Florida State University Libraries
Electronic Theses, Treatises and Dissertations The Graduate School
2014 Integrating Mainstream Criminological Theory into the Biosocial Perspective: An Empirical Analysis Joseph A. Schwartz
Follow this and additional works at the FSU Digital Library. For more information, please contact [email protected] FLORIDA STATE UNIVERSITY
COLLEGE OF CRIMINOLOGY AND CRIMINAL JUSTICE
INTEGRATING MAINSTREAM CRIMINOLOGICAL THEORY INTO THE
BIOSOCIAL PERSPECTIVE: AN EMPIRICAL ANALYSIS
By
JOSEPH A. SCHWARTZ
A Dissertation submitted to the College of Criminology and Criminal Justice in partial fulfillment of the requirements for the degree of Doctor of Philosophy
Degree Awarded: Summer Semester, 2014 Joseph A. Schwartz defended this dissertation on May 22, 2014. The members of the supervisory committee were:
Kevin M. Beaver
Professor Directing Dissertation
Stephen J. Tripodi
University Representative
William B. Bales
Committee Member
Thomas G. Blomberg
Committee Member
The Graduate School has verified and approved the above-named committee members, and certifies that the dissertation has been approved in accordance with university requirements.
ii
To my beautiful and loving wife Jen, you are my best friend, greatest advocate, and the source of my happiness.
iii
ACKNOWLEDGMENTS
First and foremost, I would like to acknowledge my mentor, Dr. Kevin Beaver. I am sure that whatever feeble attempt I make to demonstrate just how much of an impact Dr. Beaver has had on my life will be fully inadequate. Dr. Beaver’s patience, commitment, intellectual prowess, and generosity are simply unparalleled. I truly admire his endless hunger for knowledge and his fearless pursuit of it. Dr. Beaver has taught me so much more than the fundamental aspects of research and teaching, he has taught me to aggressively pursue my goals, the true value of hard work, and how to answer the questions I seek to answer. Perhaps even more importantly, he has taught me how to be a better person. Through example, he has shown me how to be a better husband, and one day, how to be a better father. I am certain that my efforts will pale in comparison, but I hope that I will be able to pass some of the knowledge and experiences he has instilled in me to my future students. Thank you so much for taking me under your wing, your efforts have not gone unnoticed and I will never forget the guidance and support you have provided.
I would also like to acknowledge the scholars who were generous enough with their time to serve on my dissertation committee: Dr. William Bales, Dean Thomas Blomberg, and Dr. Stephen Tripodi. Your support, guidance, and advice through this process has been invaluable not only in regard to the improvement of my dissertation but also to my development as a scholar and a criminologist. To receive input from scholars of this caliber is truly an honor and I am deeply grateful for the time and energy they have contributed to this project and my scholarly evolution. I would like to especially acknowledge Dean Thomas Blomberg for his influential advice and guidance during the latter portion of my graduate career. I sincerely appreciate each and every conversation we have engaged in, and I thank you for your perpetual support.
I would also like to acknowledge the multitude of other scholars who I have had the honor of interacting with and learning from. First, I would like to thank the faculty in the College of Criminology and Criminal Justice at FSU for challenging me to reach my potential and for providing the necessary support along the way. In addition, I would like to sincerely thank the faculty of the Department of Criminal Justice at California State University, San Bernardino for sparking my interest in criminology and for encouraging me to pursue my goals.
iv
I would like to especially acknowledge the assistance of Dr. Gisela Bichler, Dr. Larry Gaines, Ms. Mary Schmidt, Dr. Pamela Schram, and Dr. Stephen Tibbetts. Without the encouragement of Dr. Schram I likely would not have pursued a graduate degree and without the guidance of Dr. Tibbetts I may still be unaware of the causal influences of biology on the underlying etiology of behavior. Thank you for all of your advice, support, and guidance during my entire career as a student, I am certain that I would not have been able to accomplish my goals without you.
I would also like to thank my fellow graduate students at Florida State. I have been fortunate enough to have forged many friendships during my time in Tallahassee. I cannot express how grateful I am for the times I was able to share with many of you. I can honestly say that I will look back fondly over my time as a graduate student at FSU, and I have plenty of stories to share with my future graduate students and colleagues (most of which revolve around conversations or experiences in the Hecht House basement). A special acknowledgement is in order for Eric Connolly and Joe Nedelec. I am certain that my time as a graduate student would have been far less productive and certainly less entertaining had I not been fortunate enough to cross paths with the two of you. Thank you both for your friendship, support, and guidance.
Words simply cannot describe the amount of support I have received from my family. I would like to sincerely thank my father, Joe Schwartz, and mother, Kim Schwartz, for their unwavering support and guidance over the course of my entire life. They both have instilled in me the importance of hard work and striving to achieve excellence in all I do. Thank you both for being such amazing parents. I would also like to thank my younger brother Allan Schwartz, whose strength, tenacity, and drive are enviable (as difficult as that is for an older brother to admit). I am excited to see what your future, with my future sister-in-law Brittany Hernandez, holds. I would also like to thank my younger sister, Shavaun Schwartz. I am certain that my trips back home would have been far less entertaining without our conversations. In addition, my fashion sense has certainly benefited from your advice (and ridicule, for that matter). I would also like to thank my grandmother, Ann West, and grandfather, Al Schwartz, both of whom have provided me with invaluable advice and support. Thank you both so much, I love you very much. Finally, I would like to thank my wife’s family who have taken me in as one of their own, and allowed me to move their daughter across the country. I am extremely grateful for their patience, understanding, and support over the past several years.
v
I have saved my final acknowledgement for my beautiful and loving wife, Jen. I simply cannot imagine going through this entire process without her love, support, and friendship. You are my rock, and I can say with certainty that I would not have been able to attempt, let alone accomplish, this achievement without you. I am fully aware of the sacrifices you have made in order for me to pursue my dreams. How you were able to tolerate all of the times I have babbled on about something I was working on, or the long nights and weekends I had to spend working, or the holidays we had to spend away from family and friends back home without even one complaint is nothing short of remarkable and exemplifies just how simply amazing you are. The fact that you were so willing to sacrifice your own goals and dreams so that I could selfishly pursue my own speaks directly to your generous and loving nature. Simply put, you are the most beautiful and amazing woman I have ever met, as long as I have your support I know that I can accomplish anything. Thank you from the bottom of my heart.
I’m sure that I have omitted several others who have guided or supported me along the way. I have been fortunate enough to have come across many, many people who have impacted my life in so many different ways and to name them all here would require far more space than has been allotted. I would like to sincerely thank each and every one of you.
This research uses data from Add Health, a program project directed by Kathleen Mullan Harris and designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill, and funded by grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations. Special acknowledgment is due to Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Information on how to obtain the Add Health data files is available on the Add Health website (http://www.cpc.unc.edu/addhealth). No direct support was received from grant P01-HD31921 for this analysis.
vi
TABLE OF CONTENTS
List of Tables ...... xi
List of Figures ...... xiii
Abstract ...... xiv
CHAPTER 1 STATEMENT OF THE PROBLEM ...... 1
1.1 Research Questions ...... 4
1.2 Outline...... 7
CHAPTER 2 THE BIOSOCIAL PERSPECTIVE ...... 10
2.1 Behavior Genetic Methodologies ...... 15
2.1.1 Twin Studies ...... 19
2.1.2 Monozygotic Twins Reared Apart (MZA) ...... 29
2.1.3 Adoption Studies ...... 31
2.1.4 Family-Based Studies ...... 33
2.1.5 Nonshared Environment Studies ...... 35
2.2 Behavior Genetic Research and Antisocial Behavior ...... 37
2.2.1 Meta-Analyses of Behavior Genetic Research on Antisocial Behavior ...... 38
2.2.2 Discussion of Findings ...... 46
2.3 Gene-Environment Interplay ...... 50
2.3.1 Gene-Environment Interaction (G × E) ...... 51
2.3.2 Gene-Environment Correlation (rGE) ...... 53
2.3.3 Gene-Environment Interplay and Antisocial Behavior ...... 56
2.4 Summary and Discussion ...... 60
CHAPTER 3 CONVENTIONAL CRIMINOLOGICAL THEORY: EMPIRICAL FINDINGS AND CONCERNS...... 64
vii
3.1 Rational Choice Theory ...... 67
3.1.1 Key Assumptions and Theoretical Concepts of Rational Choice Theory ...... 69
3.1.2 The Empirical Status of Rational Choice Theory ...... 73
3.2 Social Learning Theory ...... 81
3.2.1 Key Assumptions and Theoretical Concepts of Social Learning Theory ...... 83
3.2.2 The Empirical Status of Social Learning Theory ...... 87
3.3 Classic Strain Theory ...... 93
3.3.1 Key Assumptions and Theoretical Concepts of Classic Strain Theory ...... 94
3.3.2 The Empirical Status of Classic Strain Theory ...... 98
3.4 Social Bonding Theory ...... 101
3.4.1 Key Assumptions and Theoretical Concepts of Social Bonding Theory ...... 104
3.4.2 The Empirical Status of Social Bonding Theory ...... 106
3.5 Summary and Discussion ...... 111
CHAPTER 4 SITUATING EXISTING CRIMINOLOGY THEORIES WITHIN THE BIOSOCIAL PERSPECTIVE ...... 114
4.1 Precursors to the Biosocial Integration Model...... 118
4.1.1 Biosocial Theories of Antisocial Behavior ...... 119
4.1.2 Preliminary Models of Biosocial Integration ...... 122
4.2 The Biosocial Integration Model ...... 126
4.3 Summary and Discussion ...... 129
CHAPTER 5 METHODS ...... 135
5.1 Data ...... 135
5.1.1 Description of the Analytic Sample ...... 140
5.2 Measures ...... 142
viii
5.2.1 Antisocial Behavior ...... 142
5.2.2 Prevalence of Illicit Drug Use...... 143
5.2.3 Alcohol Use ...... 144
5.2.4 Rational Choice Theory ...... 145
5.2.5 Social Learning Theory...... 146
5.2.6 Classic Strain Theory ...... 148
5.2.7 Social Bonding Theory ...... 151
5.2.8 Demographic Covariates ...... 155
5.3 Analytic Plan ...... 155
5.3.1 Univariate ACE Decomposition Model ...... 156
5.3.2 Multivariate Regression ...... 160
5.3.3 DeFries-Fulker Analysis ...... 160
5.3.4 Gene-Environment Interaction ...... 163
5.3.5 Research Question 1 ...... 166
5.3.6 Research Question 2 ...... 167
5.3.7 Research Question 3 ...... 167
5.3.8 Research Question 4 ...... 168
CHAPTER 6 RESULTS ...... 173
6.1 Research Question 1 ...... 173
6.1.1 Antisocial Behavior ...... 174
6.1.2 Alcohol Use ...... 176
6.1.3 Illicit Drug Use ...... 177
6.2 Research Question 2 ...... 178
6.2.1 Antisocial Behavior ...... 179
6.2.2 Alcohol Use ...... 181 ix
6.2.3 Illicit Drug Use ...... 181
6.3 Research Question 3 ...... 182
6.3.1 Antisocial Behavior ...... 183
6.3.2 Alcohol Use ...... 184
6.3.3 Illicit Drug Use ...... 184
6.4 Research Question 4 ...... 185
6.4.1 Antisocial Behavior ...... 185
6.4.2 Alcohol Use ...... 187
6.4.3 Illicit Drug Use ...... 189
CHAPTER 7 DISCUSSION ...... 203
7.1 Summary of Results ...... 203
7.2 Limitations ...... 211
7.3 Future Directions ...... 214
APPENDICES ...... 220
A. ITEMS FOR ANTISOCIAL BEHAVIOR MEASURES ...... 220
B. UNIVARIATE ACE AND THRESHOLD LIABILITY MODEL RESULTS ...... 222
C. IRB APPROVAL FORMS...... 237
REFERENCES ...... 240
BIOGRAPHICAL SKETCH ...... 266
x
LIST OF TABLES
Table 2.1: Levels of Genetic Relatedness for Family Studies ...... 34
Table 5.1: Description of the Final Analytical Sample by Level of Genetic Relatedness ...... 170
Table 5.2: Overview of the Waves at which Each Included Measure is Assessed ...... 171
Table 6.1: Descriptive Statistics for all Included Measures ...... 191
Table 6.2: Results from Models Examining Antisocial Behavior at Wave 1 ...... 192
Table 6.3: Results from Models Examining Antisocial Behavior at Wave 2 ...... 193
Table 6.4: Results from Models Examining Antisocial Behavior at Wave 3 ...... 194
Table 6.5: Results from Models Examining Antisocial Behavior at Wave 4 ...... 195
Table 6.6: Results from Models Examining Alcohol Use at Wave 1 ...... 196
Table 6.7: Results from Models Examining Alcohol Use at Wave 2 ...... 197
Table 6.8: Results from Models Examining Alcohol Use at Wave 3 ...... 198
Table 6.9: Results from Models Examining Alcohol Use at Wave 4 ...... 199
Table 6.10: Results from Models Examining Illicit Drug Use at Wave 1 ...... 200
Table 6.11: Results from Models Examining Illicit Drug Use at Wave 2 ...... 201
Table 6.12: Results from Models Examining Illicit Drug Use at Wave 3 ...... 202
Table 7.1: Tabulation of Significant Associations between Theoretical Measures and Antisocial Outcomes Organized by Theory and Model Estimated ...... 218
Table B.1: Univariate ACE Model Results for the Wave 1 and Wave 2 Antisocial Behavior Measures ...... 222
Table B.2: Threshold Liability Model Results for the Wave 3 and Wave 4 Antisocial Behavior Measures ...... 223
Table B.3: Univariate ACE Model Results for the Wave 1 and Wave 2 Alcohol Use Measures ...... 224
Table B.4: Univariate ACE Model Results for the Wave 3 and Wave 4 Alcohol Use Measures ...... 225
Table B.5: Univariate ACE Model Results for the Wave 1, Wave 2, and Wave 3 Illicit Drug Use Measures ...... 226
xi
Table B.6: Univariate ACE Model Results for the Thoughtfully-Reflective Decision Making (TRDM) Measure...... 227
Table B.7: Univariate ACE Model Results for Peer Antisocial Behavior, Peer Illicit Drug Use, and Peer Alcohol Use Measures ...... 228
Table B.8: Univariate ACE Model Results for Classic Strain Measures ...... 229
Table B.9: Univariate ACE Model Results for Wave 1 and Wave 2 Parental Attachment Measures ...... 230
Table B.10: Univariate ACE Model Results for Wave 1 and Wave 2 School Attachment Measures ...... 231
Table B.11: Univariate ACE Model Results for Wave 1 and Wave 2 Neighborhood Attachment Measures ...... 232
Table B.12: Threshold Liability Model Results for Wave 1 and Wave 2 Peer Attachment Measures ...... 233
Table B.13: Univariate ACE Model Results for Wave 1 and Wave 2 Parental Involvement Measures ...... 234
Table B.14: Threshold Liability Model Results for Wave 3 Community Involvement Measures ...... 235
Table B.15: Univariate ACE Model Results for Wave 1 and Wave 2 Religious Commitment Measures ...... 236
xii
LIST OF FIGURES
Figure 3.1: Graphical Example of a Gene-Environment Interaction ...... 63
Figure 4.1: Visual Representation of Existing Biosocial Theories ...... 132
Figure 4.2: Visual Representation of Preliminary Models of Biosocial Integration ...... 133
Figure 4.3: Visual Representation of the Biosocial Integration Model ...... 134
Figure 5.1: Path Diagram of a Univariate ACE Model ...... 172
Figure 7.1: Visual Representation of the Percentage of All Significant Associations by Model Estimated...... 219
xiii
ABSTRACT
Within the field of criminology, there is no shortage of theoretical perspectives. While these theoretical perspectives are quite diverse, they do share a common thread: an exclusive focus on social factors that contribute to criminal behavior. Despite the limited explanatory power of both classic and more recent criminological theories, an overt sociological focus persists. In direct contrast, the biosocial perspective offers a more comprehensive explanation of behavior, with a focus on both biological and environmental influences. Despite the contributions of the biosocial perspective in elucidating the underlying etiology of antisocial behavior, there is currently a paucity of theories which can be effectively situated within the biosocial perspective. In an effort to spark theoretical development within the biosocial perspective, this dissertation proposes a biosocial integration model which allows for various forms of theoretical development and integration. In addition, four mainstream criminological theories—rational choice theory, social learning theory, classic strain theory, and social bonding theory—were empirically examined using genetically sensitive research designs in an attempt to fit such theories within the biosocial perspective. The results revealed three key findings. First, nearly all (more than 80 percent) of the measures examined were significantly influenced by genes. Second, while multivariate regression models identified a large number of significant associations between key theoretical concepts and antisocial behavior, many of these associations fell from statistical significance after controlling for genetic and shared environmental influences. Third, even after controlling for genetic influences, some theoretical concepts were significantly associated with antisocial behavior and substance use. Additional models revealed that several theoretical concepts also significantly moderated genetic influences on the examined outcomes. The findings are contextualized within the extant literature and suggestions for future research and theoretical development are discussed.
xiv
CHAPTER 1
STATEMENT OF THE PROBLEM
We are like the blind man who tries to describe the whole elephant after having only touched its trunk. Our theories and empirical explanations of crime as it varies over time, space, people, and the life course are woefully incomplete.
-Daniel S. Nagin (2007, pp. 260-261)
The above quote was offered by Nagin (2007) during his 2006 Sutherland Address to the
American Society of Criminology and exemplifies one of the most pervasive and concerning issues within the field of criminology. More specifically, the theories which constitute the very bedrock of criminological research tend to be limited in focus and leave the vast majority of the variation in antisocial and criminal behaviors unexplained (Bernard & Snipes, 1996; Elliott,
Ageton, & Cantor, 1979; Liska, Krohn, & Messner, 1989; Weisburd & Piquero, 2008). As with any scientific discipline, theoretical perspectives are critical in shaping both empirical research and knowledge regarding phenomena directly related to the discipline (Kuhn, 1996). Clearly, there is no shortage of criminological theories and an even greater number of refinements and reformulations of such theories (for example, see Lilly, Cullen, & Ball, 2011). Despite the numerous theories aiming to explain variation in criminal behavior, quantity cannot sufficiently supplant quality and the theoretical development within the field of criminology to date has resulted in “a million modest little studies that produce a million tiny conflicting results”
(Bernard & Snipes, 1996, p. 302).
In an attempt to contextualize this observation, Weisburd and Piquero (2008) examined all studies published in Criminology, widely regarded as the leading journal in the field of
1 criminology, between 1968 and 2005 in an effort to determine how well criminologists explain criminal outcomes. While over 1,000 articles were published in Criminology over this period of time, only 259 empirically assessed criminological theory in some way and only 169 included the information necessary to estimate the statistics included in the study. The authors then estimated the percentage of variance explained in criminal outcomes examined in each of the included articles using R2 coefficients. The study yielded two sets of findings that are directly relevant to this dissertation. First, the average percentage of variance explained by studies assessing existing criminological theories was modest at best and tended to hover between 20 and 30 percent of the variance in the examined criminological outcomes. The authors reiterate this finding by pointing out that “many models leave 80-90 percent of the variance unaccounted for” (pp. 472-473). This first set of findings provides a direct indication of the inability of existing theoretical perspectives to adequately explain criminological phenomena. Second, the findings indicated that the percentage of variance explained by criminological studies has not improved over time and may actually be declining. When summarizing their results, the authors point out that “[c]riminology does not appear to be evolving along a science-like track over time, one centered on improvement” (p. 474).
Based on this overall pattern of findings, it seems quite clear that existing theoretical perspectives, and the further refinement of such perspectives, may have taken the field of criminology as far as possible. While Weisburd and Piquero (2008) call for “[a] two-pronged approach of better theory and better and more sophisticated theory testing” (p. 494), such a call is basically a simple reiteration of the procedures the field of criminology has relied upon in the past. In this way, the root cause of the overall lack of explanatory power among existing criminological theory seems to stem directly from one of two sources: (1) either the theories
2 themselves; or (2) the paradigm from which the theories were developed. In either case, even extensive revisions to such theories are not likely to result in a noticeable change in the proportion of variance explained by concepts central to such theories. Rather, a more dramatic and sweeping change may be necessary to increase the explanatory power of the theories which drive research in the field of criminology. While existing criminological theories vary in their scope and emphasized concepts, the vast majority of these theories do share a common characteristic—they focus exclusively on the social and environmental factors that contribute to criminal behavior.
A developed and rigorous line of research in other disciplines, such as psychology, has revealed that the most powerful and effective explanations of antisocial behavior are comprised of both environmental and biological factors (Moffitt, 2005a; Pinker, 2002; Rowe, 1994).
Recently, this concerted emphasis on both environmental and biological influences on antisocial and criminal behavior has manifested within the field of criminology in the form of biosocial criminology (Beaver, 2013a; Walby & Carrier, 2010; Wright & Cullen, 2012). The biosocial perspective does not contradict the current criminological paradigm, rather, and as this aims to illustrate, the conventional criminological perspective can be situated within the larger biosocial perspective. In this way, conventional theoretical perspectives may still hold merit and can be augmented using the biosocial perspective to explain a greater portion of the variance in antisocial outcomes (Walsh, 2002). Despite the promise of the biosocial perspective and the theoretical development stemming from the incorporation of biosocial concepts into behavioral theories in other disciplines, the field of criminology has been slow to embrace the biosocial perspective (Beaver, 2013a; Wright & Boisvert, 2009; Wright & Cullen, 2012).
3
The need for a more integrated and biologically-conscious criminology is nothing new.
At least three previous ASC presidents have directly voiced the need for a greater integration of biology within the field of criminology (Cullen, 2011; Rafter, 2008; Zahn, 1999). While these calls have been largely ignored by the vast majority of criminologists, such apathy comes at the price of theoretical and empirical stagnation (as directly evidenced by Weisburd & Piquero,
2008). At the current time, the field of criminology finds itself in a peculiar situation: highly lauded theoretical perspectives only explain a menial proportion of the variance in criminal behavior; a massive amount of time and pages have been spent attempting to modify and revise existing theories with virtually no progress; and some of the most prominent criminologists continue to call for more of the same (see for example Sampson, 2013). In order to progress criminology as a scientific discipline, it is first necessary to discard the underlying assumption that all variation in human behavior is the direct result of socialization. Failing to make such a change endangers the future of the field of criminology.
1.1 Research Questions
Despite the recent traction gained by biosocial criminology, there has been virtually no formal effort made to integrate the existing criminological theories into the biosocial perspective.
What makes this revelation all the more surprising is the numerous calls for such integration over the past several decades within the field of criminology. In her address to the American Society of Criminology (ASC), Margaret Zahn (1999) openly called for the incorporation of biology into the field of criminology. More specifically, she noted “[i]t is important, if not imperative, that we become aware of the biological and biochemical bases for behavior and incorporate them, where relevant, into our theory about violent crime and its consequences” (emphasis in original,
4 p. 3). Even after such a direct call for integration, existing criminological theories remain nearly devoid of biological or biosocial factors related to criminal and delinquent behavior (for an exception see Moffitt, 1993). This dissertation attempts to directly address this gap in the literature by examining whether mainstream criminological theories fit within the biosocial perspective. More specifically, four mainstream criminological theories—rational choice theory, social learning theory, classic strain theory, and social bonding theory—will be empirically examined within the confines of genetically sensitive research designs in an effort to determine whether such theoretical perspectives fit within the larger biosocial perspective.
The current study will examine four research questions directly related to the task of fitting existing criminological theory within the biosocial perspective. These four research questions are:
Rational Choice Theory:
1. Does adequately controlling for genetic influences alter the overall association between
thoughtfully reflective decision making (TRDM) and antisocial behavior?
Social Learning Theory:
2. Does exposure to delinquent peers significantly predict subsequent antisocial behavior
even after taking genetic influences into account?
Classic Strain Theory:
3. After controlling for genetic influences, does strain, as defined by Merton (1934),
significantly predict antisocial behavior?
Social Bonding Theory:
4. Does controlling for genetic influences reduce or eliminate the association between
attachment, involvement, and commitment and antisocial behavior?
5
Each of these four Research Questions, and the existing research relevant to each, will be discussed in more detail in subsequent chapters. These four Research Questions are aimed at attempting to fit each of the examined theories within the biosocial perspective. As discussed more thoroughly in subsequent chapters, a necessary prerequisite to integrating each theory into the biosocial perspective is to isolate the effects of each theory’s primary concepts on antisocial behavior net of the effects of genetic influences. Concepts which fail to predict antisocial outcomes in such a context are likely subject to genetic confounding and may have no real influence on antisocial behavior when such confounding is properly controlled. Alternatively, theoretical concepts which significantly explain variance in a given antisocial outcome within the confines of a genetically sensitive modeling strategy should be viewed as highly robust predictors of antisocial behavior. Biosocial integration can only be carried out with theories comprised of concepts that fall into the latter category. In this way, the four Research Questions examined in the current dissertation aims to accomplish two related goals: (1) examine whether concepts central to each examined theory continue to predict a wide range of antisocial outcomes while controlling for genetic influences; and, more importantly (2) whether the examined conventional theories themselves can be integrated into the larger biosocial perspective.
Based on these goals and the proposed Research Questions, the current dissertation aims to make three distinct contributions to the extant literature. First, no previous studies have examined the predictive ability of the four examined theoretical perspectives within a genetically informed modeling strategy. In this way, the current empirical status of each perspective remains relatively unknown. While it remains fully possible that each perspective explains a significant portion of the overall variance in antisocial behavior, it also remains possible that previously examined associations are the direct result of genetic confounding. Second, the
6 current dissertation provides a novel biosocial integration strategy which provides multiple strategies aimed at incorporating existing criminological theories into the larger and more comprehensive biosocial perspective. The biosocial integration model builds on previous forms of theoretical integration and directly addresses many of the limitations present within previous efforts. Third, the current dissertation attempts to situate each of the four theoretical perspectives examined into the biosocial perspective in an effort to increase their explanatory power and scope.
1.2 Outline
Based on the overarching goals of the current dissertation, a thorough review of multiple lines of research is necessary and will be offered. Due to the large amount of information reviewed, it is necessary to provide an overview of the general organization of this dissertation.
Chapter 2 provides an overview of the biosocial perspective and is organized into four sections.
First, an overview of standard social science methodologies (SSSMs) and the methodological limitations they possess will be presented. Second, a comprehensive overview of the most common behavior genetic (BG) modeling strategies will be presented. Importantly, BG modeling strategies allow for the estimation of both genetic and environmental contributions to various behavioral outcomes. Even more importantly, these modeling strategies also allow for the isolation between environmental influences on a given outcome while adequately controlling for genetic influences. Third, the literature examining genetic and environmental influences on antisocial behavior will be reviewed. Fourth, the various ways in which genetic and environmental influences combine to generate variation in behavioral outcomes, a phenomenon referred to as gene-environment interplay, will be outlined.
7
Chapter 3 provides an overview of the current state of rational choice, social learning, classic strain, and social bonding theories. The chapter is divided into four sections, each of which provides an overview of the key assumptions and theoretical concepts implicated in each of the examined theories. In addition, each section provides an overview of the empirical evidence bearing on each of the examined theories. The primary purpose of Chapter 3 is to provide an overview of the empirical validity of each of the examined theories, and to provide an overview of the development of each theory over time.
Chapter 4 is divided into two sections and formally proposes the biosocial integration model. The first section provides an overview of theoretical integration and the various forms of integration that have been previously proposed. In addition, the first section of Chapter 4 provides a summary of precursors to the biosocial integration model including previous biosocial theories (Ellis, 2005; Robinson & Beaver, 2009) and preliminary models of biosocial integration
(e.g., Fishbein, 1990; Walsh, 2000; 2002). The second section of Chapter 4 presents the biosocial integration model and outlines the primary advantages of this model over previous models of biosocial integration.
Chapter 5 presents the methods used to address the four Research Questions and is divided into three subsections. The first section provides an overview of the data and analytic sample used in the current dissertation. The second section provides a detailed description of the measures used. The third and final section provides a detailed description of the analytic techniques used to answer each of the four Research Questions.
Chapter 6 presents the findings of the analyses described in the previous chapter and is divided into four sections. Each section presents the findings related to an individual theory and the related Research Question. In this way, the findings related to rational choice theory are
8 presented in the first section, social learning theory in the second section, classic strain theory in the third section, and social bonding theory in the fourth section.
The seventh and final chapter provides a discussion of the findings and is divided into three sections. First, the results of the analyses performed are summarized. In addition, the findings are also contextualized within the extant literature and the biosocial integration model proposed in Chapter 4. The second section provides an overview of the limitations of the current dissertation and how such limitations may impact the results reported. Finally, the third section discusses the implications of the findings of the current dissertation for future research and theoretical integration.
9
CHAPTER 2
THE BIOSOCIAL PERSPECTIVE
To understand human individuality, social science must be bio-social in its outlook…Scientific advance depends on a relentless effort to reduce complex phenomena to their less complex elements and to build back and see how the elements contribute to the whole.
-David C. Rowe (2001, pp. 73-74)
The role of environmental and social influences in the etiology of criminal and delinquent behavior has been extensively examined in the extant criminological research. Nearly all major criminological theories implicate either a specific social/environmental factor or a combination of such factors in the etiology of criminal and delinquent behaviors. Influential studies have examined the manner in which a wide range of social and environmental factors, including neighborhoods (Anderson, 1999; Sampson, Raudenbush, & Earls, 1997), peer networks (Warr,
2002), and families (Hirschi, 1969) ultimately influence antisocial and criminal behavior. The influence of such social institutions on the development of criminal or delinquent behavior has been extensively examined using a wide range of methodologies and statistical techniques.
Studies have detected significant associations between these purely social influences and antisocial behaviors within a large number of samples, over extended periods of time, and for a wide assortment of antisocial behaviors.
While this culmination of findings is quite impressive, such findings are far from unequivocal and should still be viewed with caution. Nearly all criminological research examining the association between social factors and antisocial behaviors has relied on the use of standard social science methodologies (SSSMs; Beaver, 2013a; Harris, 2009; Walsh, 2002).
SSSMs refer to any methodology or analysis that disallows researchers from appropriately
10 accounting for genetic influences. Such methodologies involve the selection of one member per household for inclusion in the analytic sample. Information pertaining to the individual’s behavior and environment is then collected from each respondent and is then analyzed in statistical models to test various hypotheses. This procedure of collecting behavioral and environmental information from one individual per household implicitly assumes that each of the measures included in the model, including the outcome measure, is purely social and is not under any genetic influence. This underlying assumption of SSSMs supports the entire weight of these methodologies and reveals an important limitation of such modeling strategies: the presence of any genetic influence on any of the variables analyzed will violate this assumption and result in significant limitations including biased parameter estimates and standard errors.
While SSSMs have been used extensively in criminology and a number of other social science disciplines, such methodologies may suffer from a serious methodological limitation that is commonly overlooked. Since SSSMs are unable to estimate and partial out the influence of genetic factors on observed associations, it remains possible that uncontrolled genetic influences may confound any observed association between social/environmental factors and antisocial behaviors. More specifically, the potential presence of genetic influences on both socialization variables and antisocial behaviors effectively prevents SSSMs from adequately isolating the influence of a given social influence on antisocial behavior, resulting in a spurious association
(Johnson et al., 2009; McGue et al., 2010). In this way, studies that employ SSSMs inherently assume that genetic factors have no influence on the independent and dependent variables of interest. As long as this assumption is not violated, any observed association will not be confounded by uncontrolled genetic influences and the use of an SSSM is appropriate. However, mounting evidence indicates that this assumption is becoming increasingly untenable, regardless
11 of the combination of independent and dependent variables examined (Ferguson, 2010; Mason &
Frick, 1994; Miles & Carey, 1997; Moffitt, 2005b; Rhee & Waldman, 2002). In this way, existing research employing SSSMs has likely produced findings that are biased due to the inability to adequately partial out the confounding influence of genetic factors. To state the problem as clearly as possible: there is a substantial amount of evidence which indicates that many of the previously identified associations between concepts central to conventional criminological theories and antisocial behaviors are spurious.
Despite a significant amount of theoretical evidence suggesting that genetic confounding is a widespread methodological limitation plaguing the vast majority of the extant criminological literature, it is currently unclear how and to what extent such limitations may bias criminological research. The primary goal of this dissertation is to clarify this significant gap in the existing literature by examining whether several of the most influential and prominent theories of criminal behavior can be empirically integrated into the biosocial perspective. In other words, do conventional theoretical explanations of criminal behavior hold after accounting for genetic confounding? In order to address this question, along with the previously specified Research
Questions, it is necessary to employ methodologies which properly account for genetic influences and effectively isolate the influence of theoretical concepts on the outcome of interest.
Collectively, such methodologies are referred to as behavior genetic (BG) research methodologies and are becoming increasingly popular in a number of social science disciplines including criminology (Barnes & Boutwell, 2013; Jaffee, Price, & Reyes, 2013).
Prior to providing an overview of the various methodological techniques which adequately control for genetic confounding and result in fully-specified models, it is first necessary to clarify one particular issue: biosocial criminology represents an overarching
12 paradigm or perspective, not a single theory of human behavior. This point is stated explicitly since even the most recent editions of some of the most widely read and required textbooks on criminological theory made the mistake of narrowly classifying biosocial research and findings as theories as opposed to an overarching perspective which can encompass many theories. For example, in the sixth edition of their widely read textbook Criminological Theories:
Introduction, Evaluation, and Application, Akers and Sellers (2013) repeatedly refer to central issues within the biosocial perspective—such as genetic influences or neuropsychological functioning—as aspects of a theory. However, this particular view is far too narrow, as these particular findings and concepts stem not only from their own theories, but also from their own disciplines in some instances. Behavior genetics is a subfield of psychology with several universities offering degrees (including doctoral degrees) in the field of behavior genetics.
Neuroscience and molecular genetics are similar. These are not theoretical concepts that fit cleanly within the overarching theory of biosocial criminology. Rather, these are perspectives in their own right—which include theories, concepts, and other organizing ideas—which fall under the much larger perspective of biosocial criminology. In other words, the biosocial perspective borrows concepts, theories, and methodologies from other perspectives in an effort to better understand individual variation in human behavior.
This distinction is an important one since paradigms—at least in the Kuhnian sense— often organize multiple theories and it is often times impossible to present a single theory which adequately represents an entire perspective or paradigm (Kuhn, 1996). While some biosocial criminologists have attempted to develop a biosocial theory of criminal behavior (Ellis, 2005;
Robinson & Beaver, 2009), such efforts have been largely unsuccessful with the resulting theories being overly complex in an effort to take so many sets of influences into account (both
13 biological and environmental). In addition, the resulting theories consist of multiple intricate facets which, even individually, would be quite difficult or even impossible to test empirically.
While such efforts should be lauded, they are likely far too ambitious. Antisocial behavior is a highly complex phenotype that is likely the result of a complex combination of environmental and biological influences. Attempting to capture the gradations of these influences (and the combinations between them) in a single theory is a task that is likely impossible. Rather, there are likely multiple theoretical perspectives that fit nicely within the biosocial perspective and collectively specify the underlying etiology of antisocial and criminal behavior.
In order for theoretical explanations of criminal behavior to effectively fit within the biosocial perspective, the concepts contained within such theories must have a causal influence on antisocial phenotypes after controlling for genetic influences. Theories which emphasize concepts that do not significantly explain variance in antisocial outcomes after taking genetic influences into account do not fit within the biosocial perspective. The primary goal of this dissertation is to empirically test whether concepts specified within conventional criminological theories fit within the biosocial perspective. In order to perform this particular task, it will be necessary to utilize BG modeling strategies to effectively remove genetic influences and isolate the potential association between each theoretical concept and the outcome of interest. In this way, it is necessary to provide an overview of BG modeling strategies and the logic underlying such methodologies. The remainder of this chapter will be devoted to providing an overview of
BG methodologies and the findings from studies employing such methodologies to better specify the extent to which genetic and environmental influences explain individual level differences in antisocial phenotypes.
14
2.1 Behavior Genetic Methodologies
In an attempt to further integrate the biosocial perspective within the field of criminology,
Walsh (2000) suggested “it is time for mainstream criminology to at least pull back its blinders and peek at what behavior genetics has to offer” (p. 1077). Unfortunately, based on the limited number of criminologists who actively and consistently employ genetically sensitive research designs, few have followed Walsh’s suggestion (Wright et al., 2008; Wright & Cullen, 2012). In fact, since Walsh’s suggestion was published in 20001, only 1.4% (five total) 2 of all articles published in Criminology, the flagship journal of the American Society of Criminology, employ research designs that effectively estimate and partial out the influence of genetic factors. A closer examination reveals that a total of eight criminologists have collectively contributed to the five articles published in Criminology in a period of time that spans over a decade. This preliminary investigation further solidifies the observation that the vast majority of criminologists are unfamiliar with BG methods and the assumptions and procedures that accompany such methodologies. For this reason, it is necessary to provide an overview of BG methods.
BG methods, at their most fundamental level, are specialized analytic techniques that allow for the estimation of both genetic and environmental influences on virtually any phenotype or measureable characteristic. More specifically, BG techniques partition the variance within a given phenotype (e.g., height, weight, delinquency, or self-control) into three latent components:
(1) heritability (symbolized as h2), (2) shared environmental influences (symbolized as c2), and
1 A total of 357 articles contained within 51 issues of Criminology have been published in the timeframe mentioned. The five articles which employ behavior genetic methodologies comprise only 1.4% of this overall total. 2 Importantly, a closer inspection of one of the studies included in this count (Haynie & McHugh, 2003) reveals that while behavior genetic methods were employed (i.e., DeFries-Fulker analysis), such methods were not executed properly and the resulting findings are rendered uninterruptable due to the authors’ misunderstanding of the analytic strategy. Despite these problems, the authors did indeed attempt to utilize a genetically informed analysis and the study is retained in the count presented. 15
(3) nonshared environmental influences (symbolized as e2). The heritability component of a BG analytic model provides an estimate of the proportion of the variance in the examined phenotype that can be explained by additive genetic influences. Additive genetic influences refer to the sum of the average independent effects of all genes that influence the examined phenotype (Neale,
2009). This type of genetic influence is juxtaposed against dominance genetic contributions and epistasis. Dominance effects take into account the influence of interactions between alternative copies of a single gene (which are referred to alleles) on the examined phenotype (Beaver,
2013a). Epistasis takes into account the influence of interactions between alleles of two or more different genes on the examined phenotype (Neale, 2009). Additive genetic influences operate under the assumption that each of the genes that collectively contribute to the examined phenotype operate independently of one another. Alternatively, dominance and epistasis assume nonadditive effects on the examined phenotype in that multiple alleles and genes are working interactively with one another. Dominance and epistasis are hypothesized to have relatively small effects on most human phenotypes and are therefore typically omitted from most BG models. Rather, the proportion (or percentage) of overall variance in the examined phenotype that can be explained by additive genetic influence is typically estimated and referred to as heritability.
BG modeling strategies also estimate the proportion of overall variance in a given phenotype that can be explained by environmental influences. However, as alluded to above,
BG modeling strategies distinguish between two separate types of environmental influences.
First, shared environmental influences are comprised of environmental experiences that are the same between siblings from the same household and work to make siblings more closely resemble one another across the examined phenotype. The underlying logic assumes that if a
16 particular environment explains a significant portion of the variance in the examined phenotype and this environment is experienced by both (or all) siblings, then the environment should have a similar impact on both siblings making them more closely resemble one another. For example, growing up in poverty would impact both siblings from a given household and would thereby be considered a shared environmental influence. Other examples include experiencing a divorce, family abuse or neglect, and neighborhood-level factors that are experienced by both siblings. In direct contrast to shared environmental influences are nonshared environmental influences which are comprised of environmental experiences that are differentially experienced between two siblings from the same household. In this way, any environment that is different between siblings and differentiates one sibling from the other is considered a nonshared environment.
Common examples of the nonshared environment include differential peer groups, different school environments, and different parenting strategies (such as parents treating their children different from one another). Importantly, nonshared environments can also arise when siblings subjectively interpret an objective event (Turkheimer & Waldron, 2000). For example, siblings may interpret and react differently to an event that they both experience such as their parents divorcing. In this way, it is often difficult to positively differentiate shared environments from nonshared environments, with some behavior geneticists even going as far as to question whether shared environments even exist in the purest sense (Turkheimer & Waldron, 2000).
BG modeling strategies are designed to partition 100 percent of the variance in a given phenotype into these three categories (h2, c2, and e2) as latent factors. In other words, each of the components estimated indicate the collective influence of all genetic and environmental influences that culminate into the phenotype of interest. This characteristic of BG modeling strategies is highly attractive in that such models account for all environmental and genetic
17 influences that significantly influence variation in the phenotype examined are included in the model, providing a distinct advantage over SSSMs. In this way, BG modeling strategies effectively partial out genetic influences on the examined phenotype and isolate the association between independent and dependent variables of interest, resulting in a much more conservative quasi-experimental design (Johnson et al., 2009). SSSMs, on the other hand, are incapable of removing genetic influences and make it impossible to effectively isolate any potential association. The fundamental difference between BG modeling strategies and SSSMs which provides the former with distinct advantages over the latter, is that BG modeling strategies analyze samples which include more than one child per household, while SSSMs rely on sample with only one child per household. This inclusion of more than one child per household in the analytic sample allows for the decomposition of the overall variance in a given phenotype into the components described above. Since SSSMs do not analyze samples with more than one child per household they cannot differentiate between genetic and environmental influences making it impossible to adequately account for genetic confounding and preventing any causal associations from being detected.
Below, a more detailed description of the procedures that are used to effectively partition phenotypic variance into each of the aforementioned categories will be provided. As mentioned above, the majority of these procedures require analytic samples that include more than one sibling per household and cannot be performed on samples which only include one sibling per household. The reasoning and logic underlying this particular data requirement will also be discussed. Finally, an overview of the most common BG modeling strategies, along with the advantages and limitations of each, will be provided. This discussion will commence with the most common modeling strategy—the twin study.
18
2.1.1 Twin Studies
BG modeling strategies require analytic samples that possess two specific characteristics.
First, as discussed previously, the analytic sample must include more than one child from the same household. Second, the level of genetic relatedness between children from the same household must be accurately recorded. In other words, most BG modeling strategies require that the amount of genetic material that the household members share with one another is known and recorded. Importantly, samples of twin pairs satisfy both of these requirements. By definition, samples that consist of twin pairs include multiple children from the same household.
In addition, there are two differing types of twins in the general population: monozygotic (MZ; commonly referred to as identical twins) and dizygotic (DZ; commonly referred to as fraternal twins) twins. MZ twins share 100 percent of their genetic material and occur in approximately 1 in every 250 births. The MZ twinning process is quite similar to non-twin pregnancies in which a single sperm fertilizes a single egg, but then, for reasons that remain unknown currently, the fertilized egg splits to form two identical embryos which eventually develop into two identical fetuses. In contrast, DZ twins only share 50 percent of their genetic material, which is the same amount of genetic material that is shared by non-twin full siblings. DZ twins occur in approximately 1 in every 125 births and are the result of two separate sperm fertilizing two separate eggs. Since the fertilization occurs at roughly the same time, both fertilized eggs eventually develop into fetuses that are carried during the same pregnancy. Since twins are typically reared in the same household by the same parents and the procedures involved with assessing the level of genetic relatedness between twins is fairly simple and straightforward, samples of twin pairs have become a staple in BG research.
Twin studies analyze samples that consist of MZ and DZ twins to estimate the heritability
(h2), shared environmental influences (c2), and nonshared environmental influences (e2) on a 19 phenotype of interest by examining how similar MZ twins are to one another relative to DZ twins across the phenotype of interest. Since MZ twins share twice as much genetic material as
DZ twins but both types of twins’ rearing environments are identical, the only factor that would result in MZ twins more closely resembling one another than DZ twins are genetic influences.
For example, if a research team were interested in determining whether genetic factors have a significant effect on body mass index (BMI), they could simply collect BMI information from a sample that consisted of both MZ and DZ twin pairs. There would be preliminary evidence that genetic factors have a significant influence on BMI if each MZ twins’ BMI scores more closely resemble their co-twins’ BMI score relative to the similarity between each DZ twin and their co- twin.
While the underlying logic of twin studies is fairly straightforward, there are additional details regarding the procedures used to compare twins from the same pair with one another.
The most common procedure involves the estimation of an intraclass correlation coefficient. An intraclass correlation coefficient expresses the direction and the magnitude of the association between one twin’s score on the measure of interest (e.g., BMI) with their co-twin’s score on the same measure. While traditional correlation coefficients focus on the association between two different variables, intraclass correlation coefficients estimate the association between the same measure but for two different members of the same household (e.g., a twin and their co-twin).
MZ intraclass correlation coefficients can then be compared DZ intraclass correlation coefficients to garner estimates of h2, c2, and e2 on the examined phenotype. Keep in mind that the majority of twins included in these samples are raised in the same household by the same parents, grow up in the same neighborhoods, and often attend the same schools as their co-twin.
In this way, environments between MZ and DZ twin pairs should be relatively similar and the
20 only reason that MZ twins would more closely resemble one another than DZ twins would be because MZ twins share twice as much genetic material with one another than DZ twins.
MZ and DZ intraclass correlations can be used to calculate direct estimates of the proportion of overall phenotypic variance explained by genetic, shared environmental, and nonshared environmental influences (Neale & Maes, 1992). The proportion of variance that can be explained by genetic influences can be estimated as:
h2 = 2(rMZ – rDZ) (2.1) where rMZ refers to the intraclass correlation for MZ twins, and rDZ refers to the intraclass correlation for DZ twins. As is indicated in Equation 2.1, the intraclass correlation for DZ twins is subtracted from the intraclass correlation for MZ twins which effectively removes all of the variance in the phenotype that is due to shared environmental factors. In addition, the difference term included in Equation 2.1 also removes half of the variance in the examined phenotype that is due to genetic influences since MZ twins share twice as much genetic material as DZ twins.
In order to correct for this underestimation, the resulting difference term is doubled. The resulting coefficient is a direct estimate of the proportion of overall phenotypic variance that can be explained by genetic influences (Neale & Maes, 1992; Plomin, 1990; Walsh, 2002).
The same intraclass correlations for MZ and DZ twins can be used to estimate the proportion of variance that is explained by shared environmental influences. The coefficient for c2 can be estimated as:
c2 = 2(rDZ) – rMZ (2.2) where rDZ still refers to the intraclass correlation for DZ twins and rMZ still refers to the intraclass correlation for MZ twins. However, there are two noticeable differences regarding
Equation 2.2 relative to Equation 2.1. First, only the rDZ term is doubled. Doubling rDZ fixes
21 the proportion of variance explained by genetic influences to be equal for MZ and DZ twins.
Second, the rMZ term is subtracted from the rDZ term (after it is doubled). The resulting coefficient provides a direct estimate of the proportion of overall phenotypic variance that can be explained by shared environmental influences.
Finally, a direct estimate of the proportion of phenotypic variance that is explained by nonshared environmental influences can be calculated as:
e2 = 1 – (h2 + c2) (2.3) where h2 refers to the heritability estimate obtained from Equation 2.1 and c2 refers to the estimate of shared environmental influence obtained from Equation 2.2. Both terms are summed and then subtracted from 1 to estimate the residual variance in the examined phenotype that is not explained by genetic or shared environmental influences. Importantly, since e2 is estimated as the residual variance that is not explained by h2 and c2, measurement error is also included in the e2 estimate.
Two additional points regarding the equations presented above and their resulting coefficients warrant additional attention. First, the sum of h2, c2, and e2 will always be equal to
1.00, indicating that all three components collectively explain 100 percent of the overall variance in the phenotype examined. Second, the equations presented above are somewhat antiquated and have been updated to increase accuracy and to be used with samples that include a wider range of kinship pairs other than twins. While the inclusion of additional types of kinship pairs increases the versatility of BG modeling strategies, the modified equations become far more computationally advanced. Despite these recent developments, the underlying logic of twin- based studies remains the same: if MZ twins more closely resemble one another than DZ twins
22 across a variable of interest, genetic influences explain a significant portion of the variance in the examined variable.
While the logic underlying twin studies is fairly straightforward, there are three assumptions that must be met for the estimates obtained from twin samples to be considered unbiased. The first assumption is referred to as the equal-environment assumption (EEA) and refers to the assumption that twins from the same MZ pair experience environments that are no more similar than twins from the same DZ pair. In other words, if MZ twins experience environments that are significantly more similar than environments that are experienced by DZ twins, estimates from BG models may be biased. Critics of twin studies have argued that a violation of the EEA would result in upwardly biased heritability coefficients. More specifically, critics have argued that MZ twins more closely resemble one another relative to DZ twins since they experience environments that are more similar. For example, it is not uncommon for parents to dress MZ twins similarly or for MZ twins to be mistaken for one another (Loehlin & Nichols, 1976). While these observations may seem to provide preliminary evidence which suggests that the EEA is commonly violated, it is important to point out that the only violations of the EEA that would potentially result in biased heritability estimates are those that increase the similarity of MZ twins on the measures of interest. For example, the fact that
MZ twins have a more similar physical appearance than DZ twins may violate the EEA, but such a violation would not upwardly bias the influence of genetic factors on a measure of delinquency. This is primarily due to the fact that there is no empirical evidence suggesting that the physical appearance of a given individual is related to their propensity to engage in delinquency, to argue otherwise would be to argue in favor of Lombroso’s (2006 [1876]) theory
23 of the criminal man. Regardless, violations of the EEA remain a significant point of contention between critics and advocates of twin-based research designs.
Fortunately, there has been a significant amount of research examining how often the
EEA is violated and whether violations of the EEA result in biased heritability estimates (Cronk et al., 2002; Eaves et al., 2003). Providing an empirical examination of the potential repercussions of violating the EEA is not a simple task to carry out and researchers have had to rely on innovative approaches. A number of studies have examined samples in which a subsample of MZ and DZ twin pairs has been incorrectly classified wherein MZ twins were mistakenly classified as DZ twins and vice versa (Conley et al., 2013; Gunderson et al., 2006;
Kendler, 1983; Kendler et al., 1993). This type of misclassification typically occurs when researchers are forced to rely on physical similarities to assess zygosity as opposed to genotyping. Such cases provide the opportunity to examine the potential effect of violating the
EEA on heritability coefficients. More specifically, if the EEA is violated, MZ twins who were mistakenly classified as DZ twins should be less similar to one another than MZ twins who were correctly classified. Along the same lines, DZ twins who were misclassified as MZ twins should be more similar to one another than DZ twins who were correctly classified.
Collectively, the results of these studies found no evidence in favor of upwardly biased heritability coefficients. In addition, the results of a recent study (Conley et al., 2013) which analyzed misclassified twins from three separate samples found that violation of the EEA actually resulted in a significant reduction in heritability coefficients indicating that such estimates are actually downwardly biased. Importantly, Conley and colleagues’ findings extended to a wide range of phenotypes including height, weight, BMI, ADHD, depression, and even delinquency. Taken as a whole, the results of these studies seem to indicate that violation
24 of the EEA does not make MZ twins more similar to one another across a wide range of behavioral phenotypes, many of which are considered important causes or correlates of antisocial behavior.
The second assumption underlying twin studies is an absence of assortative mating. The concept of assortative mating refers to the phenomenon of human mate selection practices that result in mating couples that closely resemble one another across a wide range of physical, psychological, and behavioral characteristics (Boutwell et al., 2012; Krueger et al., 1998; Lykken
& Tellegen, 1993; Shackelford et al., 2005). For example, studies have revealed that mates tend to possess similar demographic characteristics (Mare, 1991; Rhule-Louie & McMahon, 2007), similar levels of educational attainment (Lefgren & McIntyre, 2006) and intelligence (Mascie-
Taylor & Vandenberg, 1988), and even political affiliation (Kandler, Bleidorn, & Riemann,
2012). In addition, a corresponding line of literature has revealed that mates tend to display similar levels of antisocial and delinquent behaviors, wherein delinquent individuals are significantly more likely to select a delinquent mate than a nondelinquent individual (Beaver,
2013b; Boutwell et al., 2012; Capaldi et al., 2008; Krueger et al., 1998; Rhule-Louie &
McMahon, 2007). For example, in a recent study, Boutwell and colleagues (2012) analyzed a nationally representative sample of mates from the Early Childhood Longitudinal Study, Birth
Cohort (ECLS-B) and found that the propensity for criminality (measured as antisocial behavior and substance use) between mates was strongly correlated, with a correlation coefficient of .54.
Taken together, the results of this line of research provide overwhelming evidence that individuals tend to select mates that are similar to themselves.
The presence of assortative mating may potentially bias results from BG models because this process results in a systematic increase in the genetic similarity of all siblings that possess
25 any dissenting genetic material (i.e., all siblings other than MZ twins). Since mates tend to more closely resemble one another across phenotypes that have been found to be influenced by genetic factors than individuals selected at random, there is also a greater probability that such mates will also have genotypes that are more similar than two individuals chosen at random. For example, delinquency has been found to be significantly influenced by genetic factors (Mason & Frick,
1994; Miles & Carey, 1997; Rhee & Waldman, 2002; Ferguson, 2010) and mates tend to engage in similar rates of offending (Beaver, 2013b). In this way, there is a high probability that mates will share many of the polymorphisms that are related to delinquency. This, in turn, may result in an increase in the amount of genetic material that their non-MZ twin offspring possess. Based on the research outlined above, there is good reason to believe that the assumption of no assortative mating is consistently violated in studies employing BG methods.
While the violation of this assumption does result in biased parameter estimates, studies have revealed that the presence of assortative mating actually results in significantly underestimated heritability estimates (Keller et al., 2013; Vinkhuyzen, van der Sluis, Maes, &
Posthuma, 2012). For example, Beauchamp and colleagues (2011) found that specific BG models are highly sensitive to violations of the assortative mating assumption and models which adequately account for assortative mating resulted in a dramatic increase in the estimated heritability estimate (from .31 to .91) and a significant decrease in the shared environmental estimate (from .59 to .00). Taken together, it seems that the assortative mating assumption is consistently violated in studies employing BG modeling strategies. However, violation of this assumption actually results in more conservative heritability estimates.
The third and final assumption underlying twin studies is that the phenotype of interest must be measured accurately and without error. Since the nonshared environment coefficient
26 estimates the proportion of the phenotypic variance that is explained by the nonshared environment and measurement error, it is possible that measures with higher amounts of error may yield less reliable estimates. In other words, measures that are less reliably measured may result in inflated e2 estimates, which, in turn, may artificially deflate estimates of h2 and c2. Once again, a violation of this assumption would result in an artificially deflated heritability coefficient.
It is worth noting that these three assumptions are commonly violated in studies employing BG research. In many instances there is simply no way to avoid a violation of at least one of the assumptions. Critics of BG research may point to this fact in an effort to discredit the findings garnered from such studies and claim that violations of these assumptions result in biased heritability estimates which calls into question the finding that genetic factors significantly influence examined phenotypes. Such critics are simply misinformed and do not fully understand the repercussions that accompany the violation of each of these assumptions.
While studies have revealed that a violation of any of the assumptions does result in biased heritability coefficients, the vast majority of the studies that have examined the extent to which heritability estimates are biased have indicated that these estimates are downwardly biased, indicating that heritability estimates are more conservative than they should be (Beauchamp et al., 2011; Conley et al., 2013). In other words, BG models are designed to err on the side of caution and consistently produce highly conservative heritability estimates and inflated estimates of environmental influence. In this way, it is far more likely that genetic factors explain a significantly greater proportion of the variance in phenotypes related to antisocial behavior than previous studies have reported.
27
In addition to violations of the underlying assumptions of BG models, critics have also pointed out that BG modeling strategies rely on samples of twins and that such samples may not be representative of samples of singletons, making it impossible to extrapolate findings from genetically sensitive research designs to any larger population. A recent study by Barnes and
Boutwell (2013) explored this critique by comparing the effects of a wide range of covariates on antisocial behavior (which included delinquency, drug use, and victimization) between twins and non-twins from a nationally representative sample. Some of the covariates included in the analysis included: low self-control, depression, delinquent peers, popularity, physical attractiveness, GPA, and a host of parenting measures. The results revealed that twins did not significantly differ from non-twins across levels of antisocial behavior and each of the examined covariates. In addition, the association between each of the covariates and the antisocial behaviors examined did not significantly differ between twins and non-twins. The results of this study provide strong evidence indicating that twin samples are representative of non-twin samples, effectively dampening any argument to the contrary.
Despite the overwhelming amount of evidence suggesting that the above concerns regarding BG modeling strategies are not warranted, additional research strategies have been devised to directly address such limitations. While all of these strategies are distinct from one another, they do share a common difference from traditional twin studies in that they do not compare common phenotypes across MZ and DZ twin pairs. Rather, these techniques rely on alternative means to partial out genetic confounding and isolate the association of interest. Since these alternative modeling strategies do not rely on comparisons within twin-based samples, they are not subject to the assumptions outlined above. In this way, a violation of such assumptions
28 should not result in biased (either upwardly or downwardly) parameter estimates. Below, each of these alternative modeling strategies will be outlined.
2.1.2 Monozygotic Twins Reared Apart (MZA)
In an effort to eliminate any bias resulting from a violation of the EEA, researchers have obtained samples of MZ twins who were separated at birth and reared in different homes by different parents. These twins, symbolized as MZA, are unique in that they are genetically identical but have been raised in completely different environments. In this way, MZAs provide researchers with the unique opportunity to effectively separate genetic and environmental influences on any phenotype. Since MZAs share 100 percent of their genes, but experience very different environments, any similarities between them have to be solely the result of genetic influences and, conversely, any differences between them are solely the result of nonshared environmental influences. Since the twins were reared in unique environments, shared environmental influences cannot play a role. Based on this logic, we would expect that phenotypes that are significantly impacted by the environment would vary significantly between
MZAs, while phenotypes that are under significant genetic influence would be highly similar between MZAs.
One of the primary limitations of this modeling strategy is that MZAs, as one would expect, are exceedingly rare, making it difficult to generate samples with enough statistical power to generate meaningful results. However, a team of researchers at the University of
Minnesota have established the Minnesota Study of Identical Twins Reared Apart (MISTRA) in an effort to generate a functional sample of MZAs. To date, MISTRA has identified and interviewed over 100 MZA pairs. Members of the MISTRA research team have been able to analyze the data collected from the MZA pairs they have identified in an effort to examine
29 genetic and environmental influences on a wide range of phenotypes including many that are direct correlates of criminal and antisocial behavior. The results of these studies have revealed that genetic factors significantly influence nearly every measureable human phenotype, with some phenotypes being highly heritable (for a comprehensive overview see Segal, 2012). For example, Tellegen and colleagues (1988) examined a wide range of psychological and behavioral phenotypes related to criminal behavior including aggression, self-control, negative emotionality, and constraint and found that all of the examined phenotypes had heritability estimates hovering around .50. Importantly, these estimates directly converge with studies which employ the more traditional twin-based BG design. Along the same lines, additional studies have revealed that MZ twins who are reared in different homes are just as similar to one another as MZ twins who are reared in the same home (Bouchard et al., 1990; Bouchard &
McGue, 1990). This finding provides further evidence suggesting that genetic influences play a pivotal role in the development of virtually all phenotypes. Importantly, these findings are free from any bias that may result from a violation of the EEA since these studies focused on samples of MZ twins and all of the twins included in the analytic models were reared in separate homes.
While the MZA design addresses some of the most notable limitations of the traditional twin-based design, the MZA design is not without its own limitations. The most notable limitation of the MZA design is the possibility that the adoptive environments that MZ twins experience are correlated. In other words, if two MZ twins are separated at birth but are placed in two separate homes that are situated in very similar neighborhoods, attend similar schools, and have very similar parents these similar environmental experiences may work to make both twins more similar to one another and result in artificially inflated heritability estimates. To further condense this argument, critiques of the MZA approach argue that unmeasured shared
30 environmental influences may be working to make twins more closely resemble one another.
There is a fundamental flaw with this particular argument: the shared environment typically accounts for a minimal amount of the overall variance in antisocial phenotypes (Mason & Frick,
1994; Miles & Carey, 1997; Rhee & Waldman, 2002; Ferguson, 2010). The evidence regarding minimal shared environmental influence is so overwhelming that leading behavior geneticists have openly questioned whether shared environmental influences even exist (see Turkheimer &
Waldron, 2000 for an overview of this argument). Regardless of these findings, another research design—the adoption study—has been developed to take this particular limitation of the MZA design into account.
2.1.3 Adoption Studies
Just as twin pairs provide a unique situation allowing for the estimation of genetic and environmental influences on a given phenotype, adopted children also present a unique situation, in which, it is possible to tease genetic and environmental influences from one another. In addition, adoption studies are not reliant upon the EEA, alleviating concerns regarding differential treatment of siblings from the same household. The most powerful aspect of the adoption design is that since adoptees do not share any genetic material with their adoptive parents, this particular design isolates the effects of environmental influences on the phenotype of interest. More specifically, adoptees, just like any other child, share 50 percent of their genetic material with their biological father and the remaining 50 percent of their genetic material with their biological mother. However, children that are adopted shortly after birth do not share the environment with their biological parents. Any similarities that are detected between biological parents and their adopted children must be directly influenced by genetic factors. While adoptive children do not share any genetic material with their adoptive parents,
31 they do share the environment, making any similarities between adoptive parents and adoptees the result of environmental influences. Based on these unique conditions, the adoption-based design is a powerful methodology that allows for the estimation of environmental effects on the phenotype of interest net of the potentially confounding effects of genetic influences (Beaver et al., 2011; Natsuaki et al., 2013; Rutter, 2006; Rutter et al., 2001).
This somewhat elementary logic has been developed into what has been referred to as “a conservative and effective approach to examining the interplay of nature and nurture” (Natsuaki et al., 2013, p. 1753) and has been used extensively by some of the leading developmental researchers (Deater-Deckard & Plomin, 1999; Moffitt, 2005a; Rutter, 2006; Rutter et al., 2001).
The basic logic of this methodology rests on the statistical comparison between adoptees’ phenotypes to the same (or closely related) phenotypes in both their biological and adoptive parents. If the adoptee more closely resembles their biological parents then genetic influences have a greater impact on the examined trait. Conversely, if the adoptee more closely resembles their adoptive parents then environmental influences have a greater impact on the examined trait.
While adoption studies are fairly uncommon in criminology, there are a few notable examples.
For example, in one of the most well-known adoption studies, Mednick and colleagues (1984) compared the prevalence of arrest between adoptees and their biological and adoptive parents.
The results indicated that adoptees who experience both genetic and environmental risk factors
(i.e., biological and adoptive parents had been arrested) had the greatest odds of being arrested.
Adoptees that experienced only genetic risk (i.e., only their biological parents had been arrested) had the second greatest chance of being arrested. This same pattern of results has been echoed in several subsequent studies also (see for example Beaver, 2011). A comprehensive review of adoption studies examining antisocial phenotypes revealed a consistent and sizable genetic effect
32 on a wide range of antisocial behaviors (Raine, 1993). Collectively, adoption studies examining antisocial phenotypes have yielded heritability estimates that are slightly lower, but not significantly different, from estimates obtained from twin studies (Beaver, 2013a).
2.1.4 Family-Based Studies
Despite the conservative nature of the adoption-based design, some critics still argue that this particular methodology suffers from damaging limitations. Most notably, critics have previously argued that adoptees (and twins) represent “special” populations and that adoptees differ from non-adoptees limiting the generalizability of the results from such studies to any larger population. These same criticisms have been leveled against twin-based studies, and subsequently resolved (see Barnes & Boutwell, 2013). Despite empirical findings which suggest that finings garnered from twin-based and adoptee-based samples can be extrapolated to larger populations, additional BG methodologies have been developed to alleviate any lingering concerns. More specifically, family-based studies extend the logic and equations used in twin- based studies to additional non-twin kin pairs. While twin studies are limited to MZ and DZ twin pairs, family-based studies may be used to include a much larger range of kinship pairs including MZ twins, DZ twins, full-siblings, half-siblings, cousins, and even nonrelated individuals reared in the same household (e.g., stepsiblings). Due to this inclusion of a much wider range of kinship pairs, the results garnered from family-based studies are more generalizable than twin-based or adoption-based studies.
In addition to greater generalizability, another attractive feature of family-based studies is the ease in which traditional twin-based BG equations can be modified to include non-twin kinship pairs. The previously presented equations (Equations 2.1 through 2.3) can be easily modified to include varying levels of genetic relatedness. Table 3.1 displays the differing levels
33 of genetic relatedness across various kinship pairs ranging from MZ twins (r = 1.00) to unrelated or stepsiblings (r = 0.00). Despite the inclusion of a wider range of kinship groups included in the equations, the underlying logic of family-based studies is very similar to the underlying logic of twin-based studies. Namely, if kinship pairs that have a greater proportion of genetic material in common (e.g., full siblings) more closely resemble one another across the phenotype of interest than kinship pairs with less genetic material in common (e.g., half siblings), the examined phenotype is influenced by genetic factors.
Table 2.1: Levels of Genetic Relatedness for Family Studies Kinship Pair Level of Genetic Relatedness (r) Monozygotic (MZ) Twins 1.00 Dizygotic (DZ) Twins 0.50 Full Siblings 0.50 Half Siblings 0.25 Cousins 0.125 Unrelated (e.g., step) Siblings 0.00
While the family-based design presents distinct advantages over alternative BG methods
(namely greater external validity), it also presents distinct limitations that must be addressed to obtain valid and reliable parameter estimates. For example, since family-based studies include non-twin kinship pairs it is all but certain that distinct differences in age and sex may exist between individuals included in a given pair. This limitation can be easily corrected statistically by regressing the phenotype of interest on each of the potential demographic confounders and preserving the residuals prior to calculating heritability, shared environmental, and nonshared environmental estimates. This procedure would effectively remove any of the variance in the
34 examined phenotype that would be attributed to differences in age or sex, leaving the remaining variance to be decomposed into the common h2, c2, and e2 components. Research has consistently revealed that heritability estimates garnered from family-based studies which effectively account for demographic differences with pairs are highly similar to heritability estimates obtained from traditional twin-based studies (Plomin et al., 2013).
2.1.5 Nonshared Environment Studies
The BG methodologies that have been discussed thus far are primarily focused on providing an unbiased and reliable estimate of genetic, shared environmental, and nonshared environmental influences. Importantly, all three of these estimates are latent in the sense that they take into account all genetic, shared environmental, and nonshared environmental influences that contribute to the examined phenotype. While these estimates are useful and directly relevant to a wide range of both theoretical and empirical issues within criminology and other disciplines, such estimates do not identify the specific genes and environments that contribute to the phenotype being examined. A logical “next step” would be to extend traditional
BG models in both a theoretical and analytical sense to identify the specific genetic and environmental influences that contribute to particular phenotypes. In an effort to provide a more detailed description of the specific genes and environments that comprise the latent estimates of h2, c2, and e2 two additional research designs have been developed. Molecular genetic studies are aimed at identifying the specific genes that comprise the h2 estimate, while nonshared environment studies have been developed to identify the specific environmental influences which compromise e2.3 In order to isolate specific aspects of the nonshared environment, it is
3 A developing line of literature has focused on the identification of specific environments that comprise c2 (Baker, Maes, & Kendler, 2012; Walden, McGue, Iacono, Burt, & Elkins, 2004). However, the studies comprising this line of research are limited in number and have only examined a limited number of phenotypes. This limited attention in the extant literature is likely a direct result of the limited influence of the shared environment on most phenotypes. 35 first necessary to calculate the latent genetic and shared environmental influences on the phenotype of interest. This particular aspect of nonshared environment studies provides a distinct and considerable advantage over traditional SSSMs in that any statistically significant coefficient can be interpreted as an effect of the specific nonshared environmental influence on the examined phenotype net of the effect of genetic and shared environmental influences
(Beaver, 2013a; Plomin & Daniels, 1987; Turkheimer & Waldron, 2000). In this way, coefficients estimated in nonshared environment studies are immune to some of the most concerning methodological shortcomings of SSSMs such as genetic confounding. In addition, since shared environmental influences are modeled as latent factors, the probability of omitting some other potentially confounding environmental influence from the model is also greatly diminished. Taken together, nonshared environment studies make use of some of the most powerful and conservative modeling strategies currently available in the social sciences.
Perhaps the most common modeling strategy used in nonshared environment studies is the MZ difference score approach (Asbury, Dunn, Pike, Plomin, 2003; Burt, McGue, Iacono, &
Krueger, 2006; Pike, Reiss, Hetherington, & Plomin, 1996). While this particular approach is fairly straightforward, it is widely considered the “gold standard” for identifying measured nonshared environmental influences (Beaver, 2008) and has been regarded as one of the most rigorous quasi-experimental designs available (Vitaro, Brendgen, & Arsenault, 2009). The MZ difference score approach is situated on the observation that MZ twins share 100 percent of their genetic material. Since MZ twins are genetically identical to one another, any observed difference between them cannot be the result of genetic influences and must be due to nonshared environmental influences. In addition, since MZ twins are reared in the same household by the
Based on the limited nature of the existing literature, and the limited explanatory power of shared environmental influences, a more detailed explanation of the methodologies used to identify specific shared environments will not be elaborated upon. 36 same parents, the MZ difference score approach also takes shared environmental influences into account. In this way, the MZ difference score approach effectively controls for genetic and shared environmental influences and isolates observed nonshared environmental influences on the examined phenotype. Since the MZ difference score approach does not compare MZ and DZ twins (as in traditional twin studies), a violation of the EEA is not a concern (Rovine, 1994).
The MZ difference score approach is carried out by subtracting one twin’s score on a given measure from their co-twin’s score on the same measure. This same procedure is also carried out for the outcome measure of interest. For example, in order to examine whether differences in maternal attachment predict differences in delinquency during adolescence, one twin’s score on the maternal attachment measure would be subtracted from their co-twin’s score on the same maternal attachment measure (this procedure would be repeated for the delinquency measure). After both difference scores have been calculated they can then be inserted into a statistical equation and making it possible to test whether the observed nonshared environmental influence (e.g., differences in maternal attachment) has a significant effect on differences in the examined outcome (e.g., adolescent delinquency).
2.2 Behavior Genetic Research and Antisocial Behavior
Despite the overall dearth of studies which employ genetically research designs within the leading criminology journals, other disciplines have produced an overwhelming number of studies which directly examine both genetic and environmental influences on antisocial behavior.
In addition, the majority of such studies were published several decades ago, indicating that information regarding the genetic and environmental contributions to antisocial behavior has been readily available for quite some time. In a comprehensive review of the potential benefits of integrating behavior genetic methods into developmental research that is now nearly a decade
37 old, Moffitt (2005a) notes that “[m]ore than 100 studies have addressed the question of genetic influence on antisocial behavior” (p. 535). In the years since Moffitt’s review was published, there are surely dozens of additional studies that have been published. Based on the vast number of studies that have examined the genetic and environmental contributions to antisocial behavior, this section will provide a general overview of this line of research as opposed to a comprehensive review. This decision was made for two distinct reasons. First, comprehensive reviews of this line of research have been completed relatively recently (see Moffitt, 2005b), making and additional review redundant. Second, four meta-analyses reviewing this literature are currently available (Mason & Frick, 1994; Miles & Carey, 1997; Rhee & Waldman, 2002;
Ferguson, 2010), which again makes a more comprehensive review redundant. Based on the breadth and robustness of findings that accompany meta-analysis (Cooper, 2010), the majority of this section will be devoted to summarizing the content and findings of these four meta-analyses.
2.2.1 Meta-Analyses of Behavior Genetic Research on Antisocial Behavior
Meta-analysis is an analytic technique which allows for the empirical accumulation of a body of studies which may differ in various aspects, but are all aimed at addressing a similar topic of interest. More specifically, while studies that comprise a given body of research may employ different measurement strategies, analyze different samples, and employ differing analytic techniques, such studies are all aimed at gaining a better understanding of a given relationship or topic. In such situations, conflicting results may, and oftentimes do, arise making it difficult to determine whether any consensus among the findings of the studies can be reached.
Additionally, it becomes more and more difficult to determine the underlying source of conflicting findings in such instances. For example, studies with conflicting results may employ methodologies which differ in various ways and may employ dramatically different analytic
38 samples making in nearly impossible to determine whether any detected difference is a result of methodological differences between the studies or actual discrepancies in findings.
Meta-analyses are invaluable in situations wherein a voluminous literature has been produced and provide at least three important advantages. First, meta-analyses perform a systematic review of the literature and an empirical summary of the findings gleaned from such a review. In this way, it becomes possible to synthesize the findings from an overwhelming number of studies down into a single set of findings, referred to as the average effect size, which provide an overall representation of all of the individual studies in a given area of research.
Second, meta-analyses allow researchers to detect the signal through the noise in that this procedure allows for the identification of the precise methodological characteristics that may contribute to discrepancies in findings between studies. This particular characteristic of studies becomes highly valuable when attempting to draw a consensus from a body of research that is wrought with conflicting findings. In addition, meta-analyses allow for the weighting of studies with superior or rigorous methodologies which can increase the accuracy of the average effect size and provide meaningful suggestions for future research. Third, meta-analyses make it possible to detect small effect sizes and inconsistent effects. Since meta-analyses make use of studies as the primary unit of analysis (as opposed to data points within a sample), the average effect size is the result of all of the individual data points from each of the analyzed studies which results in a dramatic increase in statistical power. For example, the results of an individual study may reveal that the association between two variables is nonsignificant; however, when the results from this single study are combined with many others what appears to be an inconsistent association between the two variables may actually result in a relatively consistent association. Despite these distinct advantages of meta-analyses, such studies should
39 not be seen as a panacea. Meta-analyses are only as good as the studies that they summarize. In this way, a body of research which consistently employs careless methodologies and reports inaccurate findings will result in a meta-analysis which produces biased findings as well.
As mentioned previously, there are currently four meta-analyses which examine genetic and environmental influences on antisocial behavior (Mason & Frick, 1994; Miles & Carey,
1997; Ferguson, 2010; Rhee & Waldman, 2002). These meta-analyses synthesize findings from studies which employ various behavior genetic research methodologies to estimate the proportion of variance in antisocial behavior that be explained by genetic (h2), shared environmental (c2) and nonshared environmental (e2) influences. In this way, the average effect size of interest in each of these meta-analyses refers to each of the three components that are estimated in a behavior genetic study, h2, c2, and e2. In this way, each meta-analysis will provide an average effect size indicating the average influence of genetic influences, the shared environment, and the nonshared environment on antisocial behavior across all of the individual studies analyzed.
Prior to discussing each meta-analysis in more detail, there are two additional characteristics of these four studies that should be considered. First, the amount of time that passed between each of the studies indicates that each subsequent meta-analysis almost certainly contains novel findings that were not included in the studies that preceded it. This is an important characteristic of these studies in that the findings collectively represent a sizable segment of the literature that spans a significant period of time. In addition, due to the limited overlap in the studies included in each meta-analysis, an overall concordance in findings would seem to indicate that the overall pattern of results is highly robust. Second, the procedures used in each study were subsequently refined based on criticisms and limitations of each prior study,
40 indicating that each subsequent meta-analysis was more rigorous than the previous one.
Collectively, these characteristics likely increase the validity of the findings, particularly in the most recent studies.
Mason and Frick (1994). The first meta-analysis to synthesize results from studies examining the influences of genetic and environmental influences on antisocial behavior was performed by Mason and Frick (1994). As with any meta-analysis, Mason and Frick developed and specified explicit criteria regarding the characteristics of studies that would be included in their meta-analysis. Studies published between 1975 and 1991 which either compared MZ and
DZ twins in a classic twin-based study or analyzed a sample of adoptees (using the adoption- based methodology specified above) were considered for inclusion in the study. In regard to antisocial behavior, only studies which included various externalizing behavioral issues related to antisocial behavior (such as aggression, antisocial personality, or criminal behavior) were considered. Importantly, the authors decided to exclude studies which focused solely on substance abuse, but included studies that used substance abuse as a singular indicator of antisocial behavior more broadly. In addition, Mason and Frick (1994) only included studies in which all respondents were compared across the same measure of antisocial behavior.4 Finally, the authors excluded any studies with overlapping analytic samples and any studies in which the results could not be used to estimate an average effect size. Based on these inclusion criteria, the final sample consisted of 15 studies which included over 4,000 twin and sibling pairs.
The results of the meta-analysis revealed that the average heritability estimate across all
15 of the studies included in the analysis was .48, indicating that approximately 48 percent of the variance in antisocial behavior was explained by genetic influences. In a series of sensitivity
4 This is particularly important in regard to adoption studies, as the same measure of antisocial behavior would have to be collected for adoptees, biological parents, and adoptive parents. 41 tests, the authors also examined whether their results differed based on three different criteria.
First, the authors compared genetic influences on severe antisocial behavior in relation to nonsevere antisocial behavior. The results revealed that the heritability coefficient for severe antisocial behavior (h2 = .45) was greater than the heritability coefficient for nonsevere antisocial behavior (h2 = .00). Second, the authors examined whether demographic characteristics such as age, sex, or country of origin significantly moderated the average heritability effect size. The results indicated that these various characteristics did not moderate the results. Third, the authors examined whether the population from which the analytic sample was drawn significantly moderated the resulting heritability coefficient. The results indicated that studies which analyze samples drawn from clinical populations tend to report greater h2 estimates (h2 = .53) than studies that analyze samples drawn from the general population (h2 = .20).
Miles and Carey (1997). The second meta-analysis examining genetic and environmental influences on antisocial behavior was performed by Miles and Carey (1997). The authors limited their analysis to studies which included a measure of aggression, hostility, antisocial behavior, or juvenile delinquency. In concordance with Mason & Frick (1994), Miles and Carey
(1997) also included both twin and adoption studies in their meta-analysis. These inclusion criteria resulted in a total of 24 effect sizes gleaned from 17 studies (n = 15 twin studies and n =
2 adoption studies). The results of the meta-analysis produced an average heritability coefficient of .50, indicating that approximately 50 percent of the variance in antisocial behavior can be explained by genetic influences. As in the previous study, the authors also performed a sensitivity analysis in an effort to identify potential study characteristics which may moderate heritability estimates. The results revealed that the sex of respondents significantly moderated heritability coefficients, wherein samples comprised of males produced greater estimates of h2
42 relative to samples of females. Similarly, samples with older respondents resulted in greater heritability estimates relative to samples of younger respondents. The authors also examined whether different measurement strategies (i.e., observational measures vs. self-report measures) moderated heritability estimates. The results revealed that self-report measures of antisocial behavior produced greater estimates of h2 relative to observational measures of antisocial behavior. While estimates of h2 varied as a function of these three factors (age, sex, and measurement strategy), it is important to note that h2 estimates remained at least moderate in size across all models and ranged between .22 and .78.
Rhee and Waldman (2002). The third meta-analysis to examine genetic and environmental influences on antisocial behavior was conducted by Rhee and Waldman in 2002.
Based on the explosion of behavior genetic research focusing on antisocial behavior since the previous meta-analytic review, the authors had a larger sample of studies to choose from and were able to utilize more stringent selection criteria. The authors attempted to limit the included studies to those which included a direct measure of antisocial behavior and excluded studies which analyzed proxy measures of antisocial behavior like hostility. For example, included studies had to operationalize antisocial behavior as a clinical diagnosis, criminal behavior as measured by official records, or observations/self-reports of aggression. Studies which analyzed overlapping or nonindependent samples or reported incomplete findings were excluded from the final analytic sample. These selection criteria resulted in a total of 42 twin studies and 10 adoption studies which included over 55,000 pairs of respondents in total.
The results indicated that 41 percent of the variance in antisocial behavior was the result of genetic influences, shared environmental influences explained 16 percent of the variance, and the nonshared environment explained the remaining 43 percent of the variance. As in the
43 previous meta-analysis, Rhee and Waldman (2002) also performed a series of sensitivity analyses aimed at identifying factors which may moderate genetic influences. More specifically, the authors examined the potentially moderating role of age, sex, assessment method, zygosity determination, and concept operationalization. While heritability estimates did differ based on the technique used to determine zygosity, the difference between measured obtained from blood testing (h2 = .47) versus self-report (h2 = .43) were markedly similar. The results also revealed that self-report measures of antisocial behavior, relative to observational measures, result in smaller heritability estimates. In addition, older samples tend to result in smaller heritability estimates relative to younger samples.
Ferguson (2010). The most recent meta-analysis examining the genetic and environmental influences on antisocial behavior was conducted by Ferguson (2010). This particular meta-analysis was distinct from those previously presented in at least two ways. First,
Ferguson (2010) focused on more recent research and included studies that were published between 1996 and 2006. This particular feature of this meta-analysis is critical in that more recent studies are expected to be of higher empirical rigor in that they are more likely to employ more advanced statistical methods, draw from modern samples, and directly address limitation of studies previously published. Second, Ferguson (2010) directly attempts to bridge “the chasm between evolutionary psychology and behavioral genetics” by examining more closely the proximate (i.e., genetic influences) and ultimate causes (i.e., natural selection) of antisocial behavior (p. 163). This is another important contribution to the literature since behavior genetics and evolutionary psychology5 are directly compatible and can collectively contribute to a more
5 Evolutionary psychology refers to a field of study concerned with the application of evolutionary theory to human behavior (Barkow, 2006; Mealey, 2000). More specifically, this particular paradigm focuses on the specific processes and adaptations, operating through natural and sexual selection, which directly contribute to genetic variation, and in turn, phenotypic variation (for a more detailed overview see Wright, 1994). Importantly, 44 comprehensive understanding of the underlying etiology of nearly any human phenotype, including antisocial behavior. Despite these common interests and overlapping goals, both disciplines tend to distance themselves from one another. In addition to these two distinct contributions to the extant literature, Ferguson (2010) also employed stringent inclusion criteria in an effort to perform the most methodologically rigorous meta-analysis examining the genetic and environmental contributions to antisocial behavior to date. For example, Ferguson (2010) used the same stringent definition of antisocial behavior employed by Rhee and Waldman
(2002). In addition, only studies which employed twin-based and adoption research designs were included in the meta-analysis. These criteria ultimately resulted in a final analytic sample of 53 separate effects nested within 38 studies. Collectively, these studies analyzed nearly
100,000 twin pairs/adoptees (N = 96,918). Similar to previous meta-analyses, Ferguson (2010) also examined the potential moderating effects of sex, age, and the measurement strategy used to operationalize antisocial behavior (self-report vs. observational measures).
The results indicated that the average heritability of antisocial behavior across all of the examined studies was .56, indicating that 56 percent of the overall variance in antisocial behavior was explained by genetic influences. While genetic influences accounted for the largest proportion of the overall variance in antisocial behavior, shared environmental influences accounted for approximately 11 percent of the variance, and the remaining 31 percent of the variance was explained by nonshared environmental influences. Further, the results indicated that standardized measures of antisocial behavior (i.e., antisocial personality disorder as defined
evolutionary psychology is not a theory of individual-level behavior (Daly & Wilson, 1988), but rather a separate paradigm aimed at situating the explanations of human phenotypes within the principles of biological evolution. While evolutionary psychology can be considered a stand-alone paradigm or perspective, Barnes (2014) has previously suggested that the biosocial perspective encompasses several other perspectives including evolutionary psychology. In this way, evolutionary psychology can be viewed as one of the many explanatory tools that the biosocial perspective brings to the table all in the interest of providing a more comprehensive explanation of phenotypic variation. 45 by the DSM-IV) yield lower heritability estimates relative to studies which employ broader measures of antisocial behavior. The operationalization of antisocial behavior was also found to significantly moderate nonshared environmental influences, but not shared environmental influences. As in the previous meta-analysis performed by Rhee and Waldman (2002), the results also indicated that age significantly moderated genetic influences wherein samples comprised of older respondents tended to yield smaller heritability coefficients relative to samples of younger respondents. Relatedly, age was also found to significantly moderate nonshared environmental influences wherein older samples yielded significantly higher estimates of e2 relative to younger samples. While age did not significantly moderate the shared environment, such influences did trend downwardly across various age categories. Sex did not significantly moderate genetic, shared environmental, or nonshared environmental influences.
2.2.2 Discussion of Findings
Taken together, the results of these four meta-analyses produce at least three key findings. First, all four of the studies indicate that approximately half of the overall variance in antisocial behavior (Mason & Frick, 1994: h2 = .48; Miles & Carey, 1997: h2 = .50; Rhee &
Waldman, 2002: h2 = .41; Ferguson, 2010: h2 = .56) is the direct result of additive genetic influences. Second, nonshared environmental influences explained the second greatest proportion of the overall variance in each study. Third, the shared environment accounted for the smallest portion of the overall variance in each study. Importantly, each of these three findings directly coincide with Turkheimer’s (2000) three laws of behavior genetics. Prior to broadly contextualizing these findings within current criminological research, it is first important to reiterate a few points regarding each of the meta-analyses reviewed here. As previously mentioned, all four of these meta-analyses were conducted over an extended period of time and
46 therefore have only a limited amount of overlap in regard to the studies and effects they synthesize. In addition, each meta-analysis makes a unique contribution to the literature and effectively builds on the previous one in terms of methodological rigor. Finally, the collective findings from these meta-analyses span several decades (1975 to 2006). In this way, a consensus in findings among these studies provides a strong indication that the collective findings are highly robust.
Based on these considerations, the findings gleaned from these four studies are directly relevant to the field of criminology in at least three ways. First, the overall findings indicate that, by any reasonable standards, genes and environments significantly influence antisocial behavior.
In other words, the overall pattern of results favor a biosocial explanation of antisocial behavior, which in turn, indicates that a purely biological or sociological explanation of antisocial behavior is highly unlikely and fails to comport with a large segment of empirically-rigorous research spanning several decades. Based on these findings, the most advantageous and efficient approach to understanding the underlying etiology of antisocial and criminal behaviors would be to employ a biosocial approach.
While the studies, and the discussion of them presented above, tend to provide a heavy emphasis on the genetic contributions to antisocial behavior, this is simply a result of the overall scarcity of research and theoretical development devoted to the genetic influences on such behaviors in the field of criminology and other social science disciplines. In addition, genetic influences consistently accounted for the largest proportion of the overall variance in antisocial behavior in each of the studies. As outlined previously, most statistical models (which include multiple independent variables) presented in criminological studies typically only account for approximately 10 to 20 percent of the overall variance in the examined outcome (Weisburd &
47
Piquero, 2008). In light of these findings, it is difficult to imagine that a single variable—genetic influences—can account for 50 percent of the overall variance in such outcomes.6 Currently, no other variable known to criminologists can effectively and consistently account for one-half of the overall variance in antisocial or criminal behaviors. As a comparison, the results of the now classic meta-analysis performed by Pratt and Cullen (2000) revealed that self-control, widely considered one of the most consistent and “strongest correlates of crime” (p. 952), resulted in an average effect size that was consistently greater than .20 but less than .30 (range: .213-.278).
Based on these findings, the overall emphasis on the genetic influences on antisocial behavior becomes a bit more understandable; however, as its name implies, the biosocial perspective is not solely concerned with biological or genetic influences. While genetic influences account for approximately one-half of the overall variance in antisocial behavior, the remaining half of the variance is explained by environmental influences. Based on these findings, failing to account for either set of influences would result in suboptimal theories and statistical models which are, even under the most ideal situation, only able to account for a fraction of the overall variance in the outcomes of interest. In this way, thoroughly accounting for both sets of influences would ultimately result in more effective theories and statistical models.
Second, the results also indicate that failing to employ genetically-sensitive research designs may result in model misspecification and any resulting significant associations may be spurious due to genetic confounding (Johnson et al., 2009; McGue et al., 2010). Put differently, models which do not effectively account for genetic influences on the outcome of interest cannot effectively isolate the association between the independent variable and the dependent variable.
This particular finding, coupled with the overall lack of genetically-informed research within the
6 It is important to reiterate the explanation of the h2 estimate garnered from behavior genetic studies. While the h2 coefficient represents a single latent variable, this variable is directly informed by the complete set of additive influences from all genes within the entire genome. 48 field of criminology, raises serious concerns regarding the empirical status of the vast majority of theoretical perspectives within the field of criminology. Since such theoretical perspectives are built upon empirical foundations that are potentially wrought with spurious associations, the potential ramifications for these theories are not currently clear. This is primarily a result of the overall lack of research within the field of criminology directly aimed at integrating such theories within the biosocial perspective. While more research is needed before direct theoretical suggestions can be offered, there are potentially serious implications that may arise out of such methodological limitations.
The third way in which the findings gleaned from these four meta-analyses have direct application to criminological research is in regard to the importance of differentiating between shared and nonshared environmental influences. More specifically, the results of the four meta- analyses collectively indicate that nonshared environmental influences have far more influence on antisocial behavior relative to shared environmental influences. This particular finding indicates that the environmental influences that criminological theory and research should focus on are nonshared environmental influences. In this way, theories and research that provide an increased focus on nonshared environments (relative to shared environments) would likely result in greater explanatory power and an overall better understanding of the underlying etiology of antisocial phenotypes. In addition, this finding also reiterates the importance of the further integration of genetically-sensitive research designs into the field of criminology and other social sciences, since such methodologies present the only feasible analytic strategy allowing for the partitioning of shared environmental influences from nonshared environmental influences.
49
2.3 Gene-Environment Interplay
The study designs and research discussed above primarily focuses on decomposing the variance in a given measure into thee latent components: h2, c2, and e2. Each of these three components has been described in detail above. While a limited discussion was devoted to the different ways in which genetic effects may interact with one another (i.e., epistasis and dominance), the majority of the discussion regarding genetic and environmental influences on antisocial behavior has facetiously implied that these two sets of influences exclusively work in isolation from one another. In actuality, this is an oversimplification of reality since genetic and environmental influences frequently work in tandem to influence a given phenotype. According to Anthony Walsh (2009), “[g]enes and environments are separate entities as are hydrogen and oxygen, and can be analyzed as such, but when they have worked their magic and produced a phenotype they are not more separable than hydrogen and oxygen are when talking about the water they produce” (p. 33). To further extend this metaphor, Walsh (2011) states that “any trait or behavior of any living thing is always the result of biological factors interacting with environmental factors” (emphasis in original, p. xi). In other words, examining the independent influences of genes and the environment is an empirical endeavor that will likely result in theories and statistical models that oversimplify reality and human behavior. An alternative would be to examine how genetic and environmental influences work together in the development of antisocial behavior. This particular line of reasoning has come to be the prevailing consensus among biosocial criminologists, who now recognize the importance of specifying the processes in which genetic and environmental influences work together collectively to influence behavioral phenotypes. This culmination of genetic and environmental influences has come to be known as gene-environment interplay. The remainder of this chapter will be devoted to describing the two most common forms of gene-environment interplay—gene 50
X environment interactions (G × E) and gene-environment correlation (rGE)—and the research which applies these concepts to antisocial behavior.
2.3.1 Gene-Environment Interaction (G × E)
Criminological theories, particularly theories that focus exclusively on macro-level socializing agents (e.g., social disorganization theory), typically propose that various aspects of environment in which an individual is reared in have measurable and distinct influences on subsequent behavioral outcomes. For example, Sampson and colleagues (1997) argue neighborhoods with overall lower levels of collective efficacy are more likely to experience disproportionately high levels of violence, even after controlling for other social and organizational characteristics of neighborhoods. While studies have found support for the theoretical concept of collective efficacy (Morenoff, Sampson, & Raudenbush, 2001; Sampson,
Raudenbush, & Earls, 1997; Sampson Morenoff, Earls, 1999), this seemingly powerful theoretical concept still fails to explain the overall low prevalence of criminal behavior among individuals who reside in neighborhoods with exceedingly high levels of disadvantage, disorganization, and inefficacy.7 This observation serves as a powerful example of the wide range of heterogeneous responses of individuals who are exposed to a similar set of environmental influences.
The concept of G × E can provide a better understanding of the overall variability in response to highly similar environmental influences. Simply put, each individual has a unique composition of genes (typically referred to as a “genome”) which result in differential genetic predisposition toward certain phenotypes (as evidenced in the previous section). This genetic
7 Interestingly, Sampson and collegues (1997) do not offer descriptive statistics for their measures of violence, victimization, and homicide rates in their study. However, additional analysis of the Project on Human Development in Chicago Neighborhoods (PHDCN) data indicate that the majority of youth included in the sample (55.5%) refrained from any delinquent behavior over the study period, and an even greater proportion of youth refrained from any violent criminal behavior (67%) during the same time period (Fagan & Wright, 2011). 51 predisposition toward a given phenotype may also differ from one person to another based on differential exposure to environmental risk factors. The reverse holds true also: the influence of environmental risk factors on a given phenotype may vary as a function of different levels of genetic predisposition. In this way, a G × E may be interpreted much in the same way as a traditional multiplicative interaction term in SSSMs, the effect of one independent variable on the dependent variable is moderated by another independent variable. In other words, individuals who are most susceptible to certain phenotypes are those who exhibit the highest levels of genetic predisposition toward such phenotypes and experience a higher concentration of environmental influences that have been found to influence such phenotypes.
An example will likely provide a further clarification of the concept of a G × E. In this example, suppose that the results of a BG model examining violent criminal behavior identifies two distinct groups within a sample, one with a relatively high h2 estimate (h2 = .70) and the other with a lower h2 estimate (h2 = .30). In addition, further analyses reveal that levels of parental maltreatment during childhood significantly moderate genetic influences on violent behavior. The results of this hypothetical example are presented in Figure 3.1. The predicted probability of engaging in violent behavior is plotted on the y-axis and the x-axis presents the level of parental maltreatment each respondent experienced during childhood. The solid line represents respondents with overall higher levels of genetic risk (h2 = .70) and the dashed line represents respondents with overall lower levels of genetic risk (h2 = .30). Three specific patterns in the resulting figure warrant additional attention. First, individuals with higher levels of genetic risk seem to be at a higher risk for violent behavior than individuals with lower levels of genetic risk. In other words, across all levels of parental maltreatment, individuals with higher levels of genetic risk had a higher predicted probability of engaging in violent criminal behavior.
52
Second, parental maltreatment also seems to have a significant effect on the probability of engaging in violent behavior. In both groups (high vs. low genetic risk), the probability of engaging in violent behavior increased as levels of parental maltreatment increased. Third, while parental maltreatment has a significant effect on violent behavior across both sets of respondents, respondents with higher levels of genetic risk had a much steeper slope relative to individuals with lower levels of genetic risk. This final pattern is highly indicative of a G × E, in that we see independent effects from the genetic risk and the environmental risk measures, but the greatest effect occurs when both sets of factors are present. Keep in mind that individuals who experienced only high environmental risk or only high genetic risk are far less likely than individuals who experience both sets of risk factors to engage in violent behaviors. This is similar to the heterogeneity in behavioral response to purely social risk factors observed in the extant literature discussed above.
2.3.2 Gene-Environment Correlation (rGE)
The second type of gene-environment interplay is referred to as gene-environment correlation (rGE). Where G × Es illustrate the process in which genetic and environmental processes interact with one another to produce phenotype variance, rGEs highlight the processes wherein genetic influences can produce variation in exposure to different environments (Beaver,
2013a). The idea that an environment may be influenced by one’s genome may be somewhat difficult to grasp when viewed through the traditional criminological lens. More specifically, virtually all criminological theories and research assume that the causal arrow is drawn from the socializing agent to the individual, not the other way around. In addition, environmental influences are typically viewed a “purely social” and are free from genetic influence since
53 environments do not possess genomes and are not under genetic influence. While the idea that environments may actually be influenced by genetic influences may seem to be implausible on its face, there is a significant amount of empirical research supporting this claim (Jaffee & Price,
2007; Kendler & Baker, 2007; Scarr & McCartney, 1983). A line of research has revealed that individual-level variation in exposure to various environments is directly and significantly influenced by genetic factors. In one of the first and most comprehensive conceptualizations of rGEs, Scarr and McCartney (1983) reiterated this point directly when they observed that “the genome, in both its species specificity and its individual variability, largely determines environmental effects on development, because the genotype determines the organism’s responsiveness to environmental opportunities” (p. 424). In an effort to provide a better understanding of the underlying logic of rGEs, it may be beneficial to provide an explanation of the three types of rGEs that have been identified in previous studies: passive rGEs, active rGEs, and evocative rGEs.
The first type of rGE is referred to as a passive rGEs and is a direct result of parents providing their offspring with both their genotype and environment. Since both sets of influences emanate from a common source (i.e., parents), they will be positively correlated with one another. For example, a child with two parents who are both highly aggressive and antisocial is at an increased risk for being raised in a physically abusive or neglectful environment. In addition, this child’s parents will also pass whatever genetic predispositions toward aggression and antisocial behaviors they possess. In this way, the child’s environment
(i.e., abuse and neglect) is positively correlated with their genome (which includes a genetic predisposition toward aggressive and antisocial behavior).
54
The second type of rGE is referred to as an evocative rGE and is the direct result of one’s genome eliciting responses from an environment which are positively correlated with genetic predisposition. The concept of genes eliciting responses from the environment may seem inherently unintuitive initially since genes are not active organisms that earnestly work toward some particular goal (the sole purpose of any given gene is far more mundane: to produce proteins). However, the concept of an evocative rGE becomes far more clear when the processes which mediate the relationship between the genome and environmental responses are specified more directly. Genes to not directly elicit responses from the environment, instead, genes influence various phenotypes (e.g., personality and behavioral traits) which, in turn, elicit different responses from the environment. For example, children who display high levels of antisocial behavior, a phenotype that is directly influenced by genetic factors, may elicit more frequent and harsh punishment from their parents. In this particular example, the phenotype of antisocial behavior is influenced by genes, which ultimately are correlated with the elicited parental response of more frequent and harsh punishment.
The third and final type of rGE is referred to as an active rGE and takes into account the important role that the genome plays in the selection into specific environments that allow for optimal genetic expression. In other words, individuals will seek out specific environments (and actively avoid other environments) which most consistently align with their individual-level genetic predispositions (Plomin et al., 2013). Directly in line with the first law of behavior genetics (Turkheimer, 2000), it can be expected that selection into various environments, so called “niche-seeking,” is at least partially influenced by genetic factors. For example, more athletically-inclined youth are typically drawn to more physical activities and spend more time engaging in such activities relative to youth who are less athletically-inclined.
55
2.3.3 Gene-Environment Interplay and Antisocial Behavior
Now that the basic principles of gene-environment interplay have been introduced, it is possible to discuss the application of such concepts to the study of antisocial behavior. While the overall number of criminological studies employing these techniques and concepts may be lacking, there has been a significant amount of research outside of criminology which indirectly ties gene-environment interplay to theoretical concepts that are highly relevant within criminology. The purpose of this section is only to provide a cursory overview of this line of research, since Chapter 4 will provide a more detailed discussion of the extant literature linking the biosocial perspective to theoretical concepts pertinent within the field of criminology.
Early studies attempting to examine G × Es on antisocial phenotypes utilized adoption samples due to the ability to model both genetic and environmental influences on antisocial behavior (Cadoret, Cain, & Crowe, 1983; Cadoret, Yates, Troughton, Woodworth, & Stewart,
1995; Mednick et al., 1984). These studies operationalize genetic risk as each adoptee’s biological parents score on the antisocial behavior measure and environmental risk as each adoptee’s adoptive parents score on the same antisocial behavior measure. The results of these studies have provided preliminary support the presence of a G × E in the development of antisocial behavior, wherein adoptees who have a biological parent and an adoptive parent who possess high scores on the employed measure of antisocial behavior are most likely to display higher levels of antisocial behavior themselves. More recently, studies have extended the traditional BG modeling strategies in an attempt to determine whether genetic influences are significantly moderated by specific environmental contexts (Carlson, Mendle, & Harden, 2014).
The results of these studies have revealed specific environmental contexts that significantly moderate genetic influences on antisocial behavior such as classroom context (Lamb,
Middledorp, Van Beijsterveldt, & Boomsma, 2012), pubertal timing (Harden & Mendle, 2012), 56 preschool attendance (Tucker-Drob & Harden, 2012), and parental maltreatment (Jaffee, Caspi,
Moffitt, Dodge, Rutter, Taylor, & Tully, 2005).
The main drawback in utilizing a BG modeling strategy to study whether specific environments significantly moderate genetic influences on a given phenotype is that such methodologies model genetic influences as a latent factor and do not provide information regarding the specific genes involved in the G × E. Over the past decade, a developing line of research, typically referred to as molecular genetic research, has focused on identifying specific genes and environments that interact to directly contribute to the underlying etiology of antisocial behavior. Briefly, molecular genetics8 is a field of study which is focused on examining how genetic variation contributes to various phenotypes including antisocial behavior.
More specifically, a small percentage of genes, between 1 and 10 percent, can differ between individuals within the general population (Frazer, Murray, Schork, & Topol, 2009). These variations are referred to as alleles or polymorphisms and, at least in some cases, have been found to directly contribute to different phenotypes. A significant number of genes have been theoretically and empirically linked to antisocial phenotypes, with the majority of these polymorphisms linked to the process of neurotransmission9 and the specific neurotransmitters dopamine and serotonin (for an overview see Morley & Hall, 2003). In addition, polymorphisms implicated in the production of enzymes that metabolize neurotransmitters from the synapse after
8 This discussion of molecular genetics is highly cursory since this dissertation will not employ a molecular genetic research design. However, findings from studies employing a molecular genetic research design are directly relevant to the overall focus of the current study (i.e., situating existing theoretical perspectives within the biosocial perspective), particularly in regard to extending the concept of G × Es to phenotypes related to antisocial behavior. For a more detailed overview of molecular genetics and their application to criminological research see Beaver (2009; 2013a). 9 Neurotransmission refers to the electrical and chemical processes in which neurons (nerve cells in the brain) communicate with one another. Neurons are separated by a small gap, referred to as the synapse, requiring the use of chemical messengers (referred to as neurotransmitters) to bridge the synapse and pass the message to an adjacent neuron. After the process of neurotransmission, excess neurotransmitters remain in the synapse and must be removed. One of the primary ways excess neurotransmitters are removed from the synapse is through the release of a specialized protein called an enzyme which effectively degrades the excess neurotransmitters. 57 neurotransmission have also been linked to antisocial phenotypes (Beaver, DeLisi, Vaughn, &
Barnes, 2010; Brunner et al., 1993; Volvaka, Bilder, & Nolan, 2004; Rujescu, Giegling, Gietl,
Hartmann, & Moller, 2003).
Molecular genetic research offers the unique opportunity to examine the potential association between a measured genetic polymorphism and a given antisocial phenotype, but also to examine whether measured environmental influences significantly moderate such associations. The majority of the studies examining whether G × Es with specific genetic polymorphisms significantly contribute to the development of antisocial phenotypes have focused on the enzymatic breakdown gene monoamine oxidase A (MAOA; Caspi et al., 2002;
Foley et al., 2004; Haberstick et al., 2005; Schwartz & Beaver, 2011; Widom & Brzustowicz,
2006). Perhaps the most commonly examined moderating environmental influence examined in these studies is parental maltreatment, which the results of a meta-analysis have indicated significantly moderates the effect of MAOA on antisocial behavior (Kim-Cohen et al., 2006).
Additional combinations of measured polymorphisms and environments have also been found to significantly contribute to antisocial behaviors. For example, the influence of the dopamine transporter gene (DAT1) has been found to be significantly moderated by exposure to delinquent peers (Vaughn, DeLisi, Beaver, & Wright, 2009); family risk has been found to significantly interact with a dopamine receptor gene (DRD2) to produce early-onset of offending (DeLisi,
Beaver, Wright, & Vaughn, 2008); the serotonergic transporter gene (5-HTTLPR) has been found to significantly interact with adverse childhood events to produce antisocial personality disorder (Douglas et al., 2011).
A significant number of BG studies have examined rGEs on a wide range of environments, many of which are directly related to theoretical perspectives that are highly
58 influential within the field of criminology. While there are far too many individual studies examining the heritability of specific environments directly relevant to criminological theories, a recent meta-analysis by Kendler and Baker (2007) provides a thorough overview of the findings that can be culled from this line of research. This meta-analysis included a total of 55 studies which directly examined the genetic contributions to a wide variety of environments.
Importantly, four of the general environmental influences that Kendler and Baker examined are directly related to theoretical perspectives within criminology, including those that will be examined in the current study: parenting practices, family environment, peer interactions, and stressful life events.
First, the meta-analysis included 19 studies which examined various aspects of the parent-child relationship including: maternal/paternal warmth, maternal/paternal negativity, maternal/paternal control, and maternal/paternal monitoring. Importantly, as will be discussed in detail in the following chapter, parental socialization plays a central role in several influential criminological theories including social bonding theory (Hirschi, 1969) and Gottfredson and
Hirschi’s (1990) general theory of crime. The results of the meta-analysis revealed that environments related to the parent-child relationship are under moderate levels of genetic influence, with heritability estimates ranging between .12 and .37. Second, Kendler and Baker also included studies that examined other family environments that extended beyond parenting behaviors including family cohesion, family conflict, and family control. The results of the meta-analysis revealed that family environments are also moderately influenced by genetic influences with average heritability estimates ranging between .18 and .30.
Third, six studies examining the heritability of peer interactions were also included in the meta-analysis. Peer interactions also play a prominent role in various explanations of criminal
59 behavior, with perhaps the most influential and well-known example being Akers’ (1998) social learning theory. The results of the meta-analysis revealed that the average genetic effect size on peer interactions was .21. However, it is important to point out that while Kendler and Baker examined various aspects of peer interactions, they did not directly examine interactions with delinquent peers. Fortunately, a number of studies have directly examined genetic influences on interaction with delinquent peers, with most studies reporting heritability coefficients of .40 or higher (Cleveland, Wiebe, Rowe, 2005; Fowler et al., 2007; Harden, Hill, Turkheimer, & Emery,
2008; Iervolino, Pike, Manke, Reiss, Hetherington, & Plomin, 2002; Kendler, Jacobson,
Gardner, Gillespie, Aggen, & Prescott, 2007). Fourth, Kendler and Baker included 10 studies which examined the heritability of stressful life events. Stressful, or strain-inducing, life events have been found to be positively associated with various forms of antisocial behavior and serve as the primary causal mechanism in some of the most influential and well-known criminological theories including Merton’s (1938, 1957) classic strain/anomie theory and Agnew’s (1992) general strain theory. The results of the meta-analysis revealed heritability estimates that ranged between .24 and .47, with an average weighted heritability coefficient of .28.
2.4 Summary and Discussion
This chapter presented an overview of the basic assumptions, concepts, and methodologies that underlie the biosocial perspective. In addition, the findings from a chronologically, methodologically, and theoretically mature literature focused on the underlying etiological development of antisocial behavior were discussed. Based on this summary of the literature, four key points with direct application to existing criminological theoretical perspectives emerge and warrant additional attention. First, there is now undisputable evidence that antisocial behavior (which inherently includes criminal behavior) is the direct result of a
60 culmination of environmental and genetic influences. More specifically, genetic influences explain approximately 50 percent of the overall variance in antisocial behavior and the remaining
50 percent of the variance is explained by environmental influences, with nonshared environmental influences being far more potent than shared environmental influences. Second, the culmination of genetic and environmental influences, through G × Es and rGEs, tend to have larger effects on antisocial behavior than the independent effects of each set of influences.
Third, studies performed under the biosocial perspective, but in disciplines other than criminology, have provided evidence that any studies attempting to better specify the underlying etiological development of any phenotype but do not effectively account for genetic influences are likely misspecified and yield statistical coefficients that are essentially meaningless. Further, the theoretical implications of this particular observation are currently unknown, but there is good reason to believe that attempting to situate existing criminological theories within the biosocial perspective may significantly attenuate previously observed associations between some of the most widely-recognized theoretical concepts and antisocial behavior. For example, peer interactions, a central concept to social learning theory (Akers, 1998), have been found to be significantly influenced by genetic factors (Cleveland et al., 2005; Fowler et al., 2007; Harden et al., 2008; Iervolino et al., 2002; Kendler et al., 2007). Along the same lines, some studies indicate that specific genetic polymorphisms (i.e., a dopamine receptor gene [DRD2]) cluster within peer groups, suggesting that youth tend to select peers with genetic predispositions similar to their own (Fowler, Settle, & Christakis, 2011). A sizable literature has also indicated that the association between parental socialization, a key theoretical concept in self-control and social control theories, and various antisocial outcomes becomes highly attenuated or nonsignificant after taking shared genetic influences between parents and their children into account (Harris,
61
1995, 2009; Jaffee et al., 2004; Lytton, 1990; Wright & Beaver, 2005). In addition, based on the first law of behavior genetics (Turkheimer, 2000), all other key theoretical concepts would be under at least some genetic influence.
While such observations provide some preliminary evidence which suggests that any previously observed association between theoretical concepts and antisocial outcomes will be at least attenuated when genetic influences are properly controlled, the direct results of tests examining these associations within the confines of a genetically informed model remain unknown. This is a direct result of the overall lack of criminological research that has been done under the biosocial perspective. The current project aims to provide a more thorough understanding of the predictive ability of theoretical concepts central to rational choice theory, social learning theory, classic strain theory, and social bonding theory after properly accounting for genetic influences. Once this information has been gathered, it will then be possible to determine how to situate each of these theoretical perspectives within the biosocial perspective.
62
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Predicted Probability Violenceof 1 2 3 4 Level of Parental Maltreatment
High Genetic Risk Low Genetic Risk
Figure 3.1: Graphical Example of a Gene-Environment Interaction.
63
CHAPTER 3
CONVENTIONAL CRIMINOLOGICAL THEORY: EMPIRICAL FINDINGS AND CONCERNS
The biosocial perspective is rapidly developing, such that results from hundreds of studies now indicate that both biological and environmental influences work—both independently and interactively—to generate variation in antisocial phenotypes (Ferguson, 2010;
Mason & Frick, 1994; Miles & Carey, 1997; Rhee & Waldman, 2002). While a small number criminologists remain openly and vehemently opposed to a more “biological” criminology, this underlying blind opposition is becoming increasingly indefensible and unpopular. Despite a few flamboyant examples (see Carrier & Walby, 2014; Walby & Carrier, 2010), criminologists now seem to be more accepting of biosocial explanations of antisocial behavior than ever before.
While the overall presence of biosocial research may be lacking in many of the leading journals within the field of criminology, the number of presentations at professional conferences, the number of freshly minted PhDs specializing in biosocial criminology, and the number of criminology programs that offer some sort of biosocial criminology course continues to grow each year. Whether this growth is the result of a fundamental shift in the underlying ideology within the field of criminology or is driven by demand from curious students, practitioners, and researchers remains an open question. Regardless of the source of the shift, the present represents an exciting time within the field of criminology.
Despite the substantial gains in the underlying understanding of the contributions of biological and environmental influences on antisocial behavior made over the past decade or two, the processes that eventually culminate into antisocial and criminal behaviors are still relatively unknown. In other words, while a large swath of research has indicated that genetic
64 and environmental influences directly contribute to antisocial phenotypes, the underlying processes that specify the manner in which both sets of influences eventually lead to antisocial behavior have not been explicitly detailed. This lack of specificity is likely a direct result of the overall absence of any theoretical progress within the biosocial perspective. As previously mentioned, the few attempts to specify a biosocial theory of criminal behavior (Ellis, 2005;
Robinson & Beaver, 2009) have been largely unsuccessful. However, these particular theories were concerned with distilling the entire biosocial perspective down into a single theory as opposed to fitting a single theory within the confines of the biosocial perspective. For example,
Ellis (2005) attempts to specify a biosocial theory of antisocial behavior by drawing on concepts from multiple disciplines including evolutionary psychology, neuroscience, endocrinology, psychiatry, and psychology while also attempting to incorporate known environmental factors that have previously been shown to increase predisposition toward antisocial behavior such as exposure to cigarette smoke in utero. Ellis also attempts to incorporate aspects of genetic risk and concentrations of various neurochemicals including cortisol and various neurotransmitters.
Even this cursory overview illustrates the primary limitation of the theory: it is overly complex.
In an effort to integrate all of the findings from biosocial research, Ellis proposes a theory that collapses under its own weight.
An alternative approach has been embraced by a handful of biosocial criminologists.
More specifically, both Walsh (2000, 2002) and Fishbein (1990, 2001) have argued that formal and novel biosocial theories of criminal behavior are unnecessary and instead propose that biosocial concepts can be integrated into existing criminological theory. In perhaps the most direct example of this form of integration, Walsh (2000) attempted to integrate findings from the field of behavior genetics within Merton’s (1938) strain theory. Walsh (2000) explained that the
65 purpose of his inquiry was to “explore how behavior genetics can complement and expand traditional social science understanding of criminal behavior in the context of anomie/strain theory, a respected and long-lived criminological theory that focuses on cultural and structural causes of crime” (p. 1075). While this particular suggestion has greatly influenced the current dissertation, there is a fundamental flaw in the logic extended by Walsh and others calling for the integration of biosocial concepts within existing criminological theory. More specifically, this observation assumes that the direction of integration should point from biosocial concepts to existing criminological theory, which inherently assumes that concepts central to conventional theories have causal influences on antisocial behavior even after taking biological influences
(including genetics) into account. While this may be the case, the overall lack of studies examining associations between concepts central to such theories and antisocial outcomes within the confines of genetically informed models leaves this question open. Alternatively, if after controlling for genetic influences, concepts central to mainstream criminological theories are no longer associated with antisocial outcomes, the integration of biosocial concepts into such theories would be counterproductive, effectively diminishing the explanatory power of such theories even further. Based on these observations, the current dissertation proposes that the direction of integration should be reversed and should point from existing criminological theory to the biosocial perspective. This model of integration will be more thoroughly specified in
Chapter 4, but the underlying logic is presented here in an effort to effectively situate the discussion of the following theoretical perspectives within this context.
Prior to providing a more thorough discussion of the manner in which conventional criminological perspectives can be situated within the biosocial perspective, it is first necessary to provide an overview of the theories that will be examined in the current dissertation. The
66 remainder of this chapter will be devoted to providing an overview of each of the following theories: rational choice theory; social learning theory; classic strain theory; and social bonding theory. In this way, the current chapter has two primary objectives. First, a detailed discussion of each theory, and its central concepts, will be provided. While logical and empirical limitations noted in the literature will be noted and discussed, the primary purpose of the current chapter is to provide a detailed overview of each theoretical perspective. In this way, critiques of each perspective will be limited and focused only on criticisms noted by the author(s) of each perspective or those that have been noted in the literature. Second, an overview of the current empirical status of each theory will be provided. Importantly, this overview will be limited to only the most influential, comprehensive, and empirically-robust studies examining each particular theory. The theories examined in the current dissertation are some of the most influential theories of antisocial behavior within the social sciences and have resulted in literally thousands of studies. In this way, an exhaustive review of the empirical research surrounding each theory would result in hundreds of pages and would detract from the primary objectives of the current project.
3.1 Rational Choice Theory
Many of the conventional criminological theoretical perspectives that have received the most attention over the past several decades tend to shift focus away from processes that occur within the individual and, instead, attempt to specify the mechanisms that lead from various aspects of socialization or environmental influences to antisocial behavior. Many of these theories focus on concepts that exert some sort of influence on the agent in some way, primarily through socialization. Therefore, the underlying focus of such perspectives is on the
67 environmental influences as opposed to the individual. While some notable exceptions exist10, most theoretical perspectives tend to treat the internal responses to specific environmental influences as an afterthought. For example, while Gottfredson and Hirschi’s (1990) general theory of crime focuses on individual-level factors that predict variation in antisocial behavior, this perspective does not detail the specific internal mechanisms that bridge the gap between parental socialization, the development of self-control, and antisocial behavior. What processes sparked by inadequate parental socialization ultimately result in overall lower levels of self- control? What internal mechanisms lead from possessing lower levels of self-control and ultimately engaging in criminal or antisocial behavior? These questions remain largely unanswered, not just in regard to self-control theory, but also within the majority of the most commonly studied criminological theories. Simply establishing an association between theoretical concepts and antisocial behavior does not sufficiently capture the complex internal processes that ultimately contribute to variation in antisocial behavior.
Deterrence theory, and other theories falling under the rational choice perspective, attempts to provide a more thorough specification of the internal process that ultimately results in antisocial behavior. In other words, rational choice theories attempt to better specify the mediating processes that connect external influences and variation in antisocial behavior. More specifically, the rational choice perspective attempts to specify the role of choice in the decision- making process that underlies all behavioral outcomes, including antisocial behavior. Why do some individuals eventually engage in criminal behavior while others, facing the same or similar
10 For example, labeling theory (Becker, 1963; Cooley, 1902; Lemert, 1951; Mead, 1934) specifies the process from which external events (typically contact with the criminal justice system) results in internal modifications of one’s perception of self. The labeling perspective is unique in that it details both the external and internal processes that eventually result in antisocial behavior. Another notable example, would be Agnew’s (1992; Agnew et al., 2002) general strain theory (GST), which details the process in which external strain-inducing experiences trigger negative emotional states, which in turn, ultimately result in antisocial coping mechanisms. Another notable and relevant example was offered in the now classic work of Matza (1969), who specified the process by which individuals “become deviant.” 68 circumstances, refrain? Proponents of rational choice theories argue that in order to adequately answer this question, criminologists must take into account human agency, or the process in which human beings make decisions. In addition to providing a more robust and thorough understanding of criminal behavior, proponents also argue that providing a greater emphasis on this decision-making process will result in at least one other benefit. Criminal law is built upon the fundamental notion that individuals play a central role in the behaviors they ultimate engage in. In this way, individuals are punished for misconduct and rewarded for prosocial behavior.
This foundational concept operates under the assumption that humans actively choose their own actions and are therefore responsible for such actions. In this way, a failure to consider the decision-making process that underlies criminal behavior drives a wedge between current explanations of criminal behavior and the legal code (Nagin, 2007). In addition, this disconnect between theoretical explanations of criminal behavior and criminal law likely has negative repercussions for the extent to which criminological research can inform public policy. Below, a discussion of the key theoretical assumptions and concepts that comprise rational choice theories will be provided. In addition, findings from the empirical literature focusing on the rational choice perspective will be discussed.
3.1.1 Key Assumptions and Theoretical Concepts of Rational Choice Theory
The basic assumption that human behavior is directly influenced by the expected consequences, whether they be good or bad, of such behaviors is a commonly held belief that has spilled over into criminological theory. Based on the pervasiveness of this assumption, it is not all that surprising that some of the first formally proposed criminological theories focused on the role of expected consequences in the decision-making process that ultimately resulted in behavior. The majority of the guiding ideas and logic underlying the rational choice perspective
69 were initially proposed throughout the 18th century by Cesare Beccaria (1963 [1764]) and Jeremy
Bentham (1948). Many of the underlying ideas of the rational choice perspective can even be traced back to Thomas Hobbes proposal of a “social contract” during the 16th century. During these periods of rapid social change, scholars began to recognize the utility of introducing proportional, yet adequately harsh, punishments for criminal behavior in an effort to deter potential offenders from committing such acts. The organizing idea underlying the concept of deterrence was that prior to committing a given act, human beings weigh the potential costs and benefits that accompany engaging in (or failing to engage in) the act in question. In this way, many scholars hypothesized that it would be possible to deter potentially harmful and violent behaviors by attaching consequences to these behaviors that fit specific criteria. More specifically, punishments that accompany a specific behavior must be carried out as quickly as possible (celerity), have a high probability of being carried out (certainty), and must be proportional to the behavior committed (more severe behaviors should be responded to with more severe sanctions; certainty). Punishments that adequately balance these three characteristics were lauded by Beccaria, Bentham, and others as being the most likely to deter potentially harmful behaviors.
In order for deterrence to be possible, three basic assumptions regarding human nature must be present. First, human behavior must be the direct result of free will. In other words, humans possess the ability to freely choose whether they ultimately engage in a given behavior.
In direct opposition to free will is the classic assumption of determinism—the probability of the occurrence of a given event is directly related to and dependent upon the occurrence of other antecedent events (Walsh, 2014). Anything more strongly worded would constitute fatalism, not
70 determinism.11 In an attempt to further differentiate determinism from fatalism, Walsh (2014) stated that “[s]cientific determinism does not state that X will lead to Y absolutely and unerringly; rather, it says that given the presence of X, there is a certain probability that Y will occur” (p. 5). Regardless, the concept of deterrence rests on the assumption that humans possess free will, at least to some extent. Scholars have argued that free will and determinism (or fatalism) do not reflect a natural dichotomy, instead various gradients of each perspective are possible. For example, a number of scholars have argued in favor of constrained agency or bounded free will (for an overview see Swidler, 1986), wherein the potential menu of behavioral choices available to an individual is directly related to the environmental circumstances to which the individual is exposed to. In other words, individuals are only able to choose from the options that are available to them, and some individuals simply have more potential options to choose from than others. In this way, we cannot assume that humans have free will in the purest sense, but, humans still possess the ability to make decisions based on the (limited) available options.
The second assumption underlying the concept of deterrence, and the rational choice perspective, is that humans are hedonistic in that they tend to favor pleasure over pain. All else equal, humans will choose options that increase pleasure and minimize pain. Deterrence directly exploits this particular aspect of human nature by actively manipulating the potential pain and pleasure that accompanies various behaviors. For example, attaching a harsher penalty to a given criminal act increases the potential pain that an offender may experience if they are caught engaging in said act. This particular aspect of deterrence is directly related to the third assumption of the rational choice perspective and deterrence: humans are rational beings and
11 This is a common misconception among criminologists who assume, and often argue, that determinism dictates that “the offender’s behavior is ‘determined’ by something other than his or her free-willed choice” (Brown, Esbensen, & Geis, 2013; p. 218). This description would more closely reflect fatalism as opposed to determinism, which Walsh (2014) argues underlies any form of science and is a position held by all scientists. 71 weigh the potential costs and benefits of a given act before acting. More specifically, Bentham
(1948), and later others, argued that prior to the commission of a given act, humans undergo a
“hedonistic calculus” in which they weigh the probability of pleasure maximization against the pain of punishment. This particular process, which eventually culminates into a choice, is one of the defining features of the rational choice perspective and details the internal processes that eventually culminate into antisocial behavior. The assumption that humans are rational beings stems primarily from the field of economics and the principle of expected utility (for an overview see Schoemaker, 1982) which asserts that individuals will make rational decisions with the intent to maximize profits and minimize costs or losses. According to the rational choice perspective, this same rationale can be directly applied to all other behavioral outcomes including criminal behavior. More specifically, offenders will engage in criminal behavior if they perceive the potential benefits of such behavior outweigh the potential costs.
The underlying assumption that humans are rational and hedonistic beings who freely choose whether to participate or refrain from criminal activity directly aligns with the concept of deterrence. More specifically, deterrence asserts that the proper manipulation of the costs and benefits which underlie various behaviors will significantly increase or decrease the likelihood that individuals will engage in such behaviors. Sliding the weights of celerity, certainty, and severity of punishment alters the information that is processed by potential offenders and ultimately results in a decrease in the likelihood of engaging in antisocial or criminal behaviors.
Based on these observations, it is no surprise that the concept of deterrence was highly influential in the development and evolution of the American justice system. This form of deterrence— typically referred to as formal deterrence—posits that formal social controls (e.g., police, judges, legal codes) are responsible for any potential deterrent effect. Alternatively, informal deterrence
72 focuses on the potential social (as opposed to legal) consequences that accompany misbehavior.
Importantly, a findings from a number of studies have revealed that informal deterrent effects are far more influential than formal deterrent effects (Green, 1989; Grasmick & Bursik, 1990; Pratt,
Cullen, Blevins, Daigle, & Madensen, 2006; Williams & Hawkins, 1989; Zimring & Hawkins,
1973). However, these findings are far from unequivocal and a number of additional studies have found that the formal deterrent effects outweigh informal deterrent effects (Nagin &
Paternoster, 1991; Nagin & Pogarsky, 2001). Importantly, some early studies have suggested that any deterrent effect stemming from formal punishment is likely indirectly influenced by the informal or social consequences that also stem from formal interventions (Williams & Hawkins,
1989; Zimring & Hawkins, 1973). In either case, the rational choice perspective, and subsequently the concept of deterrence, has resulted in a significant number of empirical studies.
The section below will provide an overview of the empirical literature examining the rational choice perspective within the context of antisocial and criminal behavior.
3.1.2 The Empirical Status of Rational Choice Theory
The rational choice perspective has a storied past within the field of criminology. Most criminologists would agree that the fundamental building blocks for the field of criminology were directly connected with the rational choice perspective (Brown et al., 2013). The primary and most influential organizing principles of the Classical School of Criminology directly aligned with the main premises and assumptions of the rational choice perspective and deterrence theory (Akers & Sellers, 2013). Needless to say, the rational choice perspective—and deterrence theory specifically—was quite influential during criminology’s infancy. While the rational choice perspective was eventually replaced by the positivist perspective in the early
1900s, and then later by purely sociological theories of criminal behavior beginning in the 1920s,
73 a renewed interest in the rational choice perspective was sparked by advances in the field of economics, specifically in regard to the principle of expected utility described above during the late 1960s (Pratt et al., 2006; Lilly et al., 2011; Nagin, 1998). Since this time, the rational choice perspective has remained a mainstay within the literature. While the rational choice perspective no longer carries the same level of influence on criminological theory as it once did, this perspective still actively influences modern criminological research. The longevity of this perspective, along with the steady level of interest associated with it, has generated an impressive number of studies. To put the overall size of the literature in perspective, there have been at least five different systematic reviews of the literature related to the rational choice perspective within criminology (Blumstein, Cohen, & Nagin, 1978; Nagin, 1998, 2013; Paternoster, 1987; Pratt et al., 2006). In their meta-analytic review of the literature, Pratt and colleagues (2006) argue that deterrence theory “is one of the most extensively tested theories of crime” (p. 374). Below, a summary of this expansive literature will be offered. This summary will focus on four distinct areas of the existing literature: (1) the rationality of offenders; (2) the effectiveness of deterrence;
(3) the effectiveness of formal and informal deterrence; and (4) modern developments in the rational choice perspective.
Rationality of Offenders. Despite the pervasive influence of the rational choice perspective on the field of criminology and the legal code, there has been little research actually focusing on the rationality of offenders. This particular observation is somewhat surprising since the assumption of rationality is one of the primary, and inherently necessary, assumptions of the rational choice perspective. In addition, one of the primary advantages of the rational choice perspective is the identification of the mechanisms within the individual which ultimately explain variation in offending. The majority of the studies that have attempted to specify the decision
74 making process that underlies criminal behavior analyze qualitative data, typically in the form of interviews with successful offenders. Tunnell (1990, 1992) interviewed repeat property offenders in an effort to gain a better understanding of the decision making process that eventually culminated into the choice of criminal behavior. Offenders tended to believe that their overall chances of being caught were relatively low, and that if they were caught, they would receive a lenient punishment. Importantly, this particular finding directly aligns with more recent studies which indicate that offenders with higher levels of contact with the criminal justice system (or are in direct contact with others who have higher levels of contact) are significantly more likely to engage in subsequent criminal behavior compared to their counterparts (Paternoster & Piquero, 1995; Piquero & Paternoster, 1998; Piquero & Pogarsky,
2002). This somewhat counterintuitive finding has been referred to as the experiential effect
(Paternoster, 1987) or the resetting effect (Pogarsky & Piquero, 2003).
Despite this congruence in findings, this particular pattern of behavior does not necessarily constitute rational decision making, at least not in the purest sense. While the offenders interviewed by Tunnell (1990, 1992) described the measures they would take to avoid detection, their overall assessment of the risk of detection was highly unrealistic. A closer examination revealed that these offenders, did not accurately assess the probability of arrest, performed only menial levels of planning prior to the commission of a particular crime, and were uninformed of the most likely legal penalties for the crimes they committed. A similar set of findings emerged from an ethnographic study of burglars performed by Cromwell and colleagues
(1991), who concluded that opportunity and other situational factors played the most important role in previous burglaries. More recently, De Haan & Vos (2003) interviewed a sample of offenders who have previously committed street robberies, and argued that the potential rewards
75 stemming from these criminal behaviors extend beyond the money or goods they obtain in such crimes. Rather, additional factors figured into the decision making process include emotional and psychological factors such as the release of tension and desperation. Another recent study
(Dugan, LaFree, & Piquero, 2005), focused on the potential role that recent increases in the certainty and severity of punishment played in overall decreases in airplane hijackings. These modifications to the potential punishments which accompany hijackings were found to be more effectively deter offenders, providing support for the rational choice perspective.
Previous studies have attempted to account for this discrepancy in findings by arguing for models of partial or bounded rationality (De Haan & Vos, 2003; Matsueda, Kreager, & Huizinga,
2006). This explanation takes into account variability in the potential choices available to offenders, which stem from various social influences. Rather than assuming that individuals are perfectly rational, more modern theorists have argued that the underlying mechanisms driving human behavior are far more complex and subject to many moderating and mediating processes.
In an effort to better capture this notion of partial rationality, Paternoster and Pogarsky (2009;
Paternoster, Pogarsky, & Zimmerman, 2011) introduced the concept of thoughtfully reflective decision making (TRDM), which refers to “intelligent or ‘competent’ decision making because it is the best way to sift among alternative courses of action for the most effect way to reach one’s desired goals” (Paternoster et al., 2011, p. 2). An empirical examination of TRDM revealed that adolescents who scored higher on the TRDM measure were more likely to be in better health, were less likely to use drugs and alcohol, were engaged in more community involvement, and were less likely to engage in criminal behavior (Paternoster et al., 2011). The results of this study provide support for the rational choice perspective since respondents who were more likely
76 to thoughtfully assess the potential options available were less likely to engage in delinquent or risky behaviors and more likely to engage in prosocial behaviors.
Effectiveness of Deterrence. In direct contrast to the minimal amount of criminological research examining the rationality of offenders, a sizable number of studies have directly examined the effectiveness of various aspects of deterrence or deterrence theory. Fortunately, a relatively recent meta-analysis performed by Pratt and colleagues (2006) directly attempted to empirically synthesize this line of research in an effort to directly examine the current state of deterrence theory. Pratt and colleagues examined all studies published before 2003 and utilized stringent inclusion criteria which ultimately resulted in the inclusion of 40 studies and a total of
200 effect sizes. A total of four “predictor domains” directly related to deterrence theory were included in the study and included: certainty of punishment, severity of punishment, deterrence theory composites, and the threat of non-legal sanctions. Importantly, the authors of the study included a number of statistical procedures in an effort to estimate unbiased effect sizes including weighting estimates by sample size, correction for multiple estimates from each study, and correction for multivariate estimates.
The results of the meta-analysis produced five specific sets of findings directly related to deterrence theory. First, the overall mean effect sizes of the examined predictor domains on outcomes related to crime and/or deviance are relatively weak, ranging between .00 and -.20.
The authors observe that these effect sizes are weaker than previous meta-analyses examining the potential association between alternative theoretical concepts and antisocial outcomes
(Andrews & Bonta, 1998; Pratt & Cullen, 2000). Second, even when significant mean effect sizes were observed, such effects become significantly attenuated, oftentimes to zero, in multivariate models. Third, studies which employed more rigorous methodological approaches
77 were significantly more likely to detect attenuated or nonsigificant effects. In addition, many of the strongest and most consistent effects were detected using bivariate modeling strategies, making such findings highly susceptible to confounding. Fourth, the certainty of punishment was among the strongest predictors of deterrence, particularly when predicting white-collar offenses such as fraud and tax violations. Fifth, non-legal sanctions were also among the most robust predictors of deterrence and severity of punishment was among the weakest. Pratt and colleagues (2006) interpret these results as direct evidence of the limited explanatory power of deterrence theory, even going as far as to state “[d]eterrence theory is likely to persist in much the same way despite the relatively poor predictive capacity of the variable specified in the theory” (p. 386). While the results of this study may be somewhat dated, a recent systematic review of the literature by Nagin (2013) seems to echo the findings. While Nagin praises the utility of the deterrence theory, he also recognizes the limitations of the theory, particularly in regard to punishment severity. While Nagin does not perform a meta-analysis, the overall conclusions drawn from his review directly coincide with the results reached by Pratt et al.
(2006). Taken together, these results suggest that empirical evidence illustrating the predictive ability of deterrence theory is—at best—mixed.
Formal and Informal Deterrence. Directly in line with one of the primary findings of the meta-analysis performed by Pratt and colleagues (2006), a parallel line of research has compared the effectiveness of formal versus informal sanction costs. In other words, deterrence theory— and, more broadly, the rational choice perspective—is concerned with the “pains of punishment” and the resulting influences such pains (whether real or perceived) have on the overall decision making process. However, the vast majority of early studies failed to recognize the importance of informal sanction costs and focused nearly exclusively on costs associated with formal
78 sanctions. During the 1970s and 1980s, when the rational choice perspective and deterrence theory received renewed interest within criminology, theorists began to recognize the importance of informal sanction costs (Anderson, Chiricos, & Waldo, 1977; Paternoster & Iovanni, 1986;
Salem & Bowers, 1970; Tittle & Rowe, 1973; Waldo & Chiricos, 1972; Williams & Hawkins,
1989; Zimring & Hawkins, 1973). In addition, several theorists recognized that any association between formal sanctions and deterrent effects is likely mediated by informal sanctions, wherein formal punishment activates specific informal ramifications which ultimately results in a deterrent effect (Salem & Bowers, 1970; Zimring & Hawkins, 1973). Findings from more recent studies seem to fall directly in line with this observation. For example, a developed and methodologically sophisticated line of research has revealed that individuals with greater stakes in conformity are most likely to be deterred by informal sanction costs (Nagin & Paternoster,
1993, 1994; Nagin & Pogarsky, 2001, 2003). Subsequent studies have attempted to identify additional individual characteristics which result in variability in deterrent effects (for an overview see Piquero, Paternoster, Pogarsky, & Loughran, 2011). Some of the more salient individual characteristics that have been identified include self-control (Piquero & Tibbetts,
1996; Pogarsky, 2007), moral inhibition (Gallupe & Baron, 2014; Paternoster & Simpson, 1996), levels of intoxication and anger (Exum, 2002), and religiosity (Spivak, Fukushima, Kelley, &
Jenson, 2011).
A related line of research indicates that informal deterrent effects are not limited to tangible social relationships and costs. Rather, internalized ramifications, such as feelings of shame and guilt, have also been found to have deterrent effects (Grasmick & Bursik, 1990;
Grasmick, Bursik, & Arneklev, 1993; Klepper & Nagin, 1989; Nagin & Paternoster, 1993;
Tibbetts & Meyers, 1999). Importantly, studies have found that the anticipation of shame has a
79 significant mediating effect on deterrence across both males and females, but females seem to be more susceptible relative to males (Tibbetts & Herz, 1996). The results of a recent study revealed that anticipated feelings of shame completely mediated the association between certainty of punishment and the intent to engage in delinquent behavior (Rebellon, Piquero,
Piquero, & Tibbetts, 2010). The results of this line of research seem to indicate that informal social costs seem to be driving the majority of any detectable deterrent effects. Furthermore, a complex process in which external and internal (e.g., feelings of shame) informal social costs differentially influence the underlying decision makings processes which culminate into behavioral outcomes including antisocial behavior.
Modern Developments. Due to the limited empirical evidence favoring deterrence theory and the rational choice perspective, several theorists have advocated for theoretical integration between deterrence theory and other popular theoretical perspectives (Gray, Ward, Stafford, &
Menke, 1985; Nagin & Paternoster, 1994; Paternoster et al., 2011; Piquero & Tibbetts, 1996;
Stafford & Warr, 1993).12 Two particular instances of theoretical integration in regard to deterrence theory have received a significant amount of empirical attention. First, Stafford and
Warr (1993) attempted to apply concepts from social learning theory (Akers, 1998) to deterrence theory. More specifically, offenders who have been detected and punished—or who are aware of others who have been—are significantly less likely to offend again in the future. However, offenders who engage in delinquent or criminal acts but are not detected or punished—or have knowledge of others who were failed to be detected—undergo “punishment avoidance” which
12 Importantly, the majority of these studies simply advocate for the “cross-fostering” of theoretical concepts among deterrence theory and additional theoretical perspectives as opposed to specifying an explicit integration model. For example, several theorists have recognized that various individual characteristics which are directly tied to additional theoretical perspectives (e.g., self-control) significantly predict variance in deterrent effects, but make no effort to specify the process in which integration between these two perspectives could be carried out. The majority of these studies do not even specify the direction of integration (whether deterrence theory should be integrated into other perspectives or vice versa), resulting in an incomplete explanation of the nature of integration proposed. 80 acts as a negative reinforcer which, in turn, results in a significant increase in the probability of future offending. In this way, the decision making process underlying antisocial behavior is shaped by the presence of punishment, the lack of punishment, and the modeling of others’ behavior (based on information regarding the detection or lack of detection passed on to the offender). Importantly, this reconceptualization of deterrence theory takes into account the direct experiences of the offender, but also accounts for information the offender has regarding others’ experiences. Importantly, subsequent studies have revealed that others’ experiences play an important role in the decision making process underlying antisocial behavioral outcomes (Anwar
& Loughran, 2011; Matsueda et al., 2006; Pogarsky, Piquero, & Paternoster, 2004).
Second, Piquero and Tibbetts (1996) integrated concepts from self-control and deterrence theories. More specifically, individual’s perceptions of the potential costs and benefits that accompany criminal behavior, in addition to their perception of being sanctioned, were significantly associated with self-control. In this way, individual characteristics significantly predict variation in the assessment of risk, but also in the probability of detection and both formal and informal sanction costs. A substantial number of additional studies have observed findings which illustrate the potentially important role of the trait of self-control within deterrence theory (Nagin & Pogarsky, 2001, 2003; Poarsky, 2002, 2007).
3.2 Social Learning Theory
As noted previously, many criminological theories simply identify associations between various sets of social influences and antisocial behavior, but fail to identify the underlying process and specific mechanisms driving the observed associations. Obviously, the rational choice perspective is one notable example, but so are learning theories. While many texts and studies errantly operationalize social learning theory as “a theory of bad companions or a
81
‘cultural deviance’ theory” (Akers & Jensen, 2006, p. 38), such a view is likely far too narrow, as social learning theory was proposed as a general theory of crime. According to Akers and
Jensen (2006), social learning theory was proposed as “a general theory that offers an explanation of the acquisition, maintenance, and change in criminal and deviant behavior that embraces social, nonsocial, and cultural factors operating both to motivate and control criminal behavior and both to promote and undermine conformity” (p. 38). Based on this explanation, social learning theory as a theoretical perspective has much more to offer than an explanation of the role that “bad companions” play in the development of antisocial behavior. Rather, social learning theory seeks to specify, explicitly, the processes which ultimately result in all antisocial and criminal behaviors.
Directly in line with this particular goal, social learning theory operates under the assumption that all behaviors are the direct result of a learning process. More directly, all behavior—including antisocial and criminal behavior—is learned just like anything else (Akers
& Sellers, 2013). However, social learning theory does not simply identify “learning” as the process in which criminal behaviors are passed from one person to another, the theory also specifies the processes which both promote and control criminal behavior. In addition, and perhaps most importantly, social learning theory also specifies the process in which criminal behavior is learned. This is an important departure from previous perspectives in that the social learning perspective explicitly identifies the manner in which external socializing processes are internalized by the individual and eventually influence behavioral patterns. In this way, social learning theory provides a greater level of detail in the description of the underlying etiology of antisocial and criminal behavior. For a number of reasons, social learning theory is one of the most frequently tested criminological perspectives (Akers & Sellers, 2013; Stitt & Giacopassi,
82
1992) and the criminological theory that is most favored among criminologists (Cooper, Walsh,
& Ellis, 2010; Ellis & Walsh, 1999). Below, a discussion of the main theoretical concepts which comprise social learning theory will be provided. The actual learning process which eventually culminates into behavior will also be specified.
3.2.1 Key Assumptions and Theoretical Concepts of Social Learning Theory
While modern social learning theory is considered a distinct theoretical perspective, the overarching concepts and processes specified in the theory are rooted in Edwin Sutherland’s
(1947) theory of differential association, which is widely recognized as one of the first purely sociological explanations of criminal behavior (Brown et al., 2013). Differential association operated under the assumption that all criminal behavior is learned in the same way anything else is learned. Based on this assumption Sutherland argued that differential association theory was a general theory of criminal behavior and possessed the ability to explain both street and white- collar crimes. While Sutherland’s explanations of criminal behavior focused on explaining variation at the individual level, the theory also indirectly accounted for macro-level socializing agents such as neighborhoods. Sutherland argued that individuals who reside within disorganized neighborhoods are more likely to engage in criminal behavior due to the socialization processes that are more common in such neighborhoods. In this way, differential socialization does not run counter to more macro-level theories of criminal behavior, rather, this theoretical perspective is more concerned with specifying the micro-level processes that occur in such neighborhoods or other contexts.
Sutherland (1947) presented differential association theory as nine interconnected principles, with each principle building on the previous one. First, the foundation of the entire perspective is that criminal behavior is learned. This principle effectively rules out any process
83 internalized within the offender such as biological influences or human nature (Brown et al.,
2013). The second and third principles specify that criminal behavior is learned through interaction with others, with intimate personal groups—such as family and friends—providing the most potent influence. The fourth principle specifies that learning the motives, drives, rationalizations, and attitudes that accompany crime are just as important as learning the actual techniques of criminal behavior. In this way, learning the motivations that underlie criminal behavior is just as important as learning how to physically engage in such behaviors. The fifth principle is concerned with the driving force of the drives and motives specified in the fourth principle. More specifically, drives and motives are the direct result of either favorable or unfavorable definitions of legal codes which are acquired during the socialization process specified in the second and third principles.
The sixth principle is typically referred to as the principle of differential association
(Akers & Sellers, 2013) and indicates that criminal or delinquent behavior is a result of an excess of definitions favorable toward such behaviors over definitions unfavorable to the violation of laws. In other words, when definitions favorable toward antisocial behavior outweigh definitions unfavorable toward antisocial behavior individuals will engage in delinquent or criminal behaviors. The seventh principle specifies that differential associations that occur earlier
(priority), last longer and take up more time (duration), occur more often (frequency), and involve others with whom the individual more closely identifies with (intensity) will be more influential (Akers & Jensen, 2006). The eighth and ninth principles indicate that the process which ultimately results in criminal behavior involves the same mechanisms that are involved in any other learning, and that the same general needs and values underlie both criminal and non- criminal behavior.
84
While Sutherland’s (1947) theory of differential association is largely regarded as a standalone theoretical perspective and was lauded as the most thorough and plausible explanation of criminal behavior of the time, some theorists argued that this perspective is highly vague, particularly in regard to the actual learning mechanisms that result in behavior. Burgess and Akers (1966) attempted to better specify such mechanisms in their “differential association- reinforcement” theory. Rather than offer a competing theoretical perspective, Burgess and Akers offered a thorough revision and modernization of differential association theory. More specifically, the reformulated theory attempted to reframe differential association within the learning principles of operant and respondent conditioning popular in behavioral psychology at the time. In light of various critiques and empirical findings, Akers (1973, 1977, 1985, 1998) modified the theory numerous times eventually resulting in what is now commonly referred to as social learning theory. Importantly, social learning theory is fully complementary with
Sutherland’s theory of differential association, but instead offers a “broader theory that retains all of the differential association process in Sutherland’s theory (albeit clarified and somewhat modified) and integrates it with differential reinforcement and other principles of behavioral acquisition, continuation, and cessation” (Akers & Sellers, 2013, pp. 80-81).
Social learning theory includes four major concepts which detail the learning process which eventually results in criminal or antisocial behavior: differential association; definitions; differential reinforcement; and imitation. Differential association refers to the process specified by Sutherland (1947), including the social interactions which develop definitions that are either favorable or unfavorable toward the commission of criminal behavior. In addition, the modalities of differential association (Akers & Sellers, 2013) which refer to the situations in which differential associations are most influential (i.e., priority, duration, frequency, and
85 intensity) are also retained. Definitions refer to internal attitudes and the meanings that are attached to specific behaviors. In this way, definitions comprise the internal orientations, rationalizations, justifications, and attitudes that indicate whether a given act is viewed as right or wrong (Akers & Jensen, 2006). General definitions include moral, religious, or other conventional values and norms that encourage prosocial behavior and discourage antisocial behaviors. Specific definitions are unique to the individual and define whether certain situations provide opportunities for antisocial behavior.
Differential reinforcement refers to the balance of anticipated or actual rewards and punishments that are the direct result of a given act or set of acts. When balancing the potential rewards and punishments that accompany a particular act, individuals consider the past, present, and even the anticipated future ramifications that accompany the act. Similar to the rational choice perspective, acts which are expected to result in the greatest reward and least amount of punishment are most likely to be carried out and persist over time (Akers & Sellers, 2013).
Rewards may be nonsocial (e.g., the physical effects of drugs), but social rewards are considered to be the most powerful. Finally, imitation refers to the process in which an individual will engage in a given act after witnessing similar behavior in others. Several factors play an important role in the imitation process including the characteristics of the model, the behavior observed, and the observed consequences (or lack thereof) of the act (Akers & Jensen, 2006).
Imitation plays a more important role during the initialization of a given act, but is expected to continue to have some influence on the maintenance of behavioral patterns.
Akers (1998; Akers & Sellers, 2013) also specifies the manner in which these different concepts culminate into the actual process of learning. Importantly, this process is highly complex and follows a distinct sequence but also features feedback effects which differentially
86 emphasize various concepts (Akers & Sellers, 2013). The initiation of criminal behavior is a direct result of the organization of learned definitions, imitation of deviant models, and the expected balance of reinforcement at that current moment (Akers & Jensen, 2006). The actual or perceived rewards and punishments which accompany the act directly influence whether the act will be committed again. In addition, all four of the concepts central to social learning theory influence both the initiation and the maintenance of antisocial behavior, however, it is expected that imitation will play less of a role in the latter process. As previously mentioned, this process features prominent feedback effects wherein specific environmental factors may influence definitions (or some other aspect of the process), which in turn, results in an overall modification of the other aspects of the process. In addition, Akers’ (1998) most recent modification to social learning theory also recognizes the potential role that both meso-level and macro-level factors play in the learning process. In this social structure and social learning (SSSL) model, Akers argues that structural factors have an indirect influence on behavior via the influence of such factors on the learning process. More specifically, structural factors influence differential associations, definitions, differential reinforcement, and imitation, which in turn, influence behavior.
3.2.2 The Empirical Status of Social Learning Theory
A significant portion of the literature has been devoted to empirically assessing the primary concepts and processes that are directly implicated in social learning theory. As previously mentioned, the popularity of social learning theory has consistently increased since its initial proposition (Burgess & Akers, 1966), with recent inquiries revealing that the popularity of the theory has increased (Cooper et al., 2010; Ellis & Walsh, 1999). Based its long history and the overall empirical interest in social learning theory within the field of criminology, it is not all
87 that surprising that an immense literature related to this particular theory has taken shape.
Fortunately, two recent and comprehensive reviews of the literature have been completed (Akers
& Jensen, 2006; Pratt et al., 2010). Rather than focus on the hundreds of independent studies that comprise this literature (e.g., the trees), the most pertinent and informative findings gleaned from both comprehensive reviews of the literature will be offered below (e.g., the forest).
The first comprehensive review was offered by Akers and Jensen (2006), and was aimed at providing an overall summary of the empirical literature bearing on social learning theory.
Importantly, Akers and Jensen provide a comprehensive literature review and extrapolate findings from the existing literature, but they do not quantitatively synthesize the findings from the literature with meta-analytic techniques. While Akers and Jensen (2006) caution that even the richest survey data and most advanced modeling strategies “cannot fully reproduce the behavioral process envisioned in social learning theory,” they also acknowledge that “[i]f the theory is correct, then empirical findings…derived from or consistent with the theory, that approximate or provide a snapshot of the underlying process, should be supported by the data when subjected to proper statistical analysis” (p. 43). Akers and Jensen (and later Akers and
Sellers, 2013) categorize findings from the extant literature into three distinct categories: (1) general findings from studies examining the association between social learning variables and delinquent or antisocial behavior; (2) research on social learning within a family context; (3) research on social learning within peer groups or other group context. Empirical findings from each of these categories will be outlined below.
Perhaps the largest literature stemming from social learning theory is comprised of studies examining associations between the conceptual variables that comprise the theory and various delinquent, criminal, and antisocial behaviors. While this line of research exhibits
88 significant levels of variability in regard to a number of pertinent characteristics including the operationalization of key theoretical concepts, the population from which the analytic sample was drawn, and the outcome or dependent variables examined, Akers and Jensen (2006) conclude that this extensive line of research provides overwhelming evidence in favor of social learning theory. More specifically, the authors cite numerous studies from the 1970s, 1980s,
1990s, and the early 2000s, which provide direct support for the concepts and underlying processes specified in the theory. This same pattern of results persisted even when controlling for competing theoretical concepts, with social learning variables typically exerting stronger and more consistent effects on the outcomes of interest relative to competing theoretical concepts
(Catalano et al., 1996; Jang, 2002; Kaplan, 1996; Thornberry et al., 1994; but see Pratt & Cullen,
2000). Importantly, the majority of these findings have been garnered from studies examining samples from the United States, but a limited number of studies have also revealed similar findings in international samples (Bruinsma, 1992; Junger-Tas, 1992; Lopez et al., 1989; Wang
& Jensen, 2003; Zhang & Messner, 1995). A significant number of studies have also examined the underlying processes specified by Akers (1998) by which delinquent and criminal behavior is learned (Akers & Lee, 1996; Andrews & Kandel, 1979; Elliott & Menard, 1996; Empey &
Stafford, 1991; Esbensen & Deschenes, 1998; Esbensen & Huizinga, 1993; Sellers & Winfree,
1990; Warr, 1993). The overall pattern of findings flowing from this line of research has provided support for the learning process specified in social learning theory.
Akers (1998) has previously argued that the family is a key socializing agent with which individuals are differentially associated, wherein families provide individuals with definitions either favorable or unfavorable toward antisocial behavior, provide differential reinforcement for or against such behaviors, and also act as models for imitation. Albeit indirectly, a limited line
89 of research has also provided support for the importance of families within the context of social learning theory. Previous studies have revealed that antisocial behavior concentrates within families (Fagan & Wexler, 1987; McCord, 1991) with parents modeling antisocial behaviors for their children to imitate. In addition, parent-child interactions have been previously shown to have a direct influence on both prosocial and antisocial behavior in children (Snyder &
Patterson, 1995; Wiesner et al., 2003). Finally, previous studies have also revealed that younger siblings are at an increased risk for delinquent behaviors when they have older delinquent siblings (Lauritsen, 1993; Rowe & Farrington, 1997; Rowe & Gulley, 1992). Taken together, the findings flowing from this line of literature suggest that learning processes occurring within a family context have a significant influence on subsequent behavioral outcomes.
The influence of peer associations is so closely tied to social learning theory that many have mistakenly classified the theory as only a theory of peer influence. While social learning theory openly recognizes the peer group socialization as a critically important group context within which the social learning process occurs, the theory is not limited to socialization among peers. A substantial number of studies have reported significant associations between the behavior of one’s peers and their own behavioral patterns (Huizinga et al., 1991; Loeber &
Dishion, 1987; Loeber & Stouthamer-Loeber, 1986; Warr, 2002). The evidence is so overwhelming that Akers and Jensen (2006) observed that “[o]ther than one’s own prior deviant behavior, the best single predictor of the onset, continuance, or desistance of crime and delinquency is differential association with conforming and law-violating peers” (p. 51). A related line of research has revealed that the influence of gang membership, affiliation, and participation is also highly associated with antisocial behavior, even when taking into account peer behavior (Battin et al., 1998; Curry et al., 2002; Winfree et al., 1994). Taken together, the
90 social learning processes that occur within various social contexts, particularly in regard to peer groups, have a significant and consistent influence on subsequent behavioral patterns.
The second, and most recent, comprehensive review of the social learning literature was performed by Pratt and colleagues (2010) and consisted of a meta-analytic review of the literature. As previously discussed, meta-analyses are powerful quantitative tools which allow for the synthesis of a large number of research findings into a more digestible set of average effect sizes which make overall patterns more discernable. Pratt and colleagues limited their meta-analysis to empirical articles published after the initial proposal of social learning theory
(Burgess & Akers, 1966), studies which deliberately intended to test social learning theory as proposed by Akers (1998; Burgess & Akers, 1966), estimated the independent effect of at least one of the concepts included in the theory, and examined an outcome measure that represented some form of delinquency or criminal behavior. These inclusion criteria yielded 133 studies published between 1974 and 2003, with 246 statistical models resulting in a total of 704 effect sizes and representing 118,403 individual cases (Pratt et al., 2010, pp. 773-775). Importantly, studies were included if they examined social learning theory in isolation, in competition with competing theoretical perspectives, or within an integrated theoretical perspective. The analysis also took into account various sample characteristics including the sampling frame used (general vs. school sample) in addition to the demographic composition of the sample. Finally, information on a number of additional factors related to the model specification used in each study was also collected. These characteristics include whether variables from competing perspectives were included in the models, whether the study used a cross-sectional or longitudinal research design, and the dependent variable used in the study.
91
The meta-analysis performed by Pratt and colleagues (2010) yielded three main findings directly relevant to social learning theory. First, the explanatory power of social learning theory seems to be directly in line—if not greater than—with previous theoretical perspectives that have been subjected to meta-analytic review. More specifically, the mean effect size of both differential association (Mz = .225) and definitions (Mz = .218) seem to be directly in line with the average effect sizes garnered from Pratt and Cullen’s (2000) meta-analysis of the self-control literature, and significantly greater than the average effect sizes stemming from Pratt and colleagues’ (2006) meta-analytic review of deterrence theory. In addition, while some of the methodological and sample characteristics examined in the meta-analysis did significantly moderate these effect sizes somewhat, the overall pattern of results revealed that both differential association and definitions had consistent and moderate to large effects on the examined outcomes.
The second key finding garnered from the meta-analysis revealed that some aspects and concepts related to social learning theory have received more empirical attention than others.
More specifically, differential association and definitions have received the greatest amount of attention, while differential reinforcement and imitation have received the least amount of attention. Directly in line with this finding, the third key finding to emerge was that some components of social learning theory are more strongly associated with antisocial outcomes than others. Differential association and definitions exert the strongest average effect sizes, while differential reinforcement (Mz = .097) and imitation (Mz = .103) seem to exert the weakest average effect sizes. Importantly, differential reinforcement and imitation are the two theoretical concepts that Akers (1998) added to Sutherland’s (1947) original conceptualization of differential association. Despite these relatively weak average effect sizes, Pratt and colleagues
92 conclude that the overall pattern of results stemming from their analysis provided strong support in favor of social learning theory. This same overall conclusion was more recently echoed by
Akers and Sellers (2013) in their summary of the empirical literature pertaining to social learning theory when they noted that “[t]he relationships between social learning variables and delinquent, criminal, and deviant behavior found in the research are typically strong to moderate and there has been very little negative evidence reported in the literature” (p. 89).
3.3 Classic Strain Theory
As the fields of sociology and criminology transitioned from early biological explanations of criminal and antisocial behavior in the early decades of the 20th century, theoretical perspectives focusing on macro-level influences, such as the neighborhood (Shaw &
McKay, 1942), became the leading explanations of criminal behavior. While these perspectives were highly influential in the development of the field of criminology—and remain influential at the current time—they fail to account for variability in behavioral response to macro-level influences. As previously mentioned, in even the most disorganized or criminogenic neighborhoods a significant portion of the residents within each neighborhood refrain from criminal behavior. For example, in his now classic ethnographic study of neighborhoods in
Philadelphia, Anderson (1999) noted that most of the families he observed could be classified as
“decent” families as opposed to “street” families. Failing to account for this differential response to macro-level socializing agents and processes is a significant shortcoming of many of these theoretical perspectives, and one that is commonly glossed over.
One of the earliest, and perhaps the most well-known, attempts to account for the variability in response to macro-level social conditions and experiences was offered by Robert
Merton (1938). Drawing from concepts originally proposed by Emile Durkheim (1951 [1857]),
93
Merton offered an explanation of criminal behavior that focused on the process that was originated by experienced social conditions which differentially impact individuals based on their current standing within the class structure. Various macro-level conditions, which Merton attempted to specify in subsequent modifications of his theory (1957, 1959, 1968), ultimately resulted in feelings of strain or pressures to engage in antisocial or criminal behaviors.
Individual-level responses to these feelings or pressures can vary between individuals, effectively resulting in measurable differences in resulting behavioral responses. The theoretical perspective offered by Merton quickly became extremely popular within the field of criminology, particularly during the 1960s (Lilly et al., 2011). The overall influence of Merton’s strain theory continues within the contemporary literature and is largely regarded as the point of departure for two modern theoretical perspectives in Agnew’s (1992) general strain theory (GST) and Messer and Rosenfeld’s (1994) institutional anomie theory (IAT). In short, there are few theoretical perspectives that have had a greater influence on the theoretical and empirical landscape of the field of criminology than Merton’s strain theory. Below, a summary of the primary assumptions and concepts that comprise classic strain theory and the current empirical status of the theory will be provided.
3.3.1 Key Assumptions and Theoretical Concepts of Classic Strain Theory
Merton’s (1938, 1957) strain theory is rooted in assumptions and concepts first proposed by Durkheim (1951 [1857]) in his influential work “Suicide.” More specifically, Durkheim proposed that human beings possess unlimited appetites or desires which must be socially controlled in order for a society to function properly. Societies which do not properly socialize individuals to place reasonable limits on desires will suffer from higher rates of deviant and criminal behaviors due to the overall pressure to fulfill unlimited desires and the lack of
94 legitimate opportunities or means to fulfill such desires. The result of the disjunction between desires and the ability to achieve goals ultimately culminates into what Durkheim referred to as a state of normlessness or anomie. Anomie is present when societies fail to properly implement the proper social regulation of appetites through the emphasis of the use of legitimate means to achieve goals. Durkheim also recognized other social circumstances which may ultimately result in anomie, such as wars or political revolutions, but the disjunction between aspirations and realized goals is the most pertinent to criminological inquiry.
Merton (1938) applied the concept of anomie to the industrialized societies which were modern at the time, and the United States in particular. Merton recognized American culture places a significant amount of emphasis on economic success and achievement across all levels of society. In this way, the richest of the rich and the poorest of the poor are equally socialized to strive toward the “American Dream” of wealth and success. While individuals are primed to emphasize the overall goal of wealth and success, far less emphasis is placed on the approved social means necessary to achieve these goals (e.g., college education, hard work, etc.). In other words, American culture places an overwhelming amount of emphasis on the goals that people should strive toward, but little to no emphasis on the appropriate means that are required to achieve these goals. Merton argued that this particular situation is highly problematic since an integrated society would equally emphasize goals and socially-approved means. This problem becomes even more salient within members of the lower class who experience even more limited access to socially-acceptable means due to the current nature of the social structure. Despite limited access to legitimate means, the goal of wealth and success is equally salient at all levels of American society. This combination of an unlimited desire for wealth and success coupled with an overall lack of the desire or opportunity to utilize legitimate means results in a
95 suboptimal society. More specifically, Merton argued that societies which overemphasized goals at the expense of socially-acceptable means would experience higher rates of anomie, and in turn, higher levels of criminal and antisocial behavior.
Merton (1938) also recognized the important role that the disconnect between expectation and payoff in the development of criminal behavior. American culture emphasizes the already unlimited desires of humans to achieve wealth and success without providing adequate levels of opportunity to achieve such lofty goals. Merton argued that this disconnect, which is even more emphasized at lower social classes, ultimately resulted in an anomic society which produces feelings of strain or pressure to achieve socially universal goals through alternative means. Since
American society does not heavily emphasize the importance of utilizing legitimate means to achieve universal social goals, individuals are primed to explore alternative avenues to realize such goals. In this way, some individuals may resort to illegitimate means—including deviant or criminal behaviors—in an effort to effectively close the gap between expectations and payoff.
Directly in line with his theory, Merton argued that societies which equally emphasize means and goals, or societies which place less overall emphasis on the accumulation of wealth, are more integrated, do not experience anomie resulting in diminished levels of strain, and have overall lower levels of crime.
While a significant portion of Merton’s (1938) strain theory was devoted to specifying the structural processes which ultimately culminate into criminal behavior, he also specified the processes which result in individual-level variation in response to these structural influences (see
Baumer, 2007). More specifically, Merton (1968) proposed that individuals responded to and attempted to alleviate feelings of strain with one of five adaptations. The first adaptation, conformity, refers to continuing to believe in the goals of wealth and success while also
96 continuing to endorse the use of conventional means to achieve such goals. Merton argued that the default response to feelings of strain was to conform and to continually value both socially- accepted goals and legitimate means. While the majority of members within society are conformists, many others ultimately find the feelings of strain stemming from the disconnect between expectation and payoff to be overwhelming and attempt to alleviate their feelings of strain in other ways.
The second adaptation, innovation, is the most common criminal response and refers to individuals who continue to believe in the goals of wealth and success, but abandon conventional means in favor of unconventional means. Most offenders utilize this adaptation by substituting criminal behavior for conventional means in an effort to achieve highly valued goals. Third, ritualists simply adjust of the overall goal in an effort to fit the conventional means that are available. By scaling down their expectations, these individuals do not have to turn to illegitimate means and also alleviate feelings of strain. Fourth, retreatists, abandon the goal of economic success and the conventional means to achieve such goals. Merton hypothesized that individuals who utilize this adaptation typically suffer from alcoholism or drug addiction.
Finally, rebellion refers to a complete rejection of the current system in favor of an alternate one with newly defined means and goals. While Merton goes to great lengths to describe each particular adaptation to strain in an effort to explain variation in response to structural influences, no attention is given to the process which influences variability in the adaptation utilized to alleviate feelings of strain. This is a significant oversight, particularly due to the significant differences between each of the identified adaptations. Surprisingly, this is a criticism that has not been addressed in subsequent presentations of strain theory and has not been addressed within the extant literature.
97
3.3.2 The Empirical Status of Classic Strain Theory
Despite the resounding presence strain theory holds in the history of both sociology and criminology, the concepts and processes presented by Merton (1938, 1967) have resulted in a surprisingly limited line of research. This observation is particularly surprising after taking into account the dramatic influence of classic strain theory on the historical and contemporary theoretical landscape of the field of criminology. The roots of classic strain theory extend into
Cohen’s (1955) subcultural theory of “delinquent boys,” Cloward and Ohlin’s (1960) theory of opportunity, Agnew’s (1992) general strain theory (GST), and Messner and Rosenfeld’s (1994) institutional anomie theory (IAT). Needless to say, the underlying concepts and ideas proposed by Merton in his conceptualization of classic strain theory have remained relevant throughout the theoretical evolution of the field of criminology and remain today. In this way, the limited amount of empirical attention researchers have devoted to testing the key aspects of classic strain theory is surprising. Regardless, the limited literature that has been devoted to classic strain theory will be reviewed below.
In his now classic study, Hirschi (1969) analyzed a sample of junior and senior high school students in an early test of Merton’s strain theory. Based on the assumptions of classic strain theory, Hirschi expected that a disjunction in educational aspirations and expectations would result in higher levels of strain and, in turn, higher levels of delinquency. The results of the analyses provided three sets of findings directly relevant for classic strain theory. First, only a small proportion of the sample experienced a disjunction between aspirations and expectations, indicating that this issue may not be as salient as Merton initially indicated. Second, respondents who reported higher aspirations than expectations were no more likely to engage in delinquency than individuals who reported identical levels of aspirations and expectations indicating that the disjunction between aspirations and expectations—a central concept of the theory—did not 98 significantly predict delinquency. Third, only a limited number of respondents reported aspirations which exceeded their expectations—the situation that would most likely result in feelings of strain. Based on these findings, Hirschi concluded that classic strain theory does not significantly predict delinquency.
Another study performed by Farnworth and Leiber (1989) attempted to address one of the more notable limitations of Hirschi’s (1969) study by focusing on economic (as opposed to educational) aspirations and educational expectations. The results of the study revealed that a disjunction between economic aspirations and educational expectations was strongly associated with both serious and minor delinquency, providing direct support for classic strain theory.
However, this overall pattern of findings has not been supported in the overall literature examining Merton’s strain theory. Burton and Cullen (1992) reviewed approximately 50 studies published between 1963 and 1991 testing classic strain theory. The majority of these studies operationalized strain as a disjunction between aspirations and expectations, but also failed to find a significant association between strain (operationalized in this way) and delinquency. In this way, the majority of the extant literature—at least during the time period examined—failed to provide support for classic strain theory. However, Burton and Cullen recognize that an alternative operationalization of strain relates more directly to individual perceptions of blocked opportunities, a definition that more closely resembles Merton’s original conceptualization of strain (Kubrin, Stucky, & Krohn, 2009). While the majority of studies included in the review did not operationalize strain in this way, approximately 60 percent of the studies that did report significant associations between strain and delinquency operationalized strain as the presence of blocked opportunities.
99
Subsequent studies have also advocated for the operationalization of strain as relative deprivation, which refers to comparisons between one’s own standing and others (Burton &
Dunaway, 1994). Findings flowing from this limited line of research have revealed significant and positive associations between relative deprivation and various forms of criminal behavior.
In an effort to more directly examine the direct implications of various operationalizations of strain, Burton and colleagues (1994) utilized three distinct measurement strategies: perceptions of blocked opportunities; a disjunction between aspirations and expectations; and relative deprivation. The results of the study revealed significant bivariate associations between all three operationalizations of strain. However, subsequent analyses involved the inclusion of variables related to competing theoretical perspectives including social control and social learning theories into the multivariate equations. The results of this step in the analysis failed to identify any significant associations between any of the measures of strain and criminal behavior, indicating that competing theoretical perspectives are better suited for explaining variability in criminal behavior than classic strain theory.
A handful of subsequent studies have attempted to offer a more refined measurement strategy aimed at tapping strain as originally proposed by Merton (for example see Agnew,
Cullen, Burton, Evans, & Dunaway, 1996; Wright, Cullen, Agnew, & Brezina, 2001). In an effort to examine the current empirical status of macro-level theories and predictors of criminal behavior, Pratt and Cullen (2005) performed a meta-analysis of over 214 articles containing over
500 statistical models and nearly 2,000 effect size estimates. The authors identified a total of seven macro-level theoretical perspectives, one of which was classic strain theory (referred to as anomie/strain theory), which have been subjected to empirical investigation in the extant literature. Despite previous emphasis on both the macro- and individual-level aspects of
100
Merton’s strain theory (Messner, 1988; see also Baumer, 2007), Pratt and Cullen note that “[f]ew direct tests of anomie/strain theory, however, exist at the macro level” (p. 392). While the authors conclude that the results of the meta-analysis provide moderate support for classic strain theory, they also acknowledge that “anomie/strain theory has not been adequately tested to confirm its empirical status” (p. 410).
Taken as a whole, the current empirical status of classic strain theory is surprisingly underdeveloped and mixed at best. Studies have yet to attempt to test both the individual and structural aspects of the theory within a single study, despite Baumer’s (2007) comprehensive proposal of classic strain theory as a multilevel theory. In addition, with only one exception
(Menard, 1995), no studies consider the potential role that adaptations specified by Merton play in the development of criminal behavior. Adaptations to strain occupy a critically important feature of classic strain theory, making it quite surprising that the vast majority of studies examining the perspective have completely overlooked this particular facet of the theory. Also related to the adaptation to strain, Merton failed to specify the mechanisms and/or processes which account for variability in adaptation selection. In other words, the processes that ultimately result in conformity over innovation are not specified or even alluded to. Subsequent research has yet to offer any additional insight in this regard either.
3.4 Social Bonding Theory
Perhaps the most influential, in terms of both empirical and theoretical attention, theoretical perspective within the field of criminology is the social control perspective. While competing assumptions may underlie the two perspectives, both social control theories and strain theories stem from Durkheim’s (1951 [1857]) argument that human beings possess unlimited appetites or desires. Strain theories—such as Merton’s (1938) classic strain theory—build upon
101 this argument by cautioning against the potentially anomic consequences which accompany over-emphasizing goals over conventional means within modern industrialized societies. Social control theories emphasize the importance of social controls aimed at limiting individual’s appetites through socialization practices all in the better interest of both the social as a whole and the individuals who comprise the social. These overall goals stem directly from Durkheim’s
(1951 [1857]) original propositions which indicate that more integrated societies require some form of control in an effort to prevent the development of anomie. While social control and strain theories interpret Durkheim’s observations quite differently and concentrate such observations into vastly different theoretical perspectives, both sets of theories stem from
Durkheim’s assumptions regarding human nature.
Based on these underlying assumptions regarding human nature, social control theories operate under the assumption that a loosening of social controls will result in increases in criminal and antisocial behavior. In other words, human beings are motivated to offend and will do so unless the proper social controls are implemented and enforced. This particular assumption runs directly counter to many of the other mainstream theoretical perspectives within the field of criminology including social learning theories and strain theories (Brown et al.,
2013). In conjunction with these assumptions, some of the main proponents of the social control perspective have advocated for a fundamental shift in the way criminologists examine offenders.
More specifically, rather than attempting to understand the underlying mechanisms which ultimately result in criminal and delinquent behavior, advocates of the social control perspective argue that understanding the mechanisms and processes that prevent criminal and delinquent behaviors is far more pertinent. For example, Hirschi (1969) noted that “[t]he question ‘Why do they do it?’ is simply not the question the theory is designed to answer. The question is ‘Why
102 don’t we do it?’” (p. 34). Since all humans are believed to be motivated to engage in criminal behavior under the proper social conditions, Hirschi—and his predecessors—argue that attempting to specify the processes that underlie such behaviors is a meaningless endeavor.
Rather, attempting to understand why some individuals actually refrain from criminal or delinquent behaviors is a far more interesting question.
Perhaps the most influential and well-known social control theory was proposed by
Travis Hirschi (1969) in his now classic and noteworthy book “Causes of Delinquency.” While several notable control theories were proposed during the 1950s and 1960s (Matza, 1964; Nye,
1958; Reckless, 1961; Reiss, 1951; Sykes & Matza, 1957; Toby, 1957), Hirschi’s social bonding theory quickly became the most popular and, arguably, the most impactful social control theory.
The overwhelming popularity of social bonding theory is likely the direct result of two contributing factors. First, Hirschi proposed a concise, parsimonious, and coherent theory which incorporated elements from previous control theories. In this way, social bonding theory was a digestible and easily comprehended theory which summarized and built upon the theoretical foundation set by earlier theories. Second, Hirschi not only specified the assumptions and concepts of social bonding theory, he also offered an empirical operationalization for each concept. In addition, he also performed an empirical test of social bonding theory using a sample of adolescents from California. The manner in which Hirschi initially proposed social bonding theory not only made a convincing argument in favor of the theory, but also set the standard for criminological research in the subsequent decades (Akers & Sellers, 2013). Below, a discussion of social bonding theory will be offered in addition to a summary of the overall findings from the empirical literature focused on testing the theory.
103
3.4.1 Key Assumptions and Theoretical Concepts of Social Bonding Theory
As previously mentioned, some of the more attractive aspects of social bonding theory are the clarity with which the theory is proposed and the underlying simplicity of the concepts and processes included. Social bonding theory proposes that individuals who refrain from criminal and delinquent behaviors are more strongly bonded to society relative to individuals who give into their natural tendency to offend. When social bonds are weakened or broken, individuals are significantly more likely to engage in delinquent or criminal behaviors. Hirschi
(1969) further specified that individuals are not bonded to some larger latent entity known collectively as the social, but rather, individuals are bonded to specific socializing agents within the social such as peers, parents, teachers, and family members. The greater the extent to which social bonds between the individual other socializing agents are weakened, the greater the likelihood that the individual will engage in criminal or antisocial behavior. Hirschi proposed that the social bond was comprised of four separate, yet related, elements: attachment; commitment; involvement; and belief. While each of the four elements of the social bond are viewed separately, they are also dependent upon one another wherein the weakening of one element of the social bond will likely result in the weakening of other elements.
The first element of the social bond, attachment, refers to the affection that characterizes a relationship between people and is viewed as the most important social bond. Higher levels of attachment are inversely related to delinquency since this element of the social bond takes into account the admiration and identification with others who hold specific behavioral expectations and opinions. Lower levels of attachment would constitute a greater level of insensitivity regarding other’s opinions, which in turn decreases the constraints on behaviors that are viewed as unconventional or delinquent (Akers & Sellers, 2013). Hirschi (1969) also argued that concepts previously implicated in earlier control theories—such as self-control, internal control, 104 and indirect control—are actually encompassed under the concept of attachment preventing the need for further conceptualization. While Hirschi emphasized parental attachment in his proposition of social bonding theory, he also acknowledges the importance of attachment with other important socializing agents including peers. Along the same lines, Hirschi argued that the specific socializing agents to whom one is attached to does not really matter. The most important aspect is that the relationship is characterized by a strong level of attachment, which ultimately contributes to the overall social bond and diminishes the likelihood of offending.
Subsequently, strong levels of attachment to delinquent peers or even antisocial parents are expected to result in decreases in the likelihood of offending since individuals are attached to the person not necessarily their character (Akers & Sellers, 2013).
The second element of the social bond is commitment and is directly related to Toby’s
(1957) stakes in conformity. More specifically, commitment refers to the investment in conventionality that is potentially at risk by engaging in criminal or delinquent behaviors. The more committed an individual is to conventionality, the more they have to lose by engaging in delinquent behaviors. For example, individuals may lose their job, damage their relationship with a significant other, or diminish their overall social standing by engaging in criminal or delinquent behaviors. The potential risk of losing previously acquired social investments is much greater for individuals with greater levels of such investments, resulting in greater levels of commitment to conformity. Alternatively, individuals with lower overall levels of social investment have far less to lose and, in turn, are far less committed to conformity. Importantly, this particular aspect of social bonding theory implicates the principles of the rational choice perspective in that human beings are largely rational and will be more likely to refrain from
105 criminal activity when the potential costs far outweigh the potential benefits (Akers & Sellers,
2013).
Involvement is the third element of the social bond, and simply refers to being too occupied with conventional activities to engage in unconventional activities. For example, conventional activities such as spending time with family, studying, and participation in community activities may comprise the majority of one’s attention, leaving little time for delinquent or criminal behavior. The fourth and final element of the social bond is belief, which
Hirschi (1969) described as the endorsement of conventional rules, values, and beliefs, with the most important being the belief that laws and social rules represent moral standards and should be obeyed. Importantly, individuals do not have to necessarily agree with each and every law or rule implemented within a society, but rather, they should generally accept the underlying belief that laws and rules in general are meaningful and valid. In this way, individuals who no longer hold such beliefs are expected to be more loosely bonded to society and are more likely to engage in criminal activity.
3.4.2 The Empirical Status of Social Bonding Theory
Since its formal introduction in 1969, Hirschi’s social bonding theory “has been one of the most, if not the most, tested theories in criminology” (Lilly et al., 2011, p. 119; emphasis in original). There are likely two specific factors accounting for the overwhelming amount of empirical attention that social bonding theory has received. First, Hirschi deliberately proposed a theory that could be easily tested with secondary data by proposing concepts and processes that were clearly specified and explicit. Hirschi (1969) recognized the overly complex nature of many of the existing theoretical perspectives and the problems that accompany attempting to empirically test such complex theories. Based on these assessments, Hirschi sought to propose a
106 concise and parsimonious theory that was amenable to empirical assessment. Second, Hirschi offered a direct operationalization of the central concepts in social bonding theory in his initial proposal of the theory. This particular practice offered researchers a unique opportunity to test various aspects of social bonding theory using the proper operationalization of the primary theoretical concepts as intended by Hirschi. Taken together, these aspects of social bonding theory, and the manner in which it was proposed, have culminated into an impressive body of research.
The majority of the studies examining social bonding theory have attempted to directly assess the predictive ability of each of the four elements of the social bond (i.e., attachment, commitment, involvement, and belief). Several early tests of social bonding theory attempted to create a general measure of the social bond comprised of individual items tapping each of the four elements of the bond (see Krohn & Massey, 1980). The overall pattern of findings stemming from this line of research indicates that the social bond is significantly and negatively associated with delinquency, but the effect sizes were largest when predicting minor forms of delinquency. Subsequent studies have advocated for the examination of each individual element of the social bond in an effort to provide a more thorough understanding of the mechanisms which underlie the association between the social bond and subsequent criminal or delinquent behavior (see Wiatrowski, Griswold, & Roberts, 1981). Studies have found significant associations between various combinations of the elements of the social bond and a wide range of antisocial and criminal outcomes including white collar crime (Lasley, 1988), marijuana use in adolescence (Akers & Cochran, 1985), and other forms of delinquency (Rankin & Kern,
1994).
107
In a systematic review of the literature, Kempf (1993) analyzed a total of 71 studies conducted between 1970 and 1991 examining social bonding theory. Unfortunately, the review did not yield an easily interpreted pattern of results. While Kempf concluded that the overall findings from the analysis “professed at least conditional support for control theory” (1993, p.
167), much of the literature suffered from significant methodological limitations making it difficult to assess the overall pattern of findings more directly. The results revealed a diverse set of measurement strategies aimed at tapping the four elements of the social bond as specified by
Hirschi (1969). This variability in the operationalization of the social bond makes it difficult to generalize findings across studies, particularly since some studies offer much stronger measures while other studies offer much weaker measures. In addition, the majority of studies failed to include measures that tapped all four elements of the bond making it difficult to draw conclusions regarding the predictive ability of the theory as a whole. Studies conducted after
Kempf’s review have provided additional support for social bonding theory (Costello & Vowell,
1999; Rankin & Kern, 1994; Wright, Caspi, Moffitt, & Silva, 1999). In addition, a recent meta- analysis revealed a significant and negative association between parental attachment and delinquency, with an overall mean effect size of r = .18 (Hoeve et al., 2012).
The vast majority of the early studies examining various aspects of social bonding theory were cross-sectional, only examining associations between measures of the elements of the social bond and antisocial outcomes at the same time. Some critics argued that longitudinal studies of social bonding theory were necessary to effectively establish causal order (see Brown et al., 2012). Agnew (1985) was one of the first criminologists to test social bonding theory within the confines of a longitudinal study. Prior to performing the longitudinal analysis, Agnew examined cross-sectional association between eight items used to tap the social bond and
108 delinquency. The results revealed significant and moderately sized effects on the delinquency measures examined. Agnew then examined the potential effect of the social bond measures on the change in delinquency between two time points. The results of the longitudinal models yielded dramatically different results with only a limited number of significant associations between the included social bond measures and delinquency. In addition, the overall effect sizes had dropped significantly wherein the included social control measures only explained between 1 and 2 percent of the variance in changes in delinquency between the two time points.
Subsequent longitudinal analyses have yielded similar results suggesting that the association between the social bond and subsequent delinquency may not be causal (Agnew, 1991; Burkett
& Warren, 1987; Elliott, Huizinga, & Ageton, 1985; Massey & Krohn, 1986; McCarthy & Hoge,
1984; Paternoster & Iovanni, 1986). Based on the pattern of findings stemming from his analysis, Agnew (1985) concluded that “cross-sectional studies have greatly exaggerated the importance of Hirschi’s social control theory” (p. 58).
A complementary line of research has also attempted to better specify the mechanisms which ultimately result in each of the four elements of the bond. Arguably, the potential role of peers in the development and maintenance of social bonds has received the most attention within the extant literature (see Kubrin et al., 2009 for a more comprehensive overview). Recall that
Hirschi argued that stronger social bonds should be inversely related to delinquency. In this way, individuals who are strongly bonded to peers should be less likely to engage in criminal or delinquent behavior, even if their peers are themselves delinquent. This overall pattern has not been borne out in the literature. Rather, a substantial line of research has consistently revealed the exact opposite pattern of findings: individuals who are more strongly attached to delinquent peers have been found to be significantly more likely to engage in delinquent and criminal
109 behavior (Krohn, Thornberry, Rivera, & LeBlanc, 2001; Warr, 2002). In addition, after accounting for peer influence, studies tend to report attenuated or nonsignificant associations between social bond measures and delinquency (Agnew, 1991; Marcos, Bahr, & Johnson, 1986;
Massey & Krohn, 1986). Taken together, this line of research indicates that Hirschi’s (1969) original conceptualization of social bonding theory is, at the very least, incomplete and requires modification in order to effectively explain variation in delinquent and criminal behavior.
Directly related to studies examining the potential role of peers within social bonding theory, several studies have attempted to identify the specific sources to which one becomes bonded. Several studies have implicated factors directly related to parenting such as being attached to both parents (Rankin & Kern, 1994) and parental supervision (Simons, Simons, Burt,
Brody, & Cutrona, 2005). Additional studies have also found that higher levels of school attachment and commitment result in lower levels of antisocial behavior (Cernkovich &
Giordano, 1992; Stewart, 2003). Akers and Sellers (2013) make the argument that “[a]dherence to religious practices clearly indicates commitment to conventionality, involvement in conventional activities, and attachment to others” (p. 121). A significant number of studies have identified a negative association between attachment to religious beliefs and delinquency, providing evidence in favor of religion as a potential source of attachment (Cochran & Akers,
1989; Evans, Cullen, Dunaway, & Burton, 1995; Welch, Tittle, & Grasmick, 2006). In addition, the results of a meta-analysis conducted by Baier and Wright (2001) revealed moderate and negative associations between religiosity and delinquency. A series of more recent reviews of nearly 300 studies revealed similar findings (Johnson, De Li, Larson, & McCullough, 2000;
Johnson & Jang, 2010; Johnson, 2011).
110
3.5 Summary and Discussion
The four theoretical perspectives presented in this chapter constitute some of the most influential and well-known explanations of criminal and delinquent behavior within the field of criminology. With that said, each of these perspectives have encouraged the development of a vast and impressive literature aimed at empirically analyzing the predictive ability of each theory or attempting to further specify aspects of each perspective. Each individual perspective has been subjected to a remarkable number of empirical tests with some studies finding support for each perspective and other studies failing to find support. Throughout all of the iterations of statistical models, the swapping of independent and dependent variables, the innovation in measurement strategies, and the analysis of new and innovative samples, the overall findings stemming from the literature are decidedly mixed. While some of the reviewed perspectives have received much more empirical support (i.e., social learning theory) than others (i.e., rational choice and classic strain theories), all of the perspectives have received at least some support and remain relevant within the field of criminology currently. In addition, each of the presented theoretical perspectives remain staples in the most popular criminological theory textbooks and are taught in virtually every criminology course to date.
This overall pattern of findings raises two important questions regarding the current landscape of the field of criminology. First, what is the primary source of the overall lack of consensus regarding each of the reviewed theories? In other words, what factors account for the lack of consensus in findings stemming from the voluminous literature surrounding each perspective? Despite the impressive number of studies examining each of these perspectives, subsequent theoretical modifications, and an overall methodological sophistication, the overall explanatory power of each perspective has not significantly increased over the past few decades
(Weisburd & Piquero, 2008). The overall stagnation in theoretical progress can only be 111 contributed to one of two sources: (1) our methodologies or (2) the theories themselves. Second, what potential role does the fixation on purely social explanations of antisocial behavior play in the mixture of findings present in the extant literature? Despite the presence of a directly relevant, methodological advanced, and highly developed literature (reviewed in Chapter 2) directly implicating the role of biological factors in the development of antisocial behaviors, criminologists have yet to draw from this line of research. Whether this oversight is directly related to the lack of consensus among findings remains an open question at this time. However, it remains fully possible that this overall preference for purely social explanations of antisocial behavior is directly tied to the mixed empirical findings and the overall limited explanatory power of the leading theoretical perspectives within the field of criminology.
Observing empirical evidence both in favor of and opposing existing conventional criminological theories is not unique to the overview and discussion presented in this project.
While previous researchers have also identified this overall pattern of findings, there is lack of consensus regarding the subsequent actions that should be taken (Bernard & Snipes, 1996; Elliot et al., 1985; Hirschi, 1979). This particular issue remains unresolved currently; however, two competing perspectives have been offered in the extant literature. First, Hirschi (1979) advocated for the falsification and rejection of theories that do not perform well. Put simply, theories that do not adequately explain variation in criminal and delinquent behaviors are of limited utility and should be swept into the dustbin of criminology. Hirschi argued that theoretical integration is largely unnecessary and, perhaps more importantly, not possible since different theoretical perspectives offer competing incompatible assumptions about human nature.
On the other side of the debate, Benard and Snipes (1996; see also Elliott et al., 1985) argued that theoretical integration is a far more effective and feasible approach than falsification. Since
112 all theories explain at least some variance in criminal behavior some of the time, falsification, at least in the most literal sense, becomes largely impossible. In addition, Bernard and Snipes observed that focusing on concepts, and the relationships between concepts, that significantly explain variation in criminal behavior is a much more effective approach than pitting one theory against another. In this way, concepts do not belong to specific theories and can be combined in an effort to explain more variance in criminal behavior.
Theoretical integration has a storied past within criminology (see Bernard & Snipes,
1996), but much of the attention attributed to integration has focused only on purely social concepts specified in theories which were created under the untenable assumption that social influences on behavior work in isolation. While a few notable exceptions exist (Ellis, 2005;
Fishbein, 1990; Robinson & Beaver, 2009; Walsh, 2000, 2002), integration involving concepts and empirical findings stemming from the biosocial perspective is a task that has yet to be attempted. In addition, the few examples that do currently exist either suffer from taking far too broad of an approach (Ellis, 2005; Robinson & Beaver, 2009), or from attempting to integrate concepts that are potentially incompatible with existing theoretical concepts (Fishbein, 1990;
Walsh, 2000). In either case, additional attention regarding the potential integration of existing theories, or at the very least the concepts specified within such theories, into the biosocial perspective is necessary since this type of integration can potentially boost the explanatory power of theories aimed at explaining antisocial behaviors. The next chapter will provide a brief overview of the biosocial integration models that have been offered in previous studies and then offer a new and unique integration strategy.
113
CHAPTER 4
SITUATING EXISTING CRIMINOLOGY THEORIES WITHIN THE BIOSOCIAL PERSPECTIVE
As more biologically informed research appears, the theoretical predictions of the sociological paradigm will be revealed as limited, if not misspecified in important ways. Eventually, it will become commonplace to ask: How can any theory that ignores the human body be complete?
-Francis T. Cullen (2009, p. xvi)
The abundance of criminological theories that have been proposed over the preceding decades is quite staggering and directly contributes to a significant problem within the field of criminology—the proper way to handle to overwhelming number of theories that have and continue to be introduced. The problem is further compounded by the large number of studies, producing conflicting results, which examine each individual theory. Criminologists have observed that the overwhelming number of theories within the field of criminology impedes scientific progress by flooding the literature with research attempting to better specify concepts within a perspective which, even taken as a whole, only explains a minimal amount of variation in criminal behavior (Bernard & Snipes, 1996).
While there seems to be a consensus among criminologists that there are simply too many theories within criminology, and this oversaturation is harmful to empirical progress within the field, there is far less consensus regarding the proper way to address this problem. This particular issue ultimately resulted in a now classic debate within the literature with some criminologists advocating for the falsification of theories (Hirschi, 1979, 1989) and others calling for theoretical integration in an effort to reduce the overall number of theories (Bernard &
114
Snipes, 1996; Elliott et al., 1985; Liska et al., 1989; Thornberry, 1989). Travis Hirshi (1979,
1989) was the most adamant proponent of the falsification process in which one theory should be pitted against another and the one that explains more variance within the outcome of interest should be retained and the other should be discarded. Hirschi argued that this is the only possible way to decrease the overall number of theories within criminology since theories tend to be contradictory with one another, not just in regard to the concepts they emphasize but also in regard to the processes they specify. Proponents of theoretical integration, most notably Elliott and colleagues (1985), argued that theoretical perspectives within criminology are not contradictory in nature and can be integrated in an effort to emphasize those concepts and processes which best explain variation in criminal behavior. In this way, the concepts and variables specified within theories should be independently evaluated and the concepts which explain the greatest proportion of variance in criminal behavior should be retained and integrated into a separate theory.
“Theoretical integration” is a fairly broad term that may encompass any of a number of different processes aimed at combining various aspects of multiple theories into a single, larger theory with greater explanatory power. Importantly, Liska and colleagues (1989) specified various forms of theoretical integration directly relevant to criminological theory. Two primary forms of theoretical integration are possible: propositional and conceptual integration.
Propositional integration involves linking two or more separate theories, but retaining the concepts and processes contained within each individual theory. Theories are ultimately combined into a single larger theory using one of three different strategies based on the individual characteristics of the theories. First, end-to-end integration is developmental in the processes and associations specified in one theory precede those that occur within the other. In
115 this way, this form of propositional integration attempts to impose a causal order among theoretical processes by utilizing the dependent variable specified in one theory as the independent variable in the next. Second, side-by-side integration focuses on integrating theories which explain variation within related outcomes. For example, theories aimed at explaining various forms of delinquency in adolescence, may also explain minor forms of criminal behavior in adulthood. Third, up-and-down integration focuses on conceptually broadening one theory such that it encompasses other perspectives and the concepts contained within them. This form of integration is relatively rare since most theories cannot be broadened conceptually.
Conceptual integration is less concerned with retaining the processes and concepts specified within each theory and more focused on identifying conceptual overlap between separate theories. In this way, conceptual integration focuses on integrating theories which contain similar concepts and attempt to explain similar behavioral outcomes. Liska et al. (1989) caution that this form of theoretical integration may have limited utility and may simply result in a modification of the underlying meanings of previously specified theoretical concepts.
However, conceptual integration is an important first step toward the ultimate goal of propositional integration. An important prerequisite for any form of propositional integration is the identification of overall patterns present within theories and their underlying concepts. Once these patterns have been identified, it is only then possible to integrate individual theories into a more unified and more powerful theory using one of the integration strategies specified above.
Another form of theoretical integration was initially proposed by Thornberry (1989) and appears to be an alternative to more formal forms of theoretical integration. More specifically,
Thornberry advocated for theory elaboration which involves the logical extension of a single theory in an attempt to enhance its overall predictive ability. Theory elaboration essentially
116 specifies the process by which theorists perform various theoretical and conceptual modifications directly in line with advances within the extant literature. The process specified by Thornberry very closely resembles the process which is carried out within the field of criminology currently:
(1) a theoretical perspective is formally presented; (2) limitations of the theory are noted; (3) various empirical tests are performed in an effort to quantify the validity of the concepts and processes included in the theory; (4) specific theoretical and conceptual modifications are offered based on the results of the empirical tests carried out in the previous step; (5) the proposed modifications are either carried out or a rebuttal against each modification is offered; (6) the modified theory is once again subjected to empirical tests and further modifications are offered.
In this way, the overall advantage of theoretical elaboration as specified by Thornberry is not clearly specified. In addition, as observed by Bernard and Snipes (1996), “[r]egardless of one’s orientation, elaboration logically leads to integration” (p. 309).
Despite numerous calls for theoretical integration into the biosocial perspective—or more biologically-informed theories—very few criminologists have attempted to undertake this particular task (Fishbein, 1990). The few attempts to more closely integrate the biosocial perspective and existing criminological theory can be divided into two separate categories. First, some criminologists have attempted to present biosocial theories of antisocial behavior (Ellis,
2005; Robinson & Beaver, 2009; Vila, 1994). Such theories attempt to combine concepts and empirical findings from multiple academic fields including criminology, biology, evolutionary psychology, psychiatry, neuroscience, and molecular genetics. Second, two formal attempts have been made to integrate biological and traditional criminological concepts (Walsh, 2000,
2002). The primary goal of these integrative efforts was to properly situate biological concepts within or alongside existing criminological theories and concepts. All of these previous attempts
117 represent a step in the correct direction, but they also suffer from several important limitations.
This chapter will provide an overview of previous biosocial integration strategies in an effort to specify the contributions made by each previous attempt, but also to highlight the limitations of each previous attempt. Following this discussion, an innovative biosocial integration model focused on integrating existing criminological theory within the biosocial perspective will be presented.
4.1 Precursors to the Biosocial Integration Model
Based largely on the empirical findings presented in Chapter 2, biosocial criminologists have criticized the general practice of ignoring the potential importance of considering the role of biological and genetic contributions to antisocial behavior (for an example see Wright & Cullen,
2012). Meanwhile, conventional criminologists have leveed a series of criticisms against biosocial research including the dangers of determinism, a lack of plausible policy implications, and generalizability issues stemming from analyzing samples of twin and sibling pairs (Akers &
Sellers, 2013; Kubrin et al., 2009; Lilly et al., 2011). While these specific concerns have been directly addressed by a large number of researchers working across a wide range of academic disciplines (Beaver, 2013a; Pinker, 2002; Ridley, 1999; Rowe, 1994; Wright, 1994), conventional criminologists continue to advance new critiques of the biosocial perspective.
Perhaps the most prevalent and consistent of these criticisms is the overall lack of biosocial theories of criminal behavior. For example, Kubrin and colleagues (2009) argue that
“heritability and genetic anomaly studies can only establish whether biology plays a role in determining behavior. They cannot explain how biology influences criminal behavior”
(emphasis in original; p. 50). While this statement is highly debatable, the recognition of an overall lack of biosocial theories of antisocial behavior is a valid one.
118
The overall lack of theoretical development within the biosocial perspective has been noted for quite some time and is widely acknowledged by both conventional and biosocial criminologists (Agnew, 2011; Ellis, 2005; Fishbein, 1990; Robinson & Beaver, 2009; Vila,
1994; Walsh, 2000, 2002). Efforts to address this particular criticism have only been attempted by a relatively small number of biosocial criminologists and can be divided into two separate categories. First, some biosocial criminologists have attempted to develop biosocial theories of criminal and antisocial behavior which take into account all of the known biological and environmental influences and risk factors which have been found to be associated with antisocial and maladaptive behavioral patterns (Ellis, 2005; Robinson & Beaver, 2009). Second, other biosocial criminologists have taken a completely different approach and have advocated for various forms of theoretical integration, typically involving existing conventional criminological theories (Fishbein, 1990; Walsh, 2000, 2002; Vila, 1994). Each of these two approaches will be detailed below.
4.1.1 Biosocial Theories of Antisocial Behavior
Many mainstream criminology textbooks refer to “biosocial theories” of criminal and antisocial behavior (for examples see Akers & Sellers, 2013; Bernard, Snipes, & Gerould, 2010;
Kubrin et al., 2009). Kubrin and colleagues (2009) even go as far as to proclaim that “[a] wide variety of biosocial theories that suggest specific ways that biology influences criminality have been developed” (p. 50). The “biosocial theories” discussed extensively in such texts are not formally-presented and empirically testable explanations of antisocial behavior, rather, the majority of these “theories” are simply empirical findings stemming from various lines of research. More specifically, the findings mistaken for theories are the primary tenets or concepts that are organized by the biosocial perspective.
119
To date, only two biosocial theories of antisocial and criminal behavior have been proposed—evolutionary neuroandrogenic theory (ENA; Ellis, 2005) and Integrated Systems
Theory (IST; Robinson & Beaver, 2009). While both theories are distinct from one another, they closely resemble one another and emphasize many of the same concepts and processes. Broadly, both theories attempt to integrate empirical findings from both conventional criminology and the biosocial perspective in an effort to propose a theory with maximum explanatory power. In addition, both theories attempt to integrate concepts from conventional theoretical perspectives.
For example, IST directly integrates concepts from several theories including labeling, strain, social learning, and social disorganization theories (see pp. 374-380). Both theories also emphasize findings from several academic disciplines which more directly align with the biosocial perspective such as evolutionary psychology, molecular genetics, psychiatry, developmental psychology, and neuroscience. While IST more directly integrates findings from neuroscience research, ENA more directly emphasizes findings from evolutionary psychology.
For example, the first main proposition of ENA which states “that aggressive and acquisitive criminal behaviour evolved as an aspect human reproduction, especially among males” (Ellis,
2005, p. 288).
Both theories attempt to conceptually integrate empirical findings, concepts, and entire theoretical perspectives into a singular biosocial theory of human behavior. Figure 4.1 offers a visual representation of the type of integration that both theories attempt to perform. The circle represents the entire biosocial perspective which encompasses several other perspectives and disciplines including behavior genetics, molecular genetics, and evolutionary psychology (see
Barnes, 2014 for a more detailed discussion). The biosocial perspective also encompasses all of the empirical studies which directly implicate both biological and environmental influences in
120 the etiological processes which ultimately result in antisocial behaviors. The rectangles within the circle represent all of the concepts and empirical findings stemming not only from biosocial research, but also from conventional criminological studies since both theories attempt to integrate theories and findings from both sets of literature. On its face, this form of theoretical integration seems to be quite successful in that all of the theoretical concepts specified in both conventional criminological and biosocial research are neatly compacted within the biosocial perspective. However, the processes which underlie this particular form of theoretical integration are highly complex and difficult, if not wholly impossible, to test empirically. Both theories describe a highly interconnected and complex relationship not just between the theoretical concepts and antisocial behavior, but also between the concepts themselves. Such relationships involve multiple interactions between various concepts in addition to potential two- way causal relationships between two or more concepts. Robinson and Beaver also note the importance of taking into account individual-level variation in exposure to various risk factors based on the frequency, regularity, intensity, and priority of exposure (p. 380).
While both attempts should be lauded for their ambition, novelty, and underlying creativity, it is highly unlikely that either theory can be assessed empirically and breaking each theory into smaller, more digestible pieces would simply result in rehashing the previous research on which each theoretical concept is built. In this way, both theories tend to be examples of Thornberry’s (1989) alternative integration approach of theoretical elaboration
“taken to its logical extreme” (Bernard & Snipes, 1996). Put more simply, the theories are so broad and attempt to take into account so many concepts from such a diverse set of sources that the overall conclusion is that “everything causes everything,” which is not all that useful when attempting to specify the underlying mechanisms which ultimately explain variation in antisocial
121 behavior. Based on these limitations, it is not surprising that neither of these biosocial theories have been subjected to empirical tests.
Directly in line with the comprehensive biosocial theories proposed by Ellis (2005) and
Robinson and Beaver (2009) is Vila’s (1994) general paradigm for understanding criminal behavior which is described as an extension of Cohen and Machalek’s (1988) theory of evolutionary ecology. Vila does not describe his approach as a theory, but rather as a paradigm in which new, interdisciplinary, and more powerful theories can be developed. The general paradigm is grounded in three levels of measurement: ecological; macro-level; and micro-level.
Vila argues that any useful theory must be ecological, integrate concepts from various perspectives, and developmental. Biological factors are argued to play a significant role at the micro-level, but Vila also stresses the importance of examining interactions between factors stemming from each of the three levels of measurement. While Vila’s general paradigm does not represent a theory in the formal sense, it is directly related to the comprehensive biosocial theories described in this section. Since Vila does not propose a formal theory, there are no concepts or processes which can be subjected to an empirical test. Rather, theories developed within the paradigm can be tested.
4.1.2 Preliminary Models of Biosocial Integration
One of the main driving forces behind the movement toward biosocial theories of criminal and antisocial behavior is the limited explanatory power of conventional theoretical perspectives (Weisburd & Piquero, 2008). This observation is one that has been made by both biosocial and conventional criminologists and has been one of the primary factors motivating the call for theoretical integration (see Bernard & Snipes, 1996 for an overview). To date, Bernard and Snipes (1996; see also Bernard et al., 2010) have proposed the most comprehensive and
122 versatile integration strategy. This particular strategy involves five points of consideration.
First, Bernard & Snipes advise criminologists to shift their attention from whole theories to the concepts which comprise each theory. Rather than focusing on entire theories, and all of the concepts and processes which they organize, Bernard and Snipes suggest that criminologists should focus only on those concepts and processes which best predict criminal behavior, regardless of the perspective they are organized by. Second, theories should be assessed based on their overall contribution to the scientific process rather than their overall validity. Third, rather than attempting to falsify theories, criminologists should work toward identifying risk factors with the most potent effects on criminal behavior. This process is far more welcoming toward the process of integration and will allow for more nuanced explanations of criminal behavior. Fourth, contradictory theories cannot be integrated, but few criminological theories are contradictory with one another allowing for broad integration. Fifth, theories should be interpreted and organized based on the policy implications that should directly stem from them.
Following these guidelines, Bernard and Snipes argued that all existing conventional theories of criminal behavior could be divided into one of two broad categories: individual difference theories or structure/process theories.
While the integration strategy proposed by Bernard and Snipes (1996) was focused almost exclusively on conventional criminological theories, some of the elements of their strategy have been used in an attempt to incorporate elements of the biosocial perspective into conventional criminological theories. For example, several biosocial criminologists have argued that the field of criminology does not need standalone biosocial theories of criminal behavior, rather, biosocial concepts should simply be integrated into existing theories in an effort to increase their explanatory power (Fishbein, 1990; Walsh, 2000, 2002). This particular strategy is
123 directly related to Bernard and Snipes’ suggestion of shifting attention away from existing theories and instead focusing on concepts and processes which best explain criminal behaviors.
In this way, concepts drawn from both the biosocial perspective as well as conventional theories of crime can be combined into a more comprehensive and effective explanatory model. Walsh
(2000, 2002) has perhaps been the most well-known advocate of this form of biosocial integration and has explicitly demonstrated the manner in which biosocial concepts can be directly integrated into several conventional theories of crime.
Figure 4.2 provides a visual representation of the integrative strategy proposed by Walsh
(2000, 2002) and others. As represented by the single-headed arrows, the direction of theoretical integration runs from theoretical concepts and the biosocial perspective to each of the theories examined in the current project. Importantly, and directly in line with Bernard and Snipes’
(1996; Bernard et al., 2010) suggestions, theoretical concepts are allowed to vary freely and may differentially contribute to each of the four theories included. In this way, concepts are not directly tied to a single theory, but may be included in any or all of the theories presented. In addition, concepts stemming directly from the biosocial perspective are also allowed to contribute to any or all of the included theories. This particular integration strategy is appealing since it makes use of existing conventional theories, but fortifies them with biosocial concepts and research findings preventing the need to construct new theories from scratch. In addition, the integration strategy specified by Bernard and Snipes is still highly relevant, allowing for the further integration of any combination of conventional and biosocial concepts in an effort to maximize the explanatory power of any resulting theoretical models.
This form of biosocial integration possesses several attractive attributes and also directly addresses many of the concerns which accompany the comprehensive biosocial theories
124 discussed above. For example, this form of integration is far less complex than the comprehensive theoretical models previously proposed (Ellis, 2005; Robinson and Beaver,
2009). In addition, this form of integration also retains the aspects of criminological theory conventional criminologists are most familiar with, likely minimizing resistance among the more sociologically-minded. However, with these advantages also comes two important limitations.
First, this form of theoretical integration is directly reliant upon assumptions that, based on the literature summarized in Chapter 2, seem to be untenable. For example, in order to integrate biosocial concepts into existing theories one must inherently assume that the primary concepts, processes, and assumptions specified within each theory are directly in line with the biosocial perspective. Failing to satisfy this assumption would result in a nonsensical theory which offers competing explanations for the outcome of interest. For example, social learning theory is built on the underlying assumption that human beings have no predisposition toward crime or conformity and can effectively be described as “blank slates” (Agnew, 2011). This is a fundamental assumption of social learning theories which postulate that all behavior is learned and effectively transcribed onto each individual’s blank slate. This assumption directly contrasts with decades of research resulting in hundreds of individual studies which provide unequivocal evidence indicating that genetic influences differentially predispose individuals to different patterns of behavior (Mason & Frick, 1994; Miles & Carey, 1997; Moffitt, 2005; Rhee &
Waldman, 2002; Ferguson, 2010). This major discrepancy in even the most fundamental aspect of each perspective (the assumptions they make about human nature) suggests that any form of integration would not only be impossible, but would also result in a highly problematic theory which may actually detract from the overall scientific process.
125
The second limitation of this form of biosocial integration is that the underlying problems and issues stemming from the theories themselves are not addressed. Rather than refine or even discard portions of the existing theories that are ineffective or problematic, this form of integration would retain all aspects of the existing theories but also add an additional layer of complexity in the form of biosocial concepts. For example, as reviewed in Chapter 3, the extant literature seems to indicate that strain operationalized as the disconnect between expectations and payoff is an ineffective predictor of delinquent and criminal behavior (Burton & Cullen, 1992).
In his attempt to integrate the biosocial perspective into classic strain theory, Walsh (2000) intricately describes the biological processes which contribute to variation in response to strain as described by Merton (1938) and Agnew (1992). Walsh’s effort should be viewed as a critical turning point in the field of criminology and has significantly influenced the current project, but missing from his discussion is the absolution of the logical and empirical issues that underlie classic strain theory. In this way, the integration of biosocial concepts may increase the explanatory power of existing theories, but this is simply a function of the concepts included in the theory making the overall utility of the theory itself highly questionable. Simply integrating various aspects of the biosocial perspective does not address the other problematic aspects of the theory. In addition, the improved explanatory power stemming from the integration of biosocial concepts may actually detract from the overall scientific process since it may mask significant problems in the underlying logic or assumptions of each theory. While this form of integration may represent a step in the right direction, it still leaves many problems unaddressed.
4.2 The Biosocial Integration Model
The two general biosocial integration strategies utilized in the extant literature do not present a viable solution to the overall problem of limited explanatory power within existing
126 criminological theories. While neither approach presents a definitive solution to this problem, both strategies represent a step in the right direction and open the door for the development of a more effective integration strategy. Any biosocial integration strategy aiming to boost the explanatory power of existing criminological theories, while avoiding the limitations of the existing biosocial integration strategies, would have to meet four specific criteria: (1) allow biosocial and conventional theoretical concepts to exist in a side-by-side fashion; (2) allow for resulting theoretical perspectives and concepts to be subjected to empirical tests; (3) prevent problematic theoretical development stemming from the integration of concepts organized by theoretical perspectives which hold conflicting assumptions regarding human nature; and (4) allow for theoretical modification and refinement to more closely align with empirical findings.
This dissertation attempts to specify an integration strategy which meets each of these four criteria while also providing a general scaffolding for the integration of both existing and future criminological theories. In addition, various forms of integration can be performed at lower levels of theoretical development prior to taking the necessary steps for biosocial integration. This allows for theoretical refinement during the early stages of the integration process in an effort to maximize the explanatory power of each theory. Conversely, the biosocial integration model also allows for theoretical development directly within the biosocial perspective, however, such theories are held to the same standards as theories which are eventually integrated into the perspective. The biosocial integration model presented in the current dissertation also allows for the integration of both complex and simplistic theoretical perspectives into the biosocial perspective. In this way, the integration models specified by
Liska and colleagues (1989) and Bernard and Snipes (1996) are still relevant and can be performed in the lower levels of the biosocial integration model.
127
The biosocial integration model is presented in Figure 4.3. As can be seen from the figure, the direction of integration leads from existing criminological theory—and the concepts which comprise them—to the biosocial perspective. Importantly, the direction of integration suggested by Walsh (2000, 2002) and Fishbein (1990) is inverted in the biosocial integration model. In this way, the various issues which underlie each of the theoretical perspectives can be addressed at the lower levels of integration. For example, the integration process specified by
Bernard and Snipes (1996) can take place at the bottom levels of the integration model. After conceptual and logical issues have been addressed and theories have been refined to an acceptable level, they can then be integrated into the biosocial perspective. The biosocial integration process involves additional theoretical development, but also requires empirical evidence suggesting that the concepts and processes specified in the original theory significantly predict the outcome of interest within the confines of a genetically-informed analysis. In other words, theorists must ensure that the concepts and processes specified in their theories are not subject to genetic confounding and effectively isolate the association between concepts/processes and the outcome of interest. Theoretical processes and concepts that fail to predict antisocial outcomes within the confines of a genetically-informed analysis should either be refined further or discarded.
This form of integration is directly related to Akers’ integrative technique of conceptual absorption (Akers & Sellers, 2013), which refers to one theory engulfing another due to overlap in the concepts specified in each theory. In addition, this form of theoretical integration is also directly in line with up-and-down integration as discussed by Liska et al. (1989). This type of integration is possible in this instance since the biosocial perspective engulfs both biological and environmental influences on behavior, making it possible to situate theories of antisocial
128 behavior which are comprised of both sets of influences neatly within the perspective. In this way, the biosocial perspective is similar to Vila’s (1994) general paradigm but is far more developed due to the exponential increase in the amount of biosocial research performed during recent years. Due to the broad coverage of the biosocial perspective, virtually any theory can be integrated into the biosocial perspective regardless of whether such a theory emphasizes biological influences, environmental influences, or a combination of the two. The only caveat is that all theories must continue to explain a significant amount of variance in antisocial phenotypes after taking genetic influences into account. This particular aspect of the biosocial integration model directly addresses the limitations of earlier integration models. Namely, subjecting conventional theoretical models to genetically-informed tests prevents the possibility of including competing concepts within the same theoretical model.
4.3 Summary and Discussion
One of the most recent critiques of the biosocial perspective is the overall lack of theoretical development within the perspective and the atheoretical nature of the majority of the studies which comprise the perspective (see Kubrin et al., 2009 as an example). While this particular critique is not without merit, the biosocial perspective—at least in its modern form—is relatively young and has only recently developed into a perspective which can support theoretical development. This observation is evidenced by the lack of consensus among biosocial criminologists regarding the most appropriate strategies for developing biosocial theories. One camp of biosocial criminologists have advocated for large, interdisciplinary, and complex theories which attempt to explain all of the empirical findings presented within the biosocial perspective (Ellis, 2005; Robinson & Beaver, 2009; Vila, 1994). Others have argued that the development of novel theoretical perspectives is unnecessary and specific aspects of the
129 biosocial perspective should be directly integrated into existing conventional theories (Fishbein,
1990; Walsh, 2000, 2002). Both perspectives address limitations of the other, but also leave much to be desired by introducing concerning limitations of their own.
The biosocial integration model presented here is an attempt to directly address the limitations of the previously presented integration models. This model of theoretical integration draws directly from various aspects of the previous models, but also makes necessary modifications in an effort to address the limitations of these previous models. The end result is an integration model which allows for theoretical development and integration across multiple levels. In addition, the biosocial integration model allows for the development of new theories, various forms of integration among existing theories, and the integration of existing theories within the biosocial perspective. By emphasizing the importance of utilizing behavior genetic modeling strategies to effectively isolate associations between theoretical concepts and outcomes of interest, the presented model allows for both the theoretical and empirical integration of various theories within the biosocial perspective. In addition, the biosocial perspective is well- suited for this type of integration since it takes into account not only the independent influences of both biological and environmental factors, but also the intricacies of the interplay between both sets of influences (see Vila, 1994).
The previous chapters present a large amount of information directly bearing on the current state of the field of criminology. Two primary patterns of findings can be gleaned from the large amount of literature summarized in the previous chapters. First, antisocial phenotypes—and many of the concepts directly implicated in popular conventional theories of crime—are significantly influenced by genetic factors which, in turn, may result in genetic confounding and sources of spuriousness. In this way, the current empirical standing of the vast
130 majority of the mainstream criminological theories remains unknown, since previous studies did not properly control for genetic confounding and any resulting associations may be spurious.
Second, while the findings garnered from SSSMs has yielded somewhat mixed results for each of the four theoretical perspectives reviewed in the current dissertation (rational choice theory, classic strain theory, social learning theory, and social bonding theory), these perspectives have persisted indicating that the field of criminology views them as useful in at least some capacity.
Based on this observation, the next step in the current project is to attempt to integrate each of these four theories into the biosocial perspective using the biosocial integration model. Since all four theories have already undergone a significant amount of theoretical refinement, the subsequent analysis will focus more exclusively on examining the potential association between the primary concepts of each theory and antisocial behavior within the confines of a genetically- informed model. The next chapter will provide an overview of the data, measures, and analytic strategies used in the current dissertation.
131
Biosocial Concepts
Biosocial Concepts Biosocial Concepts
Biosocial Theory
Conventional Concepts Conventional Concepts
Conventional Concepts
Figure 4.1: Visual Representation of Existing Biosocial Theories
132
Figure 4.2: Visual Representation of Preliminary Models of Biosocial Integration
133
Figure 4.3: Visual Representation of the Biosocial Integration Model
134
CHAPTER 5
METHODS
The previous chapters have provided an overview of the biosocial perspective (Chapter 2) and summarized the key assumptions and current empirical status of four of the most influential theoretical perspectives within the field of criminology (rational choice theory, social learning theory, classic strain theory, and social bonds theory; Chapter 3). In addition, Chapter 4 proposed a biosocial integration model aimed at developing biosocial theories of antisocial behavior which allows for the integration of existing criminological theories into the biosocial perspective. As previously discussed, in order for a theory to be integrated into the biosocial perspective, the primary processes and concepts organized by the theory must continue to explain a significant portion of the variance in antisocial behavior within the confines of a genetically-informed analytic model. Theories that do not meet this minimal standard cannot be integrated into the biosocial perspective and should either be discarded or, if possible, refined with a more careful eye toward biosocial integration. The primary goal of the current project is to determine whether the four theories discussed in Chapter 3 can be integrated into the biosocial perspective. This task will be performed by addressing each of the four research questions proposed in Chapter 1. The current chapter will provide an overview of the sample analyzed in the current project, the operationalization of the examined theoretical concepts, and the analytic methods performed to test the proposed research questions.
5.1 Data
The current dissertation analyzed data from the National Longitudinal Study of
Adolescent Health (Add Health), which is a nationally representative sample of American youth enrolled in middle or high school during the 1994-1995 school year (Harris et al., 2009). The
Add Health employed a four wave panel design and followed respondents for approximately 14 135 years, covering a substantial portion of the life course. Details of the sampling procedures have been specified previously (Harris, 2011; Harris et al., 2009; Harris, Halpren, Smolen, &
Haberstick, 2006), but the study employed a multi-stage cluster design wherein all middle and high schools in the United States had a known probability of selection, resulting in a nationally representative sample of youth. In total, 132 schools (80 of which were high schools) were selected and stratified by region, urbanicity, school type, ethnic mix, and size. Students attending each of these schools were asked to complete a self-report questionnaire during one class period on a day designated by the research team with no make-up day being offered for absent students. This in-school portion of the study was performed between September 1994 and
April 1995. Student’s parents were notified of the survey and the date it would be administered.
Participation was completely voluntary and all respondents provided written informed consent.
In total, the in-school survey was completed by more than 90,000 students, who provided information on a wide range of topics including overall levels of health, school activities, delinquent behaviors, and friendship networks.
The in-school portion of the study also includes information regarding each respondent’s friendship network. More specifically, each respondent was provided with a roster of students enrolled at their current school and were asked to select their five closest male friends and five closest female friends. Based on this information it is possible to identify specific friendship networks comprised of respondents nested within the Add Health sample. In addition, responses from members of each respondent’s nominated peer groups can be used to glean information regarding each respondent’s peer group and each of the individual members of each friendship network. Importantly, this information has been utilized in previous studies which analyze the
Add Health data (Haynie, 2001, 2002).
136
After completion of the in-school portion of the study, a subsample of respondents was selected to participate in the in-home portion of the study. More specifically, students attending each of the 132 schools included in the in-school portion of the study were stratified by grade and sex and approximately 17 students within each stratum were selected for the in-home portion of the study. This process led to the selection of approximately 200 students from each of the
132 schools included in the study, resulting in a sample of 12,105 “core” respondents. In addition to the core sample, a number of additional special populations were oversampled. After taking into account all of the oversampled populations, the final sample included in the wave 1 in-home portion of the study consisted of 20,745 adolescents. All respondents were asked to complete a 90 minute in-home survey which collected detailed information on a wide range of topics including relationships with parents, delinquent behavior, school performance, and personality traits. In addition, information from 17,700 primary caregivers (most commonly the mother) was also collected. The survey was conducted using several computer-based interviewing techniques based on the content of the questions being asked (Czaja & Blair, 2005).
The majority of questions were asked verbally by a trained interviewer and respondents were asked to record their answer on a laptop computer. For more sensitive questions, such as those dealing with sexual or delinquent behaviors, respondents read questions privately and then responded privately on the provided computer. Importantly, this interviewing technique has been found to produce more truthful answers and reduce the overall amount of bias which accompanies collecting information on more sensitive topics (Turner et al., 1998).
One of the subpopulations oversampled during the first wave of data collection were twins and sibling pairs. Adolescents included in the core subsample were asked whether they had a co-twin or lived with a sibling. Cousins, unrelated siblings, half-siblings and co-twins who
137 lived in the same household as a respondent included in the core sample were added to the sample with certainty. In addition, a probability sample of full-siblings was also added to the sample. In total, over 3,000 sibling pairs of adolescents with varying levels of genetic relatedness are nested within the full Add Health sample (Harris et al., 2006).
Twin zygosity was determined using a three-stage process. First, all opposite sex pairs were classified as DZ twins. Second, all same-sex twins were given a self-report survey aimed at assessing confusability of appearance (Rowe & Jacobson, 1998). These items asked respondents to report how often they were confused for their co-twin by family members, strangers, and teachers. While this process allowed for the classification of the vast majority of the same sex twin-pairs, a small number of pairs could not be classified with certainty. In order to assess the zygosity of such pairs, respondents were asked to provide a buccal cell sample which was used for genotyping. Specifically, members of unknown pairs were compared with one another across seven highly polymorphic genotypes. Pairs matched on at least five polymorphisms were classified as MZ twins and pairs matched on fewer than five polymorphisms were classified as
DZ twins. Importantly, this zygosity determination process has been found to be highly valid, with estimates indicating an accuracy rate of over 95 percent (Reitveldt et al., 2000).
Approximately one year after the completion of the first wave of the study, the second wave of the study commenced. Data collection was carried out over approximately five months
(April-August) in 1996. Respondents who were enrolled in grades 7 through 11 were interviewed again at wave 2, while respondents who were in 12th grade during wave 1 interviews were not reinterviewed at wave 2. Participating respondents were interviewed in their home by trained interviewers. Nearly 15,000 respondents (N = 14,738) participated in the second wave of the study, resulting in a retention rate of nearly 90 percent (Harris et al., 2006). The surveys used
138 during the second wave of data collection were highly similar to those used during the first wave of the study due to the limited amount of time that elapsed between each wave.
The third wave of data collection was carried out between August 2001 and April 2002 when respondents were between 18 and 28 years old. Since most respondents had begun to transition into adulthood at the time the third wave of data was collected, the surveys used for the first and second waves of the study were no longer appropriate. For this reason, the questionnaires were modified to tap more age-appropriate topics such as romantic relationships, employment experiences, civic participation, community involvement, and contact with the criminal justice system. During wave 3 interviews, respondents were also asked to provide a saliva sample and a urine sample. These samples were used to obtain biomarker information for a number of different outcomes including the prevalence of a number of sexually-transmitted diseases (including HIV) and the concentration of various genetic polymorphisms. Due to the amount of time that had elapsed between the second and third waves of data collection, additional effort was made to locate and interview respondents. For example, respondents who were incarcerated were located and interviewed while in jail or prison. In total, 15,197 respondents participated in the third wave of the study and yielded an overall response rate of 77 percent (Harris et al., 2006).
The fourth and final wave of data collection was carried bout between 2007 and 2008 when respondents were between 24 and 34 years old. Once again, the questionnaire administered to respondents was modified to tap more age-appropriate topics and also to link responses to items collected earlier in the study to various life domains in adulthood (Harris,
2011). For example, during wave 1 interviews, respondents were asked to report how likely they thought they would experience various situations including graduating from college, getting
139 married, and getting HIV or AIDS. At wave 4 respondents were asked to report their current marital status, their highest level of education, and were also asked to provide saliva samples which were tested for the presence of HIV antibodies. Respondents were asked about a wide range of topics tapping social, psychological, health, and economic domains. In addition, respondents were asked to provide additional biomarker indicators which allowed for the collection of a large number of measures including: blood pressure, pulse, lipid levels, glucose levels, inflammation levels, height, weight, waist circumference, and genetic polymorphisms.
Due to extended efforts to minimize attrition and additional sources of missingness, a response rate of over 80 percent was achieved (N = 15,701; Harris, 2011).
5.1.1 Description of the Analytic Sample
As was thoroughly discussed in Chapter 2, the twin-based design is the most commonly used behavior genetic methodology, which can be used to isolate the influence of a given environmental influence on the outcome of interest net of the effect of genetic and shared environmental influences (Plomin et al., 2013). In addition, family studies are based on the underlying logic of twin studies but offer modified equations which properly adjust parameter estimates to take into account non-twin sibling pairs. Extending behavior genetic modeling strategies to incorporate twin pairs and singleton sibling pairs provides at least two advantages over traditional models which are limited to samples of twin pairs. First, previous studies have revealed that samples which consist of both twins and non-twin sibling pairs more accurately capture shared environmental influences on a wide range of outcomes, resulting in more conservative tests (Medland & Hatemi, 2009). Second, family studies address some of the limitations of twin studies such as the violation of the equal environments assumption (EEA) and the presence of assortative mating. Based on these advantages of a family-based approach
140 relative to a twin-based approach, the final sample analyzed in the current dissertation was restricted to MZ twin pairs, DZ twin pairs, full sibling pairs, and half-sibling pairs. Respondents were never asked to report the amount of time they resided in their current household making it impossible to determine how long they lived with their co-sibling or their parents. Since research question 4 is directly related to parent-child relationships, it was necessary to drop cousins and unrelated siblings from the final analytic sample.
Table 5.1 provides an overview of the final analytic sample including the number of individuals and pairs which comprise each category of genetic relatedness. As previously mentioned in Chapter 2, behavior genetic methodologies effectively shift the unit of analysis from individual cases to sibling pairs which allows for the comparison between siblings across measures of interest. Importantly, this particular practice effectively divides the overall sample in half which results in a significant reduction in statistical power making it more difficult to detect moderate to small effects and increasing the chances of committing a Type II error
(Plomin et al., 2013). This issue is typically addressed by creating a “double-entered” dataset wherein each sibling pair is included in the sample twice using a two-step process. First, siblings from each pair are randomly assigned the label of sibling 1 and sibling 2 and entered into the dataset in a sequential fashion (e.g., sibling 1’s score on a given measure followed by sibling 2’s score on the same measure). In this way, the resulting dataset is organized such that each row represents a single family and contains information for both siblings contained within that particular family (i.e., wide format), as opposed to each row representing each individual sibling
(i.e., long format). Second, each sibling pair was entered into the dataset a second time, but the order in which each sibling’s scores was entered was reversed (e.g., sibling 2’s score on a given measure was followed by sibling 1’s score on the same measure). This form of organization
141 accounts for the overall drop in statistical power resulting from shifting the unit of analysis from individuals to pairs, and is conventional within behavior genetic analyses (Plomin et al., 2013).
The primary limitation of this approach is that including each sibling pair into the dataset twice results in within-family clustering and may result in significant levels of heteroskedasticity and, subsequently, untrustworthy standard errors. In order address this limitation, robust standard errors were estimated within each statistical model (Kohler & Rodgers, 2001; Smith & Hatemi,
2013). The final analytic sample contained 570 MZ twins (285 pairs), 892 DZ twins (446 pairs),
2,070 full siblings (1,035 pairs), and 732 half-siblings (366 pairs), resulting in a final sample size of N = 4,264 (2,132 pairs). Importantly, the final sample size varies from model to model as a function of missing data on the measures included in each model.
5.2 Measures
5.2.1 Antisocial Behavior
Directly in line with previous studies analyzing the Add Health data (Bernat et al., 2012;
Jang & Franzen, 2013; Maimon et al., 2012; Zheng & Cleveland, 2013), 4 antisocial behavior measures, one at each wave of data collection, were included in the current dissertation. These measures were designed to tap overall levels of antisocial behavior and include both violent and nonviolent behaviors. Due to modifications to the questionnaires administered at each wave of data collection, the items used to create each antisocial behavior index vary slightly from wave to wave. However, these same measures have been used previously (Barnes, Beaver, & Boutwell,
2011; Mears, Cochran, & Beaver, 2013; Schwartz & Beaver, 2013) and have been found to load on a common factor, providing strong evidence that these items are adequately tapping the concept of antisocial behavior. During wave 1 interviews, respondents were asked to report their involvement in 15 delinquent activities over the past 12 months including shoplifting, stealing
142 something worth more or less than $50, getting into a serious physical fight, and shooting or stabbing anyone. Each of the items included in the wave 1 antisocial behavior index (and all other antisocial behavior indexes) are listed in Appendix A. The same 15 items were asked again during wave 2 interviews. Due to the modifications made to the questionnaire used during wave 3 interviews, the wave 3 antisocial behavior index was slightly different from the wave 1 and wave 2 antisocial behavior indexes. More specifically, respondents were asked 12 questions regarding the frequency in which they deliberately wrote a bad check, used someone else’s credit card without their permission, and carrying a handgun to work or school. These same 12 items were asked again during wave 4 interviews. At all four waves of data collection, responses to each item were coded such that 0 = never, 1 = 1 or 2 times, 2 = 3 or 4 times, and 3 = 5 or more times. Responses to each of the items were summed separately at each wave to create the wave
1 (α =.86), wave 2 (α =.83), wave 3 (α = .69), and wave 4 (α = .72) antisocial behavior indexes.
Higher scores on each index are indicative of higher levels of antisocial behavior at each wave.
Due to a limited amount of variation, the wave 3 and wave 4 antisocial behavior measures were dichotomized such that 0 = a score of zero on the respective antisocial behavior measure and 1 = a score of one or higher on the respective antisocial behavior measure. Table 5.2 provides an overview of the waves at which the antisocial behavior measures and all other measures included in the current dissertation are measured.
5.2.2 Prevalence of Illicit Drug Use
A total of three illicit drug use measures, one at each of the first three waves of data,13 were included in the current dissertation. During interviews at each wave, respondents were
13 While drug use measures were collected during wave 4 interviews, the employed surveys did not ask respondents to report on the same substances they were asked to report on during the first three waves of the study. Based on this discrepancy in content, and in an effort to preserve continuity in the concepts examined, the illicit drug use measures were limited to the first three waves of data. 143 asked to report the number of times they used a variety of different drugs in the past 30 days.
More specifically, respondents were asked how often they used marijuana, cocaine, and other drugs which included LSD, PCP, ecstasy, mushrooms, inhalants, ice, heroine, or non-prescribed prescription drugs. Scores on each measure were coded dichotomously where 0 = never and 1 = one or more times. Importantly, these same measures have been used in previous studies analyzing the Add Health data (Harris et al., 2006; Lessem et al., 2006). These scores were then
14 summed separately at each wave to create the wave 1 (KR20 = .46), 2 (KR20 = .39), and 3 (KR20
= .48) illicit drug use measures. These measures indicate the total number of illicit substances each respondent reported using in the past 30 days and range from 0 to 3.
5.2.3 Alcohol Use
A total of four alcohol use measures, one at each wave, were included in the current dissertation. The alcohol use measure was created using three different measures tapping different dimensions of alcohol use measured at each wave. First, respondents were asked how often they consumed any type of alcohol in the past 12 months. This particular measure is designed to tap the overall frequency of alcohol use within the past 12 months. Second, males were asked how frequently they consumed five or more drinks in a single sitting in the past 12 months while females were asked how frequently they consumed four or more drinks in a single sitting over the same period of time. This measure was designed to capture the frequency in which each respondent engaged in binge drinking over the past 12 months. Third, respondents were asked how frequently they were “drunk or very high on alcohol” over the past 12 months.
This particular measure was designed to tap the frequency in which each respondent consumed alcohol to the point of intoxication. Each measure was assessed at all four waves of the Add
14 Since all of the measured included in the illicit drug use measures are dichotomous, Kuder-Richardson (KR20) coefficients (as opposed to traditional Chronbach’s alpha coefficients) were used to assess reliability. 144
Health study and was coded such that 0 = none; 1 = 1 or 2 days; 2 = once a month or less; 3 = 2 to 3 days a month; 4 = 1 to 2 days a week; 5 = 3 to 5 days a week; 6 = every day or almost every day. All three items were summed separately at each wave to create the wave 1 (α = .92), 2 (α
=.92), 3 (α = .89), and 4 (α = .88) alcohol use measures with higher scores indicating more frequent alcohol use. Importantly, all of the alcohol measures used in the current dissertation have been used previously in studies analyzing the Add Health data (Chen and Jacobson, 2012;
Harris et al., 2006; Shin et al., 2009)
5.2.4 Rational Choice Theory
Thoughtfully Reflective Decision-Making (TRDM). Rational choice was measured using an index designed to capture variation in thoughtfully reflective decision-making (TRDM).
TRDM refers to the tendency to collect all available and relevant information regarding a given problem and thinking deliberately, carefully, and thoughtfully about possible solutions prior to acting (Paternoster & Pogarsky, 2009). In addition, TRDM refers to the application of reason and the full consideration of alternative solutions in addition to the consideration of previous decisions and what went right and wrong. Following the lead of previous researchers analyzing the Add Health sample (Paternoster et al., 2011), TRDM was measured using the following four items from the wave 1 in-home portion of the study: (1) when you have a problem to solve, one of the first things you do is get as many facts about the problem as possible; (2) when you are attempting to find a solution to a problem, you usually try to think of as many different approaches to the problem as possible; (3) when making decisions, you generally use a systematic method for judging and comparing alternatives; and (4) after carrying out a solution to a problem, you usually try to analyze what went right and what went wrong. Responses to each item were coded such that 1 = strongly disagree, 2 = disagree, 3 = neither agree nor
145 disagree, 4 = agree, 5 = strongly agree. Responses to each of the four items were summed to create the wave 1 TRDM measure (α = .74), with higher scores indicating higher levels of
TRDM.
5.2.5 Social Learning Theory
Peer Network Antisocial Behavior. During the in-school portion of the wave 1 interviews, respondents were provided with a current roster of their school and were asked to nominate their five closest male and five closest female friends. Based on this information, the
Add Health research team identified friendship networks among the respondents which allowed for the creation of friendship network variables tapping the overall concentration of various traits and behaviors within each identified network. Two separate networks are available for all respondents who provided information on this particular section of the study. First, the send network includes all individuals identified by the target respondent as friends. Second, the receive network includes all individuals who identified the target respondent as a friend. In the current dissertation, information from the send network was used to create peer network measures. Importantly, a sizable number of studies indicate that direct measures of peer behaviors tend to be more accurate than indirect measures such as asking a respondent to provide information regarding their peers’ behavior (Prinstein & Wang, 2005; Weerman & Smeenk,
2005; Young, Barnes, Meldrum, & Weerman, 2011). Based on the information included in the
Add Health data and the results flowing from this line of research, direct measures of peer antisocial behavior were analyzed in the current dissertation.
The peer network antisocial behavior measure was created using self-reported information from all subjects with available data in each respondent’s peer network. More
146 specifically, the wave 1 antisocial behavior measures (as described above and outlined in
Appendix A) were generated for each of the four subjects which comprise each respondent’s peer network. The individual scores on the wave 1 antisocial behavior indexes were then group mean centered to create the peer network antisocial behavior index.
Peer Network Illicit Drug Use. The same procedure used to create the peer network antisocial behavior score was also used to create the peer network illicit drug use measure.
Directly in line with the illicit drug use measures generated for respondents, a single peer illicit drug use measure was used in the current study. More specifically, the peer network drug use measure was designed to tap overall drug use among members of each respondent’s peer group and was created using a two-step process. First, illicit drug use measures (as described above) were created for each individual within a given respondent’s peer group. Second, the resulting measures were then averaged across all peers within a given group and indicate the average number of substances used by members of each peer group during the past 30 days. All peer network illicit drug use measures were created using responses from the first wave of data collection.15
Peer Network Alcohol Use. Directly in line with the alcohol use measures constructed for each individual respondent, a single peer network alcohol use measure was created using three separate items tapping overall alcohol consumption habits. First, the average number of times that members of each respondent’s peer network consumed alcohol in the past 12 months was created. Second, the average frequency in which members of the respondent’s peer group consumed five or more drinks in a single sitting over the past 12 months was created. Third, the
15 While these measures could be constructed from information collected during wave 2 interviews, the Add Health does not include any additional information regarding contact with the peer group identified during the first wave of the study. In this way, it is possible that respondents have transitioned to other peer groups and no longer have contact with the peers identified during wave 1 interviews. For this reason, the peer group measures were created using wave 1 items. 147 average frequency in which members of each respondent’s peer group consumed alcohol to the point of intoxication over the past 12 months was created. All three items were then summed resulting in the peer network alcohol use measure wherein higher scores indicate higher levels of alcohol use within a given peer group.
5.2.6 Classic Strain Theory
Disconnect between Aspirations and Expectations. Based on the discrepancies regarding the proper operationalization of classic strain (Burton & Cullen, 1992; Burton et al., 1994;
Kubrin et al., 2009), strain was measured in two separate ways. First, as in previous studies
(Hirschi, 1969), strain was operationalized as the disconnect between aspirations and expectations. During wave 1 and wave 2 interviews, respondents were asked two questions regarding college. Recall, during these two interview periods, the vast majority of respondents would have been in middle or high school with the mean ages at each wave ranging between 15 and 17 years old. In this way, higher education was likely an important point of consideration during this time. The first item asked respondents to indicate their desire to attend college and the second item asked respondents to indicate how likely it was that they would actually be able to attend college. Both items were coded as a 5-point scale wherein 1 = low and 5 = high.
Based on Merton’s (1938) conceptualization of strain, it would be expected that respondents who indicated a relatively high score on the first item and a relatively low score on the second would experience greater levels of strain. More specifically, respondents who greatly desired to attend college (aspirations) but felt that they did not have an adequate amount of opportunity to realize such goals (expectations) would suffer from greater feelings of strain. In order to assess levels of strain stemming from a disconnect between aspirations and expectations, the expectation item was subtracted from the aspiration item. The resulting difference score
148 indicates the level of strain each respondent experienced wherein negative scores indicate higher levels of expectations relative to aspirations indicating low levels of strain (i.e., such respondents expect to have adequate opportunity to attending college, but do not necessarily want to attend).
A score of zero on the resulting difference score indicates that aspirations matched expectations.
Finally, a positive score on the difference score indicates a strong desire to attend college, but a lack of the opportunities needed to realize such aspirations which is expected to result in greater overall levels of strain. Importantly, this operationalization not only taps the disconnect between aspirations and expectations, but also blocked opportunities.
Disconnect between Expectations and Realizations. The second operationalization of strain is meant to tap a disconnect between expectations and the realization of socially- acceptable goals. The longitudinal nature of the Add Health dataset allows for the creation of measures which directly compare expectations regarding important social milestones and the realization of such expectations. More specifically, during wave 2 interviews, respondents were asked to indicate the chances that they would experience several important experiences which included: (1) getting married by age 25; (2) graduating from college; and (3) having a middle class income by age 30. Each of these items were coded as follows: 1 = almost no chance; 2 = some chance, but probably not; 3 = a 50-50 chance; 4 = a good chance; 5 = almost certain. Each item was then collapsed into a dummy variable and coded such that 0 = less than a 50-50 chance and 1 = greater than or equal to a 50-50 chance.
During wave 4 interviews (when respondents were between 24 and 34 years old, which allows for the direct comparison between each respondent’s expectations and their realization of each of these particular goals. During wave 4 interviews, respondents were asked to report their current marital status (0 = not married; 1 = married), their current personal income (measured
149 continuously in dollars), and their highest level of educational achievement. The income measure was recoded in order to take into account a “middle-class income” which was defined as the median score on the wave 4 income measure within the full sample. The middle-class income measure was then coded dichotomously such that 0 = less than the median and 1 = greater than or equal to the median. Educational achievement was coded categorically wherein 1
= no school; 2 = 8th grade or less; 3 = greater than 8th grade, but less than high school; 4 = high school or high school equivalent; 5 = some college; 6 = four year college graduate; 7 = graduate or professional degree. This categorical measure was then collapsed into a dichotomous measure coded such that the first five categories were coded as 0 and the sixth and seventh categories were coded as 1.
After the creation of both sets of dummy variables (expectations and realizations), difference scores were calculated wherein each realization measure was subtracted from the corresponding expectation measure. The resulting difference scores range between -1 and 1 and can be interpreted as follows: -1 = expectations were exceeded; 0 = realizations matched expectations; and 1 = expectations exceeded realizations. In this way, respondents with a score of 1 on the resulting difference scores would be expected to have undergone greater feelings of strain. Finally, to capture a more global measure of strain stemming from the disconnect between expectations and realizations, all three of the difference scores were summed. Once again, higher scores indicate greater levels of strain.
150
5.2.7 Social Bonding Theory16
Attachment. While Hirschi (1969) offered an operationalization of each of the bonds identified in his initial proposal of social bonding theory, the proper measurement strategy for each bond has been a point of contention in the extant literature (Kubrin et al., 2009). Based on this observation, attachment was operationalized using four distinct measures, each aimed at tapping separate socializing agents: parental attachment; school attachment; neighborhood attachment; and peer attachment. During wave 1 and wave 2 interviews, respondents were asked how close they felt to their mother/father and how much they felt that their mother/father cares about them with all four items coded as follows: 1 = not at all; 2 = very little; 3 = somewhat; 4 = quite a bit; 5 = very much. All four items were summed separately at each wave to create the wave 1 (α = .73) and wave 2 (α = .68) parental attachment measures. Higher scores on each resulting measure indicate higher levels of parental attachment. Importantly, these same measures have been previously used in studies analyzing the Add Health data (Beaver, 2008;
Haynie, 2001; Schreck, Fisher, & Miller, 2004). Following the lead of previous studies (Schreck et al., 2004), school attachment was measured using a three item index comprised of responses to the following three questions assessed during wave 1 and wave 2 interviews: (1) feelings of closeness to people at school; (2) feeling like part of school; (3) happiness at current school.
Each item was coded such that 1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, 5 = strongly agree. All three items were summed separately at each wave to
16 The social bond of belief was not assessed in the current dissertation. This is primarily due to the fact that the Add Health does not contain measures tapping the overall construct of belief as described by Hirschi (1969). Previous studies tend to operationalize belief as one’s feelings regarding the police or breaking a given set of laws or rules (see Kubrin et al., 2009 for a thorough overview), but these items were not collected during the Add Health study. For this reason, the bond of belief cannot be properly measured and assessed in the current study. However, and as previously discussed in Chapter 3, the bonds of belief and commitment are rarely studied (for example see Schreck et al., 2004) and tend to have small overall effects on delinquency. 151 create the wave 1 (α = .77) and 2 (α = .78) school attachment measures with higher scores indicating higher overall levels of school attachment.
Neighborhood attachment was measured using the same two items collected during wave
1 and 2 interviews. More specifically, respondents were asked how safe they felt in their current neighborhood and how happy or unhappy they would be if they had to leave their current neighborhood. Responses to both items were measured using a 5-point scale and ranged from 1 to 5 with higher scores indicating greater feelings of safety and greater feelings of unhappiness
(indicating contentment with one’s current neighborhood) respectively. Both items were then summed to create the wave 1 (α = .55) and wave 2 (α = .54) neighborhood attachment measures, with higher scores on the resulting measures indicating greater feelings of neighborhood attachment. The measurement strategy for neighborhood attachment is similar to previously used measures (Manning et al., 2005). Finally, peer attachment was measured using a single item included in wave 1 and 2 interviews. More specifically, respondents were asked how much their friends cared about them with responses coded as follows: 1 = not at all; 2 = very little; 3 = somewhat; 4 = quite a bit; 5 = very much. In this way, higher scores on the wave 1 and 2 peer attachment measures represent higher overall levels of peer attachment. Importantly, this same item has been used previously as a measure of peer attachment (Schreck et al., 2004).
Involvement. As with the social bond of attachment, multiple measures of involvement, each tapping a separate facet of involvement, were included in the current dissertation. First, and following the lead of previous studies analyzing the Add Health data (Beaver, 2008; Maldonado-
Molina, Reingle, & Jennings, 2011; Prado, et al., 2009), a measure of parental involvement was included in the current dissertation. During wave 1 and 2 interviews, respondents were asked whether they had engaged in 10 different activities with their mother and/or father in the past
152 four weeks: (1) went shopping; (2) played a sport; (3) attended a religious or church-related event; (4) talked about someone they were dating or a party they attended; (5) attended a movie, play, concert, or sporting event; (6) talked about a personal problem; (7) had a serious argument about their behavior; (8) talked about work or grades; (9) worked on a project for school; (10) talked about other things they are doing in school. Each item was coded dichotomously for mothers and fathers separately such that 0 = no and 1 = yes. All 20 items were then summed separately at each wave to create the wave 1 (KR20 = .71) and 2 (KR20 = .71) parental involvement measures wherein higher scores represent greater levels of parental involvement.
The second measure of involvement included in the current dissertation was community involvement which was measured using a single item. More specifically, during wave 3 interviews, respondents were asked to indicate whether they regularly participated in volunteer or community service work when they were between 12 and 18 years old. As in previous studies, responses were coded dichotomously such that 0 = no and 1 = yes (Paternoster &
Pogarsky, 2009).
Commitment. A significant number of studies examining the potential association between commitment and subsequent delinquency have focused on educational commitment which is typically operationalized as a measure of grade point average (GPA; Agnew, 1991;
Krohn & Massey, 1980). However, a complementary line of research provides preliminary evidence suggesting that individuals with overall lower GPAs are significantly more likely to engage in criminal and delinquent behaviors relative to individuals with higher GPAs even after accounting for other important social bonds such as school attachment (for a current example see
Hoffmann, Erickson, & Spence, 2013). In addition, previous studies have found that GPA does not sufficiently capture the concept of school commitment (Cerkovich & Giordano, 1992). For
153 these reasons, GPA was not used as a measure of commitment in the current dissertation. While
Hirschi (1969) did not recognize religion as a possible source of commitment, subsequent researchers have explored the impact of religion on criminal and delinquent behavior (Johnson et al., 2000; Johnson & Jang, 2010; Johnson, 2011). In addition, religious commitment has been recently recognized as a potential mechanism linking religiosity and antisocial outcomes (Akers
& Sellers, 2013; Kubrin et al., 2009).
Based on these observations and the dearth of literature exploring the potential association between religiosity—operationalized as a social bond—and antisocial behavior, the current dissertation included a measure of religious commitment. More specifically, four identical items at wave 1 and 2 tapping overall religiosity were used to create the religious commitment measure. First, respondents were asked how often they attended religious services in the past 12 months. Second, respondents were asked how often they attended youth group,
Bible classes, or choir affiliated with their church in the past 12 months. Responses to both of these questions were coded such that 1 = never; 2 = less than once a month; 3 = once a month or more, but less than once a week; 4 = once a week or more. Third, respondents were asked how often they prayed in the past 12 months. Responses were coded on a five item scale were 1 = never; 2 = less than once a month; 3 = at least once a month; 4 = at least once a week; and 5 = at least once a day. Fourth, respondents were asked how important religion is to them with responses coded such that 1 = not important at all; 2 = fairly unimportant; 3 = fairly important; and 4 = very important. Principle components analysis revealed that all four items loaded on a common factor. Based on these results, the four items were summed separately at each wave to create the wave 1 (α = .75) and 2 (α = .75) religious commitment measures. Higher scores on the resulting measures are indicative of greater levels of religious commitment.
154
5.2.8 Demographic Covariates
In an effort to prevent confounding stemming from uncontrolled differences across demographic differences within the analytic sample, three demographic controls were included in some steps of the analysis. Age was recorded during wave 1 in-home interviews and was coded continuously. Sex was self-reported during wave 1 interviews and coded such that 0 = female and 1 = male. Finally, race was also self-reported during the in-home portion of the first wave of the study and was coded such that 0 = Caucasian and 1 = all other races.
5.3 Analytic Plan
This section outlines the analytic strategy used to address the research questions presented in Chapter 1. While the analytic plan for each research question is somewhat unique, a general four-step analytic plan was carried out for each question. First, in an effort to determine whether the examined outcome measures comport with the extant literature, it was first necessary to examine the genetic and environmental influences on each of the examined outcomes and theoretical measures. Second, prior to attempting to integrate the examined theoretical concepts into the biosocial perspective, it was necessary to examine whether such concepts significantly predict each of the antisocial outcomes net of the effects of demographic covariates. Third, a genetically-informed analytic strategy was used to effectively control for genetic effects on the examined outcome measure and isolate the association between each conceptual measure and outcome measure. As previously discussed in Chapter 2, genetic and environmental influences often work both independently and interactively to shape variation in antisocial outcomes. In this way, theoretical concepts which do not significantly predict antisocial outcomes within the confines of a genetically-informed model may still play an important role through gene- environment interplay. Based on this possibility, the fourth general step in the analytic plan was
155 to examine the potential moderating effect of examined theoretical concepts on genetic influences on each examined outcome. Below, a more technical discussion of analytic techniques employed at each step of the analysis will be provided. Importantly, this discussion will be organized by the analytic strategy used. Subsequently, the application of each analytic methodology will be applied to the four research questions proposed in Chapter 1.
5.3.1 Univariate ACE Decomposition Model
The first step in the analytic plan is to estimate the genetic and environmental contributions to the examined outcomes and theoretical measures. The overall purpose of this particular step in the analytic strategy is to establish whether genetic factors significantly influence each of the examined outcomes. While there is strong indirect and theoretical evidence suggesting that each of the examined outcomes are significantly influenced by genetic factors
(see Turkheimer, 2003), directly estimating the genetic and environmental influences on each outcome is a critical aspect of the current dissertation. Since the main methodological concern regarding theoretical integration within the biosocial perspective revolves around difficulties surrounding genetic confounding, establishing significant genetic influences on each of the examined outcomes is critically important. More specifically, genetic confounding is not a concern unless genes significantly influence both the examined independent variable and the examined outcome. In this way, genetic confounding cannot occur if the examined outcomes are not significantly influenced by genetic factors.
There are a large number of analytic techniques which allow for the direct estimation of genetic influences on a given outcome measure (see Beaver, 2013a and Plomin et al., 2013 for overviews). However, the univariate ACE decomposition model is arguably the most commonly used. This particular modeling strategy is built upon a structural equation modeling (SEM)
156 framework and used to analyze pairs of twins and siblings. The ACE model is a behavior genetic modeling strategy which allows for the decomposition of variance within a given phenotype into three separate latent constructs: additive genetic influences (symbolized as A); shared environmental influences (symbolized as C); and nonshared environmental influences
(symbolized as E). Importantly, the resulting estimates are latent in the sense that the specific genes and environments contributing to each component remain unknown and apply only to the examined sample. In this way, it is not possible to extend the findings to each of the individuals included in the sample. The ACE model is univariate in that it only examines the phenotypic variance within a single measure.
Since the univariate ACE model is built on an SEM framework, it can be presented using a path diagram which is illustrated in Figure 5.1. The rectangles presented in the figure represent the observed scores for each sibling on the measure of interest (e.g., antisocial behavior). These observed measures (and the covariance between them) are used to define A, C, and E which are latent factors. This procedure is symbolized by the single headed arrow running from each latent factor to the observed score for each sibling. Each of these single headed arrows represents a path estimate (symbolized as a1, a2, c1, c2, e1, and e2) which are collectively used to provide the final estimates of A, C, and E. The double-headed arrow running from the A estimate for one sibling to the A estimate for the second sibling symbolizes the covariance estimates for the two latent constructs. Importantly, these covariance estimates are constrained to reflect the level of genetic relatedness shared by each pair with MZ twin pairs constrained to 1.00, DZ twin and full sibling pairs constrained to .50, and the covariance between half sibling pairs constrained to .25, which is represented by the figures presented above the double-headed arrow connecting each A estimate. The second double-headed arrow, which connects both C components, symbolizes the
157 covariance for both of the latent constructs. The covariance between both C components is constrained to correlate at 1.00 for all sibling pairs since shared environmental influences equally impact both siblings. Finally, the E components of the model are allowed to vary freely and effectively capture any variance that is not explained by the A and C components, providing an estimate of nonshared environmental influences on the examined outcome plus measurement error.
As with most SEM applications, the ACE model allows for the estimation of nested models in an effort to determine whether a more parsimonious model provides a better fit to the data (Medland & Hatemi, 2009). This procedure involves constraining the A and/or C components of the full ACE model to zero in an effort to maximize model fit. Importantly, the E component cannot be constrained to zero since it also capture measurement error. This model- fitting procedure is typically carried out in two steps. First, model fit diagnostics are used to determine the overall goodness of fit. In an effort to assess overall model fit, several diagnostics were utilized in the current study. More specifically, the comparative fit index (CFI), the
Tucker-Lewis index (TLI), and the root mean square error of approximation (RMSEA) were all used to determine the adequacy of model fit. More common fit statistics, such as chi-square and the Akaike information criteria (AIC), have been found to be overly influenced by large sample sizes and model complexity and make comparing nested models difficult (Hu & Bentler, 1999).
In addition, the employed model fit diagnostics have well established cut-off points indicative of good or acceptable model fit. Values greater than or equal to .95 on the CFI and TLI indicate good fit (Hu & Bentler, 1999), while values below .05 on RMSEA indicate close fit and values up to .08 represent reasonable errors of approximation (Browne & Cudeck, 1993).
158
The second step in assessing the best-fitting and most parsimonious model involves the examination of the standard errors and corresponding p-values for each of the estimated components to determine statistical significance. Nonsignificant components are examined more closely and possibly constrained to zero in a nested model. Wald’s tests of parameter constraints are used to determine whether the constraint of a given estimate to zero results in a significant change in overall model fit. More specifically, Wald’s tests indicate whether constraining a given parameter results in a significant change in the overall chi-square estimate. A significant change in chi-square would indicate a significantly different model, whereas a nonsignificant change would indicate that the nested model is more parsimonious than the full model, and would therefore be favored (assuming all model fit statistics fall within accepted ranges).
The traditional univariate ACE model was developed for use with continuous outcome measures; however, such models have been modified to decompose the variance within dichotomous and categorical measures as well (see Neale, 2009 for a more in-depth overview).
These models, which are referred to as threshold liability models, rely on covariance estimates which are estimated using tetrachoric correlation coefficients rather than the more traditional
Pearson correlation coefficients. In addition, threshold models identify an underlying
“threshold” for each dichotomous or categorical measure included in the model which essentially represents the point in a normal distribution in which a respondent would move from one category to the next. Aside from these differences, the basic logic and interpretation of the threshold model is essentially synonymous with a traditional univariate ACE model. In addition, the model-fitting and evaluation procedures are also highly similar and carried out in the same overall pattern. All univariate decomposition models (both univariate ACE and threshold
159 liability models) were estimated using the statistical software package Mplus 7.11 (Muthén &
Muthén, 2012).
5.3.2 Multivariate Regression
The second step in the analytic plan is to examine the potential association between each of the theoretical concepts and each outcome measure. These associations were estimated using multivariate regression modeling strategies which were adjusted for age, race, and gender in an effort to eliminate common sources of confounding. In this way, each examined outcome measure was regressed on the conceptual measures and the demographic controls. Importantly, the multivariate models were conducted using linear regression (or OLS) for continuous outcome measures (e.g., wave 1 antisocial behavior measures), and logistic regression for binary outcomes (e.g., wave 4 antisocial behavior). In addition, negative binomial regression models were used to examine the illicit drug use measures. Negative binomial models were employed, as opposed to Poisson models, for two reasons. First, the standard deviation of each measure exceeded its mean, indicating the presence of over-dispersion. Second, likelihood ratio tests comparing each model to a Poisson model were estimated and revealed that negative binomial models were more appropriate than Poisson models. The overall purpose of this particular step in the analytic plan is to establish whether each examined theoretical measure is significantly associated with the examined outcomes prior to controlling for potentially confounding genetic influences.
5.3.3 DeFries-Fulker Analysis
The next step in the analytic strategy is to remove any potential sources of genetic confounding and effectively isolate the potential association between each theoretical measure and each examined outcome. While there are a number of different analytic techniques which
160 would allow for this type of procedure (many of which were reviewed in Chapter 2), DeFries-
Fulker (DF) analysis seems to be the most appropriate and straightforward. DF analysis is a regression-based statistic that is appropriate when analyzing samples of twin dyads and sibling pairs. This modeling technique is commonly used in behavior genetic research and provides accurate estimates of the proportion of overall variance in the phenotype of interest that is explained by additive genetic and shared environmental influences, with the residual variance attributed to nonshared environmental influences and error. In addition, recent studies have examined the accuracy of the estimates garnered from DF analysis relative to other modeling strategies and have concluded that DF analysis produces estimates that are directly in line with alternative modeling strategies (Medland & Hatemi, 2009; Smith & Hatemi, 2012).
The original DF equation proposed by DeFries and Fulker (1985) was developed to be used with clinical samples of twins where one twin had an extreme score on the outcome measure of interest. Rodgers and colleagues (1994) modified the original equation to allow for the use of samples drawn from the general population. The modified equation is as follows:
(5.1)