Assessing Dangerousness: Domestic Violence Offenders and Child Abusers [3 ed.] 9780826133267

Table of contents :
Dedications
Contents
Contributors
Preface
Share Assessing Dangerousness: Domestic Violence Offenders and Child Abusers, Third Edition
1. Prediction of Interpersonal Violence: An Introduction • Allison Ward-Lasher, Daniel J. Sheridan, Nancy E. Glass, and Jill Theresa Messing
2. Prediction Issues for Practitioners • Joel S. Milner, Jacquelyn C. Campbell, and Jill Theresa Messing
3. Child Physical Abuse Risk Assessment: Parent and Family Evaluations • Joel S. Milner and Julie L. Crouch
4. Evaluating Risk Factors for Fatal Child Abuse • Scott D. Krugman and Francie J. Julien-Chinn
5. Prediction of Homicide of and by Battered Women • Jacquelyn C. Campbell, Jill Theresa Messing, and Kirk R. Williams
6. Assessing Risk of Intimate Partner Violence • N. Zoe Hilton and Angela Wyatt Eke
7. Children at Risk of Homicide in the Context of Intimate Partner Violence • Peter Jaffe, Jordan Fairbairn, and Katherine Reif
Index

Citation preview

Assessing Dangerousness

Jacquelyn C. Campbell, PhD, RN, FAAN, is a professor and the Anna D. Wolf Chair at the Johns Hopkins School of Nursing. She is a national leader in research and advocacy in the field of domestic violence, or intimate partner violence (IPV). She has authored or coauthored more than 250 publications and seven books on violence, its health outcomes, and interventions for survivors. Her studies have paved the way for a growing body of interdisciplinary investigations by researchers in the disciplines of nursing, medicine, social work, and public health. Her expertise is frequently sought by national and international policy makers who are exploring IPV and its health effects on families and communities. As a nurse educator and mentor, Dr. Campbell leads by example in inspiring new generations of nurse researchers. Her BSN, MSN, and PhD are from Duke University, Wright State University, and the University of Rochester, respectively. She teaches an undergraduate and MSN elective in family violence as well as in the PhD program, and is the national program director of the Robert Wood Johnson Foundation Nurse Faculty Scholars Program. Dr. Campbell led a National Institutes of Health (NIH)-funded (T32) fellowship that provided funding for pre- and postdoctoral fellows in violence research for 15 years. Elected to the Institute of Medicine (IOM; now the National Academy of Medicine—NAM) in 2000, Dr. Campbell was also the IOM/American Academy of Nursing/American Nurses Foundation senior scholar-in-residence and was founding cochair of the IOM/NAM Forum on Global Violence Prevention. Other honors include the Pathfinder Distinguished Researcher by the Friends of the National Institutes of Health’s National Institute of Nursing Research, the American Society of Criminology Vollmer Award, and the Sigma Theta Tau Episteme Award; Dr. Campbell is one of the “20 for 20” Centers for Disease Control and Prevention Leaders in Violence and Injury Prevention and one of the inaugural 17 Gilman Scholars at Johns Hopkins University. She is on the board of directors for Futures Without Violence, is an active member of the Johns Hopkins Women’s Health Research Group, and has served on the boards of the House of Ruth Battered Women’s Shelter and four other shelters. She was also a member of the congressionally appointed U.S. Department of Defense Task Force on Domestic Violence. Jill Theresa Messing, PhD, MSW, is associate professor in the School of Social Work at Arizona State University (ASU). She earned her MSW and PhD in social welfare at the University of California, Berkeley, and went on to complete a National Institutes of Health (NIH)-funded postdoctoral fellowship in interdisciplinary violence research at Johns Hopkins University, where she studied with Dr. Jacquelyn Campbell. Dr. Messing has published over 50 articles and book chapters focused on intimate partner violence (IPV) and has been an expert witness in more than 20 domestic violence-related cases. Dr. Messing specializes in IPV risk assessment. She has evaluated the predictive validity of several forms of the Danger Assessment (DA), including the Lethality Screen and the Danger Assessment for Law Enforcement (DA-LE). She is conducting the first U.S. evaluation of the Ontario Domestic Assault Risk Assessment (ODARA) and is on a research team with Dr. Campbell that is adapting the DA for use with immigrants, refugees, and Native American victims of IPV. As a social worker, Dr. Messing is committed to evidence-based practice and is concerned with the development and testing of innovative interventions for victims of IPV. She was the principal investigator on the National Institute of Justice-funded Oklahoma Lethality Assessment Study, which examined the effectiveness of the Lethality Assessment Program (LAP), a collaborative police–social service response to IPV. She is also a coinvestigator on two studies funded by the NIH that examine the utility of Internet-based decision aids for women in abusive relationships.

Assessing Dangerousness Domestic Violence Offenders and Child Abusers Third Edition

Jacquelyn C. Campbell, PhD, RN, FAAN, and Jill Theresa Messing, PhD, MSW Editors

Copyright © 2017 Springer Publishing Company, LLC All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Springer Publishing Company, LLC, or authorization through payment of the appropriate fees to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400, fax 978-646-8600, [email protected] or on the web at www​ .copyright.com. Springer Publishing Company, LLC 11 West 42nd Street New York, NY 10036 www.springerpub.com Acquisitions Editor: Debra Riegert Compositor: Westchester Publishing Services ISBN: 9780826133267 e-book ISBN: 9780826133274 17 18 19 20 21 / 5 4 3 2 1 The author and the publisher of this work have made every effort to use sources believed to be reliable to provide information that is accurate and compatible with the standards generally accepted at the time of publication. The author and publisher shall not be liable for any special, consequential, or exemplary damages resulting, in whole or in part, from the readers’ use of, or reliance on, the information contained in this book. The publisher has no responsibility for the persistence or accuracy of URLs for external or third-party Internet websites referred to in this publication and does not guarantee that any content on such websites is, or will remain, accurate or appropriate. Library of Congress Cataloging-in-Publication Data Names: Campbell, Jacquelyn, editor. | Messing, Jill Theresa, editor. Title: Assessing dangerousness : domestic violence offenders and child abusers / [edited by] Jacquelyn C. Campbell and Jill Theresa Messing. Description: Third edition. | New York, NY : Springer Publishing Company, LLC, [2017] | Includes bibliographical references and indexes. Identifiers: LCCN 2016055478 (print) | LCCN 2016056008 (ebook) | ISBN 9780826133267 (hard copy : alk. paper) | ISBN 9780826133274 (ebook) Subjects: | MESH: Domestic Violence—psychology | Domestic Violence—prevention & control | Criminals—psychology | Homicide—prevention & control | Risk Assessment | Models, Psychological | Forecasting Classification: LCC RC569.5.V55 (print) | LCC RC569.5.V55 (ebook) | NLM WM 605 | DDC 616.85/8200112—dc23 LC record available at https://lccn.loc.gov/2016055478 Contact us to receive discount rates on bulk purchases. We can also customize our books to meet your needs. For more information, please contact: [email protected] Printed in the United States of America by Gasch Printing.

For the wonderfully strong and loving women in my family—my daughter, Christy; daughter-in-law, Nadia; and granddaughters Grace, Sophie, and Leila; my sister, Deborah, and sister-in-law, Shelley—and the strong but not afraid to be gentle men who love us—Reg, Brad, Nik, Patrik, Joe, and of course Nathan—and in tribute to the strong and loving women who came before—my mom, Dorothy, and Reg’s mother, Constance. —Jackie For my daughters.

—Jill

Contents Contributors  xi Preface  xvii Share Assessing Dangerousness: Domestic Violence Offenders and Child Abusers, Third Edition 1. Prediction of Interpersonal Violence: An Introduction   1 Allison Ward-Lasher, Daniel J. Sheridan, Nancy E. Glass, and Jill Theresa Messing Classic Clinically Based Prediction Models   2 A Victim-Service Reality: Community-Based Intervention   6 Reliability and Validity   8 An Evidence-Based Practice Model for Assessing Risk   9 Predictive Factors   13 Ethical Considerations   20 Summary  21 2. Prediction Issues for Practitioners   33 Joel S. Milner, Jacquelyn C. Campbell, and Jill Theresa Messing Clinical Versus Statistical Prediction Strategies   34 Legal Issues and Prediction   36 Ethical Issues and Prediction   37 Psychometric Issues in Clinical Practice   38 Approaches to Developing Predictive Instruments   38 Test Reliability   39 Test Validity   41 Other Measurement Issues   47 Summary  50 vii

viii C o n t e n t s

3. Child Physical Abuse Risk Assessment: Parent and Family Evaluations   55 Joel S. Milner and Julie L. Crouch Risk Assessment in Primary, Secondary, and Tertiary Prevention   56 Child Physical Abuse Risk Factors   58 Research Issues That Impact Child Physical Abuse Risk Assessment   60 Determination of Child Physical Abuse Risk   61 Summary  76 4. Evaluating Risk Factors for Fatal Child Abuse   89 Scott D. Krugman and Francie J. Julien-Chinn Context and Definition   89 Incidence of Child Fatalities   90 Causes of Child Fatalities   91 Child Death Investigation   91 Child Abuse Fatality Typologies   95 Risk Factors for Fatal Child Abuse   97 Prevention  100 Summary  101 5. Prediction of Homicide of and by Battered Women   107 Jacquelyn C. Campbell, Jill Theresa Messing, and Kirk R. Williams Homicide and Intimate Partner Violence   108 Prediction Issues   111 Published Lists of Danger Signs   114 The Danger Assessment   114 Future Directions in Lethality Risk Assessment: A Community Approach   126 Summary  129 6. Assessing Risk of Intimate Partner Violence   139 N. Zoe Hilton and Angela Wyatt Eke Risk Markers and Correlates of IPV   140 Risk Factors for Repeated Assault by IPV Offenders   144 Risk Assessment Instruments for IPV   145

C o n t e n t s ix

Frequent and Severe Assault   152 Assessing Risk After Intervention   153 Practice Issues   156 Summary  160 7. Children at Risk of Homicide in the Context of Intimate Partner Violence   179 Peter Jaffe, Jordan Fairbairn, and Katherine Reif Homicides Committed by Parents   180 Child Homicides and Child Maltreatment   182 The Overlap of Child Abuse and Intimate Partner Violence   186 Assessing Risk for Child Homicide in Intimate Partner Violence Contexts   190 Summary  199 Index  209

Contributors

Julie L. Crouch, PhD, is director of the Center for the Study of Family Violence and Sexual Assault at Northern Illinois University. She has more than 20 years of experience in conducting research on family violence, sexual assault, and trauma. Her primary research interests focus on examining the causes and consequences of child physical abuse, as well as investigating interventions that reduce the risk for child abuse. Angela Wyatt Eke, PhD, has worked with the Behavioural Sciences and Analysis Section of the Ontario Provincial Police (OPP) since 1997 and is the coordinator of research within the Criminal Behaviour Analysis Unit. Angela’s areas of research include risk assessment with a focus on intimate partner violence (IPV) and child pornography. In addition to publishing and providing training in these areas, she is on the editorial board of the Journal of Family Violence as well as Sexual Abuse: A Journal of Research and Treatment. Angela is an adjunct faculty member at Laurentian University and teaches an undergraduate forensic psychology course. She earned degrees from the University of Toronto and York University. Jordan Fairbairn, PhD, is a postdoctoral fellow with the Centre for Research and Education on Violence Against Women and Children (CREVAWC) at Western University in London, Canada. Her research focuses broadly on gender, violence, and media, with a particular interest in social responses to domestic violence and the role of social media and digital technology in violence and violence prevention. Jordan is currently the national research coordinator for the Canadian Domestic Homicide Prevention Initiative with Vulnerable Populations (CDHPIVP), led by codirectors Dr. Peter Jaffe and Dr. Myrna Dawson. xi

xii C o n t r i b u t o r s

Jordan completed her PhD in sociology at Carleton University in 2015, where her doctoral research explored how people involved in prevention programs for violence against women use and experience social media. In 2013, Jordan was the lead investigator on Crime Prevention Ottawa-funded research on sexual violence, social media, and youth, a project conducted in partnership with the Ottawa Coalition to End Violence Against Women (OCTEVAW). Jordan received her MA from the University of Guelph in 2008. This research, published in Feminist Criminology, explored Canadian news portrayals of domestic homicide and how this coverage has changed over time. Nancy E. Glass, PhD, MPH, RN, is a professor and associate dean of research, Johns Hopkins School of Nursing, and associate director, Johns Hopkins Center for Global Health. Dr. Glass conducts multidisciplinary studies across diverse global settings to test employment, economic empowerment, and safety interventions to improve the health and well-being of survivors of gender-based violence and their families. N. Zoe Hilton, PhD, C. Psych., is associate professor of psychiatry at the University of Toronto; senior research scientist at the Waypoint Centre for Mental Health Care in Penetanguishene, Ontario; and a registered psychologist. She earned degrees from the University of Southampton, the University of Cambridge, and the University of Toronto. Her applied research activities and collaborative work with policing services and other community agencies have led to several research grants and professional awards. Her research publications primarily concern domestic violence, risk assessment, and risk communication in community, correctional, and forensic psychiatric populations. Peter Jaffe, PhD, is a psychologist and professor in the faculty of education at Western University and the academic director of the Centre for Research and Education on Violence Against Women and Children (CREVAWC). He has coauthored 10 books, 24 chapters, and more than 75 articles related to the justice system and violence and abuse involving children, adults, and families. Many of his publications and professional presentations deal with domestic violence, the impact of domestic violence on children, and child custody and access disputes. He has presented workshops across the United States and Canada, as well as Australia, New Zealand, Costa Rica, and Europe to various

C o n t r i b u t o r s xiii

groups, including judges, lawyers, health and mental health professionals, and educators. Since 1999, he has been on faculty of the National Council of Juvenile and Family Court Judges in the United States for judicial education programs titled “Enhancing Judicial Skills in Domestic Violence Cases.” He was a founding member of Ontario's chief coroner’s Domestic Violence Death Review Committee. In 2009, he was named an Officer in the Order of Canada by the governor general for his work preventing domestic violence in the community. Francie J. Julien-Chinn, MSW, is a doctoral candidate at Arizona State University (ASU) in the School of Social Work. She earned her BSW at Northern Arizona University and her MSW at ASU. Ms. Julien-Chinn practiced as a social worker in child welfare for many years before beginning her doctoral education. Ms. Julien-Chinn’s research centers on workforce matters, including organizational culture and climate, and supervision within child welfare agencies and the impact of organizational factors on outcomes for children and families. Her dissertation focuses on permanency outcomes for children in out-ofhome care. Ms. Julien-Chinn is a current fellow in the Doris Duke Fellowships for the Promotion of Child Well-Being. In her research, Ms. Julien-Chinn also studies topics related to kinship families, licensed foster parents, and family resiliency. Scott D. Krugman, MD, MS, FAAP, is the chairman of the Department of Pediatrics and the director of medical education at MedStar Franklin Square Medical Center (MFSMC). He is the associate dean for medical education at MFSMC, Georgetown University School of Medicine, and professor of pediatrics at Georgetown University Medical Center, as well as clinical professor in the Department of Pediatrics and the Department of Epidemiology at the University of Maryland School of Medicine. Dr. Krugman chairs the Baltimore County Child Fatality Review Team and is a member of the Baltimore County Child Protection Review Panel. He has served on numerous statewide child abuse committees and is a faculty member of the Maryland Child Abuse Medical Providers (CHAMP). Joel S. Milner, PhD, is professor emeritus (clinical psychology), distinguished research professor, and founder director/director emeritus of the Center for the Study of Family Violence and Sexual Assault at Northern Illinois University. He is the author or coauthor of more than 200 scholarly publications, the majority of which describe

xiv C o n t r i b u t o r s

empirical studies in the areas of child maltreatment, intimate partner violence (IPV), and sexual assault. His research has focused on the description and assessment of child physical abusers, child sexual abusers, spouse abusers, and their victims. Katherine Reif, MA, is a PhD student in the School and Applied Child Psychology program in the Faculty of Education at Western University as well as a research assistant with the Centre for Research and Education on Violence Against Women and Children (CREVAWC). She has a bachelor of science degree from the University of Toronto and a master of arts degree in counseling psychology from Western University. Her master’s thesis focused on cyberbullying in the context of dating relationships among adolescent populations. She completed her internship at the London Family Court Clinic, where she worked in the Clinical Supports Program, providing counseling services to youth in conflict with the law. She currently works primarily on the Canadian Domestic Homicide Prevention Initiative with Vulnerable Populations (CDHPIVP) project with Dr. Peter Jaffe, specifically looking at children at risk from domestic homicide. She is examining professionals’ and community agencies’ responses to children at risk of domestic homicide in the context of parental separation. Daniel J. Sheridan, PhD, RN, FNE-A, SANE-A, FAAN, is a professor at the Texas A&M University College of Nursing in Bryan, Texas, where he is the director of its Forensic Health Care Education, Research, and Intervention Program. Dr. Sheridan is also an adjunct professor at the Goldfarb School of Nursing at Barnes-Jewish College in St. Louis, Missouri, primarily teaching and advising doctoral nursing students enrolled in an online program of study. In December 2013, Dr. Sheridan retired as an associate professor after teaching 12 years at the Johns Hopkins University School of Nursing, where he developed forensic nursing courses at the baccalaureate, master’s, and doctoral degree levels. Dr. Sheridan works as a consultant and expert witness for the Oregon Department of Human Services Office of Adult Abuse Prevention and Investigations, which investigates suspected abuse and neglect of the elderly and persons with profound cognitive and physical disabilities. Dr. Sheridan has more than 30 years of experience working with survivors of domestic violence, family violence, elder and vulnerable abuse/neglect, and sexual assault. Dr. Sheridan has published more than 50 clinical and research articles on nursing’s role in abuse and forensics, and has given

C o n t r i b u t o r s xv

more than 600 invited lectures on these topics nationally and internationally. Allison Ward-Lasher, MSW, is a doctoral student in the School of Social Work at Arizona State University (ASU), where she received her bachelor’s degree in psychology and her master’s degree in social work. Before returning to ASU to pursue her doctorate, Ms. WardLasher was a social worker and advocate in her community. Ms. WardLasher’s research focuses on intimate partner violence (IPV), including interactions between criminal justice and social service practitioners, risk factors for homicide, innovative interventions for survivors, and victim advocacy. She is currently assisting Drs. Campbell and Messing in the development of the Danger Assessment (DA) for immigrant women. Her dissertation examines victim advocates’ experiences working within a criminal justice context and the impact of the criminal justice system on their work with survivors. Kirk R. Williams, PhD, is professor of criminology, law, and society at the University of California, Irvine. He has published widely on the causes and prevention of violence, particularly involving youth or adult intimate partners. In addition to financial support from various private foundations, he has received numerous grants from federal and state funding sources to support his research. He has also worked extensively with community-based groups, schools, and agencies in violence prevention planning, implementation, and evaluation.

Preface

Practitioners in the helping professions (e.g., nursing, social work, psychology) often serve perpetrators and survivors of interpersonal violence, and many are asked to make predictions about the likelihood of future violence. Knowledge about risk and risk factors is increasingly expected in courts, clinics, conference rooms, shelters, hospital emergency rooms, child protective service offices, schools, research settings, batterer intervention programs, parenting programs, domestic violence advocacy programs, and child abuse and intimate partner violence (IPV) prevention programs. All of the contributors to this volume have been in practice situations during which we were asked to determine the presence, scope, and likelihood of future violence, either to thwart an offender, protect a survivor, or both. Like you, we have faced the difficult problem of predicting child abuse, IPV, or both, and are acutely aware of the implications of these determinations for our clients and the profound professional and ethical responsibilities that lie therein. Much has changed since the second edition of this book, and even more since the first edition was published more than 20 years ago. The research on IPV and child abuse risk assessment has grown in scope and sophistication, we understand more about risk and risk factors than we ever have before, and interventions that take into account risk are being used in many practice settings. The contributors to this volume are all practitioners to some extent. As such they collaborate with practitioners from across disciplines and share a profound respect for the work that practitioners are doing to end gender-based violence and child abuse. It is from this perspective that we provide the most up-to-date information on assessing dangerousness, employing language and an approach that are user-friendly. Our goal was to be helpful to practitioners and researchers with varied roles and professional experience by incorporating expertise in both research and practice. Thus, we offer you our xvii

xviii P r e fa c e

summary of the research in this area, as well as suggestions for instruments that may be helpful and the criteria by which to judge them. To this, you will add your own clinical expertise and each client will maintain his or her self-determination. We would like to thank all of the survivors of violence who have helped us understand what it is like to be in danger. There is still much to be learned about risk and safety; the bravery of women and children who survive abuse and share their stories to help others is phenomenal. We stand in tribute and remembrance to all of those who have died as a result of IPV and child abuse in this country and around the world, and to their families whose lives have been tragically altered. Thank you, also, to the practitioners who work every day with victims and survivors of abuse. Your work is so very important and your contributions are valued beyond measure. We particularly appreciate all of the practitioners we have trained, collaborated, advocated, and corresponded with throughout our respective careers. We also would like to thank all of the contributors to this volume for their responsiveness to reviews, their patience with the process, and their collective wisdom and commitment to these issues. Finally, we would like to acknowledge the help and support of our students, former students, and research colleagues. We admire and are proud to work with dedicated practitioners and colleagues from all over the globe. We are honored to stand with you to better comprehend and intervene in violence in the hope of eradicating it, particularly violence that is borne of power imbalance and fractures the sacred spaces intended to be safe and nurturing. At the same time, we strive toward a world in which peace, love, and respect are the cornerstones of family interactions; where there is gender equality; and where we celebrate the diversity of those with whom we live in this world. Jacquelyn C. Campbell, PhD, RN, FAAN Jill Theresa Messing, PhD, MSW

Share Assessing Dangerousness: Domestic Violence Offenders and Child Abusers, Third Edition

C HAPT ER 1 Prediction of Interpersonal Violence: An Introduction Allison Ward-Lasher, Daniel J. Sheridan, Nancy E. Glass, and Jill Theresa Messing

Practitioners who work with perpetrators or survivors of interpersonal violence are asked frequently to make predictions about their clients’ violent relationship.* Most notable, those who come in contact with survivors and perpetrators through the criminal justice system (i.e., law enforcement or civil or criminal court) or through child and elder protective services are often asked to predict the likelihood of future violence by alleged or convicted family violence and sexual assault perpetrators. These assessments of “dangerousness” serve the primary functions of informing interventions and developing safety strategies (often through safety planning or community referral) for survivors and controlling future violent behaviors of the perpetrator by treatment or confinement (Campbell & Glass, 2009; Campbell, Sharps, & Glass, 2001; Gondolf, Mulvey, & Lidz, 1990; Messing & Campbell, 2016). More recently, there has been a focus on collaborations between social service and criminal justice practitioners to address the dual goals of survivor safety and offender accountability; when these interventions utilize risk assessment they have been termed risk-informed collaborative interventions (Messing & Campbell, 2016; also see Chapter 5). In this chapter, we review what is generally known about the prediction of violent behavior and then discuss implications for the prediction of interpersonal violence. Succeeding chapters address the specific variables involved in the prediction of child abuse and neglect, child *Throughout this volume, the term client is used to refer to both client and patient.

1

2 ASSESSING DANGEROUSNESS

fatalities (including those that occur within the context of intimate partner violence [IPV]), IPV, and femicide. This volume represents the most current research, trends, and professional viewpoints regarding the prediction of interpersonal violence. ■■

CLASSIC CLINICALLY BASED PREDICTION MODELS

The prediction of interpersonal violence demands the use of psychometrically sound measurements and an understanding of these tools’ limitations. Classic research in clinical decision making (Benner, 1984; Harbison, 1991; Schon, 1983b) identifies three major models for prediction: (a) the linear, rationalist model; (b) the hypothetico-deductive model; and (c) the risk assessment model (Gottfredson & Gottfredson, 1988). Depending on the goal of the assessment, the practitioner may use aspects of one or more of these models. Practitioners who are ethically bound with a duty to protect or duty to report could use these models as guidelines during assessments.

Linear (Rationalist) Model Because prediction has such significant forensic implications, practitioners may use a linear model, including a decision tree or critical pathway, to guide them when making decisions that have legal ramifications. For example, Gross, Southard, Lamb, and Weinberger (1987) proposed seven steps to follow when a client makes suggestive threats. Step 1 is to clarify the threat. Many clients make vague comments that may or may not indicate a real danger. Thus, the practitioner must take the time to fully explore intent. For example, after an acute beating, a survivor may state that she wishes someone would “blow his [the abuser’s] brains out.” In this case, the practitioner needs to ask the client directly whether she intends to kill her abuser. This woman simply may be expressing her anger rather than verbalizing a true threat. Further inquiry might reveal that she does not own or have access to a firearm. The risk factor for retaliatory violence is therefore low, especially when compared with the client who tells the practitioner that she would like to kill her abuser and has borrowed her brother’s loaded handgun. Thus, if there is a clear threat, Step 2 is to assess its lethality, as well as the likelihood of the person acting on the threat. As with suicidal thoughts, not all “threats” pose a true danger or can be enacted. The incarcerated client or hospitalized patient may verbalize specific threats

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

3

of violence against someone outside of prison or the hospital but has no means to carry through on the threats. If there is evidence of danger, Step 3 is to identify a specific, intended target of violence. In family violence and family sexual assault cases, it is easy to identify intended victims. The violence is seldom random, even within homes in which multiple members reside. The practitioner working with a client who is verbalizing concerns about physically and/ or sexually assaulting a stranger may find it more difficult to identify a specific victim (by name). However, the practitioner can ask the client to indicate the gender and any specific characteristics of the intended victim. If the person can name the intended victim or specifics about the type of victim who will be sought, the threat of harm is imminent (Step 4). At this point, the practitioner needs to consider his or her duty to warn the specified victim. Specifically, according to the California Supreme Court’s Tarasoff decision, When a therapist determines, or pursuant to the standards of his [or her] profession should determine, that his [or her] patient presents a serious danger of violence to another, he [or she] incurs an obligation to use reasonable care to protect the intended victim against such danger. The discharge of this duty may require the therapist to take one or more of various steps. Thus, it may call for him [or her] to warn the intended victim, to notify the police, or to take whatever steps are reasonably necessary under the circumstances. (Tarasoff v. Regents of University of California, 1976) For more detail, the reader is referred to the body of literature on the Tarasoff decisions (Tarasoff v. Regents of the University of California, 1976). The practitioner must also take into account the client’s relationship to the intended victim (Step 5). If the intended victim is a family member, rather than a political figure, the practitioner may employ different preventive and treatment strategies. Step 6 requires the practitioner to decide whether a family or couples therapy intervention would be suitable. In cases of severe or ongoing violence, or when the survivor is afraid, the abuser is controlling, or there are risk factors for homicide (including the use of weapons) present, family therapy may increase the risk of violence to women and children (Bograd & Mederos, 1999; O’Leary, 2001). Finally, Step 7 requires the practitioner to consider whether civil commitment or involuntary hospitalization would provide the greatest

4 ASSESSING DANGEROUSNESS

good to the client and potential survivor(s). At the completion of Step 7, the practitioner needs to follow up on the results of the decisions made and may need to recycle through the decision tree at a later date. The strength of the linear model is that it provides relatively clear direction for the practitioner, as well as a “logical” argument for the decision. Using the linear model, the practitioner approaches problem solving with some notion of probability. He or she weighs outcomes according to the objective standards or theory. The weakness of this model is also its objectivity; contextually relevant information is given little consideration. In other words, factors, such as treatment outcomes, social support, and stabilization of stress, are not considered in making the prediction. The decision is driven by formula, more than by the specifics of the actual situation.

Hypothetico-Deductive Model By contrast, the hypothetico-deductive model tends to be relational and complex in assessing factors that influence professional decisions in practice. As with the linear model, the practitioner weighs different factors, but the problem is considered more in context. In addition, past experiences with similar situations provide the practitioner with patterns of cues to consider and ways to categorize the cues. In considering all of the information in the current situation, the expert is searching for a “pivotal cue” to frame all of the cues and to link with extensive theoretical and experiential knowledge (Regan-Kubinski, 1991; Schon, 1983a). After the practitioner has focused his or her questioning and assessment, he or she begins to search specifically for additional cues relevant to violence and protectiveness. In clustering the cues, the expert continually loops back to the context of the specific client and to the overall context of the community in which the situation is occurring. Finally, the practitioner arranges the cues into some hypotheses and reviews the hypotheses for completeness. He or she may seek additional cues to complete the picture, if necessary. The hypotheses then are tested for confirmation or refutation, and a final decision is made. The following case example illustrates this process: While tightly clenching his fists, a young man tells his high school counselor that his grades plummeted because his girlfriend, whom he refers to in sexually derogatory terms, broke off their relationship. The counselor knows that this student has a history of frequent alcohol abuse and fighting on school grounds. The young man’s father is in the Army on an extended

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

5

overseas assignment. The mother reports that her son refuses to accept her authority and that he has become difficult to manage in the absence of his father. Using the hypothetico-deductive model, the counselor first focuses on the cues of anger, age, rejection by the girlfriend, and alcohol abuse, reaching the pivotal cue of “potential for dangerousness.” The counselor also hypothesizes that the young man may be depressed, feeling out of control, and feeling abandoned, all of which would contribute to his potential for violence. In addition, the counselor considers such cues as the school’s location near Ferguson, Missouri, and the school’s climate after the shooting of Michael Brown, an unarmed African American teenager, by a police officer (Day, 2015); reports of several similar situations with the other young men in the school; and reports that the girls in the school have been complaining about violence by the boys. The counselor reaches the judgment that not only is this young man potentially dangerous, but also that there may be a systemic problem in the school and community. Thus, the immediate plan is to confront the young man’s anger and to recommend some structured physical activity. The counselor also determines that the girlfriend has different classes from the boyfriend and that the possibility of contact that day is slight. The young man contracts to stay away from the girlfriend and not harm her. To meet the community problem recognized by this model, the counselor consults with his female colleague and together they plan a special assembly on the topic of dating violence. They also set up gender-segregated peer groups to discuss race- and gender-based violence in the community and within dating relationships.

Risk Assessment Model A major reason for poor predictive accuracy of interpersonal violence is the assumption that violence is dichotomous and has a single dimension (Gottfredson & Gottfredson, 1988). Instead of a binary notion of violence, Gottfredson and Gottfredson propose a risk-to-stakes matrix wherein the seriousness of the action is weighed with the likelihood of repetition. Seriousness permits the assessor to consider types of harm possible across a multitude of variables. Alcohol and drug use, for example, might influence the likelihood of harm, as might a history of violence. By means of the risk assessment model, practitioners can provide assessments of risk factors or risk markers that may contribute to violence. Such a model incorporates the social and political climate, as well as the individual’s internal climate.

6 ASSESSING DANGEROUSNESS

The risk assessment model permits practitioners to weigh both the environmental and personal factors present in any given situation. The following case demonstrates this model: Convicted of felony assault on his former girlfriend, a 28-yearold male with a history of alcoholism is up for parole after serving 85% of a 12-month sentence. While in prison, he completed extensive alcohol treatment and domestic violence offender programs. Upon release, he intends to live with his mother. As the practitioner, it would be imperative to know that his mother lives less than a block from the former girlfriend and that several alcoholic siblings also live there. Releasing this man into his mother’s home, into close contact with alcoholic siblings, places him at high risk for drinking. Because his mother lives so close to his former girlfriend, further abuse and stalking are also quite possible. Instead of recommending against parole, the practitioner may advise that parole be contingent on housing arrangements that do not place the offender in such close contact with either alcohol or his prior victim. The three clinical decision-making models discussed are distinct, but not necessarily exclusive, means for deciding on interventions. In courtrooms, linear decision routes are much easier to substantiate. However, decisions are rarely so clear-cut in the clinical arena. Therefore, practitioners need to be adept in approaching (or at least justifying) these decisions from multiple perspectives. Although the prediction of interpersonal violence is a relatively young science, it is an area of utmost importance. As aptly stated by Hilton and Harris (2005), “Predicting violence is quite a different task from explaining it” (p. 3). Practitioners working in the field will always be concerned that a case with which they have worked will end in a serious injury, homicide, or homicide–suicide unless they take every possible action to avert such an outcome. Consider the following scenario. ■■

A VICTIM-SERVICE REALITY: COMMUNITY-BASED INTERVENTION 10:00 a.m.: A police officer has just called you to the scene of a domestic violence incident. The alleged perpetrator fled the scene with the children after severely assaulting his partner. The woman has multiple lacerations and bruises on her face and neck and exhibits other signs of strangulation. She tells you that 2 days ago the abuser kicked her multiple times and

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

7

threatened to harm her children if she called the police. He would not let her seek medical care. She thinks that a neighbor must have called the police because of the screaming inside the home. She tells you and the police that her husband has a permit to carry a concealed weapon and that he frequently has threatened to kill her and the children if she ever tried to end the relationship. She asks, “Do you think he would hurt my kids? Could he really kill me or my kids?” 10:25 a.m.: The police sergeant wants to know whether this ­woman’s husband poses a risk of harm to the children, his wife, and the community at large. 10:50 a.m.: A child protective services department investigator arrives at the scene and tells you she has been investigating this family because of a recent child abuse allegation filed by the school. The children did not arrive at school this morning. She asks you whether you think the children are at risk of abuse and whether the mother knows where they would go. 11:30 a.m.: You are called by another officer to come to a scene where a woman’s home was broken into by her former husband. The survivor had a current order of protection against her former husband for stalking and threatening to kill her. The responding officers ask whether you think this man is capable of coming back to kill his former wife. 1:00 p.m.: You are called by the prosecuting attorney’s office to confirm that you will testify in court later that day around 3:00 p.m. on a sexual assault case your crisis team responded to several months ago. You are informed that you will be asked to share with the court your assessment of whether the offender is likely to recidivate. The aforementioned scenario is a real-life example of a day in the life of a practitioner who works with domestic and sexual violence survivors in crisis. As illustrated, the practitioner is called on repeatedly to make assessments and predictions of risk for repeat violence, often after obtaining only a cursory history of the violent behavior and without any direct contact with the alleged perpetrator. Practitioners in acuteincident settings (e.g., hospital social workers, forensic nurses, physicians, field investigators, hotline workers, victim advocates) are especially pressed for time.

8 ASSESSING DANGEROUSNESS

■■

RELIABILITY AND VALIDITY

The accuracy (validity) and consistency (reliability) of predicting dangerousness and violence depend on multiple, complex factors (also see Chapter 2). In general, the rarer an event, the more difficult it is to predict (Campbell, Webster, & Glass, 2009; Lambert, Cartor, & Walker, 1988). For example, predicting the risk of IPV reassault has become the primary aim of the majority of IPV risk assessment instruments. IPV reassault is easier to accurately predict because it has a much higher occurrence (approximately 25%–30% of IPV cases) than does intimate partner (IP) homicide (approximately 0.04% of IPV cases; Campbell, 2004). Factors that are known to influence the accuracy, or “validity,” of predicting dangerousness include the type of violence (e.g., physical assault, sexual assault, homicide), the perpetrator’s relationship to the survivor (e.g., stranger, intimate, acquaintance), the characteristics of the perpetrator (e.g., history of violence, mental health issues), and the time period of the prediction (e.g., acute danger or chronic danger). The following chapters discuss in greater depth challenges with assessment measures and factors used to predict future violence. It is clear, however, that assessments of risk for future violence are improved when appropriately administered, psychometrically sound risk assessment scales are used. Furthermore, practitioners need to couple these objective measures with information collected on the characteristics of the perpetrator, the perpetrator’s relationship to the victim, the victim’s assessment of risk, the practitioner’s experience and judgment, and context-specific factors (e.g., poverty, unemployment, discrimination, social support).

A Poor Record of Past Prediction Previous research examining the accuracy of clinical prediction compared with statistical assessment when it comes to risk for violence, recidivism, or revictimization has been relatively consistent. Grove, Zald, Lebow, Snitz, and Nelson’s (2000) meta-analysis concluded that in some instances, professional judgment was just as accurate as statistical assessment, but not in the majority of instances. However, practitioners who assess dangerousness, in general, have a poor track record of predicting future violence (Convit, Jaeger, Lin, Meisner, & Volavka, 1988; Gondolf et al., 1990; McNeil, Binder, & Greenfield, 1988; Miller & Morris, 1988). For example, although an assessment of danger to others and/or self is a basic assessment element of involuntary confinement or psychiatric treatment,

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

9

individual practitioners have not been very successful in accurately predicting this danger for violence victims (Beigel, Barren, & Harding, 1984; McNeil et al., 1988; Meloy, 1987; Steadman & Morrissey, 1982). In fact, Ægisdóttir et al.’s (2006) meta-analysis extended Grove et al.’s (2000) work by limiting the review to predictions relevant to mental health practitioners and found that statistical prediction is more accurate than professional judgment when it comes to predicting violence (Ægisdóttir et al., 2006). Actuarial risk assessments are evidence-based predictions based on statistical analysis (Hilton & Harris, 2005). These assessments use risk factors that have been found to be predictive of violence and remove clinical judgment from the assignment of risk. Instead, professional judgment is used only when informing interventions, such as safety planning or referrals to services for survivors, or criminal sanctions for offenders, such as pretrial detention or batterer intervention programs. A compromise between unstructured professional judgment and actuarial risk assessments is risk assessments that use structured professional judgment, such as the Spousal Assault Risk Assessment Guide (SARA). These assessments differ from actuarial risk assessments because they provide practitioners with the factors necessary to assess risk but allow practitioners to use their professional judgment when determining a risk level (Cattaneo & Chapman, 2011; Messing & Thaller, 2015). In addition, predictions about violence can be made more accurately when evaluators take into account such interactive factors as age, gender, unemployment, perpetrator–survivor relationship status, perpetrator’s history of violence, use of alcohol and/or illegal substances, history of mental health issues, and availability of guns (Campbell et al., 2001; Campbell, Glass, Sharps, Laughon, & Bloom, 2007; Meloy, 1987; Segal, Watson, Goldfinger, & Averbuck, 1988). Although static risk factors, such as gender and prior history of violence, cannot be changed by intervention, dynamic risk factors, such as unemployment, perpetrator access to guns, and use of alcohol and illegal drugs, can be the focus of intervention. Risk for future violence may thereby be reduced (Campbell et al., 2003; Messing & Thaller, 2015). ■■

AN EVIDENCE-BASED PRACTICE MODEL FOR ASSESSING RISK

Using an evidence-based practice model when assessing the risk for reassault or lethality requires a practitioner to know the current research and available tools for assessing risk, to use professional judgment, and to consider client self-determination (Gambrill, 2006; Messing & Thaller, 2015).

10 A S S E S S I N G D A N G E R O U S N E S S

Evidence-Based Risk Assessment Tools An important aspect of an evidence-based practice model is the use of research to guide practice decisions, including the use of risk assessment tools with the best available evidence of their ability to predict future violence or homicide. There are several risk assessment tools available; these are discussed in depth in subsequent chapters. Briefly, these tools vary in terms of where the practitioner gathers information (i.e., case files or survivor interviews), practice setting, the skills of the assessor, and the outcome of interest (Messing & Thaller, 2015). Most risk assessments, such as the Violence Risk Appraisal Guide (VRAG; Quinsey, Harris, Rice, & Cormier, 1998; Rice, Harris, & Lang, 2013), Ontario Domestic Assault Risk Assessment (ODARA; Hilton & Harris, 2009; Hilton et al., 2004), and Domestic Violence Screening Instrument—Revised (DVSI-R; Williams, 2012; Williams & Grant, 2006), are intended for use within the criminal justice system. In the criminal justice system, documentation and/or testimony regarding the defendant’s propensity for violence may influence a variety of judicial outcomes. For example, the court may ask the practitioner to predict the likelihood of future violence when sentencing a convicted offender. The expert opinion of the practitioner may significantly affect the type and length of the offender’s sentence. Likewise, such testimony may influence the eligibility of convicted offenders to participate in new, innovative, deferred-sentencing programs or other forms of alternative dispensation, such as community-based treatment programs. In family court, practitioners’ predictions of future violence may influence the court’s ruling on an order of protection or on issues of child custody and protection (Weisberg, 2016a, 2016b). Ideally, an assessment of risk will be revisited several times, in a number of settings, and by various practitioners. When choosing an evidence-based risk assessment, practitioners must consider several things. First, and perhaps most important, is the instrument’s predictive validity, or its ability to accurately determine the risk of an outcome of interest to the practitioner. In a review of risk assessment instruments for IPV, Messing and Thaller (2013) found that, when examining reassault, the ODARA has the highest predictive validity followed closely by the SARA, Danger Assessment (DA), and Domestic Violence Screening Instrument (DVSI). A practitioner wanting to predict homicide may choose to use the DA, as the intent of this instrument is to predict lethality. Another important consideration is that the tool itself has the ability to communicate the risk posed to a survivor across disciplines (Heilbrun,

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

11

O’Neill, Strohman, Bowman, & Philipson, 2000). This allows for practitioners to use or suggest risk-informed collaborative interventions that may increase safety for survivors (Messing & Campbell, 2016; WardLasher, Messing, & Hart, in press). A practitioner’s choice of risk assessment is also dependent upon access to information. For example, the DA collects information from the survivor, whereas the ODARA, DVSI, and VRAG require access to police case files. Assessments like the SARA require access to the offender, victim, and case files (Messing & Thaller, 2015). Practitioners should also consider their training or the setting in which they work when choosing a risk assessment. For example, the DA was created for use by a victim advocate or a health care professional. Short versions of this instrument, the Danger Assessment for Law Enforcement (DA-LE) and the Lethality Screen, are to be administered by police officers or first responders (Messing & Campbell, 2016). The Domestic Violence Risk Appraisal Guide (DVRAG) is a risk assessment tool that combines the ODARA and the Hare Psychopathy Checklist—Revised (PCL-R; Hare, Clark, Grann, & Thornton, 2000). It is suggested that the DVRAG is completed in two steps: The 13 ODARA questions are completed by police officers who first respond to a domestic violence call and the PCL-R is later completed by a practitioner who will inform judicial interventions (Hilton, Harris, Rice, Houghton, & Eke, 2008). Although the ODARA was created for police officers to make an assessment of risk at the scene of an IPV incident, administering the PCL-R requires significant additional training. Thus, it takes considerable training and time to learn how to administer and evaluate findings from the DVRAG (Hilton & Harris, 2005). Nonetheless, the PCL-R is a very strong predictor of violent recidivism, in general, and specifically among more serious male abusers of women (Harris, Skilling, & Rice, 2001). When choosing a risk assessment, it is also important to attend to the sensitivity and specificity of the measure (Messing & Campbell, 2016). The most accurate risk assessments will have high sensitivity (i.e., correctly classifying a reoffender as high risk) and high specificity (i.e., correctly classifying a person who does not reoffend as low risk). Often, however, the trade-offs between sensitivity and specificity are not equal. For example, there may be a greater cost to setting an offender free who will recidivate than to detaining an offender who will not, or to providing an intervention to someone who will not be revictimized versus failing to provide an intervention to someone who is subsequently reabused. When a risk assessment has high sensitivity or specificity, a practitioner must consider the utility of the assessment for the purpose that

12 A S S E S S I N G D A N G E R O U S N E S S

he or she wants to achieve. New statistical methods, such as machine learning, allow practitioners to weight prediction errors differentially to account for these difficult trade-offs (Berk, Sorrenson, & Barnes, 2016). A practitioner might choose to use a risk assessment with high sensitivity in order to ensure that the greatest number of survivors receive an intervention. On the other hand, if a risk-informed intervention has limited resources or is only meant for the cases at highest risk, a practitioner may choose a risk assessment that has higher specificity (Messing & Campbell, 2016).

Professional Judgment A practitioner’s assessment of dangerousness can be an enormously valuable resource, highlighting a need for the development of accurate and reliable models for the prediction of interpersonal violence. Across a variety of settings, actuarial assessments are more accurate than clinical prediction (Grove et al., 2000). Hilton and Harris (2005), in their review of research literature predicting IPV against women, state that actuarial risk assessment techniques and tools are far more accurate than unstructured clinical judgments or structured clinical risk assessment tools. Practitioners who assess risk for reassault or homicide are encouraged to use their professional judgment to understand and contextualize evidence-based risk assessment tools (Messing & Thaller, 2015). Professional judgment is important when informing interventions, safety planning with survivors, and also when current literature is not available about a specific population with whom a practitioner is working.

Client Self-Determination When predicting the reoccurrence of interpersonal violence between IPs, one important source of information is the survivor of violence (Hilton et al., 2004; Messing & Thaller, 2015; Weisz, Tolman, & Saunders, 2000). Although psychometrically validated instruments are more accurate than survivor perception of risk, survivors can provide important information about dynamic risk and protective factors (Campbell, O’Sullivan, Roehl, & Webster, 2005; Messing & Thaller, 2013). Considering survivor perception of risk in conjunction with an evidence-based risk assessment may produce a more accurate picture of risk for further violence and better assist practitioners with risk-informed interventions (Connor-Smith, Henning, Moore, & Holdford, 2011; Messing & Campbell, 2016).

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

13

In addition, it is imperative that survivor self-determination is not overwhelmed by a practitioner’s determination of risk. A survivordefined advocacy approach requires that advocates recognize the survivor as the expert in her situation (Davies & Lyon, 2014). Thus, client self-determination is a foremost consideration during safety planning and other risk-informed interventions. Survivors who have a voice during criminal justice procedures are generally more satisfied with the process and more likely to engage with the criminal justice system in the future (Fleury-Steiner, Bybee, Sullivan, Belknap, & Melton, 2006; I. M. Johnson, 2007; Stover, Berkman, Desai, & Marans, 2010). Safety planning is more effective when the specific needs of a survivor are assessed and her desires are considered (Davies & Lyon, 2014; Goodman & Epstein, 2008). Therefore, risk-informed interventions for survivors should consider risk for further violence or homicide, safety planning strategies that are culturally appropriate and suit the needs of the individual, and appropriate community resources (Messing & Thaller, 2015).

PREDICTIVE FACTORS History of Violence ■■

Research into the prediction of interpersonal violence consistently shows that a history of violence is one of the best predictors of future violence (Convit et al., 1988; Janofsky, Spears, & Neubauer, 1988; Lewis, Lovely, Yeager, & Femina, 1989; McNeil et al., 1988). For example, the most important risk factor of homicide in an intimate relationship is maleperpetrated violence against a female survivor. Between 65% and 80% of IP homicides have a reported history of IPV against the female partner (Bailey et al., 1997; Campbell et al., 2003, 2007; McFarlane, Parker, Soeken, Silva, & Reed, 1999; Moracco, Runyan, & Butts, 1998; Pataki, 1997; Sharps et al., 2001; Websdale, 1999). The majority of IP homicides are committed by male perpetrators against female victims (Catalano, Smith, Snyder, & Rand, 2009; Violence Policy Center [VPC], 2012). However, when women do kill their male partners, there is a documented history of male-­ perpetrated IPV against the female in as many as 75% of cases (Campbell et al., 2007). There is limited research on risk factors for lethal or near-lethal violence in marginalized populations, such as sexual minorities. For example, data from the Federal Bureau of Investigation’s Supplementary Homicide Reports (SHR) estimates that from 1976 to 2001, 2.1% of IP homicides were between same-sex couples (Mize & Shackelford, 2008). Of those, 1.8% were male same-sex couple IP homicides in comparison to

14 A S S E S S I N G D A N G E R O U S N E S S

0.3% female same-sex partner homicides. In one study of near-lethal male same-sex couple violence, incident characteristics consistent with existing risk factors for IP homicide were reported (i.e., history of violence, strangulation, use of a weapon, partner threatens to kill, and confinement; Loveland & Raghavan, 2014). Although the literature base has expanded for same-sex IPV, researchers have yet to systematically examine risk factors for male same-sex partner homicides. To date, there has been only one study of female same-sex IP homicides and a prior history of violence was a notable risk factor to the lethal or near-lethal violence event (Glass, Koziol-McLain, Campbell, & Block, 2004).

Mental Illness Studies have produced mixed information regarding the role of mental health issues and violence perpetration; more research is needed. Perpetrators of IP homicide appear to be more likely to have a history of mental illness. In one study, 13% of perpetrators (11% of males, 15% of females) of 540 IP homicides had a reported history of mental illness, compared with 3% (not reported by gender) of nonfamily murderers (Zawitz, 1994). In a recent study, 25.6% of male perpetrators of 71 IP homicides had a history of severe mental illness, compared with 11.6% of male perpetrators of 291 non-IP homicides (Thomas, Dichter, & Matejkowski, 2011). These findings are in stark contrast to Dobash, Dobash, Cavanagh, and Lewis’s (2004) U.K. study, whose findings indicated similar proportions of mental health issues in IP homicide perpetrators and non-IP homicide perpetrators. Overall, studies demonstrate a wide range in the prevalence of mental health issues among IP homicide offenders. In one study examining male perpetrators of IP homicide from a multicity project, more than half (55.1%) of offenders were described as being in fair or poor (versus good, very good, or excellent) mental health (Sharps et al., 2001). However, in a related multivariate analysis, perpetrator mental health was not a significant predictor of IP homicide, when included with other known risk factors (Campbell et al., 2003). In a more recent study, 32% of IP homicide perpetrators were found to have a mental health diagnosis, 14% had been treated for mental health within the year prior to the offense, and 20% experienced symptoms (mostly depression) at the time of the murder (Oram, Flynn, Shaw, Appleby, & Howard, 2013). Perpetrators who have mental health diagnoses, such as personality disorders, have a high likelihood of reoffending. Although men with personality disorders represent a relatively small percentage of men who

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

15

abuse women (15%–30%), their behavioral traits of superficial charm, need for stimulation, callousness, and manipulation are quite familiar to practitioners who provide services to their victimized partners (Hilton & Harris, 2005). Offenders with personality disorders have a history of early behavioral problems, impulsivity, antisocial history, and callousness. Practitioners are often called upon to make predictions of dangerousness as a requirement of their roles in health care settings, such as psychiatric hospitals and emergency departments, or in community agencies, such as shelters. One challenge to assessing risk of violence among clients who have been diagnosed with a mental illness is the criteria used to identify “violent” acts and “violent” people. For example, psychiatric clients have traditionally been assessed for danger to self and danger to others. As a result, suicide and self-mutilation may be included in the findings, potentially overestimating the client’s level of dangerousness. Likewise, the client may demonstrate very different patterns of behavior when hospitalized (Holcomb & Ahr, 1988; Myers & Dunner, 1984) simply because he or she is receiving treatment. Because severe mental illnesses (e.g., schizophrenia, bipolar disorder) are relapsing disorders, violence is a greater factor in times of decompensation and psychosis than during stabilization (Craig, 1982; Krakowski, Jaeger, & Volavka, 1988; Tardiff & Sweillman, 1982).

Substance Misuse Most research describes a relationship between the use of alcohol and other substances with incidents of violence (Foran & O’Leary, 2008; Moore et al., 2008; Thompson & Kingree, 2004). Research indicates a bidirectional association between drug use and IPV (Hayashi, Patterson, Semple, Fujimoto, & Stockman, 2016). There is a high prevalence of substance use among IPV offenders (Ernst, Weiss, Enright-Smith, Hilton, & Byrd, 2008) as well as substance misuse co-occurring with IPV victimization (Hayashi et al., 2016; Macy, Giattina, Parish, & Crosby, 2010). A survivor’s use of substances is often an attempt to self-medicate as a result of IPV and, in some cases, women may be forced or coerced into substance use or misuse by an abusive partner (Fowler, 2009). An important barrier exists when examining the relationship between substance misuse and violence: It is difficult to identify the temporal association between a perpetrator’s substance use and misuse in relation to a violent incident. Most research relies on self-report from either the perpetrator or the survivor about the perpetrator’s history of substance use or substance use at the time of the incident (Moore et al.,

16 A S S E S S I N G D A N G E R O U S N E S S

2008). In cases in which blood or hair samples are collected (i.e., suicide, murder–suicide, or perpetrator-provided samples), more concrete knowledge about the offender’s substance use at the time of the incident is available (Campbell et al., 2007). ALCOHOL The majority of research articles on substance use and its relationship to violence focus on alcohol use, mainly of the perpetrator. Alcohol is a depressant, and is associated with arousing, euphoric effects followed by dysphoric, depressing effects. Its effect on behavior can be roughly correlated with the level of alcohol present in the body as measured by blood alcohol content, although it does not follow a simple pharmacological dose–effect relationship, and is different in each person (Miczek et al., 1994). Theoretical explanations of alcohol’s role in child and partner violence consider both proximal and distal influences. The proximal effect model indicates that there should be a temporal relationship between substance use and violence, meaning that episodes of violent acts would closely follow ingestion of the substance. The psychopharmacological effects of alcohol on cognitive processing facilitate violent behavior, in that persons having ingested significant amounts of alcohol are more likely to interpret others’ behavior as hostile and less likely to be able to problem-solve a nonviolent solution to conflict, especially if they have learned violent behaviors in their family of origin. Distal influences include individual factors, such as personality characteristics and life experiences, and contextual influences, such as relationship type, all of which may create an environment that facilitates violent behavior, especially when alcohol is consumed (Chermack & Blow, 2002). Regardless of the mechanism, alcohol use is one of the accepted risk factors for child and partner violence. In fact, there have been consistent findings suggesting the association between alcohol use and IPV (Foran & O’Leary, 2008; Smith, Homish, Leonard, & Cornelius, 2012). Between 17% and 32.9% of men who perpetrate IPV report substance misuse or alcohol-related problems; the proportion increases when examining heavy or excessive drinking (up to 50%) and binge drinking (up to 59.7%; Langenderfer, 2013). Binge drinking, alcohol misuse or dependence, and drinking frequency have been associated with IPV perpetration (Foran & O’Leary, 2008; Messing, Mendoza, & Campbell, 2016; O’Leary & Schumacher, 2003). Alcohol use during an incident of IPV increases the risk for injury to a survivor (Thompson & Kingree, 2004). It is important to note that a perpetrator’s alcohol use is a major factor in

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

17

a survivor’s assessment of her own risk for future violence (ConnorSmith et al., 2011). Furthermore, alcohol use has been identified as a risk factor for IP homicide (Campbell et al., 2007). A large proportion of men who commit IP homicide also have a history of alcohol misuse (69.6%) and used alcohol prior to the murder (81.3%; Thomas et al., 2011). COCAINE Cocaine is a central nervous system stimulant, the strongest stimulant derived from natural sources. Initially, use of this drug reduces appetite and makes the user feel more alert, energetic, and euphoric. With high doses, users can become delusional, paranoid, and even suffer acute toxic psychosis. As the drug’s effects wear off, depression sets in, leaving the user feeling fatigued and anxious. Cocaine has also been shown to increase the incidence of IPV (Chermack & Blow, 2002; Moore et al., 2008; Smith et al., 2012). In a meta-analysis examining specific drug types and their impact on risk for IPV, cocaine has been found to have the strongest association with IPV of any drug (Moore et al., 2008). Specifically, cocaine was the only drug found to be a strong predictor of psychological abuse, physical abuse, and sexual abuse (Moore et al., 2008). METHAMPHETAMINE Methamphetamine is an intense, human-made stimulant. Upon ingestion, it releases high levels of the neurotransmitter dopamine, which causes excitation, euphoria, intensification of emotions, increased alertness, and heightened sexuality. Because methamphetamine is metabolized at a slower rate, a sustained euphoric state is produced that can last up to 8 hours (Cartier, Farabee, & Prendergast, 2006). A 2006 survey of U.S. counties found that 48% of counties report methamphetamine as their primary drug problem, more significant than cocaine, marijuana, and heroin combined. In addition, 62% of counties reported an increase in domestic violence and a 53% increase in simple assaults between 2004 and 2005 as a result of methamphetamine use (National Association of Counties, 2005). Nationally, methamphetamine use continues to be a public health issue (Maxwell & Brecht, 2011; National Institute on Drug Abuse, 2013). Violent behavior has been associated with methamphetamine use and the likelihood of violence among users increases with the frequency of use (McKetin et al., 2014). In a multistate study from 1999 to 2001, 1,016 methamphetamine users were examined. Eighty percent of women participating in the study reported abuse or violence by a partner

18 A S S E S S I N G D A N G E R O U S N E S S

(Cohen et al., 2003). Similarly, a recent study found that 79% of female methamphetamine users had been physically abused by an IP (Hayashi et al., 2016). Methamphetamine, which stimulates heightened sexuality responses, has been associated with IP sexual violence and risky sexual behaviors (Hayashi et al., 2016). However, just how much methamphetamine contributes to violent behavioral patterns has been difficult to research (Brecht & Herbeck, 2013). For example, some research combines methamphetamine into a category of substances and does not examine its effects on violence independently (Moore et al., 2008). Further research is needed to evaluate the use of methamphetamine and IPV. POLYSUBSTANCE USE The interaction between two or more drugs, or polysubstance use, could impact violence (Moore et al., 2008). For example, both alcohol and cocaine use have separately been associated with IPV perpetration. Furthermore, offenders who use both cocaine and alcohol are more likely to have perpetrated IPV (Smith et al., 2012). However, a majority of the literature on the impact of substances on IPV tends to isolate alcohol use/ misuse and combine dissimilar illicit drugs (i.e., marijuana, cocaine, methamphetamine) to determine the association with violent behavior (Moore et al., 2008). Thus, there is limited research on the use of multiple substances and IPV even though it has been well established that polysubstance use is common among substance users (Smith et al., 2012). The association between alcohol and drug use in IPV cases has consistently linked substance misuse with repeat violence and homicide (Campbell et al., 2007; Connor-Smith et al., 2011). For example, alcohol and drug use by the perpetrator were found to significantly predict lethal violence in intimate relationships; drug use/misuse was a stronger predictor than alcohol use/misuse (Campbell et al., 2003, 2007). However, neither perpetrator alcohol or drug use prior to the violent event remained significant predictors of lethal violence when controlling for other important factors, such as use of a gun, threats to kill, and a nonbiological child of the perpetrator in the home. In research examining the factors related to IP homicide, approximately 70% to 80% of male perpetrators were using drugs and/or alcohol at the time of the murder (Campbell et al., 2003; Sharps, Campbell, Campbell, Gary, & Webster, 2003). Research that has compared IP and non-IP homicide has reported inconsistent findings related to differences in alcohol and drug use during homicides. Dobash et al. (2004) found that although a substantial proportion of IP perpetrators had alcohol and drug

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

19

problems (37.9% and 14.7%, respectively), alcohol and drug problems were reported among a significantly higher proportion of males who commit other types of murders (nonintimate murder). Thomas et al. (2011) also found that a large proportion of homicide offenders had a history of substance use before the homicide. However, they did not find a significant difference in the rates of substance misuse between intimate and non-IP offenders.

Gender The literature about interpersonal violence tends to be gender specific, with many of the studies seeking to identify factors related to violence by primarily examining male behaviors (Chesney-Lind & Shelden, 2014; Cho, 2012; Josephson, 1987; Lewis et al., 1989; Mulvey & Reppucci, 1988). Therefore, the research provides a limited perspective on the precursors to interpersonal violence perpetrated by females. In the case of IPV, there is a debate between family violence researchers and feminist researchers regarding gender symmetry/asymmetry in the prevalence of IPV (M. P. Johnson, 2006; Lawson, 2012; Myhill, 2015). Family violence researchers argue that men and women equally perpetrate violence toward their partners (e.g., Straus, 2010). Conversely, feminist researchers argue that IPV is a form of gender-based violence that is predominantly perpetrated by men against women (e.g., Stark, 2007; for a review of the literature on this debate, see DeKeseredy, 2011; Dutton, 2012). Conceptualizing IPV using a coercive control framework suggests that there are types of violence within relationships that are gendered. Specifically, acts intended to exert power and control, such as stalking, harassment, and intimidation, are predominantly perpetrated against women by men (M. P. Johnson, 2006; Myhill, 2015; Stark, 2007). Research also indicates gendered consequences of physical violence; women who are physically assaulted are more likely than men to report injury (Black et al., 2011; Myhill, 2015) and to be killed by an IP (Campbell et al., 2007).

Race/Ethnicity There are clear racial/ethnic disparities in rates of IPV and IP homicide. Overall, ethnic minorities are at greater risk for IPV than White women (Black et al., 2011; Powers & Kaukinen, 2012). Specifically, Native American and multiracial women experience IPV at rates higher than other racial/ ethnic groups (Black et al., 2011) and Native American and African American women are at increased risk for IP homicide (Mercy & Saltzman,

20 A S S E S S I N G D A N G E R O U S N E S S

1989; Morton, Runyan, Moracco, & Butts, 1998). The racial disparities in IPV perpetration, victimization, and homicide may be attributed to the interconnection of race and socioeconomic status (Cho, 2012). Although a significant proportion of this variation can be explained by increased rates of unemployment among Native American and African American men (Campbell et al., 2003, 2007), there is variation among studies in measurement of socioeconomic status, making it difficult to determine what specific factors account for these discrepancies. For example, some studies have found race to be significantly associated with IP homicide in bivariate analyses, but when unemployment is added in to the model, race is no longer a significant factor (Campbell et al., 2003). Other studies examining risk factors for IPV suggest that a measure of financial security is a stronger predictor than employment status or education and helps to explain racial differences (Cho, 2012). In a multicity study, Walton-Moss, Manganello, Frye, and Campbell (2005) examined risk factors for lethality among ethnic groups and found that the majority of risk factors were similar for African American, White, Hispanic, and multiracial couples. However, they also found different strengths of risk factors in certain groups and some risk factors that did not apply to some groups (Walton-Moss et al., 2005). For instance, prior arrest was found to be strongly protective against IP homicide for White and multiracial couples, but not protective at all for African American and Hispanic couples. When looking at the data more closely, this was related to the finding that abusive males in ethnic minority groups (both those who committed homicide and those who did not) were more frequently arrested than White male killers. ■■

ETHICAL CONSIDERATIONS

Although the empirical study of the prediction of interpersonal violence is important, the practitioner who must render an assessment of the probability of future violence has a responsibility to weigh several ethical issues (also see Chapter 2). Whether or not practitioners engage in research or use formal risk assessment tools, they make predictions about dangerousness. These predictions may be based on past behaviors, known risk factors, similar behaviors that have been observed in others, clinical evaluation of the alleged offender, conversation with the survivor, and/ or risk assessment instruments. In making these assessments, practitioners must not only consider the social injustice of violence perpetrated against the survivor, but must also weigh the perpetrators’ individual rights to autonomy and freedom.

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

21

When making predictions regarding interpersonal violence, there is potential for racism and classism. People of color are more likely to be prosecuted and convicted of violent crimes than White people (Spohn, 2000). Race, ethnicity, and socioeconomic status influence professional judgment, as does a practitioner’s identification with the survivor or the offender. In addition to being aware of conscious and unconscious biases, practitioners must consciously consider whether their assessments are intended to control or help the offender. For practitioners who recommend commitment of an individual judged to be “dangerous,” is this intended as a form of social control of violent behavior or is it used to alleviate the emotional and psychological distress potentially linked to violent behavior? If social control is the purpose, additional considerations about treatment, the use of psychotropic medication, and the client’s ability to provide informed consent must be explored. Some practitioners tend to underpredict the potential for danger (false negative—the person is predicted to be less dangerous than he or she actually is), others may overpredict future danger (false positive— the person is predicted to be more dangerous than he or she actually is). If practitioners underpredict the risk of further violence, they place the potential survivor at risk of being killed or seriously hurt. When practitioners overpredict the potential for danger, they lose the trust of a survivor who believed in the practitioner’s ability to identify dangerousness. A survivor may choose to ignore future assessments by the practitioner and, thus, be placed in a vulnerable position. To overpredict the potential dangerousness of an identified perpetrator may also be to participate in a process that unjustly incarcerates, labels, and/or blames a person for past behaviors. The difficult task for the practitioner is to make a judgment between the two extremes. This requires training, skill, and a willingness to weigh multiple factors using validated measures, judicious use of professional expertise, and trust in the opinions of colleagues and the survivor(s) of violence (Campbell et al., 2009).

■■

SUMMARY

In general, there is a series of behaviors that should raise concerns that violence will reoccur in any number of settings. These behaviors include the following: • Prior history of being violent • Experiencing violence as a child

22 A S S E S S I N G D A N G E R O U S N E S S • Substance misuse • History of mental illness, especially personality ­disorders

and antisocial behaviors • Failing to complete an offender treatment program • Being young and poor • Unemployment

In IP relationships, all of the aforementioned are risk factors for violence with the addition of the following as risk factors for women in heterosexual relationships: • Leaving an abusive relationship for another man • Having a child or children from a previous relationship • Stalking • His access to firearms

Assessing dangerousness for further violence by sexual offenders, batterers, and child abusers is a developing science. The following chapters explore the most current research, trends, and professional viewpoints regarding the prediction of interpersonal violence.

REFERENCES Ægisdóttir, S., White, M. J., Spengler, P. M., Maugherman, A. S., Anderson, L. A., Cook, R. S., . . . Rush, J. D. (2006). The metaanalysis of clinical judgment project: Fifty-six years of accumulated research on clinical versus statistical prediction. Counseling Psychologist, 34(3), 341–382. doi:10.1177/​ 0011000005285875 Bailey, J. E., Kellerman, A. L., Somes, G. W., Banton, J. G., Rivara, F. P., & Rushford, N. P. (1997). Risk factors for violent death of women in the home. Archives of Internal Medicine, 157, 777–782. Beigel, A., Barren, M. R., & Harding, T. W. (1984). The paradoxical impact of a commitment statute on prediction of dangerousness. American Journal of Psychiatry, 341(3), 373–377.

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

23

Benner, P. (1984). From novice to expert: Excellence and power in clinical nursing practice. Reading, MA: Addison-Wesley. Berk, R. A., Sorrenson, S. B., & Barnes, G. (2016). Forecasting domestic violence: A machine learning approach to help inform arraignment decisions. Journal of Empirical Legal Studies, 13(1), 94–115. Black, M. C., Basile, K. C., Breiding, M. J., Smith, S. G., Walters, M. L., Merrick, M., . . . Stevens, M. R. (2011). The National Intimate Partner and Sexual Violence Survey (NISVS): 2010 summary report. Atlanta, GA: National Center for Injury Prevention and Control, Centers for Disease Control and Prevention. Bograd, M., & Mederos, F. (1999). Battering and couples therapy: Universal screening and selection of treatment modality. Journal of Marital and Family Therapy, 25(3), 291–312. Brecht, M. L., & Herbeck, D. M. (2013). Methamphetamine use and violent behavior user perceptions and predictors. Journal of Drug Issues, 43(4), 468–482. Campbell, J. C. (2004). Helping women understand their risk in situations of intimate partner violence. Journal of Interpersonal Violence, 19(12), 1464–1477. Campbell, J. C., & Glass, N. (2009). Safety planning, danger, and lethality assessment. In C. Mitchell (Ed.), Intimate partner violence: A health-based perspective (pp. 319–334). New York, NY: Oxford University Press. Campbell, J. C., Glass, N., Sharps, P., Laughon, K., & Bloom, T. (2007). Intimate partner homicide: Review and implications of research and policy. Trauma, Violence, and Abuse, 8(3), 246–269. Campbell, J. C., O’Sullivan, C., Roehl, J., & Webster, D. W. (2005). Intimate partner violence risk assessment validation study: The RAVE study. Final report to the National Institute of Justice (NCJ 209731–209732). Retrieved from http://www.ncjrs.gov/ pdffiles1/nij/grants/209731.pdf Campbell, J. C., Sharps, P., & Glass, N. (2001). Risk assessment for intimate partner violence. In G. F. Pinard & L. Pagani (Eds.), Clinical assessment of dangerousness: Empirical contributions (pp. 136–157). New York, NY: Cambridge University Press. Campbell, J. C., Webster, D., & Glass, N. E. (2009). The Danger Assessment: Validation of a lethality risk assessment instrument

24 A S S E S S I N G D A N G E R O U S N E S S

for intimate partner femicide. Journal of Interpersonal Violence, 24(4), 653–674. Campbell, J. C., Webster, D., Koziol-McLain, J., Block, C. R., Campbell, D. W., Curry, M. A., . . . Laughon, K. (2003). Risk factors for femicide in abusive relationships: Results from a multisite case control study. American Journal of Public Health, 93(7), 1089–1097. Cartier, J., Farabee, D., & Prendergast, M. L. (2006). Methamphetamine use, self-reported crime, and recidivism among offenders in California who abuse substances. Journal of Interpersonal Violence, 21(4), 435–445. Catalano, S., Smith, E., Snyder, H., & Rand, M. (2009). Female victims of violence. Washington, DC: Bureau of Justice Statistics, U.S. Department of Justice. Cattaneo, L. B., & Chapman, A. R. (2011). Risk assessment with victims of intimate partner violence: Investigating the gap between research and practice. Violence Against Women, 17(10), 1286–1298. Chermack, S. T., & Blow, F. C. (2002). Violence among individuals in substance abuse treatment: The role of alcohol and cocaine consumption. Drug and Alcohol Dependence, 66, 29–37. Chesney-Lind, M., & Shelden, R. G. (2014). Girls, delinquency, and juvenile justice (4th ed.). West Sussex, England: John Wiley. Cho, H. (2012). Racial differences in the prevalence of intimate partner violence against women and associated factors. Journal of Interpersonal Violence, 27(2), 344–363. Cohen, J. B., Dickow, A., Horner, K., Zweben, J. E., Balabis, J., Vandersloot, D., & Reiber, C. (2003). Abuse and violence: History of men and women in treatment for methamphetamine dependence. American Journal on Addictions, 12, 377–385. Connor-Smith, J. K., Henning, K., Moore, S., & Holdford, R. (2011). Risk assessments by female victims of intimate partner violence: Predictors of risk perceptions and comparison to an actuarial measure. Journal of Interpersonal Violence, 26(12), 2517–2550. Convit, A., Jaeger, J., Lin, S. P., Meisner, M., & Volavka, J. (1988). Predicting assaultiveness in psychiatric inpatients: A pilot study. Hospital and Community Psychiatry, 39(4), 429–434. Craig, T. J. (1982). An epidemiological study of problems associated with violence among psychiatric inpatients. American Journal of Psychiatry, 139, 1262–1266.

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

25

Davies, J. M., & Lyon, E. (2014). Domestic violence advocacy: Complex lives/ difficult choices (2nd ed.). Thousand Oaks, CA: Sage. Day, E. (2015, July 19). #BlackLivesMatter: The birth of a new civil rights movement. The Guardian. Retrieved from https://www.the​ guardian.com/world/2015/jul/19/blacklivesmatter-birth-civil​ -rights-movement DeKeseredy, W. S. (2011). Feminist contributions to understanding woman abuse: Myths, controversies, and realities. Aggression and Violent Behavior, 16(4), 297–302. Dobash, R. E., Dobash, R. P., Cavanagh, K., & Lewis, R. (2004). Not an ordinary killer—just an ordinary guy: When men murder an intimate woman partner. Violence Against Women, 10, 577–605. Dutton, D. G. (2012). The case against the role of gender in intimate partner violence. Aggression and Violent Behavior, 17, 99–104. Ernst, A. A., Weiss, S. J., Enright-Smith, S., Hilton, E., & Byrd, E. C. (2008). Perpetrators of intimate partner violence use significantly more methamphetamine, cocaine, and alcohol than victims: A report by victims. American Journal of Emergency Medicine, 26, 592–596. Fleury-Steiner, R. E., Bybee, D., Sullivan, C. M., Belknap, J., & Melton, H. (2006). Contextual factors impacting battered women’s intentions to reuse the criminal legal system. Journal of Community Psychology, 34(3), 327–342. Foran, H. M., & O’Leary, D. (2008). Alcohol and intimate partner violence: A meta-analytic review. Clinical Psychology Review, 28, 1222–1234. Fowler, D. (2009). Screening for co-occurring intimate partner abuse and substance abuse: Challenges across service delivery systems. Journal of Social Work Practice in the Addictions, 9(3), 318–339. Gambrill, E. (2006). Evidence-based practice and policy: Choices ahead. Research on Social Work Practice, 16(3), 338–357. Glass, N. E., Koziol-McLain, J., Campbell, J. C., & Block, C. R. (2004). Female-perpetrated femicide and attempted femicide. Violence Against Women, 10, 606–625. Gondolf, E. W., Mulvey, E. P., & Lidz, C. W. (1990). Characteristics of perpetrators of family and non-family assaults. Hospital and Community Psychiatry, 41(2), 191–193.

26 A S S E S S I N G D A N G E R O U S N E S S

Goodman, L. A., & Epstein, D. (2008). Listening to battered women: A survivor-centered approach to advocacy, mental health, and justice. Washington, DC: American Psychological Association. Gottfredson, D. M., & Gottfredson, S. D. (1988). Stakes and risks in the prediction of violent criminal behavior. Violence and Victims, 3(4), 247–262. Gross, B. H., Southard, M. J., Lamb, R., & Weinberger, L. E. (1987). Assessing dangerousness and responding appropriately: Hedlund expands the clinician’s liability established by Tarasoff. Journal of Clinical Psychiatry, 48(1), 9–12. Grove, W. M., Zald, D. H., Lebow, B. S., Snitz, B. E., & Nelson, C. (2000). Clinical versus mechanical prediction: A meta-analysis. Psychological Assessment, 12, 19–30. Harbison, J. (1991). Clinical decision making in nursing. Journal of Advanced Nursing, 16, 404–407. Hare, R. D., Clark, D., Grann, M., & Thornton, D. (2000). Psychopathy and the predictive validity of the PCL-R: An international perspective. Behavioral Sciences and the Law, 18(5), 623–645. Harris, G. T., Skilling, T. A., & Rice, M. E. (2001). The construct of psychopathy. In M. Tonry (Ed.), Criminal justice: An annual review of research (pp. 197–294). Chicago, IL: University of Chicago Press. Hayashi, H. D., Patterson, T. L., Semple, S. J., Fujimoto, K., & Stockman, J. K. (2016). Risk factors for recent intimate partner violence among methamphetamine-using men and women. Journal of Psychoactive Drugs, 48(2), 135–145. Heilbrun, K., O’Neill, M. L., Strohman, L. K., Bowman, Q., & Philipson, J. (2000). Expert approaches to communicating violence risk. Law and Human Behavior, 24, 137–148. Hilton, N. Z., & Harris, G. T. (2005). Predicting wife assault: A critical review and implications for policy and practice. Trauma, Violence, and Abuse, 6(1), 3–23. Hilton, N. Z., & Harris, G. T. (2009). How recidivism affects predictive accuracy: Evidence from a cross-validation of the Ontario Domestic Assault Risk Assessment (ODARA). Journal of Interpersonal Violence, 24(2), 326–337. doi:10.1177/0886260508316478 Hilton, N. Z., Harris, G. T., Rice, M. E., Houghton, R. E., & Eke, A. W. (2008). An indepth actuarial assessment for wife assault

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

27

recidivism: The Domestic Violence Risk Appraisal Guide. Law and Human Behavior, 32(2), 150–163. Hilton, N. Z., Harris, G. T., Rice, M. E., Lang, C., Cormier, C. A., & Lines, K. (2004). A brief accurate actuarial assessment for the prediction of wife assault recidivism: The Ontario Domestic Assault Risk Assessment. Psychological Assessment, 16, 300–312. Holcomb, W. R., & Ahr, P. R. (1988). Arrest rates among young adult psychiatric patients treated in inpatient and outpatient settings. Hospital and Community Psychiatry, 39, 52–57. Janofsky, J. S., Spears, S., & Neubauer, D. N. (1988). Psychiatrists’ accuracy in predicting violent behavior on an inpatient unit. Hospital and Community Psychiatry, 39, 1090–1094. Johnson, I. M. (2007). Victims’ perceptions of police response to domestic violence incidents. Journal of Criminal Justice, 35, 498–510. Johnson, M. P. (2006). Conflict and control: Gender symmetry and asymmetry in domestic violence. Violence Against Women, 12(11), 1003–1018. Josephson, W. L. (1987). Television violence and children’s aggression: Testing the priming, social script, and disinhibition predictions. Journal of Personality and Social Psychology, 53(5), 882–890. Krakowski, M., Jaeger, J., & Volavka, J. (1988). Violence and psychopathology: A longitudinal study. Comprehensive Psychiatry, 29, 174–181. Lambert, E. W., Cartor, R., & Walker, G. L. (1988). Reliability of behavioral versus medical models: Rare events and danger. Issues in Mental Health Nursing, 9, 31–44. Langenderfer, L. (2013). Alcohol use among partner violent adults: Reviewing recent literature to inform intervention. Aggression and Violent Behavior, 18(1), 152–158. Lawson, J. (2012). Sociological theories of intimate partner violence. Journal of Human Behavior in the Social Environment, 22(5), 572–590. Lewis, D. O., Lovely, R., Yeager, C., & Femina, D. D. (1989). Toward a theory of the genesis of violence: A follow-up study of delinquents. Journal of the American Academy of Child and Adolescent Psychiatry, 28(3), 431–436.

28 A S S E S S I N G D A N G E R O U S N E S S

Loveland, J. E., & Raghavan, C. (2014). Near-lethal violence in a sample of high-risk men in same-sex relationships. Psychology of Sexual Orientation and Gender Diversity, 1(1), 51–62. Macy, R. J., Giattina, M. C., Parish, S. L., & Crosby, C. (2010). Domestic violence and sexual assault services: Historical concerns and contemporary challenges. Journal of Interpersonal Violence, 25(1), 3–32. Maxwell, J. C., & Brecht, M. L. (2011). Methamphetamine: Here we go again? Addictive Behaviors, 36(12), 1168–1173. McFarlane, J., Parker, B., Soeken, K., Silva, C., & Reed, S. (1999). Severity of abuse before and during pregnancy for African American, Hispanic, and Anglo women. Journal of Nurse Midwifery, 44(2), 139–144. McKetin, R., Lubman, D. I., Najman, J. M., Dawe, S., Butterworth, P., & Baker, A. L. (2014). Does methamphetamine use increase violent behavior? Evidence from a prospective longitudinal study. Addiction, 109, 798–806. McNeil, D. E., Binder, R. L., & Greenfield, T. K. (1988). Predictors of violence in civilly committed acute psychiatric patients. American Journal of Psychiatry, 145(8), 965–970. Meloy, J. R. (1987). The prediction of violence in outpatient psychotherapy. American Journal of Psychotherapy, 61(1), 38–45. Mercy, J. A., & Saltzman, L. E. (1989). Fatal violence among spouses in the United States 1976–85. American Journal of Public Health, 79, 595–599. Messing, J. T., & Campbell, J. C. (2016). Informing collaborative interventions: Intimate partner violence risk assessment for front line police officers. Policing: A Journal of Policy & Practice, 10(4), 328–340. Messing, J. T., Mendoza, N., & Campbell, J. C. (2016). Alcohol use and latent profiles of intimate partner violence. Journal of Social Work Practice in the Addictions, 16(1–2), 160–175. Messing, J. T., & Thaller, J. (2013). The average predictive validity of intimate partner violence risk assessment instruments. Journal of Interpersonal Violence, 28(7), 1537–1558. Messing, J. T., & Thaller, J. (2015). Intimate partner violence risk assessment: A primer for social workers. British Journal of Social Work, 45, 1804–1820.

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

29

Miczek, K., Debold, J., Haney, M., Tidey, J., Vivian, J., & Weerts, E. (1994). Understanding and preventing violence. In A. J. Reiss, Jr. & J. A. Roth (Eds.), Panel on the understanding and control of violent behavior (Vol. 3, pp. 377–570). Washington, DC: National Academies Press. Miller, M., & Morris, N. (1988). Predictions of dangerousness: An argument for limited use. Violence and Victims, 3(4), 263–283. Mize, K. D., & Shackelford, T. K. (2008). Intimate partner homicide methods in heterosexual, gay, and lesbian relationships. Violence and Victims, 23(1), 98–114. Moore, T. M., Stuart, G. L., Meehan, J. C., Rhatigan, D. L., Hellmuth, J. C., & Keen, S. M. (2008). Drug abuse and aggression between intimate partners: A meta-analytic review. Clinical Psychology Review, 28, 247–274. Moracco, K. E., Runyan, C. W., & Butts, J. (1998). Femicide in North Carolina. Homicide Studies, 2, 422–446. Morton, E., Runyan, C. W., Moracco, K. E., & Butts, J. (1998). Partner homicide victims: A population-based study in North Carolina, 1988–1992. Violence and Victims, 13(2), 91–106. Mulvey, E. P., & Reppucci, N. D. (1988). The context of clinical judgment: The effect of resource availability on judgments of amenability to treatment in juvenile offenders. American Journal of Community Psychology, 16(4), 525–545. Myers, K. M., & Dunner, D. L. (1984). Self- and other-directed violence on a closed acute-care ward. Psychiatric Quarterly, 56, 178–188. Myhill, A. (2015). Measuring coercive control: What can we learn from national population surveys? Violence Against Women, 21(3), 355–375. National Association of Counties. (2005). The meth epidemic in America: The criminal effect of meth on communities. A 2005 survey on U.S. counties. Retrieved from http://www.csam-asam​.org/sites/ default/files/pdf/misc/NationalAssociationofCounties​.pdf National Institute on Drug Abuse. (2013). Methamphetamine. Retrieved from https://www.drugabuse.gov/sites/default/files/methrrs.pdf O’Leary, K. D. (2001). Conjoint therapy for partners who engage in physically aggressive behavior. Journal of Aggression, Maltreatment and Trauma, 5(2), 145–164.

30 A S S E S S I N G D A N G E R O U S N E S S

O’Leary, K. D., & Schumacher, J. A. (2003). The association between alcohol use and intimate partner violence: Linear effect, threshold effect, or both? Addictive Behaviors, 28(9), 1575–1585. Oram, S., Flynn, S. M., Shaw, J., Appleby, L., & Howard, L. M. (2013). Mental illness and domestic homicide: A population-based descriptive study. Psychiatric Services, 64(10), 1006–1011. Pataki, G. (1997). Intimate partner homicides in New York State. Albany, NY: The New York State Governor’s Office. Powers, R. A., & Kaukinen, C. E. (2012). Trends in intimate partner violence: 1980–2008. Journal of Interpersonal Violence, 27(15), 3072–3090. Quinsey, V. L., Harris, G. T., Rice, M. E., & Cormier, C. A. (1998). Violent offenders: Appraising and managing risk. Washington, DC: American Psychological Association. Regan-Kubinski, M. J. (1991). A model of clinical judgment processes in psychiatric nursing. Archives of Psychiatric Nursing, 5(5), 262–270. Rice, M. E., Harris, G. T., & Lang, C. (2013). Validation of and revision to the VRAG and SORAG: The Violence Risk Appraisal Guide— Revised (VRAG-R). Psychological Assessment, 25(3), 951–965. doi:10.1037/a0032878 Schon, D. (1983a). From technical rationality to reflection in action. In J. Dowie & A. Elstein (Eds.), Professional judgment: A reader in clinical decision making (pp. 60–77). Cambridge, England: Cambridge University Press. Schon, D. (1983b). The reflective practitioner. New York, NY: Basic Books. Segal, S. P., Watson, M. A., Goldfinger, S. M., & Averbuck, D. S. (1988). Civil commitment in the psychiatric emergency room: The assessment of dangerousness by emergency room clinicians. Archives of General Psychiatry, 45, 748–752. Sharps, P. W., Campbell, J. C., Campbell, D. W., Gary, F. A., & Webster, D. W. (2003). Risky mix: Drinking, drug use, and homicide. National Institute of Justice Journal, 250, 8–13. Sharps, P. W., Koziol-McLain, J., Campbell, J., McFarlane, J., Sachs, C., & Xu, X. (2001). Health care providers’ missed opportunities for preventing femicide. Preventive Medicine, 33(5), 373–380. Smith, P. H., Homish, G. G., Leonard, K. E., & Cornelius, J. R. (2012). Intimate partner violence and specific substance use disorders:

1

|

PREDICTION OF INTERPERSONAL VIOLENCE: AN INTRODUCTION

31

Findings from the National Epidemiologic Survey on Alcohol and Related Conditions. Psychology of Addictive Behaviors, 26(2), 236–245. Spohn, C. (2000). Thirty-years of sentencing reform: The quest for a racially neutral sentencing process. Criminal Justice, 3, 421–501. Stark, E. (2007). Coercive control: How men entrap women in personal life. New York, NY: Oxford University Press. Steadman, H. J., & Morrissey, J. P. (1982). Predicting violent behavior: A note on a cross-validation study. Social Forces, 61, 475–483. Stover, C. S., Berkman, M., Desai, R., & Marans, S. (2010). The efficacy of a police-advocacy intervention for victims of domestic violence: A 12-month follow-up data. Violence Against Women, 16(4), 410–425. Straus, M. (2010). Thirty years of denying the evidence on gender symmetry in partner violence: Implications for prevention and treatment. Partner Abuse, 1, 332–363. Tarasoff v. Regents of the University of California, 17 Cal 3d 425; 551 P.2d 334 (1976). Retrieved from http://scocal.stanford.edu/ opinion/tarasoff-v-regents-university-california-30278. Tardiff, K., & Sweillman, A. (1982). Assaultive behavior among chronic inpatients. American Journal of Psychiatry, 139, 212–215. Thomas, K. A., Dichter, M. E., & Matejkowski, J. (2011). Intimate versus nonintimate partner murder: A comparison of offender and situational characteristics. Homicide Studies, 15(3), 291–311. Thompson, M. P., & Kingree, J. B. (2004). The role of alcohol use in intimate partner violence and non-intimate partner violence. Violence and Victims, 19(1), 63–74. Violence Policy Center. (2012). When men murder women. Washington, DC: Author. Walton-Moss, B. J., Manganello, J., Frye, V., & Campbell, J. C. (2005). Risk factors for intimate partner violence and associated injury among urban women. Journal of Community Health, 30, 377–389. Ward-Lasher, A., Messing, J. T., & Hart, B. (in press). Policing intimate partner violence: Attitudes toward risk assessment and collaboration with social workers. Social Work. Websdale, N. (1999). Understanding domestic homicide. Boston, MA: Northeastern University Press.

32 A S S E S S I N G D A N G E R O U S N E S S

Weisberg, K. (2016a). Special issue on lethality assessment. Domestic Violence Report, 21(5). Weisberg, K. (2016b). Special issue on lethality assessment. Domestic Violence Report, 21(6). Weisz, A. N., Tolman, R. M., & Saunders, D. G. (2000). Assessing the risk of severe domestic violence: The importance of survivors’ predictions. Journal of Interpersonal Violence, 15, 75–90. Williams, K. R. (2012). Family violence risk assessment: A predictive cross-validation study of the Domestic Violence Screening Instrument-Revised (DVSI-R). Law and Human Behavior, 36(2), 120–129. doi:10.1037/h0093977 Williams, K. R., & Grant, R. (2006). Empirically examining the risk of intimate partner violence: The revised Domestic Violence Screening Instrument (DVSI-R). Public Health Reports, 121(4), 400–408. Zawitz, M. W. (1994). Violence between intimates. Washington, DC: Bureau of Justice Statistics, U.S. Department of Justice.

C HAPT ER 2 Prediction Issues for Practitioners Joel S. Milner, Jacquelyn C. Campbell, and Jill Theresa Messing

Sarah sought therapy in the midst of a breakup from her abusive partner, Justin; they had two children in common and were litigating custody. During the course of their relationship, Justin had emotionally abused Sarah (calling her names, telling her that she was an unfit mother), controlled her actions (hid her car keys, did not allow her access to money), strangled her on multiple occasions, abused her during both of her pregnancies, owned multiple firearms, and had threatened to kill her. The neighbors had called the police during one violent incident and Sarah had called the police on another occasion when Justin had threatened her life. Each time, the police spoke to Justin and decided that there was not enough probable cause for arrest. On other occasions throughout their relationship, Justin took Sarah’s phone to prevent her from calling the police. Sarah was afraid of Justin and had fled the state with their children after Justin threatened to kill her. Justin filed a motion with the court and Sarah was required to immediately return the children to the state; although she was afraid, she didn’t want to be separated from her young children, and moved into her sister’s home, across town from Justin’s residence. Sarah’s therapist used the Danger Assessment, a risk assessment tool with validity data from six empirical research studies, and her practice expertise to provide evidence to the court that Sarah was in reasonable fear for her life upon fleeing. The judge took this into account, concluding that Sarah’s actions were reasonable given the risk of homicide posed by Justin. The judge then gave Justin and Sarah equal parenting time and Sarah sole legal decision-making power with regard to the minor children. 33

34 A S S E S S I N G D A N G E R O U S N E S S

Practitioners often are called upon to predict the dangerousness of their clients or the danger that their clients are in. Because the situations in which predictions are made vary greatly, some authors have suggested that distinctions be made between formal and informal predictions (e.g., Werner, Rose, & Yesavage, 1990). Formal prediction refers to views expressed by professionals in hearings and court proceedings that influence sentencing, parole, and custody decisions. Informal prediction refers to comments made by practitioners in clinical situations with criminal justice system officials, health care professionals, victim advocates, and potential victims. Regardless of whether the prediction is formal or informal, practitioners are obligated to be as accurate as possible and to have considered the ethical dilemmas of (a) confidentiality versus warning and (b) protection of individual rights versus the collective good. Although many practitioners may be reluctant to make predictions because of problems with prediction accuracy and ethical considerations, they are constantly under pressure to do so by formal systems, other professionals, clients, and clients’ families. For example, with increasing frequency, practitioners are being asked to serve as expert witnesses in cases of sexual and intimate partner violence (IPV). Although practitioners may be able to avoid becoming involved in formal predictions, there still are legal and ethical mandates for practitioners to make informal predictions of dangerousness, such as when they inform potential victims. Thus, practitioners involved in work with violent or potentially violent clients have a great need for understanding the nature, process, and research status of prediction. ■■

CLINICAL VERSUS STATISTICAL PREDICTION STRATEGIES

In addition to distinguishing between informal and formal uses of prediction, it is useful to make a distinction between clinical and statistical types of prediction. Miller and Morris (1988) describe clinical prediction as being based on professional training, professional experience, and observation of a particular client. Clinical predictions are subjective or intuitive in nature, and may or may not be organized into a structured format. Statistical prediction involves predicting an individual’s behavior on the basis of how others have acted in similar situations (actuarial) or on an individual’s similarity to members of violent groups. Such prediction is based on statistical models (e.g., additive linear models, clustering models, contingency table analysis, machine-learning approaches)

2

|

PREDICTION ISSUES FOR PRACTITIONERS

35

derived from research and includes the use of risk factor instruments (e.g., Berk, 2012; Gottfredson & Gottfredson, 1988; Miller & Morris, 1988; Ruscio, 1998). There are some risk assessment instruments that incorporate elements of clinical prediction through structured professional judgment (e.g., Kropp & Hart, 2000). Clinical predictions have been criticized for producing decisions that are inconsistent, inequitable, biased, and inaccurate (e.g., Milner, Murphy, Valle, & Tolliver, 1998; Odeh, Zeiss, & Huss, 2006). Furthermore, clinical predictions may lack accountability because the criteria and rationale used in the assessment process are not explicit. Hilton and Simmons (2001) found that clinicians’ decisions about the release of generally violent mentally ill offenders from maximum security institutions were not related to an available risk assessment score, but influenced by factors not relevant to the outcome that they were seeking (attractiveness, compliance and success with psychotropic medication, criminal history, and institutional management problems). Recidivism at followup, however, was associated with the actuarial risk assessment score; if release decisions had been made by actuarial methods, it was estimated that recidivism would have been reduced by half (Hilton, 2010). Validated risk assessment instruments have been shown to be more predictive of future behavior than clinical prediction across a large variety of settings (Ægisdóttir et al., 2006; Grove, Zald, Lebow, Snitz, & Nelson 2000). Although there still are debates (e.g., Baumann, Law, Sheets, Reid, & Graham, 2006; Johnson, 2006), the consensus of opinion is that statistical prediction is more accurate than clinical prediction. Thus, whenever possible, we strongly recommend using statistical procedures to increase the accuracy of clinical prediction. A summary of the criteria commonly used in the different types of prediction of violent behavior is presented in Table 2.1. As indicated, informal clinical prediction often occurs without the assistance of validated instruments by using the kinds of models outlined in Chapter 1. In formal prediction, however, statistical methods and/or risk instruments that meet certain psychometric standards should be used. To this end, more than two decades ago Monahan (1993) emphasized the need for “familiarity with basic concepts in risk assessment (e.g., predictor and criterion variables, true and false positives and negatives, decision rules, base rates)” (p. 247). In this chapter, we concentrate on issues related to the use of instruments for prediction. Other statistical prediction methods have been discussed elsewhere (for overviews, see Gottfredson & Gottfredson, 1988; Ruscio, 1998).

36 A S S E S S I N G D A N G E R O U S N E S S

TABLE 2.1  Approaches Commonly Used in the Formal and Informal Prediction of Violent Behavior Approach Clinical

Formal

Informal

1&2

1

3

2

Statistical

1. Clinical judgment through interview and other subjective assessments. 2. Risk factor identification is used. Construct and validity data are needed to support the use of a risk factor instrument. 3. Dichotomous/criterion/cutoff scores are used. The assessment instrument provides a score designated as the criterion or cutoff that is used to place an individual into one category (e.g., high risk) or another (e.g., low risk). Concurrent and future predictive validity data and individual classification rates are needed. Note: Although this table is a summary of procedures used in the clinical and statistical prediction of violence, in all cases multiple data sources should be used in making predictions regarding violence. Furthermore, when formal statistical prediction is possible, a single test score must never be used as the sole reason for making a prediction.

■■

LEGAL ISSUES AND PREDICTION

Practitioners involved in violence prediction must be aware of both the limitations in their ability to predict violence and the evolving legal duties to warn and protect others from the violent acts of one’s clients. Following the landmark Tarasoff v. Regents of the University of California (1976) decision, which defined a broader professional responsibility for informing and protecting individuals from possible client violence, there was a dramatic expansion by the courts of professionals’ responsibility to inform and protect third parties. In contrast, in the 1990s, there was a substantial retreat from the original Tarasoff principle (Felthous & Kachigian, 2001). A review of the Tarasoff principle and subsequent expansion and more recent restrictions and rejections of the Tarasoff principle are available elsewhere (e.g., Hafemeister, McLaughlin, & Smith, 2013; Thomas, 2009; see also Chapter 1). In addition, practitioners must be mindful of related rulings on patient–client confidentiality that include social workers engaged in psychotherapy (e.g., Jaffe v. Redmond; see Colledge, Zeigler, Hemmens, & Hodge, 2000; Mitrevski & Chamberlain, 2006). Although a description of recent changes regarding legal obligations to warn and protect is beyond the scope of this chapter, practitioners working with violent clients must continually inform themselves about legislative and judicial changes in their obligation to warn and protect third parties from a potentially violent client.

2

|

PREDICTION ISSUES FOR PRACTITIONERS

37

Cultural competence in risk assessment is an emerging trend and an important consideration for practitioners. There is a lack of culturally competent interpersonal violence assessments, though some research is being conducted in this area. The Danger Assessment (DA) has been adapted for use with (primarily Latina) women who are immigrants to the United States (Messing, Amanor-Boadu, Cavanaugh, Glass, & Campbell, 2013), and, with colleagues, Drs. Campbell and Messing are currently working to adapt the DA for use with Native American women, refugee women, and immigrant women from diverse backgrounds. ■■

ETHICAL ISSUES AND PREDICTION

In addition to the legal duty to warn, professional organizations (e.g., for psychiatrists, psychologists) often have standards of practice that state the therapist must warn potential victims. For example, the American Medical Association’s principles of medical ethics mandate that phy­ sicians protect potential victims of patients by taking action, such as notifying law enforcement agencies. Even without organizationally prescribed professional standards, clinicians have ethical responsibilities to persons in physical danger (see Gutheil, 2001, for a discussion of ethical considerations related to when a warning may be ethically justified). Prediction also involves the possibility (a) that one’s own biases will influence one’s judgment and (b) may subject a person to unfair criminal justice penalties on the basis of an inaccurate prediction. As a result of these dilemmas, clinicians often desire fail-safe prediction instruments so that no judgment is necessary. The reality, of course, is that statistical methods are in various stages of development and that their ability to correctly screen violent and nonviolent individuals will never be completely accurate. Thus, the practitioner’s knowledge of clinical assessment remains an extremely important adjunct to any statistical prediction. Instrumentation can be a valuable source of objective data, provided that prior use and testing of a measure has supported its reliability and validity. The legal and ethical responsibility of clinicians includes becoming as knowledgeable as possible about the dynamics of violence, particularly in terms of potential for further dangerousness. In addition, clinicians need to know about instruments that measure dangerousness in specific areas (e.g., child abuse and IPV, as opposed to general

38 A S S E S S I N G D A N G E R O U S N E S S

aggressiveness or pathology), as well as the appropriate setting, the needed information, the predicted outcome, and the limits of the instrument’s use, topics covered in Chapters 3 through 6. Practitioners are responsible for determining whether an assessment is appropriate for a certain client, and for providing evidence to support the application of an instrument to a new client population. In the remainder of this chapter, we discuss how one evaluates the psychometric properties of risk assessment instruments. ■■

PSYCHOMETRIC ISSUES IN CLINICAL PRACTICE

Research demonstrates that professionals working with violent clients need to increase their knowledge of measurement issues. For example, Milner (1989) surveyed 550 administrators, researchers, and practitioners in the family violence field to determine their knowledge of appropriate uses of the Child Abuse Potential (CAP) Inventory, a screening scale for child physical abuse (Milner, 1986, 1994, 2004; see also Chapter 3). The survey revealed that a substantial number of professionals suggested applications for the CAP Inventory that were inappropriate or not supported by validity data. Milner concluded that an increase in professional knowledge of the proper use of child abuse screening instruments should accompany the development of such instruments or the use of family violence screening instruments should be restricted to those who have credentials (e.g., licensed psychologists) to help ensure an adequate knowledge of measurement issues. In an attempt to increase the practitioner’s knowledge of psychometric issues, we discuss some of the traditional psychometric requirements for tests and measures. The reader should note that a comprehensive set of standards has been developed to guide the practitioner in the evaluation and use of test instruments. Psychometric and practice standards are provided in the publication Standards for Educational and Psychological Testing (American Psychological Association, 2014). This document includes sections that describe the responsibilities of the test constructor, the test publisher, and the test user. ■■

APPROACHES TO DEVELOPING PREDICTIVE INSTRUMENTS

Ideally, the development of an instrument or test is based on a welldefined, empirically validated model that describes etiological variables. Constructs from the model are used to define the content domains, or areas to be covered by the proposed instrument. Guided by the

2

|

PREDICTION ISSUES FOR PRACTITIONERS

39

content domains, a large pool of items is developed and administered to criterion and matched comparison groups. The criterion group exemplifies or “has” the characteristic being measured, whereas the comparison group does not. Then an item analysis is conducted to determine the best predictors of being in the criterion group. After item cross-validation, which involves replicating the predictive ability of the items by using another sample of criterion and comparison participants, factor analysis can be conducted to see whether groups of items describe some broader construct or descriptive factors. Then it is determined whether the total score or some configuration of scores from the factor analysis most effectively discriminates the criterion participants from comparison participants. In cases in which explanatory theories are lacking, a combination of rational and empirical approaches can be used to develop test items (e.g., Edwards, 1970). In this approach, different theoretical perspectives and empirical studies are used to develop an array of content domains to guide the construction of test items. This technique often is called a “shotgun” approach to test development, wherein all possibly relevant domains are used to guide item development. When there are no welldeveloped models, this technique can result in the successful development of a risk instrument. A drawback of developing a measure without a guiding explanatory model is that important etiological variables may be overlooked or omitted from the content domains used to develop the assessment items. Fortunately, for a screening instrument to be successful, not all content domains have to be used in the development of the test. Indeed, from a psychometric perspective, the predictive factors need not be related directly to the etiology. A subset of the descriptive factors, whether causal or marker variables, often can be found that are reliably correlated with the criterion behavior (e.g., physical abuse) and that can predict the behavior. Although this approach is not considered ideal, in reality most measures are constructed by using only a subset of the content domains related to the predicted behavior. ■■

TEST RELIABILITY

Reliability can be thought of as the consistency of an instrument. Although many types of reliability are mentioned in the psychometric literature, instrument reliability is of two major types: internal consistency and temporal stability. Internal consistency and temporal stability reliabilities are statistically represented by correlation coefficients.

40 A S S E S S I N G D A N G E R O U S N E S S

Internal Consistency Estimates of internal consistency, or item consistency, provide information on the equivalence, or homogeneity, of the test items. This equivalence provides an estimate of the degree to which the test items are all measuring the same construct. Internal consistency estimates usually are presented as alpha coefficients. High internal consistency indicates that a test is measuring a specific construct, such as depression. Low internal consistency suggests that more than one construct or a multidimensional construct, such as self-concept, is being assessed. Internal consistency estimates are affected by test length. A scale with few test items (e.g., five items) must be very homogeneous to have high (e.g., .90) internal consistency, whereas a scale with a large number of test items (e.g., 100) actually may include a variety of different but related items (e.g., distress, anxiety, impulsiveness) and still have a relatively high internal consistency. Test reliability is important because it sets the upper limits for test validity; that is, on average, test validity of an instrument cannot be higher than the internal consistency of the test. In most cases, test validity estimates will be lower, sometimes markedly lower, than the internal consistency estimates.

Temporal Stability The temporal stability of a test score indicates the degree of test score stability, or how similar test scores are over a specified time period. High levels of temporal stability suggest that the construct being measured tends to be stable over time and that the test is measuring the same construct in the same way across time. Temporal stability estimates are calculated by correlating test scores obtained from the same participant at two points in time. Thus, temporal stability estimates are represented by test–retest correlations for different time intervals, such as 1-week, 1-month, and 1-year test–retest intervals. The expected degree of temporal stability of a test score should be high if the test purports to measure personality traits because these personality characteristics are expected to be stable across time. In contrast, the level of acceptable temporal stability of a test score may be relatively modest or low if the test is designed to measure a personality state that is expected to change across time. So the level of test–retest reliability, or temporal stability, should vary as a function of the conceptualization of the construct (trait or state) under investigation. When evaluating the temporal stability of a test, the extent of gain scores is important to note. Gain scores describe test score increases

2

|

PREDICTION ISSUES FOR PRACTITIONERS

41

that occur at the second testing, relative to the first testing. Most psychological tests tend to have slightly higher scores at the second testing. Thus, when test–retest reliabilities are presented, the test score means and standard deviations obtained at each test period should be published so that the practitioner can evaluate the size of any gain score. Test–retest reliabilities and gain scores are of particular concern in certain test applications. For example, in research designs in which multiple evaluations are made at different time intervals, knowledge of the test’s temporal stability is critical. Examples of such studies include pretreatment and posttreatment evaluation research and longitudinal victim effect studies, in which the same measures are administered repeatedly, usually at fixed time intervals. In these designs, only tests with appropriate temporal stability and modest gain scores for the time periods should be used. ■■

TEST VALIDITY

Validity data provide information on the extent to which a test is adequate for the intended use, or accurate. In the validation process, the inferences made from test scores, not the actual test scores, are validated. As part of the investigation of the psychometric qualities of a test, different types of validity data should be accumulated. It is the accumulated validity evidence from many studies conducted by different investigators, not the evidence from any single study, that allows the user to determine whether the instrument measures what it purports to measure. Ideally, specific validity data for a particular type of application are accumulated so that the mass of evidence supports or does not support a given test application. Frequently, test validity data may indicate that a test is appropriate for one use with a given population, whereas data may not be sufficient to support the same application with another population or for another application with the same population. For example, a prediction instrument may have data from several sources that support predicting future child sexual abuse in White, male perpetrator samples but have little or no data supporting the use of the instrument to predict recidivism in African American groups. Furthermore, no test should ever be said to be “valid” for its intended use. Test validation is a matter of degree of validity for certain applications with specific populations. Validation is an ongoing process and never a completed task. Test reviews or advertisements that indicate a test is “valid” or has been “fully validated” are inappropriate and

42 A S S E S S I N G D A N G E R O U S N E S S

misleading. More appropriate is the review or advertisement that indicates substantial data exist demonstrating the test has some degree of validity for a specific application with a specific population or populations. Furthermore, the supporting psychometric data for a test should be available to the test user and other interested professionals in the published literature and in a technical manual. If the relevant validity information is not readily available, the author of the instrument should be contacted to determine whether such information exists. Although different types of test validity have been described, three broad categories of test validity, which contain some overlap, usually are of interest. These are content, construct, and predictive (post hoc, concurrent, future type) validity. Each of these categories is discussed in the following sections.

Content Validity The content validity of a test refers to the extent to which the test items represent a specific content domain. As previously discussed, the content domain is typically defined by the theoretical model on which the test is based. In cases in which no model is used to guide the development of test items, some rational or empirical approach is used to guide the item development, and this approach serves to define the content domains. In either case, the approach used defines the domain or domains that must be sampled adequately during test construction. The extent to which the test items represent the guiding conceptual domains indicates the degree of content validity. Thus, the procedures that guided the construction of the test items and the face validity of the items are used to inform the user about the degree of content validity. Content validity can be demonstrated by obtaining evaluations of items from experts in the field. In addition, content validity is supported if the test scores are correlated with other measures of constructs representing the content domains used to develop the test items.

Construct Validity Overlapping somewhat with content validity, construct validity refers to the extent to which the underlying constructs assumed to be measured by the test actually are measured. Construct data provide verification of what initially was theoretically or intuitively assumed during the test item construction. Construct validity is supported by the accumulation

2

|

PREDICTION ISSUES FOR PRACTITIONERS

43

of data that verify that the characteristics the instrument was designed to measure are indeed measured. In addition, development of a large array of different types of construct validity data provides a mosaic of information that allows the test user to understand what the instrument is measuring. Such information serves to assist the user in making more appropriate test score interpretations. In general, two broad types of construct validity exist: convergent validity and discriminant validity. Convergent validity data are generated when factors thought to be related to the test scores are shown to be related. Discriminant validity data are generated when factors believed to be extraneous to the test scores are found to be unrelated. Convergent and discriminant validity can be demonstrated by a variety of techniques, such as comparison of test scores with other test scores, with relationships demonstrated only where relationships are expected; demonstration of expected relationships between test scores and different criterion and comparison groups; conceptually congruent factor analysis; and test data from program evaluations that show the test scores are sensitive to treatment effects.

Predictive Validity Test predictive validity, or criterion validity, consists of three types: post hoc, concurrent, and future type (e.g., Nunnally, 1978). Post hoc prediction refers to the prediction of a condition in the past. Concurrent prediction refers to the prediction of a condition that presently exists. Future prediction refers to the prediction of a condition or event that has not yet occurred, which includes both a first occurrence and recidivism. Future prediction, therefore, involves “forecasting” the occurrence of future events on the basis of present test scores. Each of these three types of prediction is needed in the detection and prevention of violence. Post hoc prediction is very difficult, and this type of validity data is rarely available for an instrument. The major problem in post hoc prediction is that the test data are collected after the occurrence of the behavior of interest. Although a temporal separation between testing and the predicted behavior also exists for future prediction (as discussed later), more problems exist for post hoc prediction. Not only can random intervening variables affect test scores and reduce the predictive relationship, but when violence does occur, consequences often result from the occurrence of the behavior. In some cases, legal intervention will occur or treatment will be provided. In the case of family violence, children

44 A S S E S S I N G D A N G E R O U S N E S S

may be removed or an intimate partner may leave. These and other direct consequences of the violence can affect the personality and interactional characteristics of the perpetrator. Therefore, subsequent testing of the violent perpetrator may not represent the conditions present at the time of the abuse, making post hoc prediction difficult. Concurrent predictive validity is especially important when a test is used to place an individual in a particular group (e.g., currently abusive). Although concurrent predictive validity data are more often available for a test, most of these data are expressed in terms of group differences, as opposed to individual classification rates. That is, test data typically only indicate that a test shows significant group differences between a criterion group (e.g., child physical abusers) and a comparison group (e.g., nonabusers). Although group differences must exist if a test is capable of individual discrimination, the finding that a test shows significant group differences between criterion and comparison groups does not alone mean that the test has acceptable individual classification ability. Furthermore, many studies that compare groups do not match the criterion and comparison groups on demographic variables, a shortcoming that increases the likelihood that group differences will be found, but for the wrong reasons. In test validation studies in which criterion and comparison groups are not matched and group differences are found, it is not known whether the group differences are the result of the criterion variable, group demographic differences, or both factors. What is needed to demonstrate adequate individual predictive validity is the individual classification rates for test scores based on well-defined criteria and demographically matched comparison groups. Although some risk assessment procedures use demographic characteristics as risk markers (such as single-parent status, age, education level), this profiling approach results in numerous ethical problems as well as assuring the overinclusion of those with the characteristics in the risk group (even though they are not at risk) and the underinclusion of those who are at risk but do not have the selected demographic characteristics. A remaining problem is that reported individual classification rates are often based on a statistical procedure known as discriminant analysis. Although discriminant analysis is appropriate for providing initial estimates of individual predictive validity, this procedure provides optimal classification rates for each sample tested because it reweights the item scores in each new analysis. Therefore, as individual classification data accumulate, some of the data should be obtained by using the

2

|

PREDICTION ISSUES FOR PRACTITIONERS

45

test instrument standard scoring procedure designed for field use. Some decrease in the individual correct classification rates can be expected when a standard scoring procedure is used across a variety of demographically different populations. Future predictive validity is often the most desirable type of validity when violence is under study. However, future prediction is very difficult, not only because of the difficulties involved in the design of a predictive instrument, but also because numerous variables can intervene between the testing session and the predicted event. These intervening variables may increase or decrease the likelihood of the event, thus reducing the test’s predictive ability. Nevertheless, researchers have endeavored to examine the future predictive validity of risk instruments. In the case of IPV risk assessment, for example, a meta-analysis of the future predictive validity of risk assessment instruments reported that 25 studies had examined the predictive validity of one or more risk assessment instruments and five IPV risk assessment instruments had been examined for predictive validity more than one time (Messing & Thaller, 2013). When examining predictive validity of an instrument, the test user should note carefully the percentage of false-positive classifications (nonabusers labeled as abusive, nonrecidivists labeled as recidivists) and false-negative classifications (abusers labeled as nonabusive; recidivists labeled as nonrecidivists). Related to these estimates are the sensitivity and specificity of the test. Test sensitivity is the percentage of correct classifications of abusers or recidivists; test specificity is the percentage of correct classifications of nonabusers. These four outcomes expressed in terms of risk assessment for violence are presented in Table 2.2.

TABLE 2.2  Types of Individual Classification Outcomes Actual risk status Screened risk

Client at risk

Client not at risk

Client at risk

A

B

Client not at risk

C

D

A. Correct classification of at-risk status (sensitivity). B. Misclassification of risk status (false-positive classification). C. Misclassification of not-at-risk status (false-negative classification). D. Correct classification of not-at-risk status (specificity).

46 A S S E S S I N G D A N G E R O U S N E S S

The receiver operating characteristic (ROC) has been suggested as a standard measure for assessing the predictive validity of risk assessment instruments as it takes into account both sensitivity and specificity of the instrument (Douglas, Blanchard, Guy, Reeves, & Weir, 2000; Rice & Harris, 2005). The ROC curve is a plot of the sensitivity of the instrument as a function of the false-positive rate; when shown on a graph, the area under the curve (AUC), or the proportion of the graph that lies under the plotted ROC curve, provides information about predictive validity (Douglas et al., 2000; Rice & Harris, 1995). The AUC ranges from 0 to 1.0, with 0 being perfect negative prediction (in the case of violence risk assessment, this would indicate that no cases were predicted accurately), .50 indicating that the instrument predicts no better than chance, and 1.0 indicating perfect positive prediction (in the case of violence risk assessment, this would indicate that every case was predicted with perfect accuracy; Douglas et al., 2000). The AUC is interpreted as the probability that any randomly selected case would have a higher score on the risk assessment instrument than any randomly selected noncase (Rice & Harris, 1995, 2005). For example, an AUC of .65 indicates a 65% chance that a randomly selected case would have a higher score on the risk assessment instrument than a randomly selected noncase (Douglas et al., 2000; Rice & Harris, 1995). The ROC has several advantages over other measures of predictive validity in addition to providing information about sensitivity and specificity in a single measurement. It is less dependent upon the base rate (the number of cases in a sample) and the selection ratio (the proportion of the sample predicted to be a case) than the traditional statistics used to assess predictive validity (Rice & Harris, 1995, 2005). In fact, when assessing the predictive accuracy of a particular instrument, the AUC remains relatively stable as the base rates and selection ratios change (Rice & Harris, 1995). Because of this, it is possible to utilize ROCs to compare the predictive validity of risk assessment instruments when the base rates are different across samples (Rice & Harris, 1995). In addition, data do not need to be normally distributed to use the ROC (Rice & Harris, 1995). Application of the ROC is optimal with a continuous predictor variable and a dichotomous outcome (Douglas et al., 2000), also making it ideal for the prediction of violent recidivism. Rice and Harris (2005) calculated the AUC equivalent of Cohen’s (1988) effect size classifications for d. Using these calculations, an AUC of .556 corresponds to a small effect (d = 0.2), an AUC of .639 corresponds to a medium effect (d = 0.5), and an AUC of .714 corresponds to a large

2

|

PREDICTION ISSUES FOR PRACTITIONERS

47

effect (d = 0.8; Rice & Harris, 2005). This is similar to, but not wholly consistent with, other authors who have suggested that an AUC of .70 is considered moderate to large, and an AUC value of .75 or greater is considered large (Douglas et al., 2000). As an example, the average AUC across research studies of the predictive validity of IPV risk assessment instruments ranges from small to medium (AUC = .537 – .666; Messing & Thaller, 2013). ■■

OTHER MEASUREMENT ISSUES

In addition to the determination that adequate reliability and validity data exist to support a desired test application, a number of other measurement issues must be considered when a test is used to assess interpersonal violence. Several of the more important measurement considerations are (a) the possibility of participant response distortion, (b) the availability of appropriate test norms, (c) the size of the standard error of measurement (SEM), and (d) the estimated violence base rates.

Response Distortions A major issue related to the use of self-report approaches, including both interview and questionnaire assessment, is the possibility that individuals will distort their responses to questions. Response distortions include faking-good, faking-bad, and random response behavior. Faking-good behavior is related to the respondent’s attempt to distort responses in a socially desirable manner or to present himself or herself in a favorable light. Faking-bad behavior is related to the respondent’s attempt to distort responses in a socially undesirable manner and to present himself or herself in an unfavorable light. Random response behavior is related to the respondent giving responses that do not represent responses to the item content. Random responding may be the result of a variety of factors, such as a deliberate desire to avoid revealing personal data or an inability to understand item content. A more complete discussion of possible causes of the three types of response distortions is available elsewhere (e.g., Milner, 2006). Because response distortions can render test data meaningless, professionals need to include some measures of response distortions in their assessment package (e.g., faking-good, faking-bad, random response indices from existing tests such as the CAP Inventory [Milner, 1986]). This inclusion is especially important in the assessment of violence

48 A S S E S S I N G D A N G E R O U S N E S S

perpetrators because these respondents are often motivated to distort responses made to professionals investigating violence.

Test Norms The interpretation of a test score is assisted by the availability of test norms. Norm scores (e.g., test score means, standard deviations) should be available for well-defined populations. The test manual should indicate the year in which the norm data were collected and provide detailed descriptions of the methods used to collect the norm data and the demographic characteristics of the norm group. Norms may represent local or regional populations or a national probability sampling. In addition, norm data should be presented as a function of gender, ethnic background, age, marital status, educational level, socioeconomic level, number of children, and so on. If there is a test manual, the test user should expect it to provide numerous test norms so that he or she can inspect the norm data for possible moderator (demographic) variable effects. For example, on the Family Environment Scale (FES; Moos & Moos, 1986) the norm scores for the family conflict scale vary by more than 100% on the basis of family size (e.g., two family members, conflict score M = 2.11; five family members, conflict score M = 4.78). In this case, failure to consider the number of family members could result in a dramatic misunderstanding of the meaning of the FES conflict score. The problem is actually more serious on this and other tests because often several moderating variables must be considered. Although national norms are not available for most tests, norms representing the population from which the respondent was drawn usually are required in order to give meaning to the obtained test score. The norm test scores provide population-based values that enable the user to interpret the obtained test scores in relation to those obtained from a similar participant group. Finally, the types of norm values needed can vary as a function of the test application. Thus, the practitioner must determine which norm data are needed for the proper interpretation of the test score and whether these norms are available, rather than simply use whatever norms are presented in the test manual.

Standard Error of Measurement The test user must be aware that all test scores contain measurement error. An estimate of measurement error, the standard error of measurement

2

|

PREDICTION ISSUES FOR PRACTITIONERS

49

(SEM), should be available for each test scale. The SEM provides an estimate of the variation of the obtained score from the true score. Because the SEM is a standard deviation, it can be used to set a confidence interval around the obtained score so that the range of scores that includes the true score can be known. The SEM is especially important in the prediction of interpersonal violence when test cutoff scores are used to classify the examinee and the examinee’s obtained test score is near the test cutoff score. The examinee’s obtained test score should be sufficiently beyond the test cutoff score, as determined by the SEM, so that the interpretation that the test score is elevated is less likely to be the result of chance.

Violence Base Rates Professionals need to be aware of the role of base rates in the prediction of behavior. In the prediction of child abuse or any other violent behavior, the base rate affects the amount of incremental prediction added by a measure in the detection of the behavior under investigation. Optimal increases in prediction occur when the base rates are 50%, which means that 50% of the sample or population are the criterion cases (perpetrators of violence). For example, if a test has an 80% correct classification rate for both abusive and nonabusive parents and the test is used in a situation in which 50% of the participants tested are abusive (base rate of 50%), then the classification rate is 80% for each group. If the test is administered to 100 participants, then 40 abusers (80% of 50 abusers) and 40 nonabusers (80% of nonabusers) will be classified correctly, resulting in an 80% overall classification rate and an equal number of false-positive and false-negative classifications. In violence prediction, base rates are often lower than 50%. When predicting IPV recidivism, base rates may be as low as 21% or as high as 49% (Belfrage et al., 2012; Hilton, Grant, Rice, Houghton, & Eke, 2008). When base rates are lower than 50%, the usefulness of the test may decrease. For example, if the same test with an 80% correct classification rate is used in a situation in which only 5% of the participants tested are abusive (base rate of 5%) in a group of 100 participants, then the ratio of false-positive and false-negative classifications will vary dramatically. In light of the 80% correct classification rate for abusers, four of the five abusers will be classified correctly; however, only 80% of the 95 nonabusers will be classified correctly, resulting in 19 false-positive classifications. Thus, overall, 23 participants (four correct abuser classifications

50 A S S E S S I N G D A N G E R O U S N E S S

and 19 false-positive classifications) will be indicated as abusive, with only four of the 23 actually being abusive. This result means that only 17.4% of those classified as abusers are actually abusive, with 82.6% of those labeled abusive being false-positive classifications. It also is true, however, that the number of correct classifications in the 77 participants classified as nonabusers will increase. In this example, only one of the 77 participants classified as nonabusers will be a false-positive classification, resulting in a 98.7% correct classification rate for those labeled nonabusers. Thus, as the base rate decreases, the percentage of false-positive and false-negative classifications will change dramatically from the classification rates derived from studies in which 50% of the participants are abusive and nonabusive. It is always important, therefore, for the professional to estimate the base rate for the population under study and to determine the relative utility of the test for the intended application. ■■

SUMMARY

As increasing numbers of measures are developed for use in assessing violence potential and recidivism, professionals will be required to discriminate between those tests that have some utility and those that should be avoided. The responsibility for making adequate test selection and application is increasingly the responsibility of the test user. This responsibility requires that practitioners increase and maintain their knowledge of psychometric issues related to test selection and use. Clinicians can avoid making formal predictions in the courtroom, but clients who are perpetrators and victims of child, intimate partner, and sexual abuse are almost impossible to avoid, given the rates of violent behavior in our society. Thus, clinicians must be able to make reasonably accurate assessments of the potential for future dangerousness in order to fulfill their ethical and legal mandates to warn and protect potential victims. As described in Chapter 1, using an evidence-based practice framework that combines the use of risk assessment instruments with practitioner expertise (appropriate academic, clinical, and legal training; knowledge of the risk literature) and survivor selfdetermination generates the most accurate predictions at the present time (also see Messing & Thaller, 2015).

2

|

PREDICTION ISSUES FOR PRACTITIONERS

51

REFERENCES Ægisdóttir, S., White, M. J., Spengler, P. M., Maugherman, A. S., Anderson, L. A., Cook, R. S., . . . Rush, J. D. (2006). The meta-analysis of clinical judgment project: Fifty-six years of accumulated research on clinical versus statistical prediction. Counseling Psychologist, 34(3), 341–382. doi:10.1177/0011000005285875 American Psychological Association. (2014). Standards for educational and psychological testing. Washington, DC: Author. Baumann, D. J., Law, R. J., Sheets, J., Reid, G., & Graham, C. J. (2006). Remarks concerning the importance of evaluation actuarial risk assessment models: A rejoinder to Will Johnson. Children and Youth Services Review, 28, 715–725. Belfrage, H., Strand, S., Storey, J. E., Gibas, A. L., Kropp, P. R., & Hart, S. D. (2012). Assessment and management of risk for intimate partner violence by police officers using the Spousal Assault Risk Assessment Guide. Law and Human Behavior, 36, 60–67. Berk, R. (2012). Criminal justice forecasts of risk: A machine learning approach. London, England: Springer-Verlag. Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum. Colledge, D., Zeigler, F., Hemmens, C., & Hodge, C. (2000). What’s up doc? Jaffee v. Redmond and the psychotherapeutic privilege in criminal justice. Journal of Criminal Justice, 28, 1–11. Douglas, K. S., Blanchard, A. J. E., Guy, L. S., Reeves, K. A., & Weir, J. (2000). ​HCR-20 violence risk assessment scheme: Overview and annotated bibliography. Retrieved from http://escholarship​ .umassmed.edu/cgi/viewcontent.cgi?article=1362&context​ =psych_cmhsr Edwards, A. L. (1970). The measurement of personality traits by scales and inventories. New York, NY: Holt, Rinehart & Winston. Felthous, A. R., & Kachigian, C. (2001). To warn and to control: Two distinct legal obligations or variations on a single duty to protect? Behavioral Sciences and the Law, 19, 355–373. Gottfredson, D. M., & Gottfredson, S. D. (1988). Stakes and risks in the prediction of violent criminal behavior. Violence and Victims, 3, 247–262.

52 A S S E S S I N G D A N G E R O U S N E S S

Grove, W. M., Zald, D. H., Lebow, B. S., Snitz, B. E., & Nelson, C. (2000). Clinical versus mechanical prediction: A meta-analysis. Psychological Assessment, 12, 19–30. Gutheil, T. G. (2001). Moral justification for Tarasoff-type warnings and breach of confidentiality: A clinician’s perspective. Behavioral Sciences and the Law, 19, 345–353. Hafemeister, T. L., McLaughlin, L. G., & Smith, J. (2013). Parity at a price: The emerging professional liability of mental health providers. Virginia Public Law and Legal Theory Research Paper No. 2012-60. Retrieved from http://ssrn.com/abstract=2150956 Hilton, N. Z. (2010). Risk assessment for criminal justice, offender intervention, and victim services. Washington, DC: American Psychological Association. Hilton, N. Z., Grant, T. H., Rice, M. E., Houghton, R. E., & Eke, A. W.​ (2008). An in depth actuarial assessment for wife assault recidivism: The Domestic Violence Risk Appraisal Guide. Law and Human Behavior, 32, 150–163. Hilton, N. Z., & Simmons, J. L. (2001). The use of actuarial risk assessment in clinical judgments and tribunal decisions about mentally disordered offenders in maximum security. Law and Human Behavior, 25, 393–408. Johnson, W. (2006). The risk assessment wars: A commentary response to “Evaluating the effectiveness of actuarial risk assessment models” by Donald Baumann, J. Randolph Law, Janess Sheets, Grant Reid, and J. Christopher Graham, Children and Youth Services Review, 27, pp. 465–490. Children and Youth Services Review, 28, 701–714. Kropp, P. R., & Hart, S. D. (2000). The Spousal Assault Risk Assessment (SARA) guide: Reliability and validity in adult male offenders. Law and Human Behavior, 24, 101–118. Messing, J. T., Amanor-Boadu, Y., Cavanaugh, C. E., Glass, N., & Campbell, J. C. (2013). Culturally competent intimate partner violence risk assessment: Adapting the Danger Assessment for immigrant women. Social Work Research, 37, 263–275. Messing, J. T., & Thaller, J. (2013). The average predictive validity of intimate partner violence risk assessments. Journal of Interpersonal Violence, 28, 1537–1558.

2

|

PREDICTION ISSUES FOR PRACTITIONERS

53

Messing, J. T., & Thaller, J. (2015). Intimate partner violence risk assessment: A primer for social workers. British Journal of Social Work, 45, 1804–1820. Miller, M., & Morris, N. (1988). Predictions of dangerousness: An argument for limited use. Violence and Victims, 3, 263–284. Milner, J. S. (1986). The Child Abuse Potential Inventory: Manual (2nd ed.). Webster, NC: Psytec. Milner, J. S. (1989). Applications of the Child Abuse Potential Inventory. Journal of Clinical Psychology, 45, 450–454. Milner, J. S. (1994). Assessing physical child abuse risk: The Child Abuse Potential Inventory. Clinical Psychology Review, 14, 547–583. Milner, J. S. (2004). The Child Abuse Potential (CAP) Inventory. In M. L. Hilsenroth & D. L. Segal (Eds.), Comprehensive handbook of psychological assessment: Vol. 2. Personality assessment (pp. 237–246). Hoboken, NJ: John Wiley. Milner, J. S. (2006). An interpretive manual for the Child Abuse Potential Inventory. DeKalb, IL: Psytec. Milner, J. S., Murphy, W. D., Valle, L. A., & Tolliver, R. M. (1998). Assessment issues in child abuse evaluations. In J. R. Lutzker (Ed.), Handbook of child abuse research and treatment (pp. 75–115). New York, NY: Plenum Press. Mitrevski, J. P., & Chamberlain, J. R. (2006). Psychotherapist–patient privilege: Applying Jaffee v. Redmond: Communications to a psychotherapist are not privileged if they occur outside the course of diagnosis or treatment. Journal of the American Academy of Psychiatry and the Law, 34, 245–246. Monahan, J. (1993). Limiting therapist exposure to Tarasoff liability: Guidelines for risk containment. American Psychologist, 48, 242–250. Moos, R. H., & Moos, B. S. (1986). Family Environment Scale manual (2nd ed.). Palo Alto, CA: Consulting Psychologists Press. Nunnally, J. C. (1978). Psychometric theory. New York, NY: McGraw-Hill. Odeh, M. S., Zeiss, R. A., & Huss, M. T. (2006). Cues they use: Clinicians’ endorsement of risk cues in predictions of dangerousness. Behavioral Sciences and the Law, 24, 147–156.

54 A S S E S S I N G D A N G E R O U S N E S S

Rice, M. E., & Harris, G. T. (1995). Violent recidivism: Assessing predictive validity. Journal of Consulting and Clinical Psychology, 63, 737–748. Rice, M. E., & Harris, G. T. (2005). Comparing effect sizes in follow-up studies: ROC area, Cohen’s d, and r. Law and Human Behavior, 29, 615–620. Ruscio, J. (1998). Information integration in child welfare cases: An introduction to statistical decision making. Child Maltreatment, 3, 143–156. Tarasoff v. Regents of the University of California, 551 P.2d 334 (1976), vacating 529 P.2d 553 (1974). Thomas, M. (2009). Expanded lability for psychiatrists: Tarasoff gone crazy? International Journal of Mental Health and Capacity Law, 18, 45–56. Werner, P. D., Rose, T. L., & Yesavage, J. A. (1990). Aspects of consensus in clinical prediction of imminent violence. Journal of Clinical Psychology, 46, 534–538.

C HAPT ER 3 Child Physical Abuse Risk Assessment: Parent and Family Evaluations Joel S. Milner and Julie L. Crouch

More than three million child maltreatment referrals were made to social services agencies in the United States in 2014 (U.S. Department of Health and Human Services, 2016). Collectively, these referrals involved more than 6.5 million children. In 2014, 60.7% of all referrals were accepted (i.e., “screened in”) for investigation and, among accepted referrals, 19.2% were confirmed (i.e., “substantiated,” “indicated,” or “alternative response victim”) for child maltreatment (U.S. Department of Health and Human Services, 2016). In the latest national registry report, 70% of all investigated referrals were confirmed for child neglect and 17.0% were confirmed for child physical abuse. When child maltreatment referrals are accepted for investigation, child protective services workers are required to collect assessment data and render a timely decision as to whether child maltreatment occurred. If child maltreatment is substantiated, child protective services workers conduct safety assessments and attempt to identify the perpetrator(s). Unfortunately, in many child maltreatment investigations, constraints on time and resources force child protective services workers to make decisions based on limited assessment data. In situations that require the removal of children, case workers must also determine whether and when the children should be returned. The task of estimating (i.e., predicting) the likelihood of future events (e.g., reabuse) is a difficult one. Recidivism (e.g., likelihood of reabuse) risk assessments are difficult to conduct because intervening variables (i.e., factors that occur after the assessment) can affect future parenting behavior. That is, risk may change over time or after additional factors are taken into account. 55

56 A S S E S S I N G D A N G E R O U S N E S S

For example, initial assessment data may indicate a high risk of recidivism. However, later events (e.g., treatment) may reduce risk even though the high-risk determination was a valid estimate at the time of the initial evaluation. One method of reducing recidivism prediction errors caused by intervening events is to conduct additional risk assessments at future dates. ■■

RISK ASSESSMENT IN PRIMARY, SECONDARY, AND TERTIARY PREVENTION

Typically, prevention/intervention programs have been divided into three general types: primary, secondary, and tertiary prevention. Primary prevention programs target all parents (regardless of risk for child maltreatment); assessment procedures for determining risk status are not necessary in this case. Primary prevention programs, which attempt to prevent child abuse prior to its occurrence, often focus on reinforcing beliefs, practices, and conditions in the community and culture that are thought to reduce the likelihood that parents will abuse their children. Secondary prevention programs assume that some parents are more at risk for child maltreatment than other parents. Secondary prevention efforts, which also seek to prevent child maltreatment prior to its occurrence, usually involve assessing for child maltreatment risk in the general population, using checklists, clinical tools, and so forth. When parents are believed to be at risk of child maltreatment, they are offered interventions (e.g., parenting education, parent support groups, and home visitation) that are designed to reduce their risk for child maltreatment. Tertiary prevention efforts involve intervention implemented after the occurrence of child maltreatment (e.g., legal intervention, therapy). Tertiary interventions attempt to prevent recidivism. As in secondary prevention, risk assessment is usually an important part of tertiary intervention, though these assessments tend to have a different focus. After a confirmation of child maltreatment, risk assessments and safety assessments are conducted to determine whether a child should be removed or should remain with the parent. Risk assessments are also used to determine whether tertiary interventions are effective and to predict posttreatment recidivism. This section discusses the three traditional forms of intervention used to reduce risk factors for child maltreatment and the role that risk assessment plays within each approach. It should be mentioned that there are variations in these approaches. For example, some interventions

3

|

CHILD PHYSICAL ABUSE RISK ASSESSMENT

57

focus on developing individual and community strengths, as opposed to trying to reduce risk factors. In strength-based approaches, the goal is to strengthen factors that buffer or reduce the occurrence of child maltreatment. Thus, as a form of primary prevention, strength-based approaches do not screen for individual risk because the focus is not on high-risk individuals and the interventions are available to all parents in the community. As previously noted, risk assessments are commonly used in secondary and tertiary interventions. Typically, the risk factors assessed are determined by the child maltreatment models that guide the secondary and tertiary intervention programs. Models of child maltreatment vary in both breadth and purpose. At the broadest level, organizational models of child maltreatment provide frameworks that serve to organize risk/ protective factors from a variety of domains, including individual, family, community, and cultural factors (e.g., Belsky, 1980, 1993; Luster & Okagaki, 2005). Factors in each domain are believed to have either risk-potentiating or protective influences that may be transient or enduring in their effects on child abuse risk (e.g., Cicchetti & Rizley, 1981). For example, Belsky (1980) described an organizational model that used four ecological levels to organize risk/protective factors associated with child maltreatment: the ontogenic, microsystem, exosystem, and macrosystem levels. The ontogenic level refers to individual factors (e.g., parental characteristics). The microsystem level refers to family factors, such as parent–child interactions, marital discord, and the quality of family relations (e.g., adaptability, cohesion). The exosystem (community) and macrosystem (culture) levels focus on factors, such as social support, unemployment stress, and cultural values. In an extension of his organizational model, Belsky (1993) described in greater detail the various “contexts of maltreatment,” which include the immediate context (the parent–child interaction) and broader contexts (community, cultural, and evolutionary conditions) thought to influence the likelihood of child maltreatment. Within the context of Belsky’s (1980, 1993) organizational model of child physical abuse, a large number of more specific models have been developed, and most include factors from several, but not all, ecological levels. For example, a quarter of a century ago, Tzeng, Jackson, and Karlson (1991) described 25 models of child physical abuse that represented nine different paradigms (e.g., sociocultural, family system, and learning paradigms). More recent model developments have focused on the refinement of existing models, especially at the individual level, such as detailed attempts to understand social information processing in

58 A S S E S S I N G D A N G E R O U S N E S S

high-risk and physically abusive parents (Azar, 1997; Azar, Okado, Stevenson, & Robinson, 2013; Azar & Read, 2009; Azar, Reitz, & Goslin, 2008; Crouch & Milner, 2005; Milner, 1993, 2000, 2003; Seng & Prinz, 2008). ■■

CHILD PHYSICAL ABUSE RISK FACTORS

As described previously, risk assessment procedures may target a single factor at a single ecological level or may be multidimensional and assess multiple factors at different ecological levels. Reviews are available that provide detailed descriptions of risk and protective factors at the individual (Brown, Cohen, Johnson, & Salzinger, 1998; Milner, 1998; Milner & Dopke, 1997; Stith et al., 2009; Vondra, Sysko, & Belsky, 2005), family (Brown et al., 1998; Milner, 1998), community, and societal (Coulton, Korbin, Su, & Chow, 2008; Tolan & Guerra, 1998) levels. In practice, professional assessments typically focus on individual characteristics and familial factors, with less focus on community and cultural factors that may contribute to risk for child abuse. Thus, the primary focus of this chapter is on child physical abuse risk assessment at the ontogenic (parent) and microsystem (family) levels. Research has generally supported the use of individual and familial risk factors to predict abuse. For example, a 17-year prospective study found that, as the number of individual and familial risk factors increased, so did the prevalence of child abuse and neglect (from 3% when no risk factors were present to 24% when four or more risk factors were present), albeit different patterns of factors were found for child physical abuse and child neglect (e.g., Brown et al., 1998).

Individual-Level Risk Factors At the individual level, child physical abuse risk factors can be grouped into overlapping domains of demographic/social, biological, cognitive/ affective, and behavioral characteristics. Based on literature reviews (e.g., see Azar & Wolfe, 2006; Milner, 1998; Milner & Dopke, 1997; Stith et al., 2009 for reviews), risk factors in each of these domains are as follows. Demographic and social risk factors include being a nonbiological parent, being single, younger age, having lower levels of education, having a large number of children, unemployment, social isolation, and a parent’s childhood history of receipt and/or observation of violence. Putative biological risk factors include neurological, physiological, and physical health problems. For example, neurological deficits in parents that are associated with episodic dyscontrol and/or attention deficit

3

|

CHILD PHYSICAL ABUSE RISK ASSESSMENT

59

disorders may increase the risk of child physical abuse. Specific cognitive deficits, such as difficulty in verbal processing that reduces the parents’ ability to cope with family problems, are also believed to increase child abuse risk. Supporting these contentions, one study found that high-risk (compared with low-risk) parents had more cognitive deficits, as assessed by neuropsychological measures. However, after risk group differences in depression and anxiety were statistically controlled, cognitive differences were no longer significant. Thus, it is unknown whether the cognitive deficits or the negative affect, or perhaps some common third variable, is the causal factor that increases risk for child abuse perpetration. Child physical abusers are believed to possess a hyperreactive trait and to be more physiologically responsive to stimuli than nonabusive parents. However, findings from psychophysiological studies have been mixed. Some studies have found that abusive parents, compared with nonabusive parents, have higher baseline levels of physiological arousal, and other studies have found that abusers are more physiologically reactive to stressors. Still other studies have found neither of these differences. Cognitive and affective risk factors for child physical abuse include an array of characteristics, such as poor ego strength, low self-esteem, and an external locus of control (e.g., blaming others for one’s problems). Other cognitive risk factors include inappropriate expectations regarding how children should behave, negative perceptions and evaluations of children’s behavior, and “misattributions” of responsibility for behavior (e.g., attributions of hostile intent to children’s accidents). Affective risk factors for child physical abuse include distress, frustration, depression, loneliness, anxiety, and anger. It is noteworthy that in a meta-analysis that included 39 risk factors, parental “anger/reactivity” was the most predictive of child physical abuse (Stith et al., 2009). Although most child physical abusers are not mentally ill, many types of psychopathology, including personality disorders (i.e., borderline personality disorder, narcissistic personality disorder), are associated with increased risk of parenting problems, including child physical abuse. With respect to behavioral risk factors, physically abusive parents are more likely than nonabusive parents to have infants who develop attachment problems. Moreover, abusive parents, compared with nonabusive parents, engage in fewer interactions with their children and are more likely to engage in negative parenting behaviors. For example, abusive parents use more harsh disciplinary strategies, including verbal and physical assault. At the same time, abusive parents use fewer reasoning and explaining strategies and are less likely to use praise or rewards. They also are more inconsistent on the occasions when they provide

60 A S S E S S I N G D A N G E R O U S N E S S

praise and rewards for prosocial child behavior. Relatedly, abusers exhibit inadequate interpersonal skills and adult attachment problems. Physically abusive parents often report a general inability to cope with life stress. Although the frequency is debated, abusive and high-risk parents frequently have alcohol- and drug-related problems.

Familial-Level Risk Factors Familial risk factors for child physical abuse overlap with many of the aforementioned individual risk factors. Families that lack resources, have a large number of family members, and/or have inadequate living environments are at increased risk. Frequent marital discord, high levels of family verbal and physical conflict, social isolation, and a lack of family cohesion and expressiveness are also factors that increase child abuse risk. As the total number of stressors experienced by family members increases, so does the risk of child physical abuse. ■■

RESEARCH ISSUES THAT IMPACT CHILD PHYSICAL ABUSE RISK ASSESSMENT

Although the literature is replete with studies describing child abuse risk factors, there are a number of limitations that warrant consideration. A major limitation in many studies is the way that child physical abuse is operationalized. Often child physical abuse cases overlap with other forms of maltreatment (e.g., child neglect and child emotional abuse), making it difficult to isolate risk factors specific to physical abuse. Furthermore, child physical abuse may be impulsive reactions to stressful situations or premeditated acts of violence toward children. The risk factors for parents who impulsively spank their child and produce bruises may be different from those for parents who intentionally and with forethought injure their children. Because most descriptive and predictive child physical abuse research is based on poorly defined groups, results are often difficult to replicate. This difficulty adds to the likelihood of classification errors (false-positive and false-negative classifications) when attempts are made to use research findings to determine risk potential for a specific type of child abuse. Another limitation common in child physical abuse studies is the failure to incorporate demographically matched comparison groups. In the absence of demographically matched comparison groups, it is not possible to determine the extent to which differences between groups are the result of the occurrence of child abuse or to group demographic

3

|

CHILD PHYSICAL ABUSE RISK ASSESSMENT

61

differences (e.g., gender, ethnicity, age, educational level). In addition, cross-validation research using a variety of demographically different groups is needed because abuse-related factors may be found for parents from one population (e.g., mothers, Caucasian, lower socioeconomic status), but not for parents from a demographically different population. We also do not know whether risk factors for abuse by biological parents are the same as those for abuse by nonbiological caretakers. Even when concise definitions of child physical abuse and appropriate study designs are used, research challenges remain because child physical abuse is difficult to detect. The existence of abuse must be indicated by some criterion, which is often the protective services worker’s judgment, and such judgments have been shown to have errors (or biases). Moreover, most cases of child abuse remain undetected, and child abuse cases that are not reported are not studied. Most studies include only volunteer participants, which further limits the representativeness of the results. In addition, most studies use only maternal data; thus, less is known about paternal risk factors. Reliance on self-reported data is common; however, most studies do not control for participant response distortions (e.g., social desirability), which can be a significant problem in self-report studies (see Bennett, Sullivan, & Lewis, 2006; Milner & Crouch, 1997). Finally, many studies investigate and report group differences in risk factors but do not report individual prediction rates for risk criteria. Another issue that often goes unnoticed is that most risk factors are based on correlational research. Although correlational data can be used to generate risk factors that predict abuse, correlational data are not sufficient to show that a risk factor causes child abuse. Thus, when a risk factor is based on correlational data, the risk factor may or may not be related to the cause of child abuse. If clinicians provide treatment for risk factors that are correlated but not causally related to child abuse, the treatment will not be effective. ■■

DETERMINATION OF CHILD PHYSICAL ABUSE RISK

In child physical abuse investigations, assessment data are often collected through the use of interviews, personality measures, and child abuse risk assessment tools. The sections that follow provide an overview of these assessment strategies. Sources of information for these assessments include the victim, the suspected perpetrator, other family members, and collaterals.

62 A S S E S S I N G D A N G E R O U S N E S S

Interviews Ammerman and Hersen (1999b) pointed out that, in the assessment of family violence, interviews tend to be biased and are affected by respondent distortions and recall problems (also see Ægisdóttir et al., 2006; Grove, Zald, Lebow, Snitz, & Nelson, 2000). In an attempt to guide professionals who interview violent families, a number of checklists (e.g., Ayoub & Jacewitz, 1982; Louwers, Affourtit, Moll, Koning, & Korfage, 2010), risk indicators (e.g., Dalgleish & Drew, 1989; Korfmacher, 2000), and interviews (e.g., Child Abuse and Neglect Interview Schedule [Ammerman, Hersen, & Van Hasselt, 1988], Parent Interview and Assessment Guide [Wolfe, 1988]) have been developed. A major problem with existing clinical interview criteria is that they tend to lack selectivity; that is, many (and, frequently, the majority) of parents exhibiting one or more of the at-risk criteria are not abusive, resulting in an excessive number of false-positive classifications. In addition, child physical abuse often appears to be the result of the interaction of several risk factors that occur in the absence of buffering conditions (i.e., factors that serve to decrease the likelihood of child maltreatment). For example, parenting stress is associated with increased child physical abuse risk; however, this association may be especially strong among parents who have little social support and/or who believe that corporal punishment is acceptable. Checklists and structured interviews typically do not provide information about which combinations of factors are the best predictors of child abuse. Buffering variables may be overlooked in the assessment process because their role in risk prediction has not been fully explicated in the research literature. For example, family support and peer support appear to be important factors in reducing child physical abuse risk (e.g., Crouch, Milner, & Caliso, 1995; Crouch, Milner, & Thomsen, 2001; Egeland, Jacobvitz, & Sroufe, 1988; Milner, Robertson, & Rogers, 1990), but the specific buffering effects of different types of family and peer support on various risk factors have not been adequately determined. Although checklists and structured interviews are available for assessing different types of family violence, relatively few of these instruments have published validity data. Even when convergent validity data are available, concurrent validity (i.e., individual classification rates, correct classification rates, false-positive rates, and false-negative rates) is rarely examined. Although there are few data on the concurrent prediction rates of checklists and structured interviews, there are even fewer attempts to examine how well checklists/interviews predict the

3

|

CHILD PHYSICAL ABUSE RISK ASSESSMENT

63

likelihood of future child abuse (e.g., prediction of offender recidivism). Furthermore, even if a particular structured interview can be shown to have adequate psychometric characteristics, the utility of the interview procedure will vary as a function of the experience and skill of the interviewer and the cooperativeness and honesty of the respondent.

Personality Measures In addition to the use of checklists and structured interviews, general personality measures have been used to assess child abuse risk. In most assessment situations, it is believed that objective measures provide more accurate information than checklists and interview procedures. Historically, using objective measures has been viewed as superior to interview techniques because objective measures both replicate and extend the information typically obtained through interviews and because objective measures are usually less time-consuming and more economical than interviews (e.g., Nunnally, 1978). Research on the use of personality measures to assess child physical abuse risk, however, has yielded mixed results. For example, the Minnesota Multiphasic Personality Inventory (MMPI; Hathaway & McKinley, 1943) and the MMPI-2 (Butcher & Williams, 1992; Duckworth & Anderson, 1995), which are general measures of psychopathology, have been used to distinguish abusers from nonabusers. Although the MMPI initially was believed to have utility in child abuse risk assessment, efforts to replicate which MMPI clinical scales discriminate between child abusers and nonabusers have met with only limited success (e.g., Egeland, Erickson, Butcher, & Ben-Porath, 1991; Gabinet, 1979; Griswold & Billingsley, 1969; Paulson, Afifi, Chaleff, Thomason, & Liu, 1975; Paulson, Afifi, Thomason, & Chaleff, 1974). Collectively, these early studies failed to provide evidence to support use of the MMPI in child physical abuse screening. Thus, it is perhaps not surprising that (with only a few exceptions) there has been little new research in the past two decades examining the utility of using the MMPI clinical scales to identify child physical abusers, albeit there has been research examining use of the MMPI-2 validity scales to detect response distortion in parenting capacity assessments (Carr, Moretti, & Cue, 2005). Projective measures of personality have been used to distinguish abusers from nonabusers. Only a few studies, however, have reported data on the effectiveness of this application (e.g., Rorschach Inkblot Test [Cyrulnik, 2000; Derr, 1978; Lerner, 1975]). At present, even though projective measures continue to be used by some practitioners, the extent to

64 A S S E S S I N G D A N G E R O U S N E S S

which projective tests have utility in the determination of individual risk for child physical abuse has not been adequately demonstrated.

Child Physical Abuse Risk Measures As a response to the limitations of using general personality measures to assess child physical abuse risk, measures have been developed for the specific purpose of assessing child physical abuse risk in secondary and tertiary prevention programs. At present, however, most of these measures do not have adequate reliability and validity data to support their use as risk assessment measures. For example, only two measures (the Michigan Screening Profile of Parenting [MSPP; Helfer, Hoffmeister, & Schneider, 1978] and the Child Abuse Potential [CAP] Inventory [Milner, 1986b, 1994, 2004]) have multiple validation and cross-validation studies on individual (concurrent) classification rates. Only one measure (i.e., the CAP Inventory) has multiple studies reporting individual concurrent and future (prospective) predictive validity data and has supportive prospective data across ethnic groups. In the following section, psychometric data on the aforementioned CAP Inventory, one of the most commonly used child physical abuse risk assessment instruments in both secondary and tertiary risk assessment, is reviewed in detail. This review is followed by reviews of other commonly used risk assessment measures. THE CHILD ABUSE POTENTIAL INVENTORY The CAP Inventory is a 160-item, self-report screening instrument that is answered in a forced-choice, agree–disagree format (Milner, 1986b). The current CAP Inventory (Form VI) contains the Abuse scale, a 77-item child physical abuse scale that includes six descriptive subscales: Distress, Rigidity, Unhappiness, Problems with Child and Self, Problems with Family, and Problems from Others. To detect response distortions, the CAP Inventory contains three validity scales: Lie, Random Response, and Inconsistency. The validity scales are used in paired combinations to form three validity indices: Faking-Good, Faking-Bad, and Random Response. If any of the response distortion indices are elevated, the Abuse scale scores may not be an accurate representation of the respondent’s “true” score. Two special scales are part of the CAP Inventory: Ego-Strength (Milner, 1988, 2006) and Loneliness (Mazzucco, Gordon, & Milner, 1989; Milner, 2006). These special scales were developed from existing scale

3

|

CHILD PHYSICAL ABUSE RISK ASSESSMENT

65

and filler items and were designed to provide the test user with supplemental clinical information on the respondent. In addition, a number of abbreviated versions of the CAP Inventory Abuse scale have been developed; the most psychometrically sound scale was developed by Ondersma, Chaffin, Mullins, and LeBreton (2005; also see Walker & Davies, 2012). The initial development of the CAP Inventory is described in a comprehensive technical manual (Milner, 1986b). As a supplement to the technical manual, an interpretive manual for the CAP Inventory scales is available (Milner, 2006). The interpretative manual for the CAP Inventory (Milner, 2006) provides an extensive discussion of how the Abuse scale, factor scales, and validity indices should be interpreted. Separate from material in the CAP manuals, articles on psychometric characteristics of the CAP Inventory (e.g., see Milner, 1994, 2004) and on the applications and limitations of the CAP Inventory are available (Melton & Limber, 1989; Milner, 1986a, 1986b, 1989b, 1989c, 1991a, 2004, 2006). Concurrent prediction Initial Abuse scale classification rates based on discriminant analysis for child physical abusers and demographically matched comparison participants indicated overall correct classification rates in the 90% range (Milner, 1986b). In subsequent studies, for which more diverse populations have been used, the individual correct classification rates based on discriminant analysis have been in the mid-80% to the low 90% range (e.g., Milner, Gold, & Wimberley, 1986; Milner & Robertson, 1989). Because discriminant analysis provides optimal classification rates for the sample under investigation, other studies have investigated CAP Inventory Abuse scale classification rates obtained when using the standard scoring procedure. Typically, when the standard scoring procedures are used, the correct classification rates are 5% to 10% lower than those found when using discriminant analysis (Milner, 1986b; Milner & Crouch, 2012). Furthermore, in most studies using child physical abusers and demographically matched comparison participants, more falsenegative than false-positive classifications have been reported (Milner, 1986b; Milner & Crouch, 2012). This outcome suggests that it is more likely that the Abuse scale will fail to detect abusive parents (false negatives) than to misclassify demographically similar nonabusive comparison parents as abusive (false positives). It also indicates that high scores are meaningful. The CAP Inventory Abuse scale specificity (ability to correctly classify nonabusive parents) has been investigated in a variety of nonabusive

66 A S S E S S I N G D A N G E R O U S N E S S

groups with acceptable results. For example, 100% correct classification rates have been reported for low-risk mothers (Lamphear, Stets, Whitaker, & Ross, 1985), nurturing mothers (Milner, 1986b, 1989a), and nurturing foster parents (Couron, 1982). In a large-scale study (N = 1,151) of the effects of medical stressors on the Abuse scale specificity, no distortions in specificity were found in mothers of newborns with vaginal and C-section delivery, with and without complications (Milner, 1991b). However, modest distortions in Abuse scale specificity were found when parents of children with specific types of child injury (e.g., severe burns) and illness (e.g., gastric problems) were tested. Although it is possible that distortions in the Abuse scale specificity may have been caused by undetected child abuse, these data suggest that the Abuse scale specificity may be affected to some degree by a parent having a child with medical problems (Milner, 1991b). Thus, although it appears the Abuse scale can be used with mothers of newborns, additional data are needed to determine whether it is appropriate to use the Abuse scale in a medical setting with parents of children who have injuries or illness. In general, when Abuse scale classification rates have been examined for maltreatment groups other than recently identified, untreated child physical abusers, the classification rates have been lower. For example, Couron (1982) obtained data from a group of physically abusive and neglectful parents (even though the Abuse scale was not designed to detect neglectful parents) and a group of comparison parents and found that when the Abuse scale alone was used to predict group membership, the correct classification rate was 72.6%. A discriminant analysis, however, indicated an overall correct classification rate of 90.3% when the Abuse score, a stress measure, and demographic characteristics (e.g., marital status, age of parent) were used to predict group membership. Haddock and McQueen (1983) examined the utility of the CAP Inventory for detecting physically abusive caregivers employed at an out-of-home care facility for children. Abuse scale classification rates for physically abusive employees and matched nonabusive employees were examined. A discriminant analysis indicated an overall correct classification rate of 92.9% when using the CAP Abuse scale, work satisfaction items, and demographic variables to predict group membership. Although this overall classification rate for institutional abusers is encouraging, the abuser classification rate for the CAP Abuse scale alone was not reported. Collectively, these data suggest that the CAP Abuse scale may have some validity when used as a screening tool with groups other than suspected child physical abusers who are investigated by social services

3

|

CHILD PHYSICAL ABUSE RISK ASSESSMENT

67

agencies. Nevertheless, because the CAP Abuse scale was designed for use with parents, additional data (e.g., individual classification rates) are needed to determine the extent to which the CAP Abuse scale can be used with different nonparent groups. Future prediction In addition to concurrent validity data, longitudinal predictive validity data are available for the CAP Abuse scale. Milner, Gold, Ayoub, and Jacewitz (1984) found a significant relationship (p