BEYOND THE SUM OF ITS PARTS TOWARDS A NETWORK APPROACH OF RISK OF SEXUAL REOFFENDING Jan Willem van den Berg
KU Leuven Biomedical Sciences Group Faculty of Medicine Department of Neurosciences BEYOND THE SUM OF ITS PARTS TOWARDS A NETWORK APPROACH OF RISK OF SEXUAL REOFFENDING Jan Willem VAN DEN BERG Dissertation presented in partial fulfilment of the requirements for the degree of Doctor in Biomedical Sciences December 2023 Jury: Supervisor: em. Prof. dr. Luk Gijs Co-supervisors: Prof. dr. Erick Janssen dr. Daan van Beek dr. Wineke Smid Chair examining committee: Prof. dr. Inge Depoortere Chair public defense: Prof. dr. Paul Enzlin Jury members: Prof. dr. Laurence Claes Prof. dr. Kris Goethals Prof. dr. Ingeborg Jeandarme Prof. dr. Kasia Uzieblo Prof. dr. Vivienne de Vogel
Printed by: Gildeprint – The Netherlands, www.gildeprint.nl Cover design: Ilse Modder, www.ilsemodder.nl Lay-out: Ilse Modder, www.ilsemodder.nl Cover illustration: The network by ©Bas van den Berg © 2023 KU Leuven - Institute for Family and Sexuality Studies Uitgegeven in eigen beheer, Jan Willem van den Berg, Leuven (Belgium) Alle rechten voorbehouden. Niets uit deze uitgave mag worden vermenigvuldigd en/of openbaar gemaakt worden door middel van druk, fotokopie, microfilm, elektronisch of op welke andere wijze ook zonder voorafgaandelijke schriftelijke toestemming van de uitgever. All rights reserved. No part of the publication may be reproduced in any form by print, photoprint, microfilm, electronic or any other means without written permission from the publisher.
LIST OF ABBREVIATIONS BC provincial corrections system of British Columbia CBT cognitive behavior therapy CI confidence interval COSA circles of support and accountability CS-coefficient stability coefficient DSP dynamic supervision project EBIC extended Bayesian information criterium ESM experience sampling method ICC intraclass correlation coefficient METc medical ethics review board mgm mixed graphical models NBM-RSR network-based model of risk of sexual reoffending NCT network comparison test PMRF pairwise Markov random field RNR-model risk-need-responsivity model ROM routine outcome monitoring SCED single case experimental design SEM structural equation modeling SRA structured risk assessment TWEETS Twente engagement with ehealth technologies scale UMCG university medical center Groningen V
I have laboured diligently, not to mock, lament, or execrate human actions; but to understand them. Sedulo curavi, humanas actions non ridere, non lugere, neque detestari, sed intelligere. Baruch Spinoza (Tractatus Theologico-Politicus, 1670) VI
SUMMARY This dissertation aims to further our understanding of the processes by which dynamic risk factors contribute to the risk of sexual reoffending. Increased understanding of the development and nature of this risk, may, ultimately, improve effectiveness of risk management plans, treatments, and prevention initiatives aimed to help men with a history of sexual offenses to desist future offenses. The introductory chapter discusses the two main approaches to dynamic risk factors. The statistical perspective approaches dynamic risk factors through their statistically demonstrated effects on future (sexual) reoffending. In addition, the Propensities Model conceptualizes dynamic risk factors as risk-related long-term vulnerabilities that can be measured in a variety of ways and manifest themselves when triggered by an external contextual event. Research and theoretical gaps regarding dynamic risk factors are described in the introductory chapter. First, there is an absence of systematic reviews or meta-analyses of research on the predictive properties of (change scores on) dynamic risk assessment instruments developed for adult men with a history of sexual offenses. Second, studies to examine causal relationships between dynamic risk factors and sexual reoffending are typically based on group-level data. However, interindividual (group) level data cannot easily be generalized and applied at an intraindividual (person) level. Third, both studies on the predictive propensities of dynamic risk factors and current theoretical conceptualizations tend to neglect their interrelationships. Fourth, there is no theoretical account of how dynamic risk factors might give rise to the risk of sexual offending, and how sustained change in such risks may be achieved. Fifth, current perspectives on dynamic risk factors fail to recognize that they are composite constructs of interacting psychological, behavioral, and contextual characteristics. This dissertation aims to address these described research and theoretical gaps. Chapter 2 describes the results of a meta-analysis on the predictive properties of (change) scores of dynamic risk assessment instruments developed for adult men with a history of sexual offenses. Based on the data from 52 studies (N = 13,446), it was found that dynamic risk assessment instruments have small-to-moderate predictive properties for sexual, violent (including sexual) and any reoffending. The incremental predictive validity of dynamic over static risk assessment instruments was significant but modest. Change scores significantly predicted all three types of recidivism, indicating that men with a history of offenses who showed larger positive changes (reflecting a reduction in dynamic risk scores) recidivate at lower rates than those who showed a smaller change in dynamic risk factors. Chapter 3 and 4 investigated dynamic risk factors’ interrelationship through network analysis. Data from two independent samples, the dynamic supervision project (DSP; N = 803) and the provincial corrections system of British Columbia (BC; N = 4,511) was used to estimate networks of dynamic risk factors of adult men with a history of sexual VII
offending. Findings based on both samples showed high strength centralities for dynamic risk factors general social rejection/loneliness, lack of concern for others, poor cognitive problem-solving, and impulsive acts. Nodes high in strength centrality have a relatively high number of edges with high magnitudes, indicating that they have more influence on other dynamic risk factors in the network and ultimately on (sexual) reoffending. Besides, networks estimated on both samples contained distinct communities of risk factors related to sexual self-regulation, emotionally intimate relationships, antisocial traits, and self-management. Chapter 5 explored the clinical applicability and added value of collecting, generating, and reporting personalized information regarding interrelated dynamic risk factors on an intraindividual (personal) level. For this purpose, five adult men in outpatient forensic treatment for their history of committing sexual offenses and their therapist combined traditional forensic case formulation with experience sampling method (ESM) monitoring, referred to as a blended ESM procedure. In this procedure, they collaboratively explored possible dynamic risk factors, formed hypotheses about their interrelationships, monitored them through an intensive longitudinal research methodology based on a structured self‐report diary technique (ESM), discussed the resulting reports, and integrated findings in further case formulation. Participants reported increased awareness in personal patterns of dynamic risk factors and in their possible association with the risk of sexual reoffending. Further, they did not perceive the ESM procedure as burdensome, nor did they experience daily assessments as invasive. Chapter 6 introduces the network-based model of risk of sexual reoffending (NBMRSR) as a theoretical account of the causal processes by which dynamic risk factors give rise to the risk of sexual reoffending and contribute to sustained change in this risk. The NBM-RSR considers risk of sexual reoffending to involve a self-sustaining network of causally connected dynamic risk factors. Consistent with this, an increased risk of sexual reoffending is characterized through a network that contains more and stronger interconnected dynamic risk factors with a higher strength. Sustained change in risk of sexual reoffending occurs when activity in the network exceeds a critical point resulting in a new self-sustaining network. Propositions based on the NBM-RSR are introduced and translated into testable hypotheses. These propositions revolve around (a) risk of sexual reoffending resulting from the construction of a network of causally connected dynamic risk factors, (b) network stability, sudden changes, and critical transitions, and (c) dynamic risk factors’ relative influence on risk of sexual reoffending. The NBM-RSR might be a useful tool to improve effectiveness of treatments, risk management plans, and prevention initiatives aimed to assist men with a history of sexual offenses to desist. Based on the NBM-RSR, treatment providers and program managers are more able to identify and prioritize treatment targets (i.e., those dynamic risk factors with a relatively high influence on the network and thus on recidivism risk). In addition, insights from the VIII
NBM-RSR can contribute to improve existing or develop new dynamic risk assessment instruments. In conclusion, this dissertation presents the findings of several studies, culminating in the development of a new model, the NBM-RSR to further our understanding of the processes by which dynamic risk factors in adult men with a history of sexual offenses contribute to (sustained changes in) the risk of sexual reoffending. The NBM-RSR considers risk of sexual reoffending to involve a self-sustaining network of causally connected dynamic risk factors. According to this model, dynamic risk factors relative influence within this network and on sexual reoffending depends on the number and magnitudes of its causal connections with other dynamic risk factors. Network analysis on interindividual (group-level) data suggest that treatment and risk management strategies that focus on (a) social rejection/loneliness, (b) cognitive problem-solving skills, (c) impulsive behavior, and (d) callousness might have the greatest potential in reducing recidivism. However, these causal inferences derived from group-level data, cannot blindly be generalized to the intra-individual (person) level. Dynamic risk assessment instruments do have the potential to contribute to the selection of appropriate, more individually tailored treatment approaches (focusing on individually relevant criminogenic need factors) and can assist in the evaluation of treatment effects. A blended ESM procedure will provide most insight in the processes by which dynamic risk factors contribute to the risk of sexual reoffending on a personal level. IX
Ik heb mij beijverd het menselijk handelen niet te bespotten, niet te betreuren, noch te verwensen, maar te begrijpen. Sedulo curavi, humanas actions non ridere, non lugere, neque detestari, sed intelligere. Baruch Spinoza (Tractatus Theologico-Politicus, 1670) X
SAMENVATTING Dit proefschrift heeft als doel om het inzicht te vergroten in de processen waarmee dynamische risicofactoren bijdragen aan het risico op seksuele recidive. In het ultieme geval kan hierdoor de effectiviteit van behandelingen, risicomanagement- en preventieplannen gericht op het voorkomen van seksueel grensoverschrijdend gedrag van mannen die eerder aangeklaagd of veroordeeld zijn voor zedendelicten verbeteren. Het inleidende hoofdstuk bespreekt de twee belangrijkste conceptuele perspectieven op dynamische risicofactoren. Het statistische perspectief, dat dynamische risicofactoren definieert op basis van het statistisch aangetoonde directe verband met toekomstig (seksueel) grensoverschrijdend gedrag. En het Propensities Model, dat dynamische risicofactoren conceptualiseert als risico-gerelateerde langdurig aanwezige kwetsbaarheden. Deze latent aanwezige kwetsbaarheden manifesteren zich en kunnen (indirect en op verschillende wijze) waargenomen worden na activatie door een externe prikkel. Deze twee conceptuele perspectieven alsmede het empirisch onderzoek naar dynamische risicofactoren kent enkele lacunes. Om te beginnen ontbreekt het aan een systematische review of meta-analyse naar de voorspellende waarde van (veranderscores van) dynamische risicotaxatie-instrumenten ontwikkeld voor volwassen mannen die eerder aangeklaagd of veroordeeld zijn voor een zedendelict. In de tweede plaats zijn studies naar het causale verband tussen dynamische risicofactoren en toekomstig seksueel grensoverschrijdend gedrag doorgaans gebaseerd op wetenschappelijke studies op groepsniveau. Echter, conclusies gebaseerd op het interindividuele (groeps)niveau kunnen niet zonder meer worden gegeneraliseerd naar het intra-individuele (persoons) niveau. Ten derde wordt er geen rekening gehouden met de onderlinge samenhang van dynamische risicofactoren binnen zowel studies naar de voorspellende waarde van deze factoren als binnen de huidige conceptuele perspectieven op dynamische risicofactoren. In de vierde plaats ontbreekt een theoretische verklaring voor de wijze waarop dynamische risicofactoren (duurzame veranderingen in) het risico op seksueel grensoverschrijdend gedrag veroorzaken. Tenslotte gaan de huidige conceptuele perspectieven eraan voorbij dat dynamische risicofactoren samengestelde constructen zijn van op elkaar inwerkende psychologische, gedrags- en contextuele kenmerken. Dit proefschrift richt zich op het invullen van deze beschreven onderzoeks- en theoretische lacunes. Hoofdstuk 2 beschrijft een meta-analyse naar de voorspellende waarde van (veranderscores op) dynamische risicotaxatie-instrumenten ontwikkeld voor volwassen mannen die eerder aangeklaagd of veroordeeld zijn voor een zedendelict. Op basis van 52 studies (N = 13.446) bleek, dat dynamische risicotaxatie-instrumenten een lage tot matige voorspellende waarde hebben voor een seksuele, gewelddadige (inclusief seksuele) en algemene recidive. De toegevoegde voorspellende waarde van dynamische ten opzichte van statische risicotaxatie-instrumenten was bescheiden maar significant. Veranderscores op deze instrumenten bleken voorspellend voor alle drie soorten recidive. Dit wijst erop XI
dat plegers van seksueel grensoverschrijdend gedrag met een grotere afname van dynamische risicofactoren minder snel recidiveren dan met plegers die qua dynamische risicofactoren een minder grote afname, geen verandering of een toename laten zien. In hoofdstuk 3 en 4 worden de onderlinge verbanden tussen dynamische risicofactoren met behulp van netwerk analyses in kaart gebracht bij mannen die in het verleden seksueel grensoverschrijdend gedrag pleegden. Netwerken van dynamische risicofactoren zijn geschat op basis van gegevens uit twee onafhankelijke steekproeven, het dynamic supervision project (DSP; N = 803) en het provinciale correctiesysteem van British Columbia in Canada (BC; N = 4.511). Uit netwerkanalyses in beide steekproeven blijkt een hoge strength centrality voor de dynamische risicofactoren: sociale afwijzing/eenzaamheid, desinteresse in het welzijn van anderen, ontoereikende probleemoplossingsvaardigheden, en impulsief gedrag. Dynamische risicofactoren met een hoge strength centrality hebben relatief veel en relatief sterke verbanden binnen het netwerk. Hierdoor hebben zij meer invloed op andere dynamische risicofactoren en daarmee op (seksuele) recidive. Daarnaast zijn op basis van de netwerkanalyses verschillende clusters van dynamische risicofactoren te onderscheiden die getypeerd kunnen worden als: seksuele zelfregulatie, emotioneel intieme relaties, antisociale trekken en algemene zelfregulatie. Hoofdstuk 5 beschrijft onderzoek naar de klinische toepassing van het verzamelen, genereren en rapporteren van gepersonaliseerde informatie over onderling samenhangende dynamische risicofactoren op intra-individueel (persoonlijk) niveau. Hiervoor werd de traditionele forensische casusconceptualisatie gecombineerd met monitoring door middel van experience sampling method (ESM), verder benoemd als blended ESM-procedure. Vijf volwassen mannen in ambulante forensische behandeling vanwege het plegen van seksueel grensoverschrijdend gedrag onderzochten hiervoor samen met een therapeut hun dynamische risicofactoren. Ze vormden hypothesen over de onderlinge samenhang van deze factoren en scoorden deze factoren door middel van een electronisch dagboek gedurende twee weken dagelijks vijfmaal. Ze bespraken het feedback rapport wat op basis van deze gegevens gemaakt werd met hun therapeut en integreerden de bevindingen in bestaande forensische casusconceptualisatie. De blended ESM-procedure leidde bij de deelnemers tot een toegenomen bewustzijn van de eigen dynamische risicofactoren, hun onderlinge samenhang en het mogelijke verband met hun seksueel grensoverschrijdend gedrag. De ESM-procedure werd door de deelnemers niet als invasief of belastend ervaren. Hoofdstuk 6 introduceert het network-based model of risk of sexual reoffending (NBM-RSR) als een theoretisch model voor het begrijpen en verklaren van de wijze waarop dynamische risicofactoren bijdragen aan (duurzame verandering van) het risico op seksuele recidive. Volgens de NBM-RSR bestaat het risico op terugval in seksueel grensoverschrijdend gedrag uit een zelfvoorzienend, dat is een zichzelf in stand houdend, netwerk van causaal verbonden dynamische risicofactoren. Een verhoogd risico op XII
terugval in seksueel grensoverschrijdend gedrag wordt volgens dit model gekenmerkt door een netwerk dat meer en sterkere onderling verbonden dynamische risicofactoren bevat. Een duurzame verandering in het risico treedt op wanneer de activiteit in het netwerk een kritisch punt overschrijdt, resulterend in een nieuw zelfvoorzienend netwerk. Op basis van het NBM-RSR worden proposities en hypothesen geformuleerd. Deze proposities gaan over (a) het risico op terugval in seksueel grensoverschrijdend gedrag als gevolg van de structuur van een netwerk van causaal verbonden dynamische risicofactoren, (b) netwerkstabiliteit, plotselinge veranderingen en kritische transities, en (c) de relatieve invloed van dynamische risicofactoren op het risico op terugval in seksueel grensoverschrijdend gedrag. Het NBM-RSR kan een positieve bijdrage leveren aan het voorkomen van seksueel grensoverschrijdend gedrag door het verder vergroten van de effectiviteit van behandelingen, risicomanagementplannen en preventie-initiatieven gericht op het voorkomen van seksueel grensoverschrijdend gedrag van mannen die eerder zedendelicten pleegden. Met behulp van het NBMRSR zijn programmaverantwoordelijken en behandelaars namelijk beter in staat om behandeldoelen te identificeren en te prioriteren (namelijk die dynamische risicofactoren met een relatief grote invloed op het netwerk en dus op het recidiverisico). Bovendien kunnen inzichten vanuit het NBM-RSR een bijdrage leveren aan verbeteringen van bestaande of ontwikkeling van nieuwe dynamische risicotaxatie-instrumenten. Tot slot en in conclusie vergroten de in deze dissertatie beschreven de onderzoeksresultaten, die geleid hebben tot het NBM-RSR, onze kennis over en inzicht in de processen waarmee dynamische risicofactoren bijdragen aan het risico op seksuele recidive. Het NBM-RSR begrijpt het risico op terugval in seksueel grensoverschrijdend gedrag vanuit een zelfvoorzienend netwerk van causaal verbonden dynamische risicofactoren. De invloed van een dynamische risicofactor op het netwerk en daarmee op het risico op terugval in seksueel grensoverschrijdend gedrag is, volgens dit model, afhankelijk van het aantal en de sterkte van de causale verbanden met andere dynamische risicofactoren. Netwerkanalyse met interindividuele data (op groepsniveau) suggereert dat strategieën voor behandeling en risicomanagement die gericht zijn op (a) sociale afwijzing/ eenzaamheid, (b) cognitieve probleemoplossende vaardigheden, (c) impulsief gedrag, en (d) gevoelloosheid en hardvochtigheid, de grootste kans van slagen hebben om de kans op recidive te verminderen. Deze causale gevolgtrekking kan echter niet blindelings gegeneraliseerd worden naar het intra-individuele (persoons)niveau. Echter, de momenteel beschikbare dynamische risicotaxatie-instrumenten kunnen wel bijdragen aan het indiceren van passende, meer op het individu toegesneden, behandelmethoden (gericht op individueel relevante criminogene behoeften). Ook kunnen ze gebruikt worden voor de evaluatie van de effecten van de behandeling. Het toepassen van een blended ESM-procedure kan nog meer inzicht verschaffen in de wijze waarop dynamische risicofactoren bijdragen tot het risico op seksueel grensoverschrijdend gedrag bij een individu. XIII
XIV
TABLE OF CONTENTS LIST OF ABBREVIATIONS SUMMARY SAMENVATTING TABLE OF CONTENTS CHAPTER ONE Dynamic Risk Factors in Adult Men with a History of Sexual Offenses: What Do We Know and Where Are We Heading? 1.1 Introduction 1.2 Two perspectives on risk-relevant psychological and behavioral characteristics 1.2.1 Statistical perspective 1.2.2 Propensities Model perspective 1.3 Gaps in the current perspectives on dynamic risk factors 1.3.1 Research gaps regarding dynamic risk factors’ causal role 1.3.2 Theoretical gaps regarding dynamic risk factors causal role 1.3.3 Addressing the research and theoretical gaps 1.4 Towards understanding the development and nature of risk of sexual reoffending 1.5 An outline of the chapters in this dissertation CHAPTER TWO The predictive properties of dynamic risk assessment instruments developed for adult men with a history of sexual offenses: A meta-analysis 2.1 Introduction 2.2 Current study 2.3 Method 2.3.1 Search strategy 2.3.2 Eligibility 2.3.3 Final selection of studies 2.3.4 Inter-rater reliability 2.3.5 Statistical analyses 2.4 Results 2.4.1 Overall predictive properties 2.4.2 Incremental validity 2.4.3 Predictive validity of change scores XV V VII XI XV 21 23 25 25 29 30 30 31 32 33 35 39 41 42 42 42 43 43 46 47 49 49 51 53
2.4.4 Predictive validity of change scores controlled for static and initial dynamic scores 2.5 Discussion CHAPTER THREE The application of network analysis to dynamic risk factors in adult men with a history of sexual offenses 3.1 Introduction 3.2 Dynamic risk factors 3.3 Network analysis 3.4 Current study 3.5 Method 3.5.1. Participants 3.5.2 Assessment of dynamic risk factors 3.5.3 Network analysis 3.6 Results 3.6.1 Network construction and centrality of dynamic risk factors 3.6.2 Shortest paths 3.6.3 Network stability 3.7 Discussion 3.6.1 Limitations 3.6.2 Implications and strengths CHAPTER FOUR Dynamic risk factors in adult men who committed sexual offenses: replication and comparison of networks found in two independent samples 4.1 Introduction 4.1.1 Previous network analysis in adult men with a history of sexual offenses 4.1.2 The present study 4.2 Method 4.2.1 Participants 4.2.2 Assessment of dynamic risk factors 4.2.3 Network analysis 4.3 Results 4.3.1 Network construction and centrality of dynamic risk factors 4.3.2 Statistical network comparison 4.3.3 Network stability 4.4 Discussion 4.4.1 Limitations XVI 54 56 61 63 63 65 66 67 67 67 70 71 71 75 75 77 77 79 83 85 86 87 87 87 88 91 93 93 97 99 100 100
4.4.2 Future research directions 4.4.3 Clinical implications 4.4.4 Conclusion CHAPTER FIVE Personalized monitoring and feedback on risk-relevant features in forensic case formulation: A series of case-studies in men who have committed sexual offenses 5.1 Introduction 5.2 Method 5.2.1 Participants 5.2.2 Materials 5.3 Results 5.3.1 Value of feedback to case formulation 5.3.2 Evaluating the process of collecting personalized information 5.4 Discussion 5.4.1 Limitations 5.4.2 Future research directions 5.4.3 Clinical implications 5.5 Conclusion CHAPTER SIX Understanding the risk of sexual reoffending in adult men: A networkbased model 6.1 Introduction 6.1.1 The Propensities Model 6.1.2 Limitations of the Propensities Model 6.2 Network-based model of risk of sexual reoffending (NBM-RSR) 6.2.1 Risk resulting from a network of causal interacting dynamic risk factors 6.2.2 Risk determined through the network topology 6.2.3 The impact on risk of variables in the external field 6.3 Overview and discussion 6.3.1 Propositions and hypotheses to be examined 6.3.2 Detection interactions in a network of dynamic risk factors 6.3.3 Future steps and further development of the NBM-RSR 6.4 Conclusion XVII 101 102 102 105 107 110 110 110 114 114 117 120 122 123 123 123 127 129 130 131 132 133 133 135 136 137 139 141 143
CHAPTER SEVEN Discussion 7.1 Introduction 7.2 Key findings 7.2.1 Predictive properties of (change in) dynamic risk factors 7.2.2 Interrelationship of dynamic risk factors 7.2.3 The added value of personalized network of dynamic risk factors 7.2.4 Network-based model of risk of sexual reoffending 7.3 Scientific and clinical relevance 7.3.1 Scientific relevance 7.3.2 Clinical relevance 7.4 Limitations 7.4.1 Limitations of the studies in this dissertation 7.4.2 Limitations of the NBM-RSR 7.5 Recommendations for future research REFERENCES SCIENTIFIC ACKNOWLEDGEMENTS PERSONAL CONTRIBUTIONS CONFLICTS OF INTEREST STATEMENT FUNDING LIST OF PUBLICATIONS Publications in international peer-reviewed journals Publications in Dutch peer-reviewed journals Books and chapters Treatment manuals Risk assessment instruments PERSONAL ACKNOWLEDGEMENTS (DANKWOORD) XVIII 145 147 147 147 147 148 149 150 150 150 151 151 153 154 161 177 179 183 185 187 187 188 189 190 191 193
XIX
DYNAMIC RISK FACTORS IN ADULT MEN WITH A HISTORY OF SEXUAL OFFENSES: WHAT DO WE KNOW AND WHERE ARE WE HEADING? 1
1 22 CHAPTER 1 ABSTRACT Dynamic risk factors are psychological and behavioral characteristics whose change alters the risk of sexual reoffending in men with a history of sexual offenses. This introductory chapter describes the two most widely known approaches for understanding dynamic risk factors: the statistical perspective and the Propensities Model. Both approaches assert a causal role for dynamic risk factors but neither provides a theoretical explanation of the causal processes through which they give rise to the risk of sexual reoffending and contribute to sustained change in this risk. This chapter addresses these gaps by introducing a network approach application to dynamic risk factors. This approach considers dynamic risk factors as composite constructs consisting of a range of causally related psychological and behavioral characteristics interacting with each other and with contextual features. Studies to further the development of a network approach application to dynamic risk factors are introduced. These studies empirically 1) explore dynamic risk factors’ causal relationship with sexual reoffending through a meta-analysis on the predictive properties of (change scores of) dynamic risk assessment instruments developed for adult males with a history of sexual offenses (Chapter 2), 2) examine their interrelationships using statistical network analysis of dynamic risk factors as measured by the STABLE-2007 - which is a measure of dynamic risk factors for the treatment and supervision of adult males convicted of a sexually motivated offence (Chapter 3 and 4), and 3) investigate the extent to which a personalized network-based model on risk of sexual reoffending can inform forensic case formulations in clinical practice (Chapter 5). The network-based model of risk of sexual reoffending (NBM-RSR), which addresses several limitations and constraints of both the statistical perspective on dynamic risk factors as well as the Propensities Model is introduced in Chapter 6. This dissertation concludes with a discussion (Chapter 7) of the empirical studies and the NBM-RSR and outlines their scientific and clinical relevance, their limitations, and ends with recommendations for future research. KEYWORDS Dynamic risk factors, sexual reoffending, males with a history of sexual offenses
1 23 DYNAMIC RISK FACTORS 1.1 INTRODUCTION Sexual offending has major social and health-related implications. Sexual offending can be defined as any attempted or completed sexual act committed against someone without a person’s freely given consent, which might include both contact sexual offenses (e.g., rape, sexual coercion, and/or unwanted sexual contact) as non-contact offenses (e.g., sexual harassment, exhibitionism, voyeurism), whether in person or online (cfr. World Health Organization, 2021). Survivors may experience multiple longterm negative psychological, sexual, relationship, and physical health problems, further compounded by the financial costs associated with treatment, and with professional and educational difficulties (Baker et al., 2016; Dworkin, 2020, Golding, 1994; Stein & Barrett-Connor, 2000; van Berlo & Ensink, 2000). These negative consequences are not only often severe, but highly prevalent. Based on a systematic review of 32 articles, reporting on a total of 45 studies from 29 countries outside North America, Dworkin and colleagues (2021) found that prevalence rates of being survivor of sexual offending for the period from adolescence onward ranged from 0.6% to 77.6% for women, 0.3% to 65.5% for men (Dworkin et al., 2021). While The National Intimate Partner and Sexual Violence Survey, an epidemiological study conducted annually in the United States of America, indicates that 54.3%,[95% confidence interval (CI): 52.9%, 55.7%] of the women and 30.7 % [CI: 29.3%, 32.1%] of the men had a lifetime history of surviving an attempted or completed contact sexual offenses (Basile et al., 2022). Unlike large-scale national or epidemiological studies on the prevalence of having experienced and survived sexual offenses, research on sexual perpetration is more sporadic (Anderson et al., 2021). An indication on the prevalence of perpetration of sexual offending can be obtained from studies on sexual perpetration in college men. In their systematic review of empirical studies on self-reported lifetime prevalence in Canadian and American male college students (including 78 independent samples, N= 25,524), Anderson and colleagues (2021) found a prevalence rate of any sexual perpetration and rape of respectively 29.3% (SD = 16.8) and 6.5% (SD = 6.3). Crime statistics based on official records are in sharp contrasts with these self-reported prevalence rates of sexual perpetration. For example, Långström and colleagues (2015) found that 0.5% of all men living in Sweden, who were and 30 to 45 years old in 2009, were convicted for any sexual crime. Once convicted for sexual offending, long-term recidivism studies show that desistance from sexual reoffending is the norm (Hanson et al., 2018). A meta-analytic review combining the results of 20 independent scientific studies (N = 7,225) on sexual reoffending found that the sexual reoffending rate was 9.1% at 5 years, 13.3% at 10 years, 16.2% at 15 years, 18.2% at 20 years, and 18.5% at 25 years (Hanson et al., 2018). This meta-analysis showed that not only desistance is most probable, but also that the hazard rates for sexual reoffending predictably declines the longer men with a history of sexual offending remain offense-free in the community
1 24 CHAPTER 1 (Hanson et al., 2018). To effectively address sexual offending, prevention can be employed at three levels referred to as primary, secondary, and tertiary (DeGue et al., 2014; Knack et al., 2019; Laws, 2000; Orchowski & Berkowitz, 2022; Orchowsky et al., 2020; Russell et al., 2020). Primary prevention is implemented before the occurrence of initial sexual offending and generally involves wide-scale initiatives aimed at the general public (e.g., general crime deterrence, public education, sex education in schools, awareness of and changes in sexual offense supporting cultural aspects) (e.g., Armstrong et al., 2018; Bourke, 2022; Brownmiller, 1975; Nussbaum, 2021). Secondary prevention focuses on those at-risk of engaging in sexual offending through more targeted interventions (e.g., anonymous helplines for people sexually attracted to prepubescent children), which address issues known to increase the risk of sexual offending (e.g., Beier, 2021). Tertiary prevention aims to prevent sexual reoffending for those who have engaged in a sexual offense (e.g., treatment programs or probation supervision) (e.g., McCartan, et al., 2022; Richards, 2022). Although the effectiveness of treatment of men with a history of sexual offenses, in the prevention of sexual reoffending can be considered modest at best. In particular cognitive behavior therapy (CBT) has been found to reduce recidivism in people with a history of sexual offenses (Andrews & Bonta, 2010; Bonta & Andrews, 2017; Hanson et al., 2009; Holper et al., 2023; Schmucker & Lösel, 2015/2017). In addition, application of the risk-need-responsivity model (RNR; Andrews & Bonta, 2010; Bonta & Andrews, 2017) is associated with greater reductions in recidivism in offender treatment. This has been found in a wide range of correctional interventions (Andrews & Bonta, 2010; Bonta & Andrews, 2017; French & Gendreau, 2006; Landenberger & Lipsey, 2005; Lowenkamp et al., 2006; Marlowe et al., 2006; Wilson, Bouffard, et al., 2005), including those for people with a history of sexual offenses (Hanson et al., 2009; Holper et al., 2023; Olver, Nicholaichuk & Wong, 2014; Schmucker & Lösel, 2015/2017; Smid et al., 2014). The RNR model includes three general principles, also known as the “What Works principles”, for the optimization of forensic treatment. The Risk Principle dictates that the involvedness of treatment services must be proportional to the risk of reoffending; that is, individuals who have entered the criminal justice system with high-risk of reoffending should receive the most intensive treatments. Second, the Need Principle emphasizes that treatments should focus on psychological and behavioral characteristics related to the risk of (sexual) reoffending. Finally, the Responsivity Principle dictates the tailoring of treatment programs to the individual learning styles, capabilities, and the individual’s limitations. This dissertation aims to further our understanding of the processes by which riskrelevant psychological and behavioral characteristics in adult men with a history of sexual offenses contribute to the risk of sexual reoffending - in this dissertation defined as the probability of future sexual offending by men convicted of a sexual offense. Increased
1 25 DYNAMIC RISK FACTORS understanding of the development and nature of this risk, may, ultimately, improve effectiveness of treatments, risk management plans, and prevention initiatives aimed to assist men with a history of sexual offenses to desist (Gannon et al., 2019; Ward & Beech, 2015). This introductory chapter identifies current theoretical and research gaps regarding risk-relevant psychological and behavioral characteristics and introduces the network approach of psychopathology as a possible perspective to address them. The chapter will conclude with the research objectives of this dissertation. 1.2 TWO PERSPECTIVES ON RISK-RELEVANT PSYCHOLOGICAL AND BEHAVIORAL CHARACTERISTICS To help a man with a history of sexual offenses desist future ones, forensic treatment and risk assessment ideally explain the source(s) of his risk, by examining and addressing his personal risk-relevant characteristics (Mann et al., 2010). These characteristics may include biological factors (e.g., genetics, brain structures, hormone levels), (early) life experiences (e.g., childhood sexual abuse or neglect), sociocultural factors (e.g., being part of a hostile masculine (sub)culture or sexualized environment, legal variables), situational factors (e.g., access to potential victims, absence of a guardian, changes in employment), and psychological and behavioral factors (such as human agency, motivation for treatment, intelligence, extraversion, level of social emotional development). Given the focus of this dissertation on dynamic risk factors, two perspectives on psychological and behavioral characteristics contributing to the risk of sexual reoffending will be discussed in the following sections. 1.2.1 STATISTICAL PERSPECTIVE In exploring the causal effects of behavioral and psychological characteristics on reoffending, Andrews and Bonta (e.g., Andrews et al., 1990; Bonta & Andrews, 2017) introduced a statistical perspective that distinguishes between three types of covariates associated with criminal behavior, named: correlate, (dynamic) predictor, and causal (see figure 1.1). Correlates, which can be determined by cross-sectional studies on criminal past, are potential risk factors describing the size of the relationship between individual differences and criminal history. Potential, because there is no certainty that variation in the independent variable came before variation in the dependent variable. Therefore, correlates contain no causal information. Static risk factors are examples of correlates. Static risk factors, such as demographic information (e.g., age, ever lived in a legal intimate relationship), criminal history (e.g., amount and type of offenses), and victim information (e.g., sex, being relative of victims), are not amendable by deliberate treatment interventions (Bonta, 1996; Mann, Hanson & Thornton, 2010). Predictors are
1 26 CHAPTER 1 Figure 1.1: Three types of covariates according to Bonta & Andrews (2017). An arrow indicates a causal relationship, while a dashed line or dashed arrow represents respectively an undirected or directed association which might be spurious. Change in a covariate is represented by ∆.
1 27 DYNAMIC RISK FACTORS defined by an association between individual differences and future criminal behavior and can be uncovered in longitudinal studies. In other words, predictors are not only associated with criminal behavior, but their variation also precedes variation in future criminal behavior. As a result, the association of predictors could contain causal information. A predictor is called a dynamic predictor when it is amendable to change and when change of the dynamic predictor is related to variation in future criminal behavior, which can be examined in multiwave longitudinal studies. The relationship between a (dynamic) predictor or correlate and criminal behavior can be spurious, meaning that the covariate may falsely appear to be related due to an unseen third variable (Andrews et al., 1990; Bonta & Andrews, 2017). Controlling for third variables may be attained through experimental research designs. According to Bonta and Andrews (2017), change in a causal covariate is found to be associated with future criminal behavior in an experimental design. According to Andrews and Bonta, due to the high level of control in this research design, causal covariates found in a randomized controlled trial leave one most confident with making statements regarding their causal influence on future criminal behavior (Bonta & Andrews, 2017; Farrington, 2013). Behavioral and psychological characteristics whose change is related to variation in future criminal behavior are referred to as dynamic predictors (Bonta and Andrews, 2017), dynamic risk factors (Bonta and Andrews, 2017; Douglas & Skeem, 2005), psychologically meaningful risk factors (Mann et al., 2010), criminogenic needs (Andrews et al., 1990), or protective factors (de Vries Robbé et al., 2015). Although there is some variation in the working definitions of these constructs, they all intent to orient and inform treatment, supervision, and rehabilitation plans (Hanson et al., 2020). For consistency reasons, in this dissertation the term dynamic risk factor is used. Table 1.1 summarizes promising and empirically supported dynamic risk factors for sexual reoffending. Traditionally, dynamic risk factors are categorized in stable dynamic and acute dynamic risk factors (Hanson & Harris, 2000a). Stable dynamic risk factors can be modified over time, for example by psychotherapy, and include personality characteristics, skill deficits, and learned behaviors (e.g., impulsivity, regulation of deviant sexual interests, social and problem-solving skills, and offense-supportive attitudes). Acute dynamic risk factors can change instantaneously (e.g., victim access, substance abuse, emotional breakdown, sexual arousal). They are supposed to related to the risk of committing a sexual offense in the short term.
1 28 CHAPTER 1 Table 1.1: Overview of promising and empirically supported dynamic risk factors related to sexual reoffending. Domain Subdomain Meta-analytic results S = Empirically-supported P = Promising Sexual interests Sexual preoccupation • Sexual preoccupation (S) • Multiple paraphilias (S) • Sexualized coping (P) • Intense impersonal sexual interests • Sexual coping • Diverse sexual outlets Offense-related sexual interests • Sexual interest in children (S) • Sexualized violence (P) • Sexual interest in prepubescent and pubescent children • Sexualized violence Distorted attitudes Victim schema • Pro-offending schema about classes of potential victims (e.g., children or women) • Pro-offending attitudes (S) • Pro-child molestation attitudes (S) • Pro-rape attitudes (S) • Generic sexual offending attitudes (S) Note that there was insufficient data to look at the predictiveness of more specific attitudes, although all three SRA categories coincided with at least one of the broader categories used in the metaanalyses Rights schema • Excessive sense of entitlement Means schema • Machiavellianism • Violent world schema Relational style Inadequate relational style • Emotional congruence with children (S) • Painfully low self-esteem was found consistently predictive in the United Kingdom, but not in other jurisdictions. • Narcissistic self-esteem hasn’t been examined in recidivism studies • Dysfunctional self-esteem (inadequate or narcissistic) • Emotional congruence with children Lack of emotionally intimate adult relationships • Lack of sustained marital type relationships (S) • Marital relationships marred by repeated violence/infidelity (S) • Lack of sustained marital type relationships • Relationships marred by violence/ infidelity Aggressive relational style • Callousness (P) • Grievance thinking (S) • Callousness • Grievance thinking Selfmanagement Social deviance • Childhood behavior problems (S) • Juvenile delinquency (S) • Non-sexual offenses (S) • Non-compliance with supervision (S) • Violation of conditional release (S) • Antisocial personality disorder (S) • Impulsivity/recklessness (S) • Employment instability (S) • Early onset and pervasive resistance to rules and supervision • Lifestyle impulsiveness Dysfunctional coping in response to stress/problems • Poor coping (externalizing) (P) • Poor problem-solving • Poor emotional control Promising: significant predictive value found in at least one study and other relevant supportive evidence; Supported: significant predictive value found through meta-analytically integration of at least three studies (Mann et al., 2010; Thornton, 2002; Thornton, 2013).
1 29 DYNAMIC RISK FACTORS 1.2.2 PROPENSITIES MODEL PERSPECTIVE In response to the outcome of scientific research, the statistical perspective on dynamic risk factors has been a topic of discussion. Relevant questions concern the distinction between stable and acute dynamic risk factors, the extent to which acute dynamic risk factors are in- or external to individuals, and the relationship between dynamic and static risk factors (Mann et al., 2010; Thornton, 2016; Ward & Beech, 2006; Ward & Beech, 2015). First, research of acute dynamic risk factors does not support the idea that these factors necessarily change instantaneously (Hanson et al., 2007). For example, Babchishin and Hanson (2020) found that although acute dynamic risk changes across time, the pattern of change varies between individuals. Second, theoretical discussion concentrates on the question to what extent acute dynamic risk factors are considered triggering contextual events and therefore required to be external to the individual (Mann et al., 2010; Ward & Beech, 2006). For example, “victim access” can be considered both an external trigger as well as resulting from a dynamic risk factor. That is, unintended victim access can activate deviant sexual interest, while on the other hand, deliberate contact with potential victims may also result from deviant sexual interest. Third, questions can be raised regarding the interrelationship between static and dynamic risk factors. That is, do observed correlations between static and dynamic risk factors exist a) because they are caused by the same underlying latent variable, or because b) static risk factors act as markers of the past operation of dynamic risk factors (i.e., dynamic risk factors cause static risk factors; Beech & Ward, 2004; Ward & Beech, 2015)? Based on these findings and theoretical observations and questions, Thornton and colleagues introduced a latent variable perspective on dynamic risk factors, referred to as the Propensities Model (Thornton, 2016). The Propensities Model reconceptualized dynamic risk factors as long-term vulnerabilities (Knight & Thornton, 2007; Thornton, 2002; Thornton & Knight, 2015) or enduring propensities (Mann et al., 2010) which may be activated in response to environmental triggers (Thornton, 2016). When triggered by an external contextual event, these latent variables manifest and can be measured in a variety of ways (Mann et al., 2010; Thornton, 2006). For example, the presence of boys might trigger the latent variable “emotional congruence with children” in an adult with a history of sexual offenses. Once triggered, he might experience a strong non-sexual affective and cognitive connection with children (“acute dynamic”), ascribe child-like characteristics to himself (“stable dynamic”), or initiate and maintain contact with children (“acute”, “stable dynamic”, “external”) (see figure 1.2; McPhail et al., 2013; McPhail et al., 2018). Once activation of this variable has resulted in a sexual offense, it will be measurable by having boys as victims (“static risk factor”).
1 30 CHAPTER 1 Child-like characteristics Boy victim Emotional congruence with children Initiation contact with children Affective connection with children Contextual trigger Static Stable dynamic Acute dynamic External Figure 1.2: Dynamic risk factor emotional congruence with children presented as a latent variable. From the Propensities Model perspective, criteria to identify a latent variable as risk-relevant include both theory and evidence (Mann et al., 2010). That is, besides empirical evidence that the variable predicts sexual reoffending, the variable a) should be considered psychologically meaningful, b) could plausibly be a cause of sexual reoffending, c) might be worth targeting in treatment or is already usually targeted in treatment, or d) is treated as plausible cause of sexual offending in criminological or social learning theories (Mann et al., 2010). However, it is precisely the notion of dynamic risk factors causal interference that leads to criticism of the Propensities Model. 1.3 GAPS IN THE CURRENT PERSPECTIVES ON DYNAMIC RISK FACTORS 1.3.1 RESEARCH GAPS REGARDING DYNAMIC RISK FACTORS’ CAUSAL ROLE Although longitudinal studies on the predictive properties of risk assessment instruments containing dynamic risk factors of adult males with a history of sexual offending increased in the last decade (Hanson et al., 2020; McGrath et al., 2011), research gaps regarding their causal relationship with sexual reoffending remain. First, current studies on the predictive properties of dynamic risk assessment instruments generally do not examine the role of change scores. Consequently, there is no certainty that change in dynamic risk factors captured by these risk assessment instruments relates to variation in the risk of sexual reoffending. Second, (multiwave) longitudinal studies on dynamic risk assessment instruments developed for adult males with a history of sexual offenses
1 31 DYNAMIC RISK FACTORS have, as yet, not been the focus of systematic reviews or meta-analyses. This constitutes an additional reason to be cautious with statements on possible causal links between dynamic risk factors and sexual reoffending (Hanson et al., 2020). Third, current studies on the predictive accuracy of dynamic risk factors on sexual reoffending do not consider dynamic risk factors’ interrelationships. This is a problem, as not considering their interactions may lead to misplaced assumptions on the causal relationship of these factors with sexual reoffending. Fourth, inferences on the causal relationship between dynamic risk factors and sexual reoffending are derived from studies based on group-level data. Although these studies provide guidance for risk assessment and treatment of individuals with a history of a sexual offense, in social science, findings from interindividual (group) level cannot blindly be generalized to the intraindividual (person) level (Fisher et al., 2018). Examining dynamic risk factors’ interrelationships at the individual level may generate working hypotheses on the etiology of the risk of sexual reoffending in a specific person (Burger et al., 2020; Epskamp et al., 2018; Kroeze at al., 2017). This personalized information may enhance treatment effectiveness aimed at desistance. However, scientific research on the clinical applicability of personalized information on interrelated dynamic risk factors in individual men with a history of sexual offenses is still lacking. 1.3.2 THEORETICAL GAPS REGARDING DYNAMIC RISK FACTORS’ CAUSAL ROLE Several theoretical gaps in current conceptual perspectives on dynamic risk factors have been discussed in the literature (e.g., Prentky et al., 2015; Thornton, 2016; Ward & Beech, 2015). First, both the statistical and Propensities Model perspectives on dynamic risk factors assert causality but do not provide an explanation for how dynamic risk factors give rise to the risk of sexual reoffending (Prentky et al., 2015, Thornton, 2016). Second, current conceptualizations of dynamic risk factors provide no theoretical account of how change in risk may be achieved (Thornton, 2016). Third, the perspectives do not explicitly recognize that dynamic risk factors are composite constructs or categories containing a range of variables like emotion, behavior, cognition, contextual characteristics, and causal strands (also referred to as lack of coherence by Heffernan & Ward, 2019; Heffernan et al., 2019; Ward & Fortune, 2016b). Whitin emotional congruence with children, for example, causal interactions can be assumed between ascribing child-like characteristics to oneself, encountering cognitive and experiencing affective connection with children, and initiating and maintaining contact with children (McPhail et al., 2013; McPhail et al., 2018). Fourth, neglecting the interrelationships between characteristics within dynamic risk factors makes it impossible to draw accurate conclusions about the potential causes of criminal behavior (lack of specificity; e.g., Fortune & Heffernan, 2021; Heffernan & Ward, 2019; Heffernan et al., 2019; Ward & Fortune, 2016b). Fifth, the statistical and Propensities Model perspectives on dynamic risk factors offer no guidance on the level of abstraction in the current perspectives on dynamic risk
www.gildeprint.nlRkJQdWJsaXNoZXIy MjY0ODMw