly predict sexual recidivism, χ²(2, N = 606) = 25.54, p < .001, Nagelkerke R² = .11. The Static-99R total score (OR = 1.42, p < .05) did significantly predict sexual recidivism, however the PMRF total score (OR = 1.22, p = .94) was not a significant predictor and did not possess an odds ratio exceeding 1.40.

(42)

DISCUSSION

In seeking to improve upon the predictive accuracy of the Static-99R by including psychologically meaningful risk factors, the present study had mixed results. Overall, the only statistically significant psychologically meaningful risk factor that was found to be predictive of sexual recidivism in regression models was sexual preoccupation. In

examining odds ratios for determinations of contribution, sexual preoccupation, residence instability, and poor problem solving were notable. Also notable was that employment instability was found to possess a rather low odds ratio, which would suggest that this item could be implemented as a protective factor, similar to older age on the Static-99R, to reflect the items relation to an absence of recidivism. However, as a block in the regression, psychologically meaningful risk factors were not significantly predictive of recidivism at either time point. For the Static-99R baseline, in the regression models, prior sexual offenses and the Static-99R total score were found to be the only statistically significant psychologically meaningful predictors. The logistic regression models were often significant as a whole, however, very few predictors were found to be statistically significant, which may be influenced by the overall low base rate of sexual recidivism (Buttars et al., 2015; Hanson et al., 2012) and the large number of predictors in the model. Although both of these attributes may have contributed to lackluster findings in logistic regression, examining correlations suggests that many factors, both static and psychologically meaningful, possess small, but statistically significant, correlations to

(43)

recidivism that can provide more nuance as to these relations.

When examining individual correlations between risk factors and recidivism at both time points, many factors possess small, but statistically significant, correlations with recidivism, although these findings may be influenced by the large sample size, the effect sizes are also small, suggesting that these are more likely to be meaningful and not just statistically significant or overpowered. For the psychologically meaningful risk factors, the only notable exceptions to this are the absence of a relationship between social relationships (intimate and platonic) with sexual recidivism at either time point, employment instability, and some aspects of sexualized violence, a subset of sexual deviancy (Casey, 2016, Ward & Beech, 2014). As for the lack of findings regarding appropriate social relationships with adults it’s also notable that there were no significant findings for either presence, which is usually indicative of atypical relationship interests and is a Static-99R risk factor (Mann et al., 2010) or quality in terms of conflict or satisfaction in the relationship, which was included as a psychologically meaningful risk factor. The relation between a lack of intimate relationships and sexual recidivism has been a fairly stable finding in the literature on sexual offending (Mann et al., 2010), so this finding is rather surprising. It may be that this finding reflects a limitation of this sample in that statutory rape and child pornography offenders were excluded. Both of these types of offenses usually reflect not only deviancy, but also potentially

inappropriate relationships in general or a preference for age inappropriate partners. In regard to the absence of findings on employment instability, it may be that other aspects of lifestyle instability that were measured here were simply more relevant to sexual reoffending. General impulsive behavior, as assessed by the treatment provider, may have