: )3: . . in“. “away L3," "3:27.44... 1... $.75. bi . .. . nu; .. . u . nigh”). u hrs? ... {In- .. .. ... o .. 3...! )Junse’yffi (A... . ; . .03. AV E. . .4. . L32? :5 9...: , .5.‘ 5h . 5:33. J .. .Wfisnnrfi. 9... a «legrfliie a? ”1””...g. Ilia (f... 2 iv... . 59...? :1 5).}. LIBRARY W Michigan State 2‘ 71! University This is to certify that the thesis entitled AN INTEGRATIVE MODEL OF EXPOSURE TO VIOLENCE, AGGRESSION, AND VIOLENT OFFENDING presented by KAYE IRENE MARZ has been accepted towards fulfillment of the requirements for the Master of Science degree in Criminal Justice y r ‘Wféio’rP’ofes r’s Signature //, (9/7 m Date MSU is an Affirmative Action/Equal Opportunity Employer PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES return on or before date due. MAY BE RECALLED with earlier due date if requested. DAIEDUE DAJEDUE DAIEDUE 5/08 K IPrQ/AchrelelRC/Dateoue indd AN INTEGRATIVE MODEL OF EXPOSURE TO VIOLENCE, AGGRESSION, AND VIOLENT OFFENDING By Kaye Irene Marz A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE Criminal Justice 2008 ABSTRACT AN INTEGRATIVE MODEL OF EXPOSURE TO VIOLENCE, AGGRESSION, AND VIOLENT OFF ENDING By Kaye Irene Marz Adolescents are exposed to violence at high rates within their homes and neighborhoods, occurring when life course trajectories form. Exposure has implications for increased risk of violent offending, including intimate partner violence (IPV). Using data from the Project on Human Development in Chicago Neighborhoods (PHDCN), this project found early exposure to violence (ETV) in the home predictive of less future IPV. However, ETV in the community supports social learning theory (SLT), indicating that IPV may be learned from models outside the home. SLT and anti-social behavior trait (ASBT) theory are supported regarding violent offending in the community. ASBT appears a larger factor for IPV offending and SLT and ASBT theory comparable for community violent offending. The interplay of ETV and ASBT on either type of offending cannot be ignored; controlling for one reduces the positive effect of the other. When investigating violent offending, both ASBT and ETV must be considered. Copyright by KAYE IRENE MARZ 2008 ACKNOWLEDGEMENT I give heartfelt thanks to my thesis committee chair, Dr. Christopher D. Maxwell, for his substantive advice and statistical expertise while conducting this project as well as his confidence in me during my years of schooling at Michigan State University. I also thank Dr. Sheila Maxwell and Dr. Vince Hoffman for serving on my thesis committee and for their guidance on the prospectus and the final thesis. I gratefully acknowledge the support of Dr. Myron Guttman and Ms. Rita Young Bantom, both with the Inter-university Consortium for Political and Social Research (ICPSR), for their investment in my education over the last four and one-half years. I also thank my friends, and especially my family, for their patience and encouragement. The challenge of graduate school while working full—time was alleviated by their understanding and support. iv TABLE OF CONTENTS LIST OF TABLES ............................................................................................................ vii LIST OF FIGURES ........................................................................................................... ix Chapter One ........................................................................................................................ 1 Statement of the Problem .............................................................................................. 1 Theories of Violence ..................................................................................................... 3 Comparison of Theories ....................................................................................... 8 Ecological Models ...................................................................................................... 10 Limitations of Previous Research ............................................................................... 13 The Present Study ....................................................................................................... 14 Chapter Two .................................... ' .................................................................................. 1 6 Literature Review ........................................................................................................ 16 Research on Aggression .............................................................................................. 17 Research on Family Violence ..................................................................................... 19 Research on Multiple Forms of Exposure to Family Violence ................................... 20 Research on Psychological Maltreatment and Supervisory Neglect .......................... 22 Research on Exposure to Family Violence and Community Violence ....................... 24 Cumulative Effects and Effects Over Time ................................................................ 26 Data from Multiple Informants ................................................................................... 28 Defining Environmental Factors with Ecological Models .......................................... 29 Chapter Three .................................................................................................................... 34 Methodology ............................................................................................................... 34 Data Collection ........................................................................................................... 35 Measures ..................................................................................................................... 37 Primary Caregiver Reports of Family Violence ......................................................... 38 Conflict Tactics Scale for Partner and Spouse ................................................... 38 Conflict Tactics Scale for Parent and Child ....................................................... 41 Subject Reports of Exposure to Violence ................................................................... 41 Subject Reports of Relationship and Community Offending ..................................... 43 Demographics and Family Environment .................................................................... 44 Additional Transformations ........................................................................................ 46 Attrition and Missing Data ................................................................................. 46 Handling Skewness ............................................................................................ 50 Multicollinearity ................................................................................................ 50 Combining Wave Measures ............................................................................... 51 Chapter Four ..................................................................................................................... 52 Results ......................................................................................................................... 52 Descriptive Results of Demographic and Model Variables ........................................ 52 Descriptive Results of Exposure to Violence Measures ............................................. 55 PC Reports of Intimate Partner Violence ........................................................... 55 PC Reports of Child-Directed Aggression ......................................................... 58 Adolescent Reports of Exposure to Violence .................................................... 60 Measures of Anti-Social Behavior ..................................................................... 64 Descriptive Results of Dependent Variables .............................................................. 66 Young Adult Intimate Partner Violence Offending ........................................... 66 Young Adult Violent Offending in the Community .......................................... 66 Results from Offending Models .................................................................................. 67 Models for Anti-Social Behavior ................................................................................ 69 Models for Young Adult Intimate Partner Violent Offending .................................... 71 The Contemporary and Cross-Sectional Models ............................................... 71 The Combined Model ........................................................................................ 77 The Wave 2 Model ............................................................................................. 80 The Wave 1 Model ............................................................................................. 85 Models of Young Adult Violent Offending in the Community .................................. 90 The Contemporary and Cross-Sectional Models ............................................... 90 The Combined Model ........................................................................................ 97 The Wave 2 Model ........................................................................................... 101 The Wave 1 Model ........................................................................................... 105 Comparisons Across Time ........................................................................................ 109 Chapter Five .................................................................................................................... 119 Discussion ................................................................................................................. 1 19 The Social Learning Theory/Anti-Social Behavior Trait Debate ............................. 125 Conclusion ................................................................................................................ 130 Limitations ................................................................................................................ 132 APPENDIX ..................................................................................................................... 135 REFERENCES ............................................................................................................... 140 vi LIST OF TABLES Table 2.1. Ecological Models ..................................................................... 31 Table 2.2. Ecological Model Comparison ....................................................... 33 Table 3.1. Independent and Dependent Violence Measures by Source and Wave. . . . . ....37 Table 3.2. Items for Repeated Measures of Violence (in Past Year) ........................ 39 Table 3.3. Attrition and Cases Responding by Instrument .................................... 47 Table 4.1. Comparison of Model Variables Across Waves ................................... 53 Table 4.2. PC Reports of Family Violence by Sex and Race or Ethnicity of Subject. . ...57 Table 4.3. Subject Reports of Exposure to Violence by Sex and Race or Ethnicity of Subject .................................................................................... 59 Table 4.4. Subject Anti-Social Behavior to Violent Offending .............................. 70 Table 4.5. Effects of Exposure to Violence on Rates of Intimate Partner Violent Offending — Contemporary and Cross-Sectional Models ........................ 73 Table 4.6. Effects of Exposure to Violence on Rates of Intimate Partner Violent Offending — Combined Model ...................................................... 76 Table 4.7. Effects of Exposure to Violence on Rates of Intimate Partner Violent Offending — Wave 2 Model .......................................................... 81 Table 4.8. Effects of Exposure to Violence on Rates of Intimate Partner Violent Offending — Wave 1 Model ........................................................... 86 Table 4.9. Effects of Exposure to Violence on Rates of Violent Offending in the Community — Contemporary and Cross-Sectional Models ...................... 91 Table 4.10. Effects of Exposure to Violence on Rates of Violent Offending in the Community — Combined Model ..................................................... 98 Table 4.11. Effects of Exposure to Violence on Rates of Violent Offending in the Community — Wave 2 Model ....................................................... 103 Table 4.12. Effects of Exposure to Violence on Rates of Violent Offending in the Community — Wave 1 Model ....................................................... 106 vii Table 4.13. Effects of Exposure to Violence on Rates of Intimate Partner Violent Offending - Comparison of Models ................................................ 11 1 Table 4.14. Effects of Exposure to Violence on Rates of Violent Offending in the Community - Comparison of Models ............................................. 114 Table A. 1. Correlation Matrix for Contemporary Models .................................... 136 Table A2. Correlation Matrix for Cross-Sectional Models ................................. 136 Table A3. Correlation Matrix for Combined Models ........................................ 137 Table A4 Correlation Matrix for Wave 2 Models ........................................... 138 Table A5. Correlation Matrix for Wave 1 Models ........................................... 139 viii LIST OF FIGURES Figure 1.1 Integrated Model of ETV, Aggression, and Violent Offending .................. 14 ix Chapter One Statement of the Problem “Youth violence exposure constitutes a major public health concern in the United States” (Hanson, Self-Brown, Fricker-Elhai, Kilpatrick, Saunders, & Resnick, 2006, p. 3; Youngstrom, Weist, & Albus, 2003; Margolin & Gordis, 2000). Research indicates that children and adolescents are exposed to violence at high rates — either experiencing or observing violence — both within their homes and in their neighborhoods (Hanson et al., 2006; Margolin & Gordis, 2000). Using a nationally representative sample of children and youth ages 2 to 17 years, Finkelhor, Ormrod, Turner, and Hamby (2005) found that 71% of the respondents reported a direct or indirect victimization in the study year — more than 50% experienced a physical assault, more than 12% a form of child maltreatment, and more than 33% witnessed violence or experienced victimization of another type indirectly. Child or youth victims experienced an average of three victimizations. Using a nationally representative sample of adolescent ages 12 to 17 years, Hanson et a1. (2006) also found nearly half reported some type of violence exposure in their lifetimes — 51% for boys and 44% for girls. Of these youth, over 22% reported physical assault or abuse — 15.5% for extrafamilial physical assault; about 10% for physically abusive punishment by a caregiver. Almost 40% reported witnessing violence — about 2% reported witnessing in home violence; over 39% reported witnessing community violence. Their data showed a strong relationship between intrafamilial violence in various forms and extrafamilial violence as well as the importance of witnessing violence as a risk factor for direct victimization. Adolescents can be exposed to interparental violence in several ways: they may see, hear, or be part of the actual event or experience outcomes, such as seeing bruises, witnessing an arrest, or leaving home to live in a shelter (Edleson, 1999; Kitzmann, Gaylord, Holt, & Kenny, 2003). Furthermore, the impact may extend beyond the initial exposure and consequences, continuing to influence the individual as an adult (Margolin & Gordis, 2000). Child abuse and marital violence typically occurs within the privacy of the home setting. However, community violence is often discussed around the neighborhood. Children may not directly witness an act of community violence, but may “form their own mental imagery of the event” based on hearing repeated accounts of the incident (Margolin & Gordis, 2000, p. 448). Victimization during adolescence occurs at a time when various life course developmental trajectories are being formed. Experiencing any type of violence during this period has implications for later psychological well-being, educational and socioeconomic achievement, and increased risk of violent offending and/or repeated victimization (Macmillan, 2001; Nofziger & Kurtz, 2005; McCloskey & Lichter, 2003; Widom, Schuck, & White, 2006; Youngstrom et al., 2003). Additionally, previous research has identified late adolescence as a critical period not only for the formation of attitudes, beliefs, and practices in romantic relationships but also for the formation of attitudes that endorse the use of violence in romantic relationships (Gorman-Smith, Tolan, Sheidow, & Henry, 2001; Vezina & Hebert, 2007; Wekerle & Avgoustis, 2003). Studies indicate that dating violence may be a precursor to spouse abuse (Simons, Lin, & Gordon, 1998). Theories of Violence Historically, research on violence and violent behavior typically investigated violence in the home and violence in the community as discrete phenomenon (Maxwell & Maxwell, 2003; Gorman-Smith et al., 2001; Fagan & Browne, 1994; Fagan & Wexler, 1987; Gelles, 1985) and tended to be descriptive or phenomenonological rather than with a theoretical base (Saunders, 2003; Margolin & Gordis, 2000; Emery, 2006). Indeed forms of family violence — child maltreatment, spousal abuse, sexual abuse, emotional abuse, witnessing interpersonal violence — were often studied in isolation as if univariate risk factors (Gelles, 1985; Williams, 2003; Hanson et al., 2006; Heise, 1998; Mohr & Tulman, 2000, Saunders, 2003). In addition, family violence and community violence have different theoretical explanations. In his article, Family Violence, Gelles (1985) described the history of family violence research and appraised its “discovery” as a both a social and sociological problem (p. 347). He identified special theories under development which were based on the View that family violence is a highly frequent event occurring within a social institution that is unique and distinct from the setting of other forms of violence. In addition, child abuse and spouse abuseI were explained by separate theories as well. as other theories were developed to explain family violence in general. Family violence, relative to other forms of violence, affects multifaceted, intertwined relationships between the persons involved in the violent event, which is shaped by the relationship that existed prior to the event and that will, in turn, shape the relationship following the event. In addition, over time and even within the same event, family members can be 1 See also Fagan and Browne (1994) for a history of theory and research about violence between adult partners. both offenders and victims of violence within the family (Tolan, Gorman-Smith, & Henry, 2006). “Family violence might be better viewed as a process not an event” (Williams, 2003, p. 443). Over 20 years later, in their article, Family Violence, Tolan et a1. (2006) wrote an updated history of family violence research. Although a sizeable amount of research had been conducted during the previous two decades into the patterns, risk factors, and interventions for family violence, much research continued to be distinct, focusing on scientific and policy interests specific only to a particular form of family violence. Likewise, conceptualization, definitions, and measurement issues remained within each of these major forms of family violence research which created difficulties in comparing findings (Tolan et al., 2006). The frequent use of retrospective designs and reliance on official records, clinical samples, or populations identified by the courts or social service agencies produced results which often differed from general population samples (Hanson et al., 2006; Margolin & Gordis, 2000; Wolak & Finkelhor, 1997; F antuzzo & F usco, 2007; Mohr & Tulman, 2000; Edleson, Ellerton, Seagren, Kirchberg, Schmidt, & Ambrose, 2007; Perry & Fromuth, 2005; Walsh & Krienert, 2007; Widom & White, 1997; Land, 2000; Holtzworth-Munroe, 2005). Yet, family violence research is coming to appreciate the extent of the common characteristics and complementary natures in the theoretical and empirical literatures regarding family violence exposure, the considerable heterogeneity in patterns of family violence, overlap in participants in the different types of family violence, (i.e. family violence in one form is a risk factor for another form to also be present in the same family) and the utility of this knowledge for investigating the degree and pathways that such exposure influences children (Tolan et al., 2006; Margolin & Gordis, 2000). However, in order to study all forms of adolescent exposure to violence (ETV) together, traditional theories on delinquency and aggression need to be tested in conjunction with family violence measures (F agan & Wexler, 1987) — indeed the “disconnect between the delinquency and the family violence literatures poses a gap that potentially misspecifies the relationships found between experiencing and/or witnessing family violence and aggression” (Maxwell & Maxwell, 2003, p. 1434). One of the “brute facts” widely recognized about crime in general is the invariance of the age-crime distribution curve over sex, race, time, place, and type of crime (Hirschi & Gottfredson, 1983, p. 552). The rate of criminal offending rises during the teen years, peaks in middle adolescence, and then begins a steady decline. In fact, Hirschi and Gottfredson (1983) argue that the age effect alone is so strong that study of juvenile vs. adult segments or the effect of other life-course factors (e.g. leaving school, marrying, or obtaining useful employment) are not only unnecessary but precluded by the effects of their correlation with the age effect. Therefore, “age is irrelevant in predicting subsequent criminality” (Hirschi & Gottfredson, 1983, p. 581). Other researchers contend that individual differences can distinguish between those who offend and those who do not, those on pathways to commit minor offenses from those who continue on to commit serious or violent offenses, as well as those who are at risk for a career of violent offending. Indeed, research should investigate “such features as the rate of offending, the pattern of offense types, and any discernible trends in offending patterns” (Blumstein, Cohen, & F arrington, 1988, p. 2; Loeber & Hay, 1997; Land, 2000). In the same vein, some researchers posit that theories and studies on child and adolescent exposure to violence and/or maltreatment would be aided by guidance from a developmental perspective about risk and impact (Margolin & Gordis, 2000; Williams, 2003; Salzinger, Feldman, Stockhammer, & Hood, 2002; Teisl & Cicchetti, 2008). Combinations in relationships between child maltreatment, interparental violence, community violence, and aggressive behavior have been explained by attachment theory, coercion theory, rejection sensitivity model, and social cognitive theory (Bradshaw, 2004; Schwartz, Hage, Bush, & Burns, 2006; Margolin & Gordis, 2000; Dodge, Bates, & Pettit, 1990; Teisl & Cicchetti, 2008). Community violence also tends to be described according to disorganization theory, theory of angry aggression (Patchin, Huebner, McCluskey, Varano, & Bynum, 2006), strain theory, and social learning theory (Patchin et al., 2006; Maxwell & Maxwell, 2003) In fact, social learning theory is one general theoretical model of violence often also applied to family or relationship violence (Gelles, 1985; Moretti, Obsuth, Odgers, & Reebye, 2006; Murrell, Christoff, & Henning, 2007; Peacock & Rothman, 2001; Chen & White, 2004; Stith, Rosen, Middleton, Busch, Lundeberg, & Carlton, 2000; Tolan et a1. 2006; Wolak & Finkelhor, 1997; Kingsfogel & Grych, 2004; Vezina & Hebert, 2007; Dodge et al., 1990). Adults fighting in the home provide real-life equivalents of models in social learning experiments (Bandura, Ross, & Ross, 1961) that demonstrated children imitating distinct aggressive behaviors of adult models (McCloskey & Lichter, 2003). Children who experience or are exposed to physical violence in their homes are socialized that violent behavior is “normal” and a legitimate means to resolve family conflict (Widom et al., 2006, p. 676; Tolan et al., 2006; Malik, Sorenson, & Aneshensel, 1997) and they will model behaviors considered likely to produce positive outcomes but avoid those likely to lead to undesirable outcomes (Simons et al., 1998). The social learning theory perspective also anticipates that children develop skill deficits and biases in processing social information, such as poor problem solving abilities, a bias to attribute hostile intent to the actions of others, difficulty managing anger, and difficulty communicating feelings, which increase the risk of using aggression or violence in response to problems (Dodge et al., 1990; Teisl & Cicchetti, 2008; O’Keefe, 2005; Nicholas & Bieber, 1996). The process is not limited to physical violence but may stem from care neglect and emotional maltreatment (Teisl & Cicchetti, 2008; Nicholas & Bieber, 1996). A family environment without reports of physical abuse should not be regarded as a non-abusive household if emotional abuse or psychological maltreatment are present (Nicholas & Rasmussen, 2006). Research has found verbal and psychological abuse to have a larger effect on victims than physical abuse; in fact, “the pathway from verbal abuse to physical violence is likely to be a focal point in understanding intergenerational violence transmission” (Langhinrichsen—Rohling, 2005, p. 113; Kwong, Bartholomew, Henderson, & Trinke, 2003). Consequently, research on aggression as an outcome from being raised in an abusive family environment needs to include measures of emotional abuse or psychological maltreatment (Teisl & Cicchetti, 2008; Nicholas & Rasmussen, 2006; Kwong etal., 2003). Early aggressive behavior is hypothesized as a mechanism in the “cycle of violence” from abusive households to violent behavior as adults by children from these households (Widom et al., 2006, p. 676). The mechanism of how exposure to violence relates to future abuse of others, however, is yet unclear. Social learning theory or a general learning model cannot explain why most individuals exposed to violence do not exhibit clinical problems, do not persistently abuse others, and do not become involved with the legal system as adults (Margolin & Gordis, 2000; Stith et al., 2000; Kwong et al., 2003; Heise, 1998; Fagan & Wexler, 1987; Heyman & Slep, 2002), how victims become offenders and the co-occurrence of offending and victimization (Widom et al., 2006; Malik et al., 1997), as well as the stability of aggression that persists regardless of the situation and across generations (Huesmann, Eron, Lefkowitz, & Walder, 1984). The exposure to violence process appears interactive — “Violence can influence development, family relationships, and support. These factors, in turn, can increase the likelihood of violence” (Williams, 2003, p. 444). In light of the variety of theories used to explain aggression, Anderson and Bushman (2002) developed the Generalized Aggression Model (GAM) as a framework to unify the “domain-limited” theories and as a mechanism to test interventions designed to reduce aggression (p. 27). Yet earlier, F agan and Wexler (1987) proposed an integrated theory of family and stranger (i.e. non-family) violence based on their assessment that both forms of violence include childhood exposure to violence as precursors to adult violence. Violence learned by children in their home environment is sanctioned and reinforced by subsequent interactions with peers and others in the community. As a natural extension, family conflict and a corresponding lack of parental monitoring weaken social bonds. Social learning theory accommodates both of these processes. Comparison of Theories Simons, Wu, Johnson, and Conger (1995) tested three perspectives, all based on social learning theory, on the mechanism of an intergenerational transmission of violence. The parenting perspective entailed “scripts” of severe and coercive treatment learned by children as normal to the role of being a parent. The family roles perspective asserted that children who are exposed to harsh discipline learn that to “hit those you love” is not only acceptable, but even at times necessary (p. 1), including not only children (as with the parenting perspective), but other family members as well. The antisocial behavior trait perspective considered antisocial behavior, like other traits, as stable across time and in various settings, which deve10ps, in part, from ineffective parenting and is manifested in deviant acts within both the family and community environments. Their analysis of data from 450 two-parent Caucasian families with children in seventh grade in rural, north-central Iowa provided support only for the antisocial behavior trait perspective as the framework able to predict associations consistent with their findings. Using data on 113 adolescent males from the same sample as above, Simons et a1. (1998) again tested the social learning perspective compared to the criminological perspective of a persistent antisocial behavior trait that accounts for male perpetrating dating violence. They found that interparental aggression did not predict dating violence but that frequent corporal punishment increased the risk of dating violence. These results were consistent with their view that corporal punishment often produces compliance, a behavior change as desired by the actor, which would lead to expectations that violence can produce positive outcomes, at least in the short term. Interparental violence, in their view, produces expectations of negative outcomes, such as threats, emotional reactions, or even violence. Consistent with the above views, their results indicated strong support for the antisocial behavior trait perspective. Low parental support and involvement was associated with the adolescent’s drug use and delinquency, which predicted dating violence. They concluded that dating violence and interparental violence are elements within a broader pattern of antisocial behavior. In contrast, Foshee, Bauman, and Linder (1999) found more support for social learning theory in their evaluation of the two theoretical perspectives of social learning theory and social control theory as explanations of the process in which exposure to family violence can lead to dating violence. They contend that children from violent households would view both forms of family violence as producing effective and positive outcomes. Based on their data of 1,965 eighth- and ninth-grade students in public school in rural North Carolina, acceptance of dating violence and an aggressive conflict- response style, as proposed by social learning theory, mediated the relationship between exposure to family violence and dating violence for both genders. For males only, the relationship was also mediated by positive outcome expectations, consistent with social learning theory, as well as beliefs in the conventional rules of society, from social control theory. Ecological Models The Fagan and Wexler (1987) integrated theory mentioned earlier accounts for influences toward patterns of violent behavior at the individual, situational, and environmental levels. Interest is growing in the use of ecological models as frameworks to structure family violence and/or community violence research (Tolan et al., 2006; Hanson et al., 2006; Little & Kantor, 2002, Salzinger et al., 2002). Ecological models were first used as multidimensional approaches to assemble and coordinate the diverse research findings about child abuse and neglect and later about battering (Heise, 1998). 10 A nested, ecological model of violence would broadly conceptualize the complex patterns and interactions between individual, family, community, and societal risk factors as well as protective factors at each level as identified by research, all within the “full complexity and messiness of real life” (Heise, 1998, p. 262; Little & Kantor, 2002). Therefore, such a model assumes that the ecology of the adolescent influences their level of exposure to violence and the differential impacts of that exposure (Hanson et al., 2006). Family violence is influenced by factors at many levels: the individual, the relationship, the home environment or other setting, and the broader neighborhood context (Tolan et al., 2006; Gelles, 1985). For example, risk factors that contribute to interparental violence may be different than for child abuse because of differing relationships between perpetrator and victim (Tolan et al., 2006). In addition, research indicates characteristics of adolescent perpetrators of dating violence are similar to those who commit general crimes (Banyard, Cross, & Modecki, 2006). An ecological model can integrate family violence factors and yet distinguish violence that occurs within the family environment from violence that occurs outside the home in order to better investigate any relationship between exposure to family violence and exposure to community violence (Tolan et al., 2006; Hanson et al., 2006). “Careful study of multiple ecological factors, with adequate consideration of relationship factors, is important to gain a more complete understanding of the various forms that arise and how they interrelate” (Tolan et al., 2006, p. 571). Ecological models, by integrating theories of exposure to violence, allowing a structural overlap in violence exposure occurrences, and incorporating shared risk and protective factors, can also lead to the design of potential intervention strategies and 11 policyz. Ecological models for interventions focus on individual and community strengths and assets rather than perceived deficits. In addition, incorporation of the ecological levels within the intervention program provides access to family, peer, community members, and community institutions for support and accountability. Some interventions that accounted for multiple determinants of violence in their application to family violence and dating violence have shown positive effects (Tolan et al., 2006; Peacock & Rothman, 2001). Whether an ecological model is used to design intervention and prevention strategies or not, a better understanding is needed of the various reactions that children may exhibit from exposure to violence, the interactions of multiple forms of violence, how reactions may also vary by developmental stage of the child and environmental characteristics, and what systems (family, community, schools, justice, and foster care) can do to protect children and teach positive approaches to handling conflict in order to effectively break the intergenerational cycle of violence (Margolin & Gordis, 2000; O’Keefe, 2005; Kwong et al., 2003; Patchin et al., 2006). The presence of one or more of the risk factors for family violence does not guarantee a family will exhibit family violence in any or multiple forms. Therefore, the ecological model needs to incorporate the paths to violence as well as those to the desistance from committing violence when framing the risk model (Tolan et al., 2006). Interventions need to happen early or as any change process is more difficult once Violent patterns have been established (Widom & Maxfield, 2001; Crockett & Randall, \ 2 Described here is the general ecological model. An ecological-transactional model aims to develop Prevention programs “to restore the individual to normative and developmentally appropriate trajectories Without ‘blaming the victim’ in the process. As a result, only those biological, psychological, and social risk factors that can be addressed in transactional terms are targeted for preventive interventions” with an ecological-transactional model (Alcantara & Gone, 2007). 12 2006) — the individual’s choice and readiness to change are vital to any lasting change (Babcock, Canady, Senior, & Eckhardt, 2005). “In this instance, individual action needs to align with the social structure to produce behavioral change and to maintain change (or stability) over the life course. Choice alone without structures of support, or the offering of support alone absent a decision to desist, however inchoate, seems destined to fail. Thus, neither [human] agency nor structural location can by itself explain the life course of crime (cf. Wikstrom 2004)” (Sampson & Laub, 2005, p. 19). Limitations of Previous Research In 2007, the final wave of data from the Project on Human Development in Chicago Neighborhoods (PHDCN) became available through the National Archive of Criminal Justice Data. In contrast to the Simons et al. (1995, 1998) and Foshee et a1. (1999) studies mentioned earlier, the PHDCN collected data from a general population sample of adolescents, not a sample drawn from a school setting from which adolescents involved in delinquent behavior may be underrepresented or absent. The PHDCN data also provide the ability to test the social learning theory/antisocial trait behavior debate with longitudinal, prospective data collected from multiple informants from an ethnically diverse urban sample of both males and females. In addition, these data allow for the adolescent’s history of offending during exposure to be controlled when examining the effect of in-home vs. community exposure to violence on subsequent offending as adults. This project utilizes the developing trends to conceptualize the study of exposure to Violence and offending within an ecological framework. The Present Study Figure 1.1 uses the three inner layers of Little and Kantor’s (2002) integrated ecological model of family violence to illustrate the ecological levels that comprise the subject’s environment as conceptualized in this paper. The figure also depicts the interrelationships between the ecological levels and their impacts on committing intimate partner and community violence as young adults. At the individual level are the subject’s characteristics of sex, race, aggressiveness, and age as well as personal histories of exposure to violence. At the family level are characteristics of the family and household. Although the exposure to violence measures are conceptualized at the individual level, due to their effects on the subjects, the community is also included in Figure 1.1 in recognition that the community, like the home, is an environment in which some of the violence the subjects reported took place. Factors in the ecological model are individually identified in Table 2.2. Figure 1.1. Integrated Model of ETV, Aggression, and Violent Offending Committing Intimate Partner Violence Committing Community Violence ° I \\ Family ’I \ ” ~-— Community l4 This project hypothesizes that adolescent exposure to violence in any setting will be positive and significant to offending as adults after statistically controlling for other factors. Specifically, the questions this project seeks to address are: 0 Does witnessing intimate partner violence between a primary caregiver and a partner or experiencing child-directed aggression during adolescence correspond to committing intimate partner violence in later romantic relationships? 0 Does witnessing or experiencing violence in the community during adolescence relate to future incidents of committing community violence? 0 Do situational crossovers occur, e.g. does exposure to caregiver-partner violence during adolescence or experiencing child-directed aggression during adolescence correspond to future incidents of committing community violence, or vise versa? Overall, this project investigates the intergenerational transmission of violence — whether “violence begets violence” while controlling for the family environment as well as the adolescent’s self-reported aggressiveness, sex, and race. In doing so, the factors of violent behavior learned through exposure to modeled violent behavior vs. violence from an inherent anti-social behavior trait will be examined individually as well as in relation to one other. Therefore, the dependent measures correspond to criminal behavior rather than aggressive behavior, which is a control variable. For this project the conceptualization of violence is not abuse, as no measure for injury could be included. Chapter Two Literature Review Despite their independent development, the theoretical and empirical literatures on interfamilial and extrafamilial violence agree substantially on many issues (Fagan & Wexler, 1987). This literature review focuses on studies that considered adolescent exposure to violence, particularly within the family, as a socializing factor in predicting involvement in future violence and uses exposure to violence as the common tie between these two theoretical perspectives. Achieving an understanding of the empirical relationship between interfamilial and extrafamilial violence can serve to enhance the independent explanatory power of the theories within each discipline and advance the development of an integrated theory of interfamilial and extrafamilial violence (Fagan & Wexler, 1987). First, studies on aggression and a history of family violence are discussed separately. Second, results from studies that investigated multiple forms of exposure, whether both within the family or in the family and community, are described. Studies that examined physical violence as well as non-physical maltreatment (e. g. psychological or emotional maltreatment and supervisory neglect) are discussed. Third, the review presents results from studies regarding issues of cumulative effects and effects over time. Fourth, studies that employed multiple informants in their investigation of exposure to Violence are presented. The review ends with the ecological framework as a mechanism to structure common risk factors between these family violence and general violence perspectives, with the goal of testing the relationship between the prevalence of 16 intrafamilial and extrafamilial violence and the risk factors or characteristics that distinguish between these groups (Fagan & Wexler, 1987). Research on Aggression A complete review of all research findings related to adolescent exposure to violence is not feasible within this project3. One starting point is to consider what is known about the development of aggression and violence by children, irrespective of setting. Huesmann et a1. (1984) analyzed data on more than 600 subjects whom they followed from age 8 to 30. They found that child predisposition and precipitating situational factors converge to explain severe antisocial aggressive behavior. Aggressiveness exhibited in school in the early years corresponded with severe antisocial aggressiveness as an adult, including committing crime and child abuse for both genders, and spouse abuse for males. Although some situations appeared to elicit aggressive behaviors more than other situations, the individual’s aggressiveness relative to their position in the population remained largely constant — “the child who is at the top of the distribution for 8-year-olds is likely to be near the top of the distribution for 30-year—olds two decades later” (Huesmann et al., 1984, p. 1131). Loeber and Hay (1997) found heterogeneity in the stability of aggression — those initially least aggressive and those initially most aggressive showed the most stability in aggression. They similarly Concluded that aggression, like other personality factors, tends to become more stable 3 Reviews or references to literature on the various aspects of violence are available elsewhere: see Loeber and Hay, 1997 regarding the biological bases of aggression and violence and surveys on juvenile aggression; see Edleson, 1999 for reviews of family violence research; see Tolan et al., 2006 for risk factor research on family violence; see Saunders, 2003 for childhood exposure to violence; see Vezina and Hebert, 2007 for risk factors for victimization in romantic relationships for young women; see Nofziger and Kurtz, 2005 for psychological outcomes of juvenile exposure to violence; see Buka, Stichick, Birdthistle, and Earls, 2001 for secondary (witnessing) community violence; see Salzinger et al., 2002 about risk for CXposure to community violence and effects of exposure on children and adolescents. 17 with age. Yet, while the level of aggression may remain stable, they found the severity of aggression may increase, remain constant, or decrease over time. Although being raised in households with a high rate of “coercive interchanges” is common to both aggressive and violent behaviors, whether violence develops may depend on the convergence of risk factors along with aggression (Loeber & Hay, 1997, p. 399). For example, Lewis, Lovely, Yeager, & Femina (1989) concluded that, “neuropsychiatrically impaired children by their hyperactivity and impulsivity seem to invite abuse” (p. 436), although this premise is controversial (Dodge et al., 1990). Lewis et al. (1989) conducted a follow-up study on 95 previously incarcerated juvenile delinquents for seven years into adulthood. Adult arrest records for aggressive offenses were found for 77% of those classified as seriously violent as juveniles, but also for 61% not classified as seriously violent as juveniles. Therefore, using early violence alone as a predictor of future adult violence misclassified 23% of the more violent juveniles and 61% of the less violent juveniles. Their results also showed that a juvenile having one kind of clinical vulnerability and exposure to abuse or violence was not more related to perpetrating adult violence than having multiple vulnerabilities only or being raised in an abusive, violent family only. Most associated with the juvenile’s perpetration of extreme aggression as an adult was the combination of multiple clinical vulnerabilities and being raised in an abusive, violent household. Therefore, stability in aggression appears to be a combination of continuity of an individual’s natural tendencies as well as continuity of situational factors (Huesmann et al ., 1984). Adolescent aggression can vary by setting, with more evidence of 18 generalization from aggression in the home to aggression also at school than the other way around (Loeber & Hay, 1997). Research on Family Violence Prior to the 19703, the scientific view and public perception of abuse and violence in the family was framed through a medical model in which abuse was suspected to be caused by psychopathology and social factors were considered irrelevant. As such, family violence was seen as rare, involving only people with mental illness or poor people, and the idea that “battered women liked being abused” (Gelles, 1985, p. 350). After decades of research the prevalence of family violence is well-documented, leading Tolan et a1. (2006) to assert that “family violence affects many persons at some point in their life and constitutes the majority of violent acts in our society” (p. 557). Hanson et a1. (2006) reported from their investigation of family environment and youth violence exposure that “many children have been the victims of and/or witnesses to several types of violence in their lifetime” (p. 4). However, research that focuses on one form of child and adolescent victimization at a time “miss a much bigger picture” (Finkelhor et al., 2005, p. 18). Therefore, over time, research began to investigate the affect of one form of family violence on the presence of another form of violence involving one or more of the same family members. In the same vein, a few studies have also investigated the effects of family violence and community violence together. Research on Multiple Forms of Exposure to Family Violence Some studies that looked at multiple forms of violence exposure draw on support from social learning theory, generalized and specific, with or without examining gender effects. Kalmuss (1984) analyzed data from a nationally representative sample of over 2,100 adults. Cases of co-occurrence of spousal assault were excluded in order to investigate sex-specific modeling. Her results indicated that being hit by a parent during adolescence in the family of origin was less associated with later involvement in severe marital aggression than witnessing their father hit their mother or their mother hit their father. Observing the father hit the mother increased the likelihood the sons and daughters would be both perpetrators and victims of severe martial aggression. Cappell and Heiner (1990) analyzed the same data, retaining cases of co-occurrence of spousal violence and limiting cases of child abuse to the past year. They found reports of spousal assault in the family of origin increased the likelihood that respondents of either sex would be a target of their spouse and parent-child aggression during childhood increased the likelihood that females would be aggressive to their children. Their results showed no sex-specific pattern of learned perpetration. They concluded the lack of association between family-of—origin aggression to the respondent’s committing spousal or child abuse implied a different source for the perpetration of these acts by males. Heyman and Slep (2002), however, found same-gender modeling effects for perpetration of violence toward a partner or child. Using over 6,000 cases from a later version of a similar nationally representative dataset, father-to-mother violence increased men’s risk of perpetration and mother-to-father violence increased women’s risk of perpetration. In 20 addition, childhood victimization increased a woman’s risk of being a perpetrator, victim, or both in an abusive partner relationship, with mother-to-daughter violence related to victimization and father-to-daughter violence related to perpetration. Maker, Kemmelmeier, and Peterson (1998) surveyed over 130 college women. Those who reported witnessing physical contact between their parents before the age of 16 also experienced more physical abuse and parental substance use during childhood than the non-witness control group. Their reports of dating violence were also higher, even after controlling for other risk factors. Carr and VanDeusen (2002) also surveyed students - 99 undergraduate men from a large Midwestern university. They found the men’s witnessing of parental physical violence, but not experiencing child abuse by a parent predicted their perpetration of dating violence. In another survey of over 600 college students conducted by Hendy, Weiner, Bakerosfskie, Eggen, Gustitus, and McLeod (2003), partner or parent models explained violence to the respondent’s current partner for both genders, with the present partner as the most powerful model. Violence the respondent received from the mother was the most powerful parental model for 7 violence in their present relationship, but for women, this model was associated with victimization whereas for men, the model was associated with both victimization and offending. Kwong et a1. (2003) found similar results with their survey of over 1,200 adults in the City of Vancouver. Abuse between parents during childhood and abuse of either parent to the respondents when they were teenagers predicted violence in the respondent’s current relationship. No gender-specific or role-specific patterns of learned behavior were noted regarding perpetration or victimization, particularly that their 21 mother’s violence predicted the respondent’s current relationship violence as did their father’s violence. Using a probability sample of adolescents from a medium-sized city in the Philippines, Maxwell and Maxwell (2003) found that the more the children witnessed aggression between their parents or experienced direct aggression from a parent, the more aggression they reported committing, even when controlling for other predictors. Witnessed aggression accounted for more of the variance in self-reports of aggression by the adolescents than did their experiences of direct aggression. The influence of neither family violence measure varied by gender. In their 20-year longitudinal study of over 520 children living in two upstate New York counties, Ehrensaft, Cohen, Brown, Smailes, Chen, and Johnson (2003) found that childhood conduct disorder for both genders posed the strongest risk for perpetrating partner violence, with exposure to violence between parents and “power assertive” punishment (p. 741) the next two risk factors. The largest independent risk factor for victimization of any partner violence was exposure to violence between parents. Conduct disorder increased the odds of partner violence victimization but was not a mediator of this effect. Research on Psychological Maltreatment and Supervisory Neglect Nicholas and Bieber (1996) studied the relationship of emotional abuse in childhood separately from physical abuse in relations to aggression and hostility in adulthood. In their sample of 216 predominately White middle-class psychology undergraduates from a western university, they found the student’s retrospective accounts of exposure to emotional and physical abuse from their parents were both related to higher aggression scores for the student as assessed by the aggression subscale to the Buss-Durkee Hostility Inventory (BDHI). Emotional and physical abuse by parents was 22 significantly related to female students’ self-reports of physical fights within the family since age 18, but not for the male students. When examining only the higher and lower abuse groups, however, the gender difference largely disappeared. Of particular importance was that these middle-class student reports of exposure to fairly low levels of emotional abuse by their parents was still significantly related to the students’ assessment scores of aggression. Using a convenience sample of 298 primarily Caucasian middle- class university students, Nicholas and Rasmussen (2006) found the effect of emotional abuse and witnessing inter-parental violence on aggression varied by gender of the child and gender of the parent. Father emotional abuse was the only significant predictor of aggression for the female students; emotional abuse by either parent did not predict aggression for the male students. For the male students, physical abuse and witnessing inter-parental violence significantly predicted the amount of reported aggression; however, higher aggression was predicted by higher reports of physical abuse by fathers and lower aggression by higher reports of physical abuse by mothers. In their studies on parental neglect, poor supervision, and punitive parenting using a sample of 2 1 8 disadvantaged children ages 4 to 8 from two states, Knutson, DeGarmo, Koeppl, and Reid (2005) found that poor supervision, i.e., inadequate awareness, significantly predicted levels of neglect and harsh discipline (ranging from yelling, scolding, and spanking to physical abuse), and harsh discipline significantly predicted levels of child aggression. These results suggest that caregiver awareness is a factor in establishing aggressiveness in childhood, even before adolescence. 23 Research on Exposure to Family Violence and Community Violence Using reports from interviews with 270 domestic violence female victims, Fagan and Wexler (1987) found that 57% of the women’s batterers had been exposed to family violence as a child — 45% were victims of child abuse and 43% witnessed spousal violence. Nearly one third had a history of both forms. A little more than half of the batterers had been “generally” violent, i.e., violent to others as well as their partner and over 80% of these generally violent men had been arrested for their violent behavior. Of the batterers who had been victims of child abuse, 67% were also violent outside of the home (vs. 46% for those that were not child abuse victims). Of the batterers that had witnessed spousal violence, 61% were also violent outside the home (vs. 44% for those that had not witnessed of familial violence). Murrell et al. (2007) analyzed data on almost 1,100 adult male batterers to study the extent that violent behavior generalized from intimate partner violence to other forms of violent and nonviolent criminal behavior depending on the man’s childhood exposure to violence: men who had witnessed, had been abused, neither had witnessed nor had been abused, or both had witnessed and had been abused. Similarly, they found the likelihood of committing general (non-intimate partner) violence and child abuse as well as the frequency and severity of intimate partner violence increased as the man’s childhood exposure to violence increased. However, non— violent criminal behavior did not increase with childhood exposure of either form. In their study of over 700 high school students, Malik et a1. (1997) found co- occurrence of victimization and perpetration among their cross-sectional sample as well as crossover effects — exposure to violence in the student’s family was associated with their involvement in community violence and dating violence, and exposure to violence 24 in their community was strongly associated with involvement in dating violence and community violence. Boys’ involvement in dating violence appeared more attributable to social modeling, but girls’ involvement in dating violence appeared more influenced by their direct personal victimization. Gorman-Smith et al. (2001) used longitudinal data on over 140 African-American and Latino males, age 15 tol9 years old, to study how the perpetration of street violence relates to partner violence in a dating or marital relationship. They identified four groups: participation in neither type of violence, partner violence only, street violence only, and participation in both partner and street violence. They found that adolescents were more likely to report partner violence if they also reported participating in street violence. Adolescents participating in both types of violence had significantly poorer functioning families than adolescents reporting no violence. These two groups were also found to be separate than the partner violence only and street violence only groups. These latter two groups were not significantly different. Using the same nationally representative sample of adolescents as Hanson et a1. (2006) (described on p. 1), Nofziger and Kurtz (2005) investigated how lifestyles contribute to adolescent exposure to violence and subsequent offending. They report that being a victim of physically abusive punishment by a parent or other caregiver increased the risk for violent offending by 70% and having been a victim of physical assault increased the risk of offending by 226%. Their data showed “all forms of exposure to violence were highly associated both with each other and offending,” indicating that lifestyles that place adolescents in situations where violence occurs increases the adolescent’s likelihood of both victimization and offending (p. 21). 25 Overall, these studies consistently show exposure to violence in adolescence is associated with abuse in the individual’s future relationships or in the community. Exposure to parental violence during childhood appears potentially to be a more important risk factor for later violent behavior than experiencing abuse by a parent. However, either association is often only weak to moderate, leading to the conclusion that, although consistent, “Social learning theory can only explain a small part of a very complex picture” (Kwong et al., 2003, p. 299; Stith et al., 2000; Sampson & Lauritsen, 1994) Cumulative Effects and Effects Over Time Although “conflict is an inevitable part of all human association. . .violence as a tactic to deal with conflict is not” (Straus, Hamby, Boney—McCoy, & Sugarman, 1996, p. 284). Research indicates that chronic or severe violence stems from cumulative effects. Murrell et a1. (2007) also found that the “generality, frequency, and severity of violence and psychopathology all increased as level of childhood exposure to violence increased” (p. 523). Similarly, Patchin et a1. (2006) found in their study of 187 youth between 9 and 15 years old in a moderately-sized Midwestern city, those who witnessed more community violence were also more likely to self-report committing assaults and carrying weapons. Regarding witnessing, the analysis by Nofziger and Kurtz (2005) found that adolescents who reported seeing half the types of violent acts listed in the survey (3 out of 6) were 486% more likely to engage in violence and those who had seen all six types were 3,335% more likely to perpetrate violence. As mentioned earlier, Lewis et a1. (1989) found that the combination of multiple clinical vulnerabilities and 26 being raised in an abusive, violent household were most associated with the juvenile’s perpetration of extreme aggression as an adult, but not either factor alone. However, the effects of exposure to violence appear to dissipate absent of continuing abuse. White and Smith (2004) reported results from their longitudinal study involving over 1,560 women about the risk of physical and sexual assault among university students. They concluded that, “physical and sexual dating violence are normative,” based on their finding that 88 percent of the women in their study reported being a victim of physical or sexual violence from adolescence to their fourth year in college. Their results indicated that the greatest risk of physical dating violence is in high school, with the percentage of victimization continuing to drop throughout the four years of college. They found that the greatest risk factor for dating violence victimization in adolescence was victimization during childhood (physical abuse, sexual abuse, or witnessing domestic violence in the home). However, the greatest risk factor of victimization in college was victimization in adolescence, independent of victimization as a child. Women, who had been victimized in childhood but not during adolescence, were at no greater risk of dating violence victimization in college than those who had not been abused as a child. Stemberg, Lamb, Guterman, and Abbot (2006) in their longitudinal study of family service case files on 110 Israeli children in middle childhood to late adolescence found that exposure to family violence in middle childhood (time 1) produce more consistent effect on the youth’s adjustment than exposure to family violence in adolescence (time 2). Children who were both victims of physical abuse and had witnessed parental violence had high problems scores, but typically did not show more 27 problems than children exposed to only one of the types of family violence studied and sometimes were not discernibly different than the control group. Also, effects of exposure to violence in childhood did not persist when abuse did not continue in adolescence. Edleson (1999) also reports that the effects of exposure to violence can decrease as the length of time from the last event increases. Data from Multiple Informants The Stemberg et a1. (2006) study previously referenced, used parents, teachers, and the child as separate informants of child adjustment. Along with these three informants, social workers also provided information about the history of family violence. They found by and large low agreement among reports by the types of informants, with correlations averaging .20 for time 1 and .26 for time 2. The stability of reporting by the informant from time 1 to time 2 was higher, with coefficients averaging .33. Reports by mothers and teachers seemed to indicate greater awareness of the child’s externalizing problems whereas children showed greater awareness of their internalizing problems, particularly those stemming’from private and subjective experiences. Many families changed in their abuse group status (Victim, Witness, Abused Witness, No Violence) from time 1 to time 2 and the children’s reports of levels of maladjustment tended to vary more than the other informants based on their recent or contemporaneous exposure to family violence. Kuo, Mohler, Raudenbush, and Earls (2000) analyzed demographic risk factors associated with children’s exposure to violence using the PHDCN data as well as how information provided about ETV varied by informant. They found that both older children and their caregivers were more likely to report higher ETV scores than younger children and their caregivers. In addition, all boys reported 28 significantly higher ETV scores than the girls. However, caregivers appeared to under- report ETV more for boys than girls, although they tended to under-report ETV for both genders. The under-reports in both cases (for boys, specifically, and both genders, generally) is perhaps due to less time spent in the home. For both studies, conclusions about the effects of ETV could have been very different if only one type of informant had been used and underscore the value of designing research that obtains information from multiple respondents and uses statistical techniques that can analyze them separately. Defining Environmental Factors with Ecological Models The control that an adolescent has over their presence in settings can affect the dynamics of their exposure to violence in those settings — youth have little control over exposure at home on in school but may have more control over their activities in the community that increase their exposure to violence and their likelihood of offending (Nofziger & Kurtz, 2005). In turn, research is needed to disentangle what factors influence if resulting aggression is general in nature or context-specific (Maxwell & Maxwell, 2003). Similarly, research is needed to identify the role of setting in adolescent exposure to violence and future victimization — “each type of victimization can lead to diverse outcomes, and alternatively, diverse types of exposure to violence can lead to the same developmental outcome” (Margolin & Gordis, 2000, p. 469). Information is also needed to understand how types of victimizations cluster, how experiencing one type may increase the likelihood of experiencing others, and why some individuals endure multiple victimizations or revictimizations (F inkelhor et al., 2005; White & Smith, 2004). The effect of one type of victimization can be overestimated unless the analysis recognizes and controls for the contemporaneous presence of other forms (Finkelhor et al., 2005). In fact, the co-occurrence of victimization and offending is known (Malik et al., 1997; Kelleher, Gardner, Coben, Barth, Edleson, & Hazen, 2006; Margolin, Gordis, Medina, & Oliver, 2003; Cappell & Heiner, 1990; Straus, 2006) and this “bidirectionality of relationship violence” makes predicting violent offending and violent victimization independently difficult (Kwong et al., 2003, p. 299) as well as understanding the dynamics of the violent event (Swan & Snow, 2006). The difficulty, then, with considering multiple forms of violence at one time is in considering multifaceted socializing experiences while disentangling what are the unique effects from any cumulative or interactive effects contributing to each outcome, controlling for the effect of the others (Margolin & Gordis, 2000; Carr & VanDeusen, 2002). The various forms of emotional maltreatment and physical abuse co-occur (Knutson, DeGarmo, & Reid, 2004; Kwong et al., 2003). As Saunders (2003) wrote, “Unfortunately, the sheer volume of what is known about different forms of childhood violence, the many potential outcomes that have been shown to be related to a history of violence in childhood, and emerging research on mediators and moderators makes conducting comprehensive research a significant theoretical and technical challenge” (p. 356) As mentioned earlier, ecological models are useful to connect and structure such varied research findings across individual, family, community, and societal levels. Table 2.1 lists several studies that designed an ecological model to address a certain aspect of family violence and/or exposure to community violence. The items listed in these models summarize many of the factors described throughout this literature review. 30 8:50:80 .593 o 8:39“ 3538.35 o $530.5 Rome—om 0 86:0: 95 $53 0 met”: 3:203 base a sewage Ewe. e5 3:230.“on 0 2888385 3392A— o 00:20? 188893:— o co 35:33. o mus—8 hogow Ema o 5325.8 28 206.8%“ a Bee: €5.83: . 5:53 we 958:3: 5:222:20 222 o AEBmmmobeEV _o>o_ 328m $28.: 322836 a mono—to £58800 0 8286 Eoom 0 03:53.86 3:5:on o 095:0 o 822:8 2 mmooo< 0 Ease; no: . 3:253 Eoom o 32:83: Be: 6255—09 0 2:5 :5 5:83 Mo 20:28. o “5:283:55. \mmm Boi— o eooeofiwez 356335 626— bE::::oO w2.to~.~2o22 “232% o fl2oaeu 8 22222.32; 0 “2332.225 3.228% 0 a2oaum~eomm£o3 2522395222 3.30 o 223292.52» A22§K o 8.528 Neoxmgeste: 0 Sec 5.2: M23: 323: 32 0 523232 2.252333% 0 22222832233 0 3283\o ms 0 392% 3.2623 «A: «ME 0 Argon 9.229220 0 SEEKS: 235 3.3 ofioofi SSEQBE o 8.228 $22822; 835%»: 382:2.E o 2.2 2283B 3.228 32: o 3.25% 6823 amfiek o ”rename“. bNEek o ESQ \Efigwexemwfi 23:22:62 339.2235 0 .3333: 2:22 azfiek 0 9.2.3.... M222m22n~ o 2: 2.2 2232.283 32: 0 33% 3.223% moHEoOmme So: 5359 a 25:6 368 .60: 0 Even mfinmecotm o 53> “ram—5930M 8:22.» 8 8:8qu 0 33% 8:882 we? own dew—236850 8:203 353: 89:35 wm§%_=-:o: .BEWE :m_m< .begw ”monmtouofimno £06850: .8: 8.233% 55m.“ $38.? .5 Emma? o AmoszoHoesz 281022 3&2 25%5 oEa—EonoD 0 “83:82:85 «:83 o 35 . . .5otoE<-:eoE< ”comm o @2285: omEoemo/x 0 232536 68038: a 3 28:0 33% wEom o \boummm ow< o :o:_:woo Eoom o .8320: .8050:th 320 a 8 32203 3.550% xom o SSS—om o ”8:322:20 RED 0 fits:— wEmmoEB o _o>o_ 126365 mazmemkm 8226.5 8236.5 «0.2222225O 22223323 3.25 32: 2229: Examw 822e,: 22¢ 22o222o232m 5.225% 2 3:8me 8239: 2252i 8222.25 625 .a a :82: age .a a 39:58 Ages :25. a. 2:5 33: as: 2282 32438: .3 «say 31 Hanson et al. (2006) designed an ecological model and tested the unique associations between family environment and different types of violence exposure: physical and sexual assault in the home or community as well as witnessed violence in the home or community. Results from their study were reported at the beginning of Chapter 1. Regarding associations with family environment, they found exposure to one type of violence was more strongly correlated with other types of violence exposure, including between intrafamilial and extrafamilial violence, than any family environment variable. Intrafamilial and extrafamilial physical assault and community violence were related to both family alcohol use and drug use, but only family drug use correlated with domestic violence. Adolescents reporting any violence exposure were more likely to also report not always living with both natural parents, and intrafamilial physical and sexual assault and witnessing violence in the home or community were associated with not always living with a natural parent. However, family income level was not significantly correlated with violence exposure. Despite all the research described above, the overlap in types of exposure to family violence as well as any overlap or qualitative differences between exposure to family violence and exposure to community violence need further investigation. In addition, the role of exposure to family violence and community violence to future offending, including gender parity in offending, remains unclear (Hamby, 2005; Raudenbush, Johnson, & Sampson, 2003; Gorman-Smith et a1. 2001; Buka, Stichick, Birdthistle, & Earls, 2001; Fagan & Wexler, 1987). To explain these issues requires data that captured all these factors from similar measures within one dataset. 32 Table 2.2 Ecological Model Comparison Hanson et al. (2006) This project Dependent variables 0 Domestic violence 0 Community violence 0 Intimate partner violence offending 0 Community violence offending Individual level (Personal History/ Characteristics) 0 Sexual assault by family member 0 Extrafamilial sexual assault 0 Physical assault by caregiver o Extrafamilial physical assault 0 Sex 0 Age 0 Race: African-American, Hispanic, Native American, Asian (White, non-Hispanic was reference group) 0 PC reports of family violence: 0 Child-directed aggression o PC-partner violence 0 Subject reports of violence exposure: 0 Experience home & community violence 0 Witness home & community violence 0 Sex 0 Race: Hispanic, African-American, other race (White, non-Hispanic is reference group) 0 Age (considered a constant) 0 Adolescent self-report of offending and YSR/YASRflgressive behavior score Family level (Microsystem) o Chemically dependent families 0 Not always living with both biological parents 0 Household income less than $1 5,000 o Chemically dependent families 0 Not living with both biological parents 0 Household income less than $20,000 0 Number of people living in the home 0 Level of supervision Table 2.2 shows how this project’s ecological model compares to the model used by Hansen et al. (2006). Using data from all three waves of the Project on Human Development in Chicago Neighborhoods (PHDCN), this project will test the above hypotheses and the social learning theory/antisocial trait behavior distinction by applying the Hanson et a1. (2006) ecological framework on exposure to violence and victimization to the longitudinal, prospective PHDCN data on adolescents in Cohort 15 followed into adulthood. This project will expand their ecological model with additional individual and family environment factors from other models described in Table 2.1 as well as apply the new model and extend the analysis by also investigating the adolescent’s exposure to violence and their aggression over the entire period to their offending as young adults. 33 Chapter Three Methodology In order to investigate associations between the various forms of adolescent exposure to violence to young adult offending, this project will utilize data from all three waves of the Project on Human Development in Chicago Neighborhoods (PHDCN). The role of intimate partner violence (IPV) and child-directed aggression as reported by the adolescent’s primary caregiver will be examined as well as the adolescent’s reports of witnessing and experiencing violence in the home and community as predictors for committing intimate partner violence or community violence by the now young adults. In other words, the aim of this project is to determine whether exposure to violence during the ages of 14 to 19 predict offending as young adults up to 22 years old. Specifically, this project is designed to address some issues noted by Hanson et al. (2006). This study will differentiate between multiple types of violence exposure: witnessing intimate partner violence between their caregiver and partner, experiencing child-directed aggression from a household member, and witnessing and/or experiencing violence, both in the home and in. the community. The project will also investigate the independent as well as combined effects of these different types of exposure on the development of the youth over the seven year period. All exposure to violence measures relate only to violence that can be learned through modeling of violent behavior in order to make the measures comparable and to not confound these measures with property or other crime types (Sampson & Lauritsen, 1994). Associations between family environment and these different types of violence exposure are also examined. The 34 investigation of an association between these factors and committing partner aggression and/or committing aggression in the community is an extension to Hanson et al. (2006). Data Collection By design, the PHDCN data were collected to study the “development of crime and violence in children and adolescents” (Liberman, 2007). The PHDCN provide a general population, rather than a school-based population, of adolescents living in Chicago from 1995 to 2002. The longitudinal nature of the data provides opportunity to evaluate the contemporary and prolonged impact of adolescent exposure to violence (Buka, Stichick, Birdthistle, & Earls, 2000). The PHDCN combined Chicago’s 847 census tracts into 343 neighborhood clusters (NCs), which were designed to be ecologically meaningful clusters based on housing density, family structure, and 21 strata representing seven racial/ethnic composition groups and three levels of socioeconomic status. Using knowledge of Chicago’s neighborhoods, the researchers created the NCs, each containing about 8,000 people, using geographically contiguous census tracts and geographic boundaries (Sampson, Morenoff, & Raudenbush, 2005; Molnar, Buka, Brennan, Holton, & Earls, 2003). The Longitudinal Cohort Study (LCS) component of the project involved a stratified probability sample of 80 of the 343 neighborhoods; three strata with less than five NCs were sampled with certainty (Molnar et al. 2003). These 80 NCs were sampled from the 21 strata with the purpose of eliminating the confounding between racial/ethnic mix and socioeconomic status by representing the 21 cells as comparably as possible“. 4 Three strata were empty of NCs. No neighborhoods were 75% or more Hispanic and high SES, none were more than 75% non-Hispanic White and low SES, and none were mixed Hispanic and African American and high SES (Molnar et al. 2003). 35 The researchers then selected block groups at random within each of the 80 chosen NCs. Using a complete listing of approximately 35,000 dwelling units within these 80 NCsS, the researchers used age and sex of household members to identify through in-person screening the pregnant women, children, and young adults in seven age cohorts — 0 (birth), 3, 6, 9, 12, 15, and 18 years — for inclusion in the LCS. Children were selected if their birthday that qualified them for the sample would occur within six months. Face-to- face interviewing was the primary method of data collection. For most cohorts (except 0 and 18), interviews were conducted with the child (aka, the subject participant or SP) and the child’s primary caregiver (PC), i.e. the person that spent the most time caring for the child. The child and the primary caregiver interviews were administered by separate research assistants. The screening identified a total of 8,347 participants (Sampson et al., 2005; Molnar et al., 2003) of which 6,226 were interviewed in Wave 1. This project used data from Cohort 15 from the PHDCN. Cohort 15 is unique in that the primary caregiver reported both IPV between them and their partner as well as violence in the household directed to the adolescent in Waves 1 and 2; the subject, now a young adult (YA), reported about their IPV in Wave 3. The subject also reported exposure to violence in the home, school, and community as well as self reports of violent offending in the community in all three waves. At Wave 3, 432 YAs responded to questions on intimate partner violence, 489 about their acts of violent offending in the community, and 486 reported on exposure to violence in various settings. 5 All dwelling units were listed in the majority of the NCs. Census blocks were sampled for enumeration using probability proportional to size for some very large NCs (Molnar et al. 2003). 36 hdeasures Table 3.1 identifies the violence measures for the analysis, which are depicted by italics, as well as their data sources from each wave of data collection. Measures from Wave 1 and Wave 2 were independent variables used as predictors for the outcome variables in Wave 3. The focus of the analysis is to investigate whether past exposure to violence, witnessed or experienced, is associated with future offending. Table 3.1 Independent and Dependent Violence Measures by Source and Wave Wave 1 I - Wave 2 Wave 3 PC reports: Intimate partner violence Young Adult IPV PC-Partner CTS I PC-Partner CTS PC reports: child-directed aggression YA/Partner CTS PC-Child CTS l PC-Child CTS Adolescent reports: exposure to violence Community offending ETV | ETV Adolescent reports: Community offending SRO SRO [ SRO All exposure to violence measures used for this project were created from self- report item responses from four instruments: Conflict Tactics Scale for Partner and Spouse (CTSP)6, Conflict Tactics Scale for Parent and Child (CTSS)7, My Exposure to Violence Scale (ETV), and Self-Report of Offending (SRO). Although self-reports on their own cannot provide a full picture of an individual’s experiences, they allow analyses from the perspective of the individual (Raudenbush et al., 2003) and can provide information on personal experiences that others may not have witnessed, particularly in the case of older adolescents who often spend time away from home, beyond the supervision of a caregiver (Youngstrom et al., 2003; Kuo et al., 2000; Buka et al., 2000). Self-reports of violence are also independent of biases in official records from the 6 The instrument in Wave 2 was the Physical Abuse Scale. 7 The instrument in Wave 2 was the Parent-Subject Conflict Tactics Scale 37 criminal justice system (Sampson et al., 2005). Table 3.2 identifies the individual response items that were combined to create the exposure to violence measures for the analysis. Primary Caregiver Reports of Family Violence Conflict Tactics Scale for Partner and Spouse The Conflict Tactics Scale (CTSP) was used to measure the extent of psychological and physical attacks engaged in by the primary caregiver (PC) of the adolescent on their partner or on them by their partner in a dating, cohabiting, or marital relationship as well as how they dealt with conflicts by using reasoning or negotiation (Straus et al., 1996). The CTSP is based on conflict theory, which is also the basis for the “symmetry in measurement” design of the CTS, recording both the behavior of the PC and their partner. The PC answered pairs of questions—one inquiring about the PC’s acts toward their partner and the other inquiring about their partner’s acts toward the PC. Therefore, the PC reported on the acts for both individuals in the relationship. The reference period for the behaviors in each wave was one year. This symmetry of measurement is an important characteristic as “research has shown that the cessation of violence by one partner is highly dependent on whether the other partner also stops hitting” (Straus et al., 1996, p. 286). From the CTSP, the Wave 1 (Earls et al., 2007a) and Wave 3 (Earls et al., 2007b) categorical response items were recoded to their midpoint values: 0=never, 1=one, two=2, 3-5=4, 6-10=8, 11-20=15, 20 or more=25 (Straus et al., 1996). For the scale used for the analysis, only items that pertained to criminal behavior were used — committing severe psychological aggression, minor physical aggression, and severe physical aggression against the partner for the PC offending scale (Wave 1: 38 so.» 88 2t 8 58: .«o 5:88 2: 25852 53ers.» 2: 88m 5255 225 55888 85» 88 Em v5 Cam _ 05>? .. 553 888 55> 85826 088 .«o 98803 ”M 263 8 5x8 n ”N o>§> 8 5x5 N 2 0225 8 me< 25:2 28088 Bum 0 288805825 2.588 8% o 2:: 2.588 26m 0 .22. 9 585 2:588 3mm 0 28 88 28088 3mm o 88 2.588 3am o 0582388 8% m 85: o aficommozotcx :23 5588 2.588 3am o _noco:=q%oxo_x >558 2.588 3am o 6855;? 252288 3.8me o .8... bmsotomhooaoumoufi o a... .2... . 88am 9 m...u._:om5>> a £3, 5x232. o 2:: o 288 9 @820 o _5:o§q\5xo§ 826% o "59522:”.— m.~._ fin.— Ewc wgw 8 50m . Bacon 8 302.5 8508:. o 5525 5:5 535‘ a 5:5 o>= 8: 28088 2: o 5853 52:: c286 o fizz—588:0 53> 2.: 2.588 :2 o 583.— ........:.:.. 2&8 .5 85m o a: .....:.:.:.. «mom 0 98588 88 :2... 8 o. b. 8 .8 . mm a i.......... E.— .5 .35 .v—BM 0 "296m 8293.— 8_mm some a 12...}. x58 5 92m o ***** 558 5 connaam 6235 o It; 8 8:588 308:. o “.382 3292: was—#588 5x02 8 :5 458888 .8585. o ***** am mama—380m 385 2: 9 58252:. c 5.5.6m Egg—2.9%.. .568 385 e 3538.: . ~58 528 £58»: a Em? a «ma—Ewe 5882m o 8582 55:0 . Sn: 82:82 85m . 2:5:th .89» 5 gm 8 28x a 55%0582E. . nay—£588 :83 5888 89A :2 9 58 .8 E.— o 8.8m a 53> 5888 80» z: 8 .535 doc—2v— 0 5.55m .58?“— 25859 So» coo—Em o ~85 8.5888 Box 5835 3158.3; o 3.58th 8?» 532m o 245888 80> 558 5 598% .5825 o 2.1283 89» 8 525 5 850.8... a ”.382 .58.»...— n.2588 8o» 2 88:23 mam—#588 5.8.809 o @8588 5x03 5 43 82888 .8508... o 85:8E 8 83588 385 .5 «E 8 58258:. o 3.3% 828.25%.— .155...?» 3 9.8:an ...._w=:.=éo a. :88..=8 S:e_8o..ww< 5525-250 35...; 5583-0; 25> 5.— EV 55.9; .8 5885—2 1252—3— ..8 2:2— .Nd «35. 39 Cronbach’s alpha=.865) and the YA offending scale (Wave 3: Cronbach’s alpha=.805) and the same aggression against the PC for the PC victimization scale (Wave 1: Cronbach’s alpha=.855). Using the severe psychological items along with the minor physical items is a departure from the typical scale-building from the CTS item but factor analysis on the logged values showed these severe psychological and minor physical aggression items produced one component. For Wave 2 (Earls et al., 2006a), the CTSP used for Wave 1 was replaced with the Physical Abuse Scale (PAS), which included all items from the physical assault subscale but with different response options. Instead of asking for the number of occurrences, the Wave 2 instrument recorded only yes/no responses to whether the behavior happened or not. Therefore, the Wave 2 CTSP scales for PC offending (Cronbach’s alpha=.857) and PC victimization (Cronbach’s alpha=.846) contained only values of O and l (and missing). The SPSS “maximum” function was used to capture the maximum value over the two items: threatened. . .with a knife or gun” and “used a knife or fired a gun” from the Wave 1 CTSP as the revised CT S retained only the second item (Straus et al., 1996). In addition, the frequency of several physically aggressive items was multiplied by a weight for the severity of the act (Straus 2004). The SPSS “maximum” function was again used across the items in producing the above offending and victimization scales. This method produces a more conservative frequency count than summing the number of times the PC or YA reported that the action occurred for each item, and was used as several behaviors could have all happened in one event. The SPSS syntax specified “max.” (with the period) as SPSS then produces the maximum of all non-missing values for an individual record and does not create missing for the entire scale if even only one item for the scale 40 is missing. The overall Wave 1 total PC offending scale items and the total PC victimization scale items used for the analysis each produced two factor components but those for the Wave 3 YA scale produced only one. Finally, for parsimonious reasons and to address multicollinearity, the SPSS maximum function was again used across the PC offending and victimization scales for Wave 1 and Wave 2 to produce one PC-partner IPV scale for the analysis. Conflict Tactics Scale for Parent and Child From the PC ’5 responses on the Conflict Tactics Scale for Parent and Child (CTSS) about their aggressive actions toward the adolescent, an aggression scale for each wave was created to be comparable to the CTSP (wave 1: Cronbach’s alpha=.730; wave 2: Cronbach’s alpha=.790). In other words, only items that pertained to criminal behavior — committing severe psychological aggression, minor physical aggression, and severe physical aggression — were used for the scales used for the analysisg. Wave 1 (Earls et al., 2007c) values were recoded to the same midpoints as used for the CTSP. Wave 2 (Earls et al., 2006b) included only yes/no responses. The SPSS “maximum” function was used to capture the maximum value across the items of the midpoint values in Wave 1 and the binary coding of affirmative response to Wave 2 items. The Wave 1 scale items used for the analysis produced three factor components; the “beat up SP” and “burn or scald SP” items produced separate components. Subject Reports of Exposure to Violence The adolescents reported their exposure to violence — experienced and witnessed — in the home, school, own neighborhood, and outside the neighborhood through the “My Exposure to Violence’ (ETV) instrument in Wave 1 (Earls et al., 2007d) and the My E TV 8 The Wave 1 and Wave 2 instruments included, “slap or spank [subject] with an open palm.” 41 instrument in Wave 2 (Earls et al., 20053) and Wave 3 (Earls et al., 2007e)9. These self- reports provide measures of exposure from the vantage point of the subject rather than being deduced from neighborhood structural and compositional indicators as often done in other studies (Buka et al., 2000)"). For all selected ETV measures in Wave 2 and Wave 3 the categorical response items were recoded to their midpoint values: 0=never, 1=one, 2- 3:2, 4—10=7, 10 or more=17.5, and were the same recode parameters used for the Wave 1 “number of times” responses, in order to create measures approximating those used for the CTSP, CTSS, and SRO. Separate scales were created for witnessed in the home, experienced in the home, witnessed in the community (in school, in own neighborhood, outside the neighborhood), and experienced in the community. Again, the maximum across the items were obtained to create each wave’s scale. The best reliability was achieved with the witness in the community scales (Wave I, witnessed in the community: Cronsbach’s alpha= .364; Wave 2: experienced in the home: Cronbach’s alpha=.643; Wave 2: experienced in the community Cronbach’s alpha=.581; Wave 2, witnessed in the home: Cronbach’s alpha=.273; Wave 2: witnessed in the community: Cronbach’s alpha=.804; Wave 3, experienced in the home: Cronbach’s alpha=.342); Wave 3, experienced in the community: Cronbach’s alpha=.473; Wave 3, witnessed in the home: Cronbach’s alpha=.321; witnessed in the community: Cronbach’s alpha=.75 8). Unfortunately, the few cases in the Wave 1 witness in the home and Wave I experienced in the community measures did not overlap in the individual items. 9 Research has demonstrated the psychometric soundness of the My E TV instrument, its construct validity, and its linking with age, gender, race/ethnicity, violent offending, and neighborhood crime in theoretically predicted ways (Selner-O’Hagan, Kindlon, & Buka, 1998). '0 Although the PCs also reported on their adolescent’s exposure to violence, this project will use only the adolescent’s self-report, which is a reasonable approach when the respondents are older adolescents (Kuo et al. 2000; Buka et al. 2000). 42 Subject Reports of Relationship and Community Offending Raudenbush et al. (2003) and Sampson et al. (2005) were used as guides for selecting items to create the violent offending measure from the subject’s Self-Report of Offending (SRO)l 1, which are listed in Table 3.2 above. A Rasch model test by Raudenbush et a1. (2003) of the nine SRO items used in their analysis revealed that removing the item, “Hit someone with whom you lived in the past year with the idea of hurting them” (p. 187) from their severity scale produced a better model fit. They determined this domestic item tapped a different latent trait then the other items. This separateness was confirmed for the present project using factor analysis. The PHDCN codebooks show that for cohort 15 subjects, the vast majority of the domestic victims were siblings or other individuals. For this project as well, this item was not included in the community violent offending scale. Of the remaining eight items, five which pertained to violent actions that could be learned through modeling were retained to construct the violent offending in the community scale (wave 1: Cronbach’s alpha=.777; wave 2: Cronbach’s alpha=.693; wave 3: Cronbach’s alpha=.718). The number of times reported for each item was recoded to the same midpoint values as used for the Wave 3 CTSP to create the IPV and community violence dependent measures on the same scale. Again, the SPSS maximum function was used to create the frequency count for the SRO scales. Factor analysis results showed the logged values of these five items all loaded as one component for Wave 1 (Earls et al., 2005b) and Wave 3 (Earls et al., 2006c) and two components for Wave 2 (Earls et al., 2005c) (“thrown objects at people” factored as a separate component). ” References to the reliability and validity of these SRO violent measures across racial groups is published in Sampson et al. (2005). 43 Demographics and Family Environment Control variables were prepared from several PHDCN studies for the analysis. Subject sex. Subject sex was analyzed as provided in the PHDCN Wave 1 Master data file (Earls et al., 2005d), coded as 0 for females and 1 for males. Subject ethnicity. During Wave 2 subjects self-identified their racial and ethnic backgrounds, which had not been obtained in Wave 1 (Sampson et al., 2005). This project used the variable providing the subject’s ethnicity from the Wave 2 Master data file (Earls et al., 2005c), with the values recoded to 0 "Hispanic," 1 "Black-not Hispanic," 2 "White—Not Hispanic," and 3 "Other-Not Hispanic." If the subject’s ethnicity was missing in Wave 2, the missing value was replaced with the derived value from Wave 1 Master file. Dummy variables were then created for each racial group for the analysis. Chemically-dependent household. From the Family Mental Health and Legal History data file (Earls et al., 20070 the number of family members with a drinking problem and those with a drug-use problem were summed separately. This information was obtained at Wave 1 only. From the PC reports, 13.4% of the subject’s mother or father had a drinking problem and 7.7% had a drug-use problem. For parsimonious reasons and to address multicollinearity, these two variables were combined into one “chemically dependent household” variable, coded as 0 for No and 1 for Yes. In all, 37.8% of the subjects had a family member with a chemical dependency at Wave 1. Living under the poverty level. In Wave 1 the salary variable was provided in the Wave 1 Master file (Earls et al., 2005d) as the maximum of total personal income and total household income obtained with the Employment and Income form. In Wave 2 (Earls et al., 2006d) and Wave 3 (Earls et al., 2006c) the salary variable was provided 44 through the Demographic form as the PC’s personal income and the income from anyone else in the household. For this project, the income categories were standardized as: less than $5,000, $5,000-$9,999, $10,000-$19,999, $20,000-29,999, $30,000-$39,999- $40,000-$49,999, more than $50,000). The national poverty level was $18,408 for a family of five in 1995 (Census, 2006). The average PHDCN household size at Wave 1 was 5.1. Therefore, for this project, a household was considered under the poverty level if the reported salary was less than $20,000. Not living with both biological parents. This measure was created from the family structure variable (FAMSTRUC=1), a created variable in the Wave 1 Master file (Earls et al., 2005d). Similar measures were created for Wave 2 (Earls et al., 2006f) from the Household Composition file and for Wave 3 (Earls et al., 2007g) by scoring 1 for subjects living with both their mother and their father. The variables were then reverse coded to create a more interpretable reference category for the regression models. Supervision scale. Using the Home Inventory obtained in Wave 1 as part of the Home Observation for the Measure of the Environment (Earls et al., 2005f), a supervision scale was created by summing the affirmative answers from Scale IV. Comparable items from the Home and Life Interview (Earls et al., 2005 g) were recoded and summed to create a similar supervision scale for Wave 2. Family size. Family size was available as a single variable in the Wave 1 Master file (Earls et al., 2005d) but was created for this project from the Household Composition data for Wave 2 (Earls et al., 2006i) and Wave 3 (Earls et al., 2007g). Aggressive behavior score. The aggressive behavior scores were provided in the Wave 1 Youth Self-Report (YSR) (Earls et al., 2007h) and the Wave 2 Young Adult Self- 45 Report (YASR) (Earls et al., 2006g). The subject indicated if an item was very true or often true of them, somewhat or sometimes true, or not true of them at the time of the survey or within the last six months. Items included, “I argue a lot,” “I am mean to others,” and “I destroy things belonging to others.”‘2 For parsimonious reasons and to address multicollinearity, the aggressive behavior scores for each wave were combined with the subject’s SRO scores (described above) using factor analysis to create one “aggression” score for each wave for the analysis. Subject age. Since the analyses uses data obtained on the same individuals over time, age is essentially a constant. For parsimonious reasons, age was not included in the negative binomial regression models. At Wave 1 (Earls et al., 2005d), the youths’ ages in Cohort 15 ranged from 13.7 to 16.9 years (mean of 15.2, s.d. of .32); at Wave 2 (Earls et al., 2005c) from 15.6 to 19.1 (mean of 17.2, s.d. of.63); at Wave 3 (Earls et al., 2006h) from 18.2 to 22.3 (mean of 19.8, s.d. of.58). Additional Transformations Attrition and Missing Data In any longitudinal study, attrition of subjects results in missing data. As shown in Table 3.3, the cohort 15 sample started with 696 subjects in Wave 1. Of these, 596 continued in Wave 2, and 491 in Wave 3. An analysis of attrition at Wave 2 using logistic regression shows that Black subjects that were less likely to be present at Wave 2, whether compared to Hispanic subjects (O.R.=2.54, p < .001), White subjects (O.R.=3. 1 7, p < .01), or other race subjects (O.R.=4.63, p < .01). The only other '2 For more information on the YSR and YASR (ASR), see the ASEBA Web site at http://www.aseba.org. 46 Table 3.3. Attrition and Cases Responding by Instrument Slgnlficant faCtor was the PC Wave 1 Wave 2 Wave 3 reports of child-directed Sample in Wave 696 596 491 aggression which indicated those Completed HOME 686 590 na who reported aggression at Wave Completed FMHLH 684 1 were more likely to be present at Completed HHC 681 593 487 wave 2 (O.R.=1 .50, p < .05) Completed CTSS 680 555 na Regarding attrition at Wave 3, C l t d SRO 680 555 489 . omp e c three of the ETV varrables were Completed YSR/YASR 679 487 na _ . , Significant. As subject reports of Completed ETV 667 555 486 . ' . . wrtnessrng vrolence 1n the home Completed CTSP 671 533 432 (O.R.=.521, p < .05) or experiencing violence in the community (O.R.=.426, p < .01) at Wave 2 increased, the subject was less likely to remain with the study at Wave 3. However, as reports of experiencing violence in the home at Wave 2 increased, subjects were more likely to remain (O.R.=2.83, p < .05). Logistic regression by wave shows that as their aggression score at Wave 1 increased, subjects were less likely to remain in the analysis sample (O.R.=.548, p < .01) as were those in households under the poverty level (O.R.=.690, p < .05), but as PC reports of child-directed aggression at Wave 1 increased, subject were more likely to be present in the analysis sample (O.R.=1.37, p < .01). Regarding the Wave 2 factors, as subject reports of witnessing violence in the home (O.R.=.577, p < .05) increased, the subject was less likely to remain with the analysis sample. However, as subject reports of experiencing violence in the home increased, subjects were more likely to remain in the sample (O.R.=2.48, p < .05). 47 The subjects from the cohort 15 sample that completed the various data components varied, as shown in Table 3.3. Although 489 young adults completed the Wave 3 SRO questionnaire, only 432 completed the Wave 3 CTSP, in part since the CTSP was administered only to those who had been in a relationship in the past year. In order to keep the same cases across all models only the 335 cases without missing data or missing data that could be estimated (see below) were retained in the analysis sample”. The final sample included 335 cases. For the analysis, missing data were considered missing at random (MAR), consistent with published analyses on the PHDCN data (Raudenbush et al., 2003) and the expectation that the mechanism for missingness in a large, well-constructed data set may be ignorable to some degree (Emery, 2006). Item non-response values for respondents who answered at least one question on the CTSP or the CTSS were recoded to zero. For the ETV variables, subject that responded yes or no to the “ever” variables were coded as zero or the appropriate value from the “past year” variables. For the SRO For wave 1, blanks in the past year frequency variables were recoded to zero for subjects who reported no lifetime prevalence. For Wave 2 and Wave 3, blanks in the past year frequency variables were recoded to zero for subjects who reported no prevalence of that activity in the past year. Missing data could not be estimated for 28 cases missing data in several of the Wave 1 variables in the models, 57 cases missing data in Wave 2 predictor variables, 9 cases missing data in both Wave 1 and Wave 2 variables, and 2 cases missing data in Wave 3 predictor variables. '3 Logistic regression showed that Hispanic subjects (O.R.=3.31, p < .05), those not living with both biological parents (O.R.=3.60, p < .01), and those in larger families (O.R.=7.15, p < .05) in Wave 1 were more likely to have a relationship at Wave 3. Hispanic subjects (O.R.=2.80, p < .05) and Black subjects (O.R.=3.29, p < .05) in the sample at Wave 2 were more likely to have a relationship at Wave 3. Subjects who witnessed more community violence in Wave 3 (O.R.=32.78, p < .001) were more likely to be in a relationship at Wave 3 and those with higher aggression scores (O.R.=.21, p < .05) were less likely to be in a relationship at Wave 3. 48 Item non-response varied across data components. OLS regression with subject sex, race (White as reference group), not living with both biological parents in Wave 2 and the Wave 1 supervision score variables was used to predict values for 5 of the 335 cases in the analysis sample that were missing Wave 2 supervision scores. Of the variables in the predictor model, only the Wave 1 supervision score showed a significant relationship to the Wave 2 supervision score (B=.480. S.E.=.3 89, p < .001). OLS regression was also used with the variables in the Wave 1 full model to predict values for 35 cases in the analysis sample missing a Wave 2 aggression score. Of the variables in the predictor model, three variables showed a significant relationship to the Wave 2 aggression score: the subject being of an other race (B=.185. S.E.=.076, p < .05), witnessing in the community at Wave 1 (B=.045. S.E.=.023, p < .05), and the Wave 1 aggression score (B=.461. S.E.=.044, p < .001). Logistic regression was used with the variables in the Wave 1 full model (see Table 4.8 or Table 4.12) to predict values for 6 cases in the analysis sample missing a Wave 2 PC-partner violence score from the CTSP and for which the PC had stated a martial status of married or living together. Of the variables in the predictor model, only the Wave 1 PC-partner violence score showed a significant relationship to the Wave 2 PC-partner violence score (O.R.=1.96, p < .001). Crosstabulations also show that 2/3 of the PCs who did not complete a Wave 2 CTSP reported no violence at Wave 1. Additionally, the completion rate for the Wave 2 CTSP was higher among those who reported violence than those that did not report violence at Wave 1. Therefore, analysis of the attrition of cases and missing data show PC reports of IPV or child-directed aggression did not deter subjects from remaining in the sample; subject reports were mixed. Yet, as the descriptive results of the final analysis sample 49 given below indicates, no indication is apparent that subjects were reluctant to report exposure to violence, including violence they had committed. Handling Skewness Rare events results in responses that cause variables to have skewed, rather than normal, distributions. This skewness characterized several variables in the PHDCN data which contained a large number of cases coded zero or in the low vales and few cases coded higher. Most statistical procedures assume the variables have normal distributions. All count variables with skewed distributions were transformed using the natural log to approximate a normal distribution for each of the independent variables. The nbreg procedure in Stata was selected to handle skewness in the dependent variables. Multicollinearity The difficulty with considering multiple forms of violence at one time is to examine the unique effects of each form controlling for and separate from the other forms, including the co-occurrence, cumulating, and interactive effects (Margolin & Gordis, 2000; Carr & VanDeusen, 2002). As mentioned above, variables were combined to create three independent variables used in the models to address multicollinearity: PC- partner IPV scale, chemically dependent household, and subject aggression score. Still tolerances produced using SPSS on the model variables at each time point show slight correlations for two variables in the Wave 2 models (see Table 4.7 or Table 4.11 below). The tolerances for two variables are under .70 ~— in Model 48: witnessing in the home (.677) and experiencing in the home (.672) in Model 4C: witnessing in the home (.675) and experiencing in the home (.661), and in the full model: witnessing in the home (.674) and experiencing in the home (.658). Since these variables did not consistently track the 50 same in being positive or negative to the dependent variables, they were run as separate independent variables in the regression models. A correlation matrix for all model variables (except the race variables) is provided in Appendix A. Combining Wave Measures In order to approximate the “lifetime” exposure measures, used by Hanson et al. (2006), the subject ETV responses for all three waves obtained “in the past year” were summed for each setting for the contemporary models (see Table 4.5 and Table 4.9) and summed for each setting from the Wave 1 and Wave 2 ETV responses for the combined analysis models (see Table 4.6 and Table 4.10). The combined measure for subject reports of experiencing violence in the home actually contains only their report at Wave 2 as no subject reported experiencing violence in the home in Wave 1. Due to the binary coding of the CTSP and CTSS Wave 2 scales, the combined measures for these scales were created by using the SPSS maximum function to retain as much variability as possible; 19 (5.7%) PCs reported PC-partner IPV and 17 (5.1%) for child-directed aggression in Wave 2 but not in Wave 1 and so a code of 1 vs. a 0 is recorded in the measure for these cases. The supervision scales for Wave 1 and Wave 2 were averaged to create the combined supervision scale for the combined analysis model. The aggression factor scores for Wave 1 and Wave 2 were averaged to create the mean aggression score used for the contemporary models and cross-sectional models (see Table 4.5 and Table 4.9) and the combined models (see Table 4.5 and Table 4.9). The logged Wave 1 aggression score ranged from -.3 to 1.8 (mean of .545, s.d. of .449), for Wave 2 from -.3 to 1.9 (mean of .562, s.d. of .441), and for the mean aggression score from -.3 to 1.9 (mean of .573, s.d. of .424). 51 Chapter Four Results This project supports past research that found high rates of adolescent exposure to violence, both within their homes and communities, both witnessed and experienced. For 98.2% of the subjects, either the subject or their caregiver reported an act of aggression witnessed or experienced in the past year across Waves 1 and 2 — 84.5% of the PCs and 92.5% of the subjects. Over these two waves, 13.7% of the subjects reported exposure to violence that occurred in the home and 92.5% in the community; across all three waves, 18.2% of the subjects reported ETV in the home and 95.5% ETV in the community. Descriptive Results of Demographic and Model Variables Univariate statistics of the demographic and other model variables are shown in Table 4.1. Although the original Wave 1 cohort 15 sample of 696 adolescents included half females and half males, the analysis sample has 178 females and 157 males, 53.1% and 46.9%, respectively. The data have information from 141 Hispanic adolescents (42.1%), 108 Black adolescents (32.2%), 45 White adolescents (13.4%), and 41 adolescents from other races (12.2%). Obtained at Wave 1 only, the PCs of 138 adolescents (41.2%) reported a household member had an alcohol or drug problem. At Wave 1, 109 adolescents (32.5%) lived under the poverty level, which this project defined as a household income less than $20,000 a year for a family of five. At Wave 2, 96 adolescents (28.7%) and at Wave 3, 69 adolescents (20.6%) lived under the poverty level. At Wave 1, 191 adolescents (57.0%) did not live with both of their biological parents, at Wave 2, 195 (58.2%) did not, and at Wave 3, 212 (63.3%) did not. At Wave 3, 11.6% no longer lived with a member of their family of origin (not shown). Of those who 52 Table 4.1. Comparison of Model Variables Across Waves (n=335)* Characteristic Wave 1 Wave 2 Wave 3 Number Percent Number Percent Number Percent Independent variables Sex of adolescent/young adult Female 178 53.1 --- --- Male 157 46.9 --- --- Race of adolescent/young adult Hispanic 141 42.1 --- --- African-American (not Hispanic) 108 32,2 --- --- White (not Hispanic) 45 13,4 --- --- Other race (not Hispanic) 41 12.2 --- --- Chemically-dependent household 138 41.2 NC NC Household income under poverty level 109 32.5 96 28.7 69 20.6 Not living with both biological parents 191 57.0 195 58.2 212 63.3 Mean SD. Mean SD. Mean SD. Supervision score (range: 0 to 23) 19.4 2.79 17.7 3.27 NC Family size 5.2 2.00 5.2 2.09 5.1 2.4] PC reports of PC-partner violence (binary in wave 2) 6.1 20.09 0.2 0.37 NC Child-directed aggr. (binary in wave 2) 5.0 7.49 0.4 0.49 NC Subject reports of violence Experienced in the home NR 0.4 1.89 0.2 1.47 Experienced in the community 0.0 0.45 0.9 2.31 0.7 2.20 Witnessed in the home 0.3 1.68 0.5 2.38 0.2 1.60 Witnessed in the community 8.2 7.24 7.7 6.69 6.3 6.67 Subject aggression score Self-report of offending 2.7 5.80 1.9 4.76 --- YSR/YASR aggression score 9.7 6.21 6.1 3.80 «- Age of adolescent/young adult 15.1 0.32 17.1 0.56 19.7 0.55 Dependent variables Mean SD. Young adult IPV --- --- 2.7 10.76 Young adult community violence --- --- 1.8 5.63 Number Percent No violence --- --- 195 58.2 IPV only --- --- 35 10.4 Community violence only --- --- 59 17.6 Both IPV and community violence --- --- 46 13.7 *The sample included the 335 cases without missing data on the variables used for the analysis. NC=Not collected; NR=Not reported 53 reported their relationship status on the CTSP at Wave 3, 58.5% of the young adults were in a dating relationship, 9.9% were married, and 7.2% were engaged. As expected, supervision, as reported by the PCs, relaxed from an average score of 19.4 (s.d. 2.79) at Wave 1 to 17.7 (s.d. 3.27) at Wave 2. On the other hand, family size did not change much over the waves, which was 5.2 people at both Wave 1 and Wave 2 (s.d. 2.0 and 2.1, respectively) and 5.1 at Wave 3 (s.d. 2.4); reflecting in part the transition of the young adults to living outside their family household but instead living with a partner and child or roommates. The average age of the adolescents at Wave 1 was 15.1 (s.d. .32) and 17.1 (s.d. .56) at Wave 2. The average age at Wave 3 of the young adults was 19.7 (s.d. .55). Regarding the ETV and aggression variables, the PCs reported an average of 6.1 (s.d. 20.1) aggressive acts between them and their partner at Wave 1. The binary coded variable at Wave 2 shows an average of .2 (s.d. .37) affirmative responses to any aggressive act at Wave 2. In Wave 1, the PCs reported an average of 5.0 (s.d. 7.5) aggressive acts by them against their adolescent. The binary coded variable at Wave 2 shows an average of .4 (s.d. .49) affirmative responses to any similar aggressive act by an adult household member at Wave 2. From the subject reports, no adolescent reported experiencing violence in the home at Wave 1, an average of .4 (s.d. 1.89) acts were reported at Wave 2, and an average of .2 (s.d. 1.47) at Wave 3. Regarding experiencing violence in the community, the adolescents reported an average of .04 (s.d. .45) acts against them in Wave 1; an average of .9 (s.d. 2.31) acts in Wave2, and an average of .7 (s.d. 2.20) acts in Wave 3. Regarding witnessing violence, the subjects reported in Wave I seeing an average of .3 (s.d. 1.68) violent acts in the home and an average of 8.2 (s.d. 7.24) in the community; an average of .5 (s.d. 2.38) acts in the home and 7.7 (s.d. 6.69) in 54 the community in Wave 2, and an average of .2 (s.d 1.60) acts in the home and 6.3 (s.d. 6.67) acts in the community in Wave 3. Finally, the average score of the subject responses to the aggression items from the Youth Self Report (YSR) in Wave 1 was 9.7 (s.d. 6.21) and to the Young Adult Self Report (YASR) in Wave 2 was 6.1 (s.d. 3.80)”. In Wave 1 the adolescents reported committing an average of 2.7 (s.d. 5.80) aggressive acts in the community, but only an average of 1.9 (s.d. 4.76) in Wave 2. As explained above, factor analysis was used to combine these latter two measures into one aggression score for the analysis. For the dependent variables, the young adults (YA) reported committing an average of 2.7 (s.d. 10.76) aggressive acts against their relationship partner in Wave 3, which was less than half of the PCs report from the CTS in Wave 1 (see Table 4.2). The young adults also reported committing an average of 1.8 (s.d. 5.63) aggressive acts in their communities, which was only slightly less than their reports in Wave 2 (see Table 4.3). Overall, 195 young adults (58.2%) reported committing no violent offending in Wave 3, 35 (10.5%) reported committing only relationship violence, 59 (17.6%) reported committing only community violence, and 46 (13.7%) reported committing both types of violence. Descriptive Results of Exposure to Violence Measures PC Reports of Intimate Partner Violence Table 4.2 displays PC reports of family violence compared to the sex and race of the adolescent across the three waves. From the PC reports of PC-partner violence ‘4 For the total cohort 15 sample, the mean Wave 1 aggression score was 10.05 (s.d. 6.31) and the mean Wave 2 score was 6.28 (s.d. 3.93). 55 through the CTSP, 35.0% of the PCs of male subjects and 47.2% of PCs of female subjects reported aggression between them and their partner that qualified as criminal behavior at Wave 1. The difference between the PC-partner IPV for daughters and sons is significant (p < .05). Reports of PC-partner conflict dropped, however, in Wave 2. Less than half as many PCs of sons (17.2%) reported PC-partner IPV in Wave 2; for daughters only a third as many (14.6%). The drop in severe aggression is actually less then for overall aggression — for PCs of daughters, 52.8% compared to 69.1%; for PCs of sons, 31.4% compared to 50.9%. In contrast to Wave 1, the PCs of daughters actually reported less PC-partner IPV than PCs of sons in Wave 2, and the difference by sex is no longer significant. No significant differences in PC-partner IPV by race are found in either wave, similar to Hanson et al. (2006). At Wave 1, PC-partner IPV ranged from 37.8% for reports from PCs of White subjects to 43.5% for reports from PCs of Black subjects. In Wave 2, PC-partner IPV dropped by 55.4% as reported by PCs of Black subjects and by 64.8% as reported by the PCs of the other subjects; however, the reports across all races still ranged from 13.3% to 19.4%. From Wave 1 to Wave 2, severe PC-partner IPV dropped over 39% as reported by PCs of Hispanic and Black subjects and over 47% for PCs of White subjects and subjects of other races. As implied by the percents in Table 4.2, severe aggression significantly co-occurred with minor aggression (p < .001) in Wave 1 — 92.9% of PCs who reported that severe violent aggression had occurred in the home in the past year also reported minor acts had occurred; however, 61.5% reported minor aggression without severe aggression. In Wave 2, 90.3% of PCs who reported severe violent aggression in the home in the past year also reported minor aggression; yet, 44.0% reported minor aggression without severe aggression (p < .001). 56 So. W 9.2.... .5. W a»... 66. W a... Butane—mom 838,—. 85:00 Mambo ”Sufism-o_mom mucofl. SEED n mmHU .35an mm Sim PE 333% umFOZ mdm * 0.2 * fem Wmm 0.0— 3...." N.wm 0.3 02 v.2 ~13 .01: 0.3 W; wsm Wmv _._v 9mm * NRV v.3 v6 QM— od— w.m .12.. v.: wdm 3.. MM— Tom _.VN Qw— ..:.:., wém 8333\— ucwuzwth 58 we m.N_ 9w Wo— Om 0.3 mum— vdm 9m— m.w_ Qw— v.2 v.2 wd— 0:2 _.: 9m— 0; 9mm odv ham 6.?” * adv a: s: * N._m “new a... T9» 3.3.. CNN flow vdm v.2. 5.00 a... odw a w.m© can :0 a: E: wd _.__ Num— Nd v.2 :: 5; * ms— ... wdm mdm 9mm Now 3: s: a N1; Wow .1. fan 5.... MAN flow wém much Eco .1. 0.31 ..:.. 540 93. ado 88 550 8E3 385 2:32: 952 252 2.25m Hem. m 933 88 350 82>? 185 253$: 83% 232 oEEon— Hem N 35$ com... 550 253 V3.2m 253:.— most 232 225m Hem _ 35$ $383; uzwhawmmus :38. .8353 a. 205m nozmm 205m 3295 SEE 32min a. 830m coxmm :30... 32min an 805m soxmm 0.85m Roux—E SE2 Roux—E a. 230m :23; mmmn: dmr—bv coco—o; 8.93-0m 8&8 =o_wmeaw< eeaeeeoeio 803.5 .3 Nag—Ea— .S 33— 3.: 5m .3 35...; :25:— uo 33%.: U.— .N... «Bah. 57 PC Reports of Child-Directed Aggression Table 4.2 also displays PC reports of child-directed aggression against the subject compared to the sex and race of the adolescent across the three waves; in Wave 1 by the PC, in Wave 2 and Wave 3 by any adult family or household member. From the PC reports of child-directed aggression through the CTSS, 67.4% of the PCs of female subjects and 73.9% of PCs of male subjects reported aggression toward their adolescent that qualified as criminal behavior at Wave 1”. Reports of child-directed aggression to their subjects dropped slightly more than 45% for both PCs of daughters and sons in Wave 2, and more than 60% for severe aggression, despite the fact that the instrument used in Wave 2 asked about aggression toward the subject by any adult member, not just by the PC. Still, 35.4% of the PCs of daughters and 40.1% of the PCs of sons reported aggression by an adult in the household toward the subject in Wave 2; with 10.1% and 13.4%, respectively, reporting severe aggression. Significant differences in child-directed aggression by race are found in both waves. The 63.8% of aggression reported by PCs of Hispanic subjects in Wave 1 is less than expected (p < .05) as is the 22.0% reported in Wave 2 (p < .001). The 80.6% reported by PCs of Black subjects in Wave 1 is larger than expected (p < .01) as is the 49.1% reported in Wave 2 (p < .01) and the 51.2% reported by PCs of the other race subjects (p < .05). The drop in reports for severe and overall aggression toward Hispanic and White subjects is comparable, 69.3% to 65.5% and 37.6% to 30.0%, respectively. However, the drop in reports of severe aggression compared to overall aggression toward Black and other race subjects was greater, 60.5% to 39.1% and 69.1% to 27.6% respectively. As with the PC reports of PC-partner IPV in Wave 1, severe child-directed aggression significantly co-occurred with minor aggression 15 The Wave 1 and Wave 2 instruments included, “slap or spank [subject] with an open palm.” 58 :5. W Q..:.:.. Jo. W a: .mo. W a» mam—Echo mo ton—om :om n 0mm 58.3sz 2QO 8:20; 9 Eamonxm n >hm .3583 3 Sim 8m 333% ”8.02 _.:.. adv MS mdN w.0N YN vfiw v.mw 06 con.— .550 wH— N.N N.NN odN ”xv a... v.0 .1. v.0 N.N 0:55 a... YNm Ow 06m KuwN 0.v a... ofiw a... ofiw c 0.0 Vflew—m .33.. N41 9m 5NN CNN VA 02. 0.3. _.N Baum—m;— 0.03% _.mm 5m 06M odN m; fivw fivw Wm 2&2 “1.... m.0_ N0 o._N hd— m6. .1. Ndh 3.. NE. QM Bacon 3333; uzuucumufi Hour no.3?» a. N._m N.N— Wm? 6.0M N.N_ WNG 5N0 MS com.— .550 3.. wk; m.m_ b.0N VVN V? a 0.3. a 0.2. v6 853 42.2.. wNm O.N_ * v.3 0.?V v.5 3.. #60 .I. “no Md xom—m as; WNN v.0 N._M wdm Th 5mm Numw 5.0 35am;— muck 0.3a v.0 ONV wdv wxm Woo Yea NM 232 .13.. CNN * o.N_ 0.Nm KuwN 56— m.vw * Numw a... _.o_ 9:...th Bow. N 9235 m.0v m.NN od Cd od Wow wa 04» com.— .550 v.34 w.: Cd od 0.0 a... N.N0 a... 900 N.N 0:55 .13., _._0 m.—N a; 0; od .12.. Woo .12., odw 5m Mom—m .12.. QVM MK; V4 V4 Cd * 0.: * N65 _.N omchmmz mung 0. ~ m m._ _ 0d 0d 66 ”mm. NAN. w.m 232 * V6? *3." _.0m NH— h._ od wdh Twh N.N BNEQL Rom. _ 93>? €33.25. Ewwcumw‘s 2:0: 00 2:0: E 3238250 2:02 bEsEEoU oEo: 028:0 ocooEom 38H 5 E :38. 5 E mmmu: Dam 8:205 was—Steam coco—o5 wEmonS, «83.5 ._e 5625m— ..o 33— a...“ new .3 3:035 8 «SSA—um— ..e 8.89: «8.33m .Qv 935—. 59 (p < .001) — 96.3% of PCs who reported severe aggression toward in adolescent in the past year also reported minor acts had occurred; however, 55.6% reported minor aggression without severe aggression. In Wave 2, 97.4% of PCs who reported severe aggression toward the subject in the past year also reported minor aggression; still, 69.6% reported minor aggression without severe aggression (p < .001). Adolescent Reports of Exposure to Violence As shown in Table 4.3, compared to the PC’s reports of PC-partner IPV, the adolescents reported witnessing far less aggression; less than 5% of female or male subjects or subjects of any race reported witnessing violence in the home in Wave 1. (The low percents can result in even small differences appearing large, and so percent differences across waves will not be reported as such.) Except for male adolescents, whose reports decreased, reports of witnessing violence in the home increased for female subjects and across all races in Wave 2, perhaps in part due to the redesign of the ETV instrument which explicitly asked whether violent acts had happened to the subject and separately whether they had witnessed the violent act happening to someone else. More than 10% of female adolescents reported witnessing violence in the home at Wave 2, significantly more than the 3.2% reported by the male adolescents (p < .01). Compared to Wave 2, reports of witnessing violence in the home dropped for both female and male subjects in Wave 3. Still, 3.9% of female subjects and 2.5% of male subjects reported witnessing violence in the home in Wave 3, with females reporting more than in Wave 1 but males reporting less than in Wave 1. Subjects of all races reported witnessing more violence in the home at Wave 2 than Wave 1, with White adolescents reporting the least at 4.4% and Black adolescents reporting the most at 9.3%. The amount of witnessed 60 violence in the home from Wave 2 to Wave 3 decreases across all races; indeed none of the other race subjects reported any witnessed violence in the home at Wave 3. At Wave 3, the amount of witnessed violence is significantly higher than expected for the Black subjects only (p < .05) and only Black subjects reported witnessing more violence in the home at Wave 3 than at Wave 1. However, the drop in violence witnessed in the home from Wave 2 to Wave 3 may be mostly due to the young adult spending less time in the home at Wave 3. Witnessed violence in the community represents the major form of violence exposure, ranging over the seven year period from a low of 73.2% for male subjects at Wave 1 to a high of 90.4% for male subjects at Wave 2 and from a low of 60.0% for White subjects at Wave 1 to a high of 93.5% for Black subjects at Wave 2. Female adolescents actually reported witnessing more community violence at Wave 1 than their male counterparts (78.1% and 73.2%, respectively, us.) but with just a 7.2% increase for females compared to a 23.5% increase for males (83.7% and 90.4%, respectively), the male adolescents reported witnessing significantly more violence in the community at Wave 2 (p < .05). Both reported witnessing less violence at Wave 3, but despite the reports by females dropping nearly twice as much as the males (11.4% to 6.3%, respectively), nearly 75% of female young adults and 85% of male young adults reported witnessing violence in the community at Wave 3. The gender difference of more reports by male young adults continues to be significant at Wave 3 (p < .01). At Wave 1, the 60.0% of White subjects who reported witnessing violence is lower than expected (p < .01) as is the 70.2% of Hispanic subjects (p < .05). The 88.9% of Black subjects making similar reports is significantly higher than expected (p < .001). At Wave 2, nearly 20% 61 more Hispanic and other race subjects and 26% percent of White subjects reported witnessing violence in the community than at Wave 1. The percent of violence witnessed by Black and other race subjects was nearly the same at Wave 2 (93.5% to 92.7%, respectively). White subjects reported a double-digit drop of 14.8% in witnessed community violence at Wave 3 while all others reported single-digit drops (ranging from 6.9% to 9.3%). However, only Black subjects reported witnessing less violence at Wave 3 than at Wave 1. At both Wave 2 and Wave 3, the White subjects reported witnessing significantly less community violence than expected (p < .05 and p < .01, respectively) the Black subjects significantly reported higher than expected (p < .01 both waves). Interestingly, the subject reports of witnessed violence in the community was the most constant of all forms of exposure to violence, with no more than a 26% change across the waves, despite the documented drop in crime during the late 19905 (Liberrnan, 2007). In Wave 1, 51.5% of subjects who reported witnessing violence in the community also reported offending in the community (p < .001), at Wave 2, 40.2% of subjects made similar reports (p < .001). Regarding experiencing violence, no adolescent reported experiencing violence in the home in Wave 1, perhaps due to the structure of the Wave 1 ETV. A redesigned ETV form was administered in Wave 2 and Wave 3, which explicitly asked whether violent acts had happened to the subject and separately whether they had witnessed the violent act happening to someone else. Similar to witnessing violence in the home, adolescents reported experiencing less violence in the home than indicated by the PCs responses to the CTSP. In Wave 2, the percent of violence experienced in the home is comparable to the percent reporting experiencing violence in the community, including over 10% of 62 female subjects reporting witnessing and experiencing violence in the home at Wave 2, both representing a significant difference than their male counterparts (p < .01 and p < .05, respectively). (Again, since the low percents can result in even small differences appearing large, percent differences across waves will not be reported as such.) Fewer adolescents reported experiencing violence in the home at Wave 3 than Wave 2, except for White subjects, whose percent remained the same. Similar to experiencing violence in the home, few adolescents reported experiencing violence in the community in Wave 1. In Wave 2, 28.7% of the female adolescents and 40.8% of the male adolescents reported experiencing community violence, representing a significant gender difference (p > .05). The reports by each dropped by 31.4% and 26.7%, respectively, from Wave 2 to Wave 3. Still, nearly 20% of female subjects and nearly 30% of male subject reported experiencing violence in their communities in Wave 3. In Wave 1, only a small percentage of Hispanic (1.4%) and Black (1.9%) adolescents reported experiencing community violence. In Wave 2, reports ranged from 24.4% for White adolescents to 42.6% of Black adolescents, with the percent reported by Black subjects being significantly more than expected (p < .05). By Wave 3, no significant differences by race are found; yet, 28.7% of Black and 26.8% other race young adults reported experiencing violence in their community and 22.0% of Hispanic and 20.0% White young adults did as well. Overall, 38.8% of adolescents who reported witnessing violence in the community also reported experiencing violence in Wave 2 (p < .001) as did 98.8% in Wave 3 (p < .001) (results not shown). 63 Measures of Anti—Social Behavior In each wave, subjects also completed the Self-Report of Offending (SRO) instrument. In Wave 1, 40.4% of female adolescents and 51.6% of male subjects reported committing at least one of the five offenses used for this project’s SRO scale that represent violent offenses that could be learned through modeling. At Wave 2, 27.0% of the female subjects and 44.6% of the male subjects reported committing such offenses. The gender difference of 22% at Wave 1 is significant (p < .05), as is the nearly 40% gender difference at Wave 2 (p < .001), which widened still to over 50% by Wave 3 (p < .001). The offending by female subjects dropped 33% from Wave 1 to Wave 2 and an additional 40% from Wave 2 to Wave 3; offending by male subjects dropped 14% from Wave 1 to Wave 2 and an additional 26% from Wave 2 to Wave 3. Still, 16.3% of female young adults and 33.1% of male young adults reported offending in the community at Wave 3. In Wave 1, the 34.0% of Hispanic young adults that reported offending in the community is significantly less than expected (p < .001), while the 61.1% reported by the Black adolescents is more than expected (p < .001). A similar percent of the White adolescents and adolescents of other races reported offending, 44.4% and 46.3% respectively. Reports of offending dropped from Wave 1 to Wave 2: 33.2% less Hispanic subjects, 13.6% less Black subjects, and 59.9% less White subjects. However, 10.6% more subjects of other races reported offending at Wave 2. At Wave 2, the differences by race are all significant: the 22.7% of Hispanic subjects and the 17.8% of White subjects who reported offending are both less than expected (p < .001 and p < .01, respectively). The 52.8% of Black subjects and the 51.2% of other race subjects who reported offending are more than expected (p < .001 and p < .05, respectively). In Wave 3, over 38% of the 64 Hispanic and Black young adults reported less violent offending in the community and 14% fewer young adults of other races reported offending; the offending by White young adults remained the same as in Wave 2. Most differences by race are significant at Wave 3: the 14.2% of Hispanic young adults who reported offending is less than expected (p < .001), whereas the 32.4% of Black subjects and the 43.9% of the other race young adults who reported offending are more than expected (p < .01 for both). The 17.8% of White young adults who reported offending is not significantly different than expected. Of subjects reporting committing violence in the community at Wave 1, 75.4% reported committing violence in the community at Wave 2, and 41.2% at Wave 3. Of those reporting committing violence at Wave 2, 50.0% reported committing violence in the community at Wave 3. Overall, 50 subjects (14.9%) reported committing violence across all three waves. The above results are all more than expected (p < .001). Self-reports of offending in the home (e. g. responding affirmatively to “hitting someone you live with”) are not included in the following regression models but are provided for comparison purposes to completely describe the possible types of violence. Similar to the reports of young adult IPV offending in Wave 3 (discussed more below), significantly more female than male adolescents reported offending in the home in Wave 1 (26.1% to 11.5%, respectively, p < .001) and in Wave 2 (12.9% to 6.4%, respectively, p < .05), although the percent of offending by either gender dropped by about 50%. By Wave 3, the percent offending in the home by either sex or race had dropped to single- digits and no significant differences are found. In Wave 1 and Wave 2, around 80% of the victims of these offenders in the home were siblings; 52% were siblings and 36% were others (not a partner) in the home at Wave 3 (not shown). In Wave 1, 62.5% of those who 65 reported hitting someone in the home also reported committing violence in the community (p < .01). However, only 42.4% of those who reported hitting someone in the home in Wave 2 also reported committing violence in the community (us). In Wave 3, 50.0% of those who reported hitting someone in the home also reported committing violence in the community (p < .01). Descriptive Results of Dependent Variables Young Adult Intimate Partner Violence Offending As shown in Table 4.2, more than twice as many female YAs reported IPV offending that qualified as criminal behavior as did their male counterparts, 38.2% to 16.6% and over four times as many females reported committing severe aggression, 17.4% to 3.8%, both indicating highly significant gender differences (p < .001 for both). Significant differences in YA IPV offending by race are also found in Wave 3. The 15.6% of IPV offending reported by White YAs is less than expected (p < .05) and the 36.1% reported by Black YAs is larger than expected (p < .05). Of the Hispanic young adults and those of other races, reports of IPV offending are similar: 25.5% for the Hispanic subjects and 29.3% for the other race subjects. Serious offending by race is not significantly different and ranged from 4.4% for White young adults to 14.6% for young adults of other races. Of the young adults who reported IPV offending in Wave 3, 62.8% also reported committing violence in the community (p < .001) (not shown). Young Adult Violent Offending in the Community As shown in Table 4.3, about twice as many male YAs reported committing in Wave 3 at least one of the five offenses used for this project’s SRO scale that represent violent offenses that could be learned through modeling. The gender difference indicated 66 by the 16.3% of female young adults and 33. 1% of male young adults who reported offending is significant (p < .001). Significant differences in violent community offending by race are also found in Wave 3. The 14.2% of aggression reported by Hispanic YAs is less than expected (p < .001) and the 32.4% reported by Black YAs and 43.9% for YAs of other races are larger than expected (p < .01 for both). The 17.8% of White YAs who reported offending is not significantly different than expected. Comparing the two dependent measures by gender, the YA IPV offending and the violent community offending are almost perfectly opposite: 38.2% of the female YAs and 16.6% of the male YAs reported committing violence against their partner; however, 16.3% of the female YAs and 33.1% of the male YAs reported committing community violence. The gender difference for both types of offending is highly significant (p < .001 for both). Both Hispanic and Black YAs reported more violent offending in the home than in the community: 44.3% and 10.2%, respectively. Both White and other race YAs reported more violent offending in the community than in the home: 12.4% and 33.3%, respectively. Of the young adults who reported committing violence in the community in Wave 3, 43.2% also reported committing violence against their partner (p < .001). Results from Offending Models Multivariate, negative binomial regression using Stata was chosen as the primary analytic strategy to compare the relative effects of exposure to violence as predicted by social learning theory (SLT) to the factor of an anti-social behavior trait (ASBT) regarding violent offending. Both types of violent offending by setting (i.e., in the home or in the community) were modeled as outcome measures. Separate models were run to examine the effects of contemporary and cumulative effects of prior exposure to violence 67 on these dependent variables. Therefore, the results for each type of violent offending will be presented to examine contemporary effects first, with and without prior exposure, and then results progressing back in time will be examined to discuss similarities and differences in explanatory effects provided by previous wave predictors. Negative binomial regression was chosen for the analysis since the dependent variables represent counts of offending over an observation length. Ordinary least squares (OLS) regression assumes a normal distribution of the dependent and independent variables. Therefore, Poisson and negative binomial regression models are standard models used for analysis of count data where the distribution of the dependent variable is not normal. Of these, negative binomial regression incorporates observed and unobserved heterogeneity into the conditional mean; therefore, the conditional variance of the dependent variable becomes greater than the conditional mean, which is not changed. Overdispersion occurs when the conditional variance of the dependent variable is greater than the conditional mean. Negative binomial regression is recommended rather than a Poisson regression model if the null hypothesis of alpha=0 is rejected (indicating the data are over dispersed) using the likelihood ratio test following the Chi-square distribution with one degree of freedom (Park, 2005; UCLA, 2008). Incident rate ratios (IRR) were chosen for the interpretation of the models rather than the standard negative binomial regression coefficients. IRR ratios have the interpretation of the expected change in the rate ratio of the dependent variable given a one unit increase in the independent variable while all other variables in the model are constant (UCLA, 2008). For example, if the IRR for the aggression score is 10.45 and the dependent variable is the number of violent offenses committed in the community in the last year, for a one unit increase in the 68 aggression score, the expected rate ratio of violent offenses committed in the community increases 10.45 times, while all other variables are held constant. An IRR above 1.0 indicates a positive relationship and an IR below 1.0 indicates a negative relationship. In other words, an IRR=.40 indicates a decrease of .40 times, or 60% less. For these data, the observation time was the same — in the past year — for all subjects; therefore, the exposure option did not need to be specified. By default, Stata performs listwise deletion of cases (Park, 2005; UCLA, 2008). Therefore, to keep the case count the same over all models, only cases without missing data (or for which values could be predicted to replace missing data) were included in the analysis file. The procedures were run using the robust option to return robust standard errors in the output, which attempt to adjust for heterogeneity in the model. For all models, except the anti- social behavior models (see Table 4.4), the reference group is White females who lived with both biological parents in non-chemically dependent households that were also above the poverty level. Models for Anti-Social Behavior According to Hirschi and Gottfredson (1983) the age effect alone is so strong that no other factors are needed to explain or predict criminal behavior. Since these analyses used only the Cohort 15 subjects, age did not have enough variability to use for analysis; in fact, it is not included in the models for parsimonious reasons since it can be considered a constant. As an alternative to using age, the subject’s aggression score, created as a factor score combining their self-reports of aggressive behavior to the YSR/YASR and their self—reports of criminal offending (see description in Chapter 3), was used in the models in Table 4.4 as a measure of their anti-social behavior trait 69 (ASBT). Table 4.4 shows the results of using the aggression score alone in negative binomial regression models run on each dependent variable. Presenting the results progressing back in time (as will be presented across all models), the mean of the Wave 1 and Wave 2 aggression score shows a positive and significant (IRR=10.45, p < .001) relationship to the rate ratio of young adult IPV offending at Wave 3. In other words, for a one unit increase in the mean aggression score, the rate ratio of IPV offending is expected to increase by 10.45 times, while hold all other variables constant. The Wave 2 aggression score also shows a positive and significant relationship to IPV offending, but an effect that is 28% less than the effect of the mean aggression score (IRR=7.46, p < .001); the Wave 1 aggression score is lower by another 27% from the Wave 2 effect and 48% less than the effect of the mean aggression score on the rate ratio of IPV offending at Wave 3. Although the effect of the adolescent’s aggressive behavior at Wave 1 and Wave 2 are relatively large and significant factors in their rate ratio of their IPV offending as young adults, these results indicate that their cumulative aggression is over double the factor in predicting their IPV offending. Table 4.4. Subject Anti-Social Behavior to Violent Offending IPV Offending Community Offending IRR S.E. IRR S.E. Waves 1 and 2 Aggression 10.45 5.94 *** 9.28 4.28 *** Log psuedo-likelihood -452.59 -397.89 Wald chi-squared 17.07 *** 23.41 *** Wave 2 Aggression 7.46 3.06 *** 4.84 2.48 ** Log psuedo-likelihood -454.91 -400.91 Wald chi-squared 24.02 *** 9.50 ** Wave 1 Aggression 5.44 2.72 *** 4.94 1.67 *** Log psuedo-likelihood -458.07 -401.84 Wald chi-squared 1 1.45 *** 22.44 *** “p5 .01, ***pS .001 70 In contrast, the mean aggression score as a predictor of violent community offending is over 10% less of a factor than its effect on IPV offending (IRR=9.28, p < .001). However, the difference between the mean aggression score and the Wave 2 score is about a 48% reduction (IRR=4.84, p < .01), but the change from the effect of the Wave 2 to the Wave 1 aggression score (IRR=4.94, p < .001) is actually about a 2% increase. Therefore, similar to the IPV offending models, the difference between the mean aggression score and the Wave 1 score is around 47%. For IPV offending, subject aggression at Wave 2 shows a larger effect than subject aggression at Wave 1. For violent community offending, the effect of subject aggression at either time point is comparable. However, for both types of offending, the mean aggression score shows as the largest predictor of the three factors, indicating the subject’s “average” level of aggression is more a factor of their offending than their aggressiveness at the individual time points. Based on the log pseudo-likelihoods, the aggression models for all violent community offending models in Table 4.4 are a better fit to the data than any of the models for IPV offending; yet all models are significant and support the theory that an anti-social behavior trait is a relatively large and significant factor in violent offending. Models for Young Adult Intimate Partner Violent Offending The Contemporary and Cross-Sectional Models Results from the young adult IPV offending models are presented in Tables 4.5 to 4.8. Table 4.5 examines contemporary effects first, with prior exposure (Model 4A and Model 48) and without (Model 5A and 5B). Model 1 shows the relationship between the demographic variables and young adult violent offending against a partner in Wave 3. Model 2 displays the results after the family environment variables at Wave 3 were 71 entered into the model. Model 3 uses the mean of the aggression scores from Waves 1 and 2 to examine the effect of the subject’s aggression independent from their reported ETV. Model 4A includes instead the summation of the three waves of subject reports of each type of ETV and most closely replicates the ecological model of individual and family environment factors used by Hanson et al. (2006); Model 4B adds the mean of the aggression scores (as described above) to examine the effects of ETV across the waves along with the subject’s report of their aggressive behavior. Model 5A analyzes the data cross-sectionally by including the ETV in the two settings from Wave 3 only; Model 5B adds the subject’s mean aggression score in order to examine ETV and aggression together. In Table 4.5, Model 1 examining the demographic variables to the young adult IPV offending shows that males are expected to have an rate ratio that is .17 times less (IRR=.17, p < .001) than females for IPV offending, while the race variables are held constant. However, in comparison to White young adults, Hispanic young adults are expected to have a rate ratio for IPV offending that is 3.34 times higher (IRR=3.34, p < .05), Black young adults 7.46 times higher (IRR=7.46, p < .001) and young adults of other races 6.41 times higher (IRR=6.41, p < .01), each while all other variables are held constant. Adding the family environment variables in Model 2 increases the expected effect of all demographic variables in magnitude as well as increasing the significance of the results for Hispanic and other race YAs relative to White YAs. Being in a chemically- dependent household and not living with both biological parents are positively related to offending and being in a household below the poverty level is negatively related to the rate ratio of offending, but none of the family environment factors are significant. 72 .00. W 9.2.2.. ..o. W 9.2.. .00. W Q... 0:80:00 0.0 00.0033 .050 0.03.. 00:00 0... :. 0000.00. 0:: . 0 .00 080.00 05 .00 000. 0.0. 0... :. 00:0..0 00.00000 39.0 803. 0.3. .:00.0:. ..0>0. 3.000 0... 0000 00.2.0000; .:00:0000 3.8.9.0095: :. 8:080 30.00.05 500 5.3 00>: 0.05 00.0800 0.0.? m. 000.0 005.0000 “MW—.02 .12.. dem— ......... 0.00— .....:.. o0.N@ ..:.:.. 00.00 3.2.. 00.00 ..:.:.. N0.N0 .22.. 0.0.00 00.0300150203 0. .000- 00.000. 00.000- 00.000. 0 0 .0 00. 00.000. 00.000. 0000:0007000000 00.. :0 2.0 a. t... S... 3.0 a. 1.... mm... 0...... .2222. 02.8.00... 2... 00.0 0.0 1.... 00.0 00.0 00.0 00.— ..... 00.0 .00 .5800 500305.05 00.0 00.0 .00 00.0 00.0 00.0 0.0 00.0 00.0.. 500085.? 00.0 00.. 00.0 004 00.0 00.. 00.0 00.. 0.0.00 5000:000me 00.0 00.. 00.0 00; 00.0 004 00.0 00.. 080: 5000500000 0.20%.. 30.320. 0: a: a: :: 00:0.03 00.00.000.00 0: a: a: 0: 0000.03 05.00-00 02090.. 0% 0.0.5.00... 00:000.; m 3.0 a: a: a: a: a: a: 0N.m 3.50... a: a: a: a: a: a: 5.03.0000 00.0 00.. 00.0 00.0 00.0 00; 00.0 0—4 .00 00.. 00.0 0... 3:030 0.005 0>...0Z 0.0 00.0 00.0 00.0 .00 00.0 00.0 00.0 00.0 00.0 00.0 00.0 .9000 30.00 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 0... 00.0 00.. 020050006000 00.3.00... 2.0.0.3002 N 3.0 .. .00 00.0 .. 00.0 00.0 .. 00.0 00.0 ... 00.0 00.0 ... 00.0 00.0 1.... 00.0 ~00 .1. 00.0 0.0 000. .050 1. 0.. 00.0 I. .0.. 00.0 ... 00.0 00.0 .2. 00.. 00.0 1. 00.0 00.0 1.... 00.0 00... 1.... 00.0 00... V.005 .. 00.. 0.0 0... 00.0 ... 00.. 00.0 ... 00.. 00.0 .. 00.. 00.0 ..:.. 2... 00.0 .. 00.. 00.0 0.00%.: 0.03% t... 00.0 5.0 1.... 00.0 0.0 1.... 00.0 5.0 .13.. 00.0 0.0 I... 00.0 00.0 1.... 00.0 0.0 .1... 00.0 5.0 0.05. .80. 2.0.0 .00 mg: .m..0 my: :00 mm. .m..0 my: 0.0 my: .00 my: .00 my: 00 000000 0.0 00000.. 00 000000 00 000000 0 30000 N 0000:. N 30000 .0005. _::0_.000-mm0.0 .0003. 00.00.5050 0.0002 .a:0.0000-80.0 0:0 0.0.2.5800 - “50:00.0 0:20...» .05....— 0.a:._.:— .00 00.3. :0 00:0.0.> 0. 0.300....— 00 300:”... .00 0.0:... 73 The major effect of adding the subject’s aggression score in Model 3 is to reduce the magnitude and significance of the race variables. Based on Model 3, a one unit increase in the aggression score is expected to increase the rate ratio of IPV offending at Wave 3 by a factor of 10 (IRR=9.98, p < .001), while the other variables are held constant. This result is similar to Table 4.4 which did not include the demographic variables. The addition of the ETV predictors from all three waves in Model 4A also reduces the effect of the demographic variables, again most notably the race variables, both in expected magnitude as well as significance. This addition also shows the effect of being in a chemically-dependent household change from positive to negative, yet still not significant. Although both experiencing violence in the home and violence in the community as reported by the subjects are positively related to the rate ratio of IPV offending, neither are significant, nor is the negative relationship of witnessing violence in the home. However, the cumulative effect of the subject’s witnessing violence in the community across all three waves is positively and significantly related to the rate ratio of relationship offending (IRR=2.5 l, p < .01) and lends support to social learning theory. In other words, for a one unit increase in the number of times violence in the community was witnessed, the expected rate ratio of the young adult’s IPV offending is expected to increase by a factor of 2.5 l , while all other variables are held constant. However, the most notable effect of including the subject’s mean aggression score in Model 4B was to reduce the effect of witnessing violence in the community in magnitude and to non- significance, indicating ASBT rather than SLT explains YA IPV offending at Wave 3 for the contemporary time period. Examining the data cross-sectionally at Wave 3 in Model 5A looks very similar to Model 4A, with the exceptions that the difference between 74 Hispanic and White young adults is no longer significant and not living with both biological parents at Wave 3 is no longer positive to offending, but negative — the latter perhaps not unexpected at the age of the young adults at Wave 3. However, in contrast to Model 48, the addition of the aggressive score into the cross-sectional model in Model SB reduces in magnitude the effect of witnessing violence in the community but the significance remains (IRR=2. 14, p < .001), leaving this variable and the subject’s mean aggression score (IRR=5.23, p < .001) and sex (IRR=.17, p < .001) as the largest and most significant factors in predicting the expected rate ratio of IPV offending in Wave 3, each while controlling for the other variables. Therefore, in the cross-sectional model, both SLT and ASBT predict IPV offending at Wave 3. Comparing across all models in Table 4.5, the effect of the sex of the young adult on their IPV offending is in large part unchanged across the models. However the effect associated with the race of the subject on their IPV offending is reduced ‘by the introduction of the ETV and mean aggression score into the models. The factor of being in a chemically-dependent household changes from positive to negative after the ETV variables are added. Living in a household below the poverty level is always negatively related to YA IPV offending. None of the family environment variables are significant in any of the models, nor is the factor of experiencing violence in either setting, although both are always positively related to offending. Witnessing violence in the home is negative to offending across the models, but lacks significance as well. On the other hand, the existence of a relationship between aggression, exposure to violence, and IPV offending is evident. The presence of both in the same model in Model 48 and Model SB shows a reduction in the effect of the mean aggression score by 40% or more when 75 .8. w .1280 a... .8. w a... 80.800 0.0 00.00.23 .050 0.0.3 .8800 05 :. 0000.08 .8: . 0 .0.. 080.00 05 ..0 008 0.8 05 :. 0082.0 00.00000 30:0 02.0.”. 0.3. 8020:. ..0>0. 3.300 05 0>000 0.000002. 800:0000 3.00.82.080: :. 080.00 30.00.05 500 5.3 00>: 003 00.080. 0.0.3 m. 9.0.0 080.0000 .0002 .12.. mmdm. ..21. NN.m: .12.. 00.00 .12.. 00.00 1.... 00.00. .12.. 00.30 .12.. 060 00.05.0045". 0.03 00.000. 0 . .000. 00.0.0. 00.000. 00.000- 0. .000. 00.000. 0005.002000st 00.. .11. 00.0 00.0 .11. 00.0 00.0. .11. 00.0 00.0. 539.00 0330.00... 00.0 0.4 00.0 0... ... 00.0 004 .8800 800308.? 00.0 00.. 00.0 .0; 00.0 00.. 080: 800305.? 00.0 00.— 00.0 004 00.0 00.— .8800 8000:000qu ... 0.0 00. 0 ... 0.0 00. 0 .00 00.0 080.. 8008000000. 3.000. 60.300 0.0 3.. 00.0 00.. ..000 0080.00.20 00.0 0... 00.0 00.. 00:0.03 08.00-00 3.000. 0.. 00.5.00... 0020.00. 8.80 00.0 004 00.0 00.. 00.0 00; 00.0 00.0 00.0 00.. 00.0 00; 00.0 0:80... 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 8.03.090 .0.0 00.. 00.0 00.. 00.0 .0.. 00.0 00.— 00.0 00.— 00.0 0... 080.00 2&3 0>.. .02 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.0 .00 00.0 3.0.60 30.00. 00.0 00.0 00.0 00.0 00.0 00.0 00.0 00.— .0.0 00.. 00.0 004 0......0:0000.80..0 0.0.3.808 .20.03-=02 2.20 00.0 00.0 .. 00.0 00.0 .. 0.0 00.0 ... 00.0 00.0 ... 00.0 00.0 .1. 00.0 00.0 .1. 00.0 0.0 000. .050 ... .0.. 00.0 .. 00.0 00.0 .1. 00.0 00.0 .1. 00.0 00.0 .1. 0.0 00.0 .11. 00.0 00.0 .11. 00.0 00.0 0.00.0. 0... 00.0 00.. 00.0 .. 00.. —0.0 00.. ~00 0... 00.0 1.00.. 00.0 .. 00.. 00.0 080%.: 000% .11. 00.0 00.0 .11. 00.0 0.0 .11. 00.0 0.0 .11. 00.0 —0.0 .11. 00.0 .00 .12.. 00.0 0.0 .12.. 00.0 0.0 0.00,. 000 2.20 .00 My: .000 My: .00 my: .000 my: .00 my: 0.0 My: .m.m my: 000000 03.0 00 300:. 00 000000 0.0 00.00:. 0 .0002 0 000000 0 0000: 0.0.0.005 0 08>? 0:0 . 0.630 .0002 0080800 .0002 0050800 -§.0:0.=0 80.05 .05....— 0...E..:— ..0 m0...“ :0 00:20; 0. 0.3.09.0 ..0 0.00.0.0 .00 0.00.0 76 controlling for the subject’s ETV, whether using the ETV from across the three waves in Model 4B or just from Wave 3 alone in Model 5B. In Model 4B, the corresponding effect of witnessing violence in the community reduces by controlling for the subject’s aggression, both in magnitude and to non-significance. However, the fact that witnessing violence in the community and aggression are both positive and significant (p < .001) in the cross-section model suggests that effect of witnessing violence in the community on IPV offending can fade over time. Another possibility is the type of community violence witnessed in Wave 1 and Wave 2 was qualitatively different from what was witnessed in Wave 3 — 58.5% of young adults who reported about their IPV offending in Wave 3 reported being in a dating relationship. The log pseudo-likelihood of the Poisson model for Model 1 is -1722.09 and for the constant only model is log pseudo-likelihood was -468.42. In comparison, the log pseudo-likelihood for the comparable aggression only model as shown in Table 4.4 is -452.59 (p < .001). The values of the log pseudo- likelihood of Model 5B, the Wave 3 cross-sectional model incorporating both ETV and the aggression score is -429.18 and shows that Model 5B is the best fit to the data (closest to zero) of all these models, all of which are significant (p < .001), which provides support for both SLT and ASBT theory. The Combined Model Again, this project sought to examine the effects of contemporary and individual and cumulative effects of prior exposure to violence on IPV offending. Therefore, additional models were run that progressed back in time to examine the effects of prior violence exposure at different time points as predictors for later offending. The results from these models are presented in Tables 4.6 to 4.8. For the progression of models in 77 Tables 4.6 to 4.8 the types of violence exposure reported by informant — Model 4A for PC as informant and Models 4B and 4C for the subject as informant — were entered separately and then combined in the final model. Model 4B most closely resembles the Hanson et al. (2006) ecological model. Table 4.6 examines the combined effect of Wave 1 and Wave 2 ETV and subject aggression on Wave 3 IPV offending. Model 1 shows the relationship between the demographic variables. Model 2 displays the results after the family environment variables at Wave 3 were entered into the model. Model 3 uses the mean of the aggression scores from Waves 1 and 2 to examine the effect of the subject’s aggression independent from their reported ETV. Model 4A and Model 4C use the maximum value across Wave 1 and Wave 2 from the PC’s reports of PC-partner violence and child- directed aggression against the subject. The maximum was used to retain as much variability as possible to the values given that the Wave 2 reports were recorded as binary only; 19 (5.7%) PCs reported PC-partner IPV and 17 (5.1%) for child-directed aggression in Wave 2 but not in Wave 1 and so a code of 1 vs. a 0 is recorded in the measure for these cases. Model 4B and Model 4C include the summation of the Wave 1 and Wave 2 subject reports of each type of ETV in order to accumulate the reports across the two different time periods. Model 4C adds the mean aggression score to examine the effects of ETV across the two waves along with the subject’s report of their aggressive behavior from the same periods. The full model incorporates all predictors into one model. Comparing across the models in Table 4.5 to Table 4.6, the effect of the sex of the young adult on YA IPV offending is in large part unchanged. Although the family environment factors in Model 2 and Model 3 in both tables are similar, the effect 78 associated with the race of the subject on their IPV offending is reduced from Table 4.5 to Table 4.6 by the introduction the family environment variables. However, the effect of race increases in Model 4B in Table 4.6 compared to Model 4A in Table 4.5 by the introduction of the ETV into the models — summed across the subject’s reports from Waves 1 and 2 only for Model 4B and across all three waves for Model 4A, indicating that the ETV from just Waves 1 and 2 absorbs less of the variance than when ETV from all three waves are used. The factor of being in a chemically-dependent household again changed from positive to negative after the ETV variables are added. Living in a household below the poverty level is always negatively related to YA IPV offending as is the PC ’8 report of supervision while the size of the YA’s family at Wave 3 is positively related to their IPV offending. However, as also seen in Table 4.5, none of the family environment variables are significant in any of the models in Table 4.6. Nor is the factor of the PC’s report of either type of family aggression, although both are always positively related to offending and reduces when controlling for the subject’s aggression. Opposite to expectations based on SLT, experiencing violence in the home as reported by the subjects is negatively related to their IPV offending, even more so when controlling for their aggressiveness in Model 4C, at which point the effect becomes significant (IRR=.3 7, p < .05). In other words, based on the full model, for each one unit increase in the subject’s reports of experiencing violence in the home, their expected rate ratio of IPV offending decreased by a factor of .39 (p < .05) — over 60% lower. Experiencing violence in the community and witnessing violence in either setting are positive to IPV offending, but only significant for witnessing in the community (IRR=1.74, p < .05) before controlling for the subject’s aggression. Interestingly, though 79 not significant, witnessing violence in the home is positive in the models in Table 4.6 (vs. negative in Table 4.5), even increasing as a factor when controlling for subject aggression. On the other hand, controlling for subject aggression reduces the positive relationship between witnessing violence in the community and IPV offending. Subject aggression is highly significant in each model it is present, although reduced by 10% when controlling for the full complement of PC and subject reports of ETV in the full model (IRR=9.42, p < .001). The log pseudo-likelihood of the Poisson model for the full model was —l310.45 and for the constant only model was -468.42. In comparison, -the log pseudo-likelihood for the comparable aggression only model as shown in Table 4.4 is -452.59 (p < .001). Fitting the full model incorporating all reports of Wave 1 and Wave 2 ETV and corresponding aggression score, the log pseudo-likelihood is -432.89, indicating the fill model is the best fit to the data (closest to zero) of all the models in Table 4.6, all of which were significant (p < .001). This full model lends support to ASBT as a predictor of IPV offending. However, the negative relationship of the subject ETV in the home corresponds to predictions of less IPV offending -— opposite to predictions of SLT. The Wave 2 Model Table 4.7 examines the effect of Wave 2 family environment, ETV, and subject aggression on Wave 3 IPV offending. Model 1 shows the relationship between the demographic variables. Model 2 displays the results afier the family environment variables at Wave 2 are entered into the model. Model 3 uses the subject aggression score from Wave 2 to examine the effect of the subject’s aggression independent from their reported ETV at Wave 2. Model 4A and Model 4C use the PC’s reports of PC-partner violence and child-directed aggression against the subject at Wave 2. As described above, 80 .00. W 0.2.2.. ..o. W 0...... .00. W a... 80.800 0.0 00.00..0> .200 0.0.3 .0..:00 0... :. 0000.08 ..8. . 0 .0.. 08080 0... ..0 0..0. 0.0. 0... :. 00:0..0 00.000x0 30...0 00...... 0.0.. .:00.0:. ..0>0. 3.050 0... 0>000 00.2.0000; .:00:0000 0..00.80..0-:0: :. 080.00 0.0.00.0... .000 5.3 00>.. 0..3 00.0800 0.... 3 0. 000.0 00:80.00 H.002 ......... mONm. ..:.:.. 00.0N. 0...... N0.N0 .12.. 00.0 ..:.:.. 00..N. ......... 00.00 0...... 00.00 030300-30 E03 :08? 00.00... 8.03.- 00.00.. 8.00.1 2.5.... 3.0.... 82:85-03... 03 .2... 8... an: .2... 2.0 00... .22. 8.0 0.... .2222. 2.00200... .. 88 .0.. $8 .00.. .2. $8 02 .828 58.353 00.8 .0.. .0... 8.. 00.0 .0.. 2.5.. £088.53 2... 00... a... .0... 8.8 S... .568 5085.590 .2. z... 02. .2. 0.8 00... om... .0... 2:2. 5085.590 0.0000. B00030 .. E... S... cm... 8... .00.. 8626.220 8... ~08 8... 0.... 8563.80.90. 0.0000. UK 0.0-03.0.30... 92.30.: 0.0.0 on... E... 8... S... 40.8 .0... 0:. 8.. .08 8... a... 0.... 001:8... 8.8 8... 8.8 8... 8.0 0.8 88 8... 8.0 S... 8... 8... 80.5.0 :8 .00.. $8 a... 00.0 00.. 008 :2 0.8 8.... $8 00.. 08.... 2e... 2:82 8.8 8.... 88 00.. an... 8.. 02. 0... 2.8 00.. $8 3.. 0.25.. 3200 $8 8.. 3.8 0.... am... 8... 8.8 z.— o..... 0... mm... 0... 28.05.00.520 0.9.0.0.ng 2.30.3th2 05.0 a... 6.0 00.0 ~00 .2. 2.0 R... .2. .0.. E... R0 8... .2. 8.0 00.0 .2. 00.0 .00 88.2.0 .. .2 $0 .. 8.. 8.0 .2. 80 8... .2. 8.. ~00 2... 00.0 .2. 0.0 3.0 .2... 8.0 0.... .020 00.. 0.... 9.... :0 .2. 0.0 0.... 00.. $0 0:. .0.. a... a... .. 8.. 3:0 25%.: 0.03% .2... 88 2... .2... 88 0.... .2... 88 0.... .2... 88 E... .22. 8... 0.... .2... 88 S... .2... 8.8 t... 2...: km” .030 .00 5: .00 5: .00 5: .00 5: 0.0 5: .00 5: .00 5: 33: 0:... .0203: 02.052 03.05: 03...: 03...: ~ 3...: 0.52223 .0002 N 0.53 - 08.80.00 .:0.0.> .055... 0.08.8. ..0 00.0w. :0 00:0.0.> 0. 0.0000030 ..0 0.000.... 0.0 0.00... 8] these Wave 2 reports were recorded as binary only. Model 4B and Model 4C include subject reports of each type of ETV from Wave 2 only. Model 4C adds the aggression factor score from Wave 2 to examine the effects of ETV along with the subject’s report of their aggressive behavior from the same period. The full model incorporates all predictors into one model. Comparing across the models in Table 4.6 to Table 4.7, the effect of the sex of the young adult on YA IPV offending is again in large part unchanged, including remaining highly significant (e.g., Model 2, IRR=.18, p < .001). Comparing Model 2 in Table 4.6 to Model 2 in Table 4.7, the effect associated with race for Hispanic and Black young adults on their IPV offending relative to White young adults is reduced by the introduction the Wave 2 (vs. Wave 3) family environment variables. The effect of race is further reduced in Model 3 for all non-White young adults, even to the point of non-significance, with the introduction of the subject’s aggression score, which itself is significant (IRR=10.40, p < .001). However, the effect associated with race increases in Model 4A between the two tables by the introduction of the PC’s reports of aggression within the family in Wave 2 (vs. summed from Wave 1 and Wave 2 in Table 4.6). Except for the small decrease for Black subjects, the effect of race also increased in Model 4B between the two tables by the introduction of the subject’s reports of ETV in Wave 2 (vs. summed from Wave 1 and Wave 2 in Table 4.6). These results indicate that family aggression and ETV from just Wave 2 absorbs less of the variance than when using these factors that were summed for Waves 1 and 2 in Table 4.6 and across all three waves in Table 4.5, perhaps in part due to the less variability allowed by the binary coding of the CTSP and CTSS measures in Wave 2. Similar to Model 3, including subject aggression into Model 4C and the full 82 model lowers the effect associated with race on IPV offending, both in magnitude and significance, except for Hispanic subjects, which shows a small, non-significant increase. The only race difference that remained significant in the full model is between Black and White subjects (IRR=3.60, p < .05). The factor of being in a chemically-dependent household (at Wave 1) again changed from positive to negative after the ETV variables are added and is highest when subject aggression is included in the models. In Table 4.7, living in a household below the poverty level at Wave 2 is always positively related to YA IPV offending at Wave 3, which shows only negatively related to IPV offending when using the household poverty level at Wave 3 in Table 4.6. The factor of not living with both biological parents at Wave 2 is always positively related to IPV offending at Wave 3, though not significant and slightly reduces when subject aggression is included in the model. The PC’s report of supervision (at Wave 2) continues to be negatively related to YA IPV offending at Wave 3. The size of the YA’s family at Wave 2 is positively related to their IPV offending at Wave 3 only when the subject’s aggression or reports of their ETV are not included in the model. Again, like Table 4.6, none of the family environment variables are significant in any of the models in Table 4.7. Nor is the factor of the PC’s report of PC-partner violence significant, although positively related to YA IPV offending in Model 4A, but negatively related when controlling for the subject’s reports of ETV and aggression in the full model. Opposite to expectations based on SLT, the PC’s reports of child-directed aggression against the subject in Wave 2 is negatively related to their IPV offending in Wave 3, which is significant (IRR=.47, p < .05) in the full model. This negative relationship is mirrored by the subject’s report of experiencing violence in the home, which becomes significant 83 when controlling for their aggressiveness in both Model 4C and the full model (IRR=.33, p < .01). In other words, based on the full model, for each one unit increase in either report of the subject experiencing violence in the home in Wave 2, their expected rate ratio of IPV offending decreases by a factor of .47 and .33, respectively — the equivalent of about 53% to 66% less. Experiencing violence in the community at Wave 2 is negatively related to IPV offending at Wave 3 and witnessing violence in the home at Wave 2 is positive to IPV offending at Wave 3, but neither is significant. Lending some support to SLT, witnessing violence in the community at Wave 2 is also positively related to IPV offending at Wave 3 and is significant in Model 4B (IRR=2.35, p < .01), yet reduces around 30% in magnitude as well as in significance in Model 4C (IRR=1.60, us.) and the full model (IRR=1.67, p < .05) when also controlling for the subject’s aggression. Interestingly, witnessing violence in the home in Wave 2 increases as a factor of IPV offending at Wave 3 when controlling for subject aggression though not significant, but witnessing violence in the community at Wave 2 as a factor of IPV offending reduces, but maintains some significance (IRR=1.67, p < .05), when controlling for subject aggression. Subject aggression remained highly significant in each model it was present (e.g., full model, IRR=12.30, p < .001). Opposite to the models in Table 4.6, subject aggression in Table 4.7 increases 25% when controlling for PC reports of IPV and child—directed aggression in the full model, indicating that subject aggression at Wave 2 explains more of the variance in YA IPV offending at Wave 3 than the other factors -— including race, despite several other factors being significant in the Table 4.7 models compared to those in the previous tables. The sex of the subject continues to be a significant factor (full model, IRR=.12, p < .001) and apparently independent of any 84 other factor included in any model. The log pseudo-likelihood of the Poisson model for the full model is -1309.90 and for the constant only model is -468.42. In comparison, the log pseudo-likelihood for the comparable aggression only model as shown in Table 4.4 is -454.91 (p < .001). Fitting the full model incorporating all reports of Wave 2 ETV and corresponding aggression score is -425.31, indicating the full model is the best fit to the data (closest to zero) of all the models in Table 4.7, all of which were significant (p < .001). This model also provides the best fit to the data of all the models investigating the subject’s IPV offending at Wave 3. The full model support ASBT theory, but provides only mixed support for SLT as witnessing ETV predicts greater IPV offending but experiencing ETV predicts less. The Wave 1 Model Table 4.8 examines the effect of Wave 1 family environment, ETV, and subject aggression on Wave 3 IPV offending. Again, Model 1 shows the relationship between the demographic variables. Model 2 displays the results afier the family environment variables at Wave 1 are entered into the model. Model 3 uses the subject aggression score from Wave 1 to examine the effect of the subject’s aggression independent from their reported ETV at Wave 1. Model 4A and Model 4C use the PC’s reports of PC-partner violence and child-directed aggression against the subject at Wave 1. Model 4B and Model 4C include subject reports of each type of ETV from Wave 1 only. Model 4C adds the aggression factor score from Wave 1 to examine the effects of ETV along with the subject’s report of their aggressive behavior from the same period. The full model incorporates all predictors into one model. 85 .8. W 9.2.2.. .—O. W 9...... .mo. W h.... .5338 2.. $52.? .650 0......» .838 2.. c. 838:. ..S. . m .0.. 089.5 2.. ..o 0...... 29. o... :. owcmzo @2898 39? mossy. 23. .520... ..0>0. 3.5.5.. 05 “$2... $5.332. 88:03.. >..mo.Eo:u-:o: :. 8:23 .ao.wo.o.n 52. 5.3 .63. 2.3 3.2.8.. 8...? m. macaw 00:20.3. .mHOZ ..:.:.. 3‘6: .3... .oéo. .12.. mumNo ..:.:.. SW; 3...." wvfih 3.2.. ver ..:.:.. v0.3» wan—Effie 3&3 no.5“? .hfimvu SUNS». hwdvfi. m .631 mod; 3531 “goofifixzuovosg. ms 2.... R. 3a 1... 2.. man 1.... 3... s... 8.22.2. 2.32%... $6 an... 2.9 N”: mNd MN; .5500 :. 60305.3 ..:..... mod ”—6 3.2.. mod on... ..:..... mod h—é DEC: 5 “5305.3 ... no. amfi ..:.. mm. wafi 3.2.. we. wad .5800 E voocotomxm E: E: a). 080... a. 805.5me 2.8%.. .83..” 2... :.. ..N... 2.. am»... 8326-250 «do mm. mm... 3... 8:20... 85.8-9. whoa? UK 3.53... 8.33.: m 55 .06 ac; wed w—A NV. an." cad aw; ovd and 0N. hm..— QNE 338m cod 6%.: cod ”a... mod vmé cod ma... cod 6%.: mod mad =O.m.>.5msm .I. .wd QQN ...... wwd NWN .1. @wd —m.N a mnd 6N.N ... find ”—.N .. mwd MVN magma. 0.93 o>= .02 end 3..— de mad mmd maé .vd 9N4 mmd em... cmd 2.4 5.525 320m «And N—.— mmd ha; omd wad mmd w—é de :.— cmd a—A @.£.U:0QDU.EU;U WNKRQNS ufiNNcmxplfigz NQSm .w... wed 3.5 x... ow.v bvfi .w... on...“ whm a"... eqm 6:6 .w... oflv mad ...... h—é med a..." cwfi 3‘6 vow.— .550 .3... wad Nam .12., :6. mNé .3. SN xvi .....:.. OWN bmfi .13.. mod Nwfi ..i... mm.m $6 .8... had wvfi V.0..—m. ...... mv. b—.m .I. on. wvd v... and o... hmfi ..3 ac.— 8.m .. om. wcfi c mm... vmfi 0.52%.: 86% .13.. bod NNd .13.. 006 —N.c 13.2.. mod m—.e .13.. mod bud .1..." mod —N.c .13.. mod m—é .13.. cod hfié 0.3). Row. 2.8m .m,.m My: .m.m My: .m.m my: .m.m my: .m.m My: .m..m my: .m.m my: V‘c: -3n~ UV 33¢: QV ~33: VV ~33: m. ~35: N Efiez N VVQ: .252 . 23$ .232 _ o>a3fl£t5¢0 2.0.3.» .85....— 3352: ..e 8.3. .8 3.3.8.» 8 chameaxm .3 38.5. d... 0...“... 86 Comparing across the models in Table 4.7 to Table 4.8, the effect of the sex of the young adult on YA IPV offending is again in large part unchanged and highly significant, although consistently a little higher when including subject aggression (e. g., full model, IRR=.22, p < .001). Comparing Model 2 in Table 4.7 to Model 2 in Table 4.8, the effect associated with the race of the young adults on their IPV offending is higher with the introduction the Wave 1 (vs. Wave 2) family environment variables. The significance remains the same as seen in Model 1; for Hispanic young adults (IRR=3.08, p < .05), for Black young adults (IRR=6.84, p < .001), and for young adults of other races (IRR=6.36, p < .01). The effect of race remains higher in Model 3 in Table 4.8 for all non-White young adults, although gaining some significance for Hispanic subjects (IRR=3.68, p < .01) with the introduction of the subject’s aggression score, which itself is significant (IRR=4.01, p < .001). Opposite to Table 4.7, however, the effect associated with race decreases in Model 4A between the two tables by the introduction of the PC’s reports of aggression within the family in Wave 1 (vs. Wave 2 in Table 4.7). Except for the increase for subjects of other races, the effect of race also decreases in Model 48 between the two tables by the introduction of the subject’s reports of ETV in Wave 1 (vs. Wave 2 in Table 4.7). These results suggest that family aggression and subject ETV from just Wave 1 absorbs more of the variance than when using these factors from Wave 2 in Table 4.7. Similar to Model 3, including subject aggression into Model 4C and the full model increases the effect associated with race on IPV offending and shows the same significance as Model 3. The factor of being in a chemically-dependent household (at Wave 1) remains largely stable across the models. In Table 4.8, living in a household below the poverty level at Wave 1 fluctuates between being positively and negatively 87 related to YA IPV offending, but never significant. However, the factor of not living with both biological parents at Wave 1 is always positively and significantly related to IPV offending at Wave 3. Therefore, compared to adolescents living with both biological parents at Wave 1, the expected rate ratio of IPV offending for adolescents that did not increases ranging from 2.18 to 2.43 times in Models 2, 3, and 4A (p < .05) and ranging from 2.39 to 2.52 times in Models 4B, 4C, and the full model (p < .01) after the addition of the subject’s reports of ETV in Wave 1. The PC’s report of supervision (at Wave 1) continues to be negatively, but not significantly, related to YA IPV offending at Wave 3. The size of the YA’s family at Wave 1 reduces as a factor of IPV offending at Wave 3 when the subject’s aggression is included in the model, even to a negative relationship in Model 3, but again, never significant. The PC’s report of PC-partner violence in Wave 1 is positively related to IPV offending in Wave 3, but is not significant. Opposite to the full model in Table 4.7, child- directed aggression in Wave 1 in the full model in Table 4.8 is positively related to IPV offending in Wave 3, but also is not significant. None of the subjects reported experiencing violence in the home so this factor could not be included in the models in Table 4.8. In support of SLT, experiencing violence in the community at Wave 1 is positively and significantly related to IPV offending at Wave 3, although reduces in both magnitude and significance from Model 4B (IRR=3.98, p < .001) to Model 4C (IRR=2.98, p < .01), and to the full model (IRR=2.39, p < .05) when subject aggression was added. Although the number of subjects who reported experiencing violence in the community is small, the full model indicates for a one unit increase in violence experienced in the community their expected rate ratio of IPV offending increases by 88 2.39 times. However, opposite to SLT, witnessing violence in the home at Wave 1 is, however, negatively related to IPV offending at Wave 3, and maintains both the same effect and significance across the three models that it is present in Table 4.8, despite the addition of subject aggression in Model 4C (IRR=.16, p < .001) and the full model (IRR=.18, p < .001). In other words, based on the full model, for a one unit increase in violence witnessed in the home, the expected rate ratio of IPV offending decreases by .18 times — equaling an 82% reduction. Witnessing violence in the community at Wave 1 is positively related to IPV offending at Wave 3 in Model 4B, but negatively related in Model 4C and the full model after subject aggression is added; none are significant. As seen in all previous IPV offending models and in support of ASBT theory, subject aggression remains highly significant in each model it is present (e.g., full model, IRR=3.64, p < .001), however displaying the lowest effect seen in any table. Similar to Table 4.6 and Table 4.7 subject aggression in Table 4.8 decreases some when controlling for the full complement of PC and subject reports of ETV in the full model, indicating these other significant factors are contributing to explaining the variance in subject IPV offending at Wave 3, more so than seen in the previous tables. The log pseudo-likelihood of the Poisson model for the full model is -l481.23 and for the constant only model is -468.42. In comparison, the log pseudo-likelihood for the comparable aggression only model as shown in Table 4.4 is -458.07 (p < .001). Fitting the full model incorporating all reports of Wave 1 ETV and corresponding aggression score is -437.03, indicating the full model is the best fit to the data (closest to zero) of all the models in Table 4.8, all of which were significant (p < .001). Although still supporting ASBT theory, the Wave 1 full model shows aggression with the smallest effect across all time periods as a factor in 89 IPV offending at Wave 3. However, support for SLT was again mixed as seen in the Wave 2 model. Models of Young Adult Violent Offending in the Community The Contemporary and Cross-Sectional Models Results from the young adult violent community offending models are presented in Tables 4.9 to 4.12. Table 4.9 examines contemporary effects first, with prior exposure (Model 4A and Model 4B) and without (Model 5A and SB). Model 1 shows the relationship between the demographic variables and young adult violent offending in the community in Wave 3. Model 2 displays the results after the family environment variables at Wave 3 are entered into the model. Model 3 uses the mean of the aggression scores from Waves 1 and 2 to examine the effect of the subject’s aggression independent from their reported ETV. Model 4A includes the summation of the three waves of subject reports of each type of ETV and most closely replicates the ecological model of individual and family environment factors used by Hanson et al. (2006); Model 4B adds the mean of the aggression scores from Waves 1 and 2 to examine the effects of ETV across the waves along with the subject’s reports of their aggressive behavior. Model 5A analyzes the data cross-sectionally by including the ETV in the two settings from Wave 3 only; Model 5B adds the subject’s mean aggression score as described above in order to examine ETV and aggression together. In Table 4.9, Model 1 examining the demographic variables to violent offending in the community shows that males are expected to have a rate ratio of violent offending in the community that is 2.52 times higher (p < .05) than females, while the race variables are held constant. In comparison to White young adults, Hispanic young adults are 90 .8. W n...:.:.. ..o. W H.0... .mo. W a... 0:30:00 0... 0030...; .050 0:55 .0..:00 05 :. 00008:. ..:.. . 0 .0.. 050.00 05 ..0 0...... 0.0.. 05 :. 0w:0..0 00.0098 305 0053. 0.3. 3020:. ..0>0. 3.03.. 05 0>05. 0.0—0:080: E00880 3.00.82.0-8: :. 8:05.. 5.0.8.0... 50.. 5.3 00>: 055 8.080.. 8...? m. 9.0% 00:80.3. ”mp-OZ .3... ww._m_ .12.. onM— 0...... N560. 0...... DEM: ..:..... .de. ..:.:.. av.wm .0 med. wohwzwm-Eo v.03 00$.”- 0. ..nm- $.05- ooth- mcdwm- wmdom- $.80- 000....00...-000:m.. w0-. . 3.. m5. .. .0.. 80 ..r. 00. 8.2 .5322. 3080...... 5:. mm... MQN .12.. .50 —~.m .2... mud N04” 1.... .5. 3:0 .EE8 50000055? .2. 3.... .md 00... .., 2.0 3.: 2.0 no... 080... 5000005.? : 3... gm .1. m... 0.... ... mm... 0.... I 0m... :5 .EES E 08859.0 on... N... S... 0... S... 3.. 00.0 3.. 082. 5085.630 0.208.. 30.50% .2. a: a: 0: ..wwm 00.000.05.30 0.. 0.. 0.. a: 00:0.0.> 005.090.. 0,200.0. UK 0.00.000... 0000.0... «- new 0.. a: a: a: a: a: 00.0 3.50”. a: a: a: a: a: 0.. 8.0.3095 and am; omd cm; mmd cm.— 0m... an; .. on... mc.~ 0.20 S..— m.:0.0n. 0.53 0>...0Z 2.. 2.0 an... ..i 0.6 mm... ... 2.0 on... ... 2.0 an... ._ 2.0 on... ... x... cm... 5.05.. 30.0m. mm... mm; mm... 3..— nvd mv.— $6 3.. $0 an. mmd S... 2......08000525 00.50.00... 30.0.... -002 N :30 va Ed ...~ mm: ... mwd 3... 3d mm.»- ..:..... ow... who ... 9mm 3... in. 35 00.... .050 and 2...: 00.0 5... mod 2..— mmd cw... .. ow. ad 0N. bu." 00.. es... 0.005 .Nd on... ... 2.0 mm... vmd 3.... .Nd em... N05 00... end em... .2. 0m... 0.5.00... 000% :5 v... med .13.. 2.. Sun .1. cwd «Wm ... mud 0N.~ .1... mm. de ... 8.. 00d .. w... Nm.~ 0.02 00% n 005 .m..w my: .m..m my: .m..m mm. .m.m my: .m..m mm. .m.m my: .m..m my: M...- ~000~< Vu- ~000~< m0 .000: V0 .000: m. .000: N .000}. ~ 300: .0002 300505-3000 .0002 50.80.8300 0.000.). 38.8%-80.5 0.... baSQESEU - bra—5:00 0... :. $55820 2.0.05 ..0 00.5- :0 00:0.0.> 0. 0.30%.”. .0 0.00....m .0... 0.5...- 91 expected to have rate ratio less, but Black and young adults of other races higher, although none of the race variables are significant. Adding the family environment variables in Model 2 did not change the effect of sex or of the young adult being Hispanic relative to White. However, the expected effect associated with the young adult being Black or of other race increases in magnitude, although becoming significant only for other race (IRR=4.22, p < .05). Being in a chemically-dependent household is negatively related to community offending and not living with both biological parents is positively related to community offending, but neither is significant. However, being in a household below the poverty level is negatively related to the rate ratio of community offending and significant (IRR=.34, p < .05). Adding the subject’s mean aggression score in Model 3 increases the magnitude of the expected effect of being male relative compared to being female and of each race variable to the rate ratio of violent offending in the community at Wave 3 as well as increasing the significance of subject sex (IRR=3.52, p < .001), being Black (IRR=2.91, p < .05) and being of other race (IRR=6.78, p < 001). Regarding the family environment factors, being in a chemically-dependent household changes to being positively related to community offending, though not significant; living in a household below the poverty level is unchanged (IRR=.36, p < .05); and not living with both biological parents increases in its effect on the rate ratio of violent community offending at Wave 3 and becomes significant (IRR=2.05, p < .05). Consistent with ASBT theory, the effect of the mean aggression score shows a large effect to the rate ratio of young adult violent offending in the community at Wave 3. Based on Model 3, a one unit increase in the mean aggression score is expected to increase the rate ratio of violent offending in the 92 community by 15.2 times (p < .001), while the other variables are held constant. In contrast, the ETV predictors created from adding the reports from all three waves in Model 4A reduces the effect of the demographic variables both in expected magnitude as well as significance, including eliminating any significance to the race variables. Including the ETV predictors also reduces the effect of being in a chemically-dependent household and not living with both biological parents, the latter to non-significance, but did not change the effect of living below the poverty level, which remains negative and significant (IRR=.38, p < .05). Although both experiencing violence in the home and in the community as reported by the subjects are both positively related to the rate ratio of violent offending in the community at Wave 3, only experiencing violence in the community is significant (IRR=2.07, p < .01). The expected effect of the subject’s cumulative witnessing violence in the home over the three waves is negative, but not significant. However, the cumulative effect of the subject’s witnessing violence in the community across all three waves is positively and significantly related to the rate ratio of relationship offending (IRR=4.31, p < .001). In other words, the rate ratio of violent offending in the community is expected to increase by 4.31 times given a one unit increase in a one unit increase in violence witnessed in the community, while all other variables are held constant. So, both community ETV measures support SLT. Adding the subject’s mean aggression score in Model 4B affects the demographic and family environment variables similar to its effect on these factors in Model 3. However, the most notable effect of including the subject’s mean aggression score in Model 43 is to reduce the effect of experiencing violence in the community, both in magnitude and significance (IRR=1.86, p < .05), reduce the effect of 93 witnessing violence in the community in magnitude but not significance (IRR=3.02, p < .001), yet to add significance to the factor of witnessing in the home (IRR=.64, p < .05). Surprisingly, the expected effect of the aggression score when present in Model 4B with the subject’s cumulative reports of ETV over the three waves reduces to one-fifth the effect and a lower significance level (IRR=3.03, p < .05) of the mean aggression score in Model 3 when not controlling for the subject’s ETV. Therefore, the contemporary model supports both SLT and ASBT theory, but the presence of both in the same model reduces the effect of the other. Examining the data cross-sectionally at Wave 3 in Model 5A shows an increase in the effect of the subject’s being male relative compared to being female to the rate ratio of committing violence in the community both in magnitude and significance (IRR=3.37, p < .001). The expected effect from the difference between Hispanic and White young adults is negative and significant (IRR=.25, p < .05). The other two race comparisons reduce compared to Model 4B and neither is significant. Living in a chemically-dependent household in Wave 1 and not living with both biological parents at Wave 3 are both positive to offending, but also not significant. However, living in a household below the poverty level in Wave 3 continues to be negative as well as significant (IRR=.35, p < .01). Regarding the ETV variables, experiencing violence in the community at Wave 3 shows its largest effect across the models in Table 4.9, indicating a 4.10 times increase in the rate ratio of committing violence in the community for each one unit increase in community violence experienced, which continues to be significant (p < .01). The expected effect of subject aggression is 25% less in Model 5A compared to Model 4A but the same in significance (IRR=3.21, p < .001). In contrast to Model 4B, the addition of 94 the mean aggression score into the cross-sectional model in Model SB results in a higher effect and significance of the subject being male (IRR=3.48, p < .001), no significance for any race variable, and continuing significance for living under the poverty level in Wave 3 (IRR=.32, p < .01). The factor of experiencing violence in the community at Wave 3 is 83.9% higher and more significant (IRR=3.42, p < .01) in Model 5B than Model 4B, but 16.5% lower than in Model 5A, in part due to the a trade-off in Model 4A compared to Model 5A in the magnitude of witnessing violence in the community, which is significant in all four models (e.g., Model 5B, IRR=2.33, p < .001). Subject aggression is a 23.7% larger factor in Model 5B than Model 4B, yet the same in significance (IRR=3.75, p < .05). These models support ASBT theory, and also SLT through the positive relationship of the community ETV variables. Comparing across all models in Table 4.9, the effect of the sex of the young adult on community offending is largest and most significant (e. g., Model SB, IRR=3.48, p < .001) when subject aggression is also included in the model. The effect associated with the race of the subject on their violent offending in the community shows little significance and, except for Model 2, is also highest when the mean aggression score is included in the model. The factor of being in a chemically-dependent household changes from positive to negative after the ETV variables are added. While not living with both biological parents in Wave 3 is positively related to community offending, living in a household below the poverty level in Wave 3 is always negatively related to community offending and significant (e.g., Model 5B, IRR=.32, p < .01). Experiencing violence in the home is always positively related to offending but not significant; witnessing violence in the home is negative to offending across the 95 models, but only significant in Model 4B (IRR=.64, p < .05). On the other hand, experiencing violence in the community (e.g., Model SB, IRR=3.42, p < .01), witnessing violence in the community (e.g., Model SB, IRR=2.33, p < .001), and subject aggression (IRR=3.7S, p < .05) are all large and significant factors and as predictors for violent offending in the community at Wave 3. As with the IPV offending, the existence of a relationship between subject aggression, exposure to violence, and community offending is evident by comparing across the models. The presence of both in the same model shows a reduction in the effect and significance of the mean aggression score by at least 75% (from Model 3, IRR=15.20, p < .001 to Model 4B, IRR=3.03, p < .001 and Model SB, IRR=3.7S, p < .05) when controlling for the subject’s ETV, whether using the ETV from across the three waves in Model 48 or just from Wave 3 alone in Model SB. The corresponding effect of experiencing or witnessing violence in the community also reduces in magnitude by controlling for the subject’s aggression in Model 4B and Model 5B. The log pseudo- 1ikelihood of the Poisson model for Model 1 is -881.23 and for the constant only model is log pseudo-likelihood was -409.33. In comparison, the log pseudo-likelihood for the comparable aggression only model as shown in Table 4.4 is -397.89 (p < .001). The values of the log pseudo-likelihood of Model SB, the Wave 3 cross-sectional model incorporating both ETV and the aggression score, is -367.54 and shows that Model SB is the best fit to the data (closest to zero) of all these models in Table 4.9, all of which except Model 1 are significant a p < .001. Again, these models examining the young adult’s current situation support both SLT and ASBT theory. 96 The Combined Model As with IPV offending, additional models were run that progressed back in time to examine the effects of prior violence exposure at different time points as predictors for later violent offending in the community. The results from these models are presented in Tables 4.10 to 4.12. For the progression of models in Tables 4.10 to 4.12, the types of violence exposure reported by informant — Model 4A for PC as informant and Models 4B and 4C for the subject and informant — were entered separately and then combined in the final model. Model 4B most closely resembles the Hanson et al. (2006) ecological model. Table 4.10 examines the combined effect of Wave 1 and Wave 2 ETV and subject aggression on Wave 3 community offending. Model 1 shows the relationship between the demographic variables. Model 2 displays the results after the family environment variables at Wave 3 are entered into the model. Model 3 uses the mean of the aggression scores from Waves 1 and 2 to examine the effect of the subject’s aggression independent from their reported ETV. Model 4A and Model 4C use the maximum value across Wave 1 and Wave 2 from the PC’s reports of PC-partner violence and child-directed aggression against the subject. The maximum was used to retain as much variability as possible to the values as Wave 2 reports were recorded as binary only; 19 (5.7%) PCs reported PC- partner IPV and 17 (5.1%) for child-directed aggression in Wave 2 but not in Wave 1 and so a code of 1 vs. a 0 is recorded in the measure for these cases. Model 4B and Model 4C include the summation of the two waves of subject reports of each type of ETV in order to accumulate the reports across the two different time periods. Model 4C adds the mean aggression score to examine the effects of ETV across the two waves along with the 97 :5. W 9.2.... ..o. W Q0... .mo. W Q... 000.0000 0.0 00.00..0> .0...0 0....3 .0..:00 0... 0. 0000.00. ..00 . 0 .0.. 0.00.00 0... ..0 0..0. 0.0. 0... 0. 0w00...0 00.00....0 30:0 0000... 0.0”. .000.0:. ..0>0. ..0>0.. 0... 0>0..0 00.000000: .000:0..00 >..00.:.0..0-00: 0. 0.020.. .00.w0.0... .00.. 0...... 00>.. 0...... 00.0.00. 8...? 0. 008w 0000.0..0... ”mt-OZ a... 0...... I... 00...... t... :0: r... 00.8 s..- 00..N. I... ..0.N.. _. 0...... 8.8.0....0 2...... 0......- 230- 02.0.- 0.0..- S.E.- 00.0..- .0. 3.1 .08....8...-o.§0.. 03 1.. N0..- 03- t... S... 2.... .1... N... ..N.... 5.8.0.. 308.00.. .3 0.... ..QN .1... 0.... :N a... .0.. 2.... .888 88.88.... .N... .0... 0.... .8... .00... .0.. 082. 88.88.... 0N... 0.... 0N... 00... 0.... ..N.. .888 8.088530 0.... N0... N0... .0... N0... ..N.. 082. 8:088:80... . 0..0..0. .0035 .0... 0... r... 00... mnN .00.. 8.8.0-206 0.... a... 2... 8... 85.238030. 0..0..0. 0.. 0.0.3.0005 00230.: 000.0 a... 0... .0.. SN N0... N0.. 0.... .0.. . N..N 3.0 0... N..N 8.....8... 0.... N0... 8... 0.... 0.... .5... 0.... 0.... 3... a... 0.... 0...... 8.0.22.0 0.... 0N. .0... 00.. 0.... 0... .0... an. _. ..0... .N.N .00... 00.. 082.... 8...... 02.82 .N... ..0... 0N... N0... N... 00... .. 0.... .0... MN... 0.... ..N... 3.... .53.. 320.. 0.... 0N. 00... .0.. 0.... 0N.. .N... S... .0... .0.. 0N... 0.... 28.058.080.00 0.0-.3000... 2.303-202 N30 .3. 0.... 000 z. a... 3.0 1. N... .00 I... 00.0 ..0... .1... 8.0 00.0 r. 0.... 00.0 .0. 3.0 88.8.0 0.... .0.. N... 0...N N... a... 0.... .....N .. N... .00 0.... 00.N 0.... 0.... .0..... 0.... 0.... .0... 0.... N0... N5. 00... 0.... 0.... 2... .0... 00... .0... 00... M88...... 000% t... 0.... 3.0 r... 0... 00.0 .. a... 00.N .. 8... SN 1... .0.. 0.... .. 0... 0E .. 0... N0.N 2...). 0.0% .005 .00 0.... .00 0.... 0.0 0.... .00 00. .00 :0. .00 00. .00 00. 0.0.5.... 0200.0... 0.0.8... 3.0.0... 00...... N38... . 00..... 0.0.0.00... N 0.00...» 000 . 0.00.5. .0005. 00030.00 .0002 00500.00 - 3.000.080 0... 0. w0.0:0....O .00.0.> ..0 00.00. :0 0000.0. > 0. 0.0009... .0 0.00am. 0.... 030,—. 98 subject’s report of their aggressive behavior from the same periods. The full model incorporates all predictors into one model. Comparing across comparable models in Table 4.9 to Table 4.10, the effect of the sex of the young adult on YA community offending is higher in magnitude in Table 4.10, and highest and most significant (e.g., Model 3, IRR=4.41, p < .001) when subject aggression is also in the model. Although the family environment factors in Model 2 and Model 3 in both tables are comparable, the effect associated with the race of the subject on their violent community offending increases from Table 4.9 to Table 4.10 by the introduction the family environment variables, opposite to IPV offending models. However, the effect associated with race is only significant in Models 4A through to the full model for young adults of other races (e.g., IRR=6.30, p < .01). The factor of being in a chemically-dependent household again changes from negative to positive after subject ETV and aggression variables are added, opposite to the IPV offending models. Living in a household below the poverty level is always negatively related to YA violent community offending as is the PC’s report of supervision, while not living with both biological parents and the size of the YA’s family at Wave 3 are positively related to their community offending. The latter two were modestly significant (IRR=2.21 and IRR=3.64, respectively, p < .05) in Model 3 when subject aggression is also in the model. Similarly, living below the poverty level in Wave 3 is only significant in Model 4A (IRR=.37, p < .05) when the PC reports of family aggression are included. Although the PC’s report of PC-partner violence is slightly negative and non- significant, child-directed aggression in the home is positively and significantly (IRR=2.35, p < .001) related to violent offending in the community at Wave 3, yet 99 reduces and loses any significance when controlling for the subject’s reports of ETV and aggression (IRR=1.96, n.s.). Experiencing violence in either setting and witnessing violence in the home as reported by the subjects are positively related to their community offending at Wave 3, but becomes negative when controlling for their aggressiveness; all continuing to be non-significant. Witnessing violence in the community is positive to violent community offending and significant (IRR=4.09, p < .001). However, controlling for subject aggression reduced the positive relationship between witnessing violence in the community and violent community offending by 47%. Still, based on the full model, for a one unit increase in violence witnessed in the community, the expected rate ratio of violent offending in the community at Wave 3 increases by 2.34 times (p < .001). Subject aggression remains highly significant (e.g., Model 3, IRR=l 8.20, p < .001) in each model it is present, although reduces by 59% when controlling for the full complement of PC and subject reports of ETV in the full model (IRR=7.44, p < .001). The log pseudo- likelihood of the Poisson model for the full model is 941 .07 and for the constant only model is -409.33. In comparison, the log pseudo-likelihood for the comparable aggression only model as shown in Table 4.4 is —397.89 (p < .001). Fitting the full model incorporating all reports of Wave 1 and Wave 2 ETV and corresponding aggression score, the log pseudo-likelihood is -370.48, indicating the full model is the best fit to the data (closest to zero) of all the models in Table 4.10, all of which except Model 1 are significant at p < .001. Again, this full model supports both SLT and ASBT theory, yet with the subject’s aggressiveness more than a three times larger factor than their witnessing violence in the community. 100 The Wave 2 Model Table 4.11 examines the effect of Wave 2 family environment, ETV, and subject aggression on Wave 3 community offending. Model 1 shows the relationship between the demographic variables. Model 2 displays the results after the family environment variables at Wave 2 are entered into the model. Model 3 uses the subject aggression score from Wave 2 to examine the effect of the subject’s aggression independent from their reported ETV at Wave 2. Model 4A and Model 4C use the PC’s reports of PC-partner violence and child-directed aggression against the subject at Wave 2. As described earlier, these Wave 2 reports were recorded as binary only. Model 4B and Model 4C include subject reports of each type of ETV from Wave 2 only. Model 4C adds the aggression factor score from Wave 2 to examine the effects of ETV along with the subject’s report of their aggressive behavior from the same period. The full model incorporates all predictors into one model. Comparing across the models in Table 4.10 to Table 4.11, the effect of being male compared to female on the rate ratio of YA violent community offending is higher in most models in Table 4.11, however, remaining the closest to the models in Table 4.10 and the most significant when aggression (Model 3, IRR=4.79, p < .001) and subject ETV and aggression are also in the model (Model 4C, IRR=3.42, p < .001 and the full model, IRR=3.08, p < .001). Across the models in Table 4.11, the effect associated with race for Hispanic young adults increases slightly (i.e., more negative) and the effect for Black and other race young adults on rate ratio of their violent community offending at Wave 3 reduces by the introduction the Wave 2 (vs. Wave 3) family environment, ETV, and aggression factors. Except for Hispanic young adults, the IRR coefficients on the 101 race predictors in Model 2 (with Wave 2 family environment variables) and Model 3 (which added the subject’s Wave 2 aggression factor score) calculate lower than the values of the IRR coefficients in comparable models in Table 4.9 and Table 4.10; however, they calculate in-between the IRR coefficients in these tables for the models with subject ETV (Model 48) and subject ETV and aggression (Model 4C), except for other race. The only race difference that remains significant in any model is between other race and White subjects (e. g., Model 4C, IRR=3.82, p < .05 and the full model, IRR=4.12, p < .01). The factor of being in a chemically-dependent household (at Wave 1) is positive to the rate ratio of violent community offending and is highest and significant (Model 48, IRR=2.25, p < .05 and Model 4C, IRR=2.43, p < .01) when subject ETV and aggression are included in the models. In Table 4.11, living in a household below the poverty level at Wave 2 is always positively related to YA community offending at Wave 3 and is highest and significant (IRR=2.22, p < .05) in Model 3 when Step 3 only included subject aggression. The factor of not living with both biological parents at Wave 2, the PC’s supervision score at Wave 2, and the size of the YA’s family at Wave 2 are essentially negatively related to the YA rate ratio of violent community offending. Of the three, only PC supervision in Model 4B, which included the subject’s reports of ETV, is significant (IRR=.89, p < .05). The factor of the PC’s report of PC-partner violence is not significant, although positively related to community offending in Model 4A, but negatively related when controlling for the subject’s reports of ETV and aggression in the full model. However, the PC’s report of child-directed aggression against the subject in Wave 2 is positively related to their violent community offending, which is significant (IRR=2.40, p < .05) 102 :5. W 9.2.2.. ._o. W 9.2.. .mo. W a... .8828 0.3 833.? 8:8 233 35:8 2: E 8385 :5. _ a 8: 083.5 2: :o 2.8 2.3 05 E owcmso 38098 305,. mosey— Bmx E020:— ._o>o_ .930: 2: 95%. $3598: 28:53: 3328298: E 8:88 _8_wo_oB 50: 5:5 :02. 0:3 8380.: BE? 3 9.8m oocobfim .mHOZ .12.. 3:: .23.. mwd: 1.... mm.::_ 1.... :53 .2. 8.3 .1. 54mm .. mv.:_ 3:33-20 Em? o:.~\.m- bwdhm- 3.2m- v9wam- n93? m9::v. 5. EV. woof—ovEéwoaq wot— ... 21 3.: ... on; mfim .12.. Sum 3.5 5320: ozmfizwwxx .12.. mm.— :53? ..:.:.. Nm.— Nhé .33.. we; N96 .8500 Eflommofima .9: 3.: mm: 3.: :9: :4 0:8: 5:88:55 3.: 5.: mm: 3.: mm: 5; .888 £82.28me 3.: mm: m9: .3: :5: :N._ oEo; :__ooo:otoaxm atone. Sufism 3.: mm; .._ 8.: :v.n .hwwm 3.8.6-220 mm: 3.: :9: m:.— 8:203 8523.0: 2.5%: U: “6.5.3... 8233; 2.3m mm: v9: em: :9: 3%: 3.: mm; hm.— om: v9: K: 3.: 86 his... 3.: 3.: v:.: 3.: .. 3.: :9: w:.: 3.: v:.: 3.: m:.: m:.: 563.59% :9: :9: .9: 3.: mm: «.9: 3...: ~:.— mm: 3.: 3.: 3.: 3:23 2&3 o>=82 .9: up; :9: :4 :9: E: .9: ma; ... mm: -.~ ..9: :9— >to>oa 320m ... m5: ~N.~ t. R: 9am .. m5: m~.~ mm: ::.— I. 3.: :9N we: m~._ Ennficoaonéoso $25.35 3303282 2.8m I. :NN «5. *EN mad I :w.~ :.m 1. Sam mm... .1. ::.~ 5.: ... a: 3.: P.— r:.m 88550 :9: mm; m9: hm; 59: :v.— 52 vmfi mm; ~9N m: an; 3.: or; xoflm 2...: :9: mm: mm: mm: 3.: :9: :9: v0: mum: and >9: 3.: :9: 253$: 83% .1... m:.. ::.m .13.. :2 ~96 1.... mo; m:.m .I. 2...— _~.m 3.2.. Q... :56 .1. em; mnd ... w: de 232 Row. 2% .m.m my: .m.m my: .m.m my: .m.m my: .m.m my: .m.m my: .m.m my: 3.3: 3E UV ~38: NV 3.3: 3. 3.3: M :45: N 3.3: ~ Ewe: Eve—ZNoga .232 N 95>.» - bEaEEcU 2: E wEEeEO 2.2:; :o 8.3— .3 3:295 8 2:89.”..— .8 38:”.— .:.v 2...; 103 only in Model 4A when PC reports are the only ETV factors in the model. This positive relationship is mirrored by the subject’s report of experiencing violence in the home in Model 48, which becomes negative when controlling for their aggressiveness in both Model 4C and the full model, though never significant. Experiencing violence in the community and witnessing violence in the home at Wave 2 both show a similar pattern. Witnessing violence in the community at Wave 2 is positively related to the YA rate ratio of violent community offending at Wave 3 and is significant in Model 48 (IRR=6.02, p < .001), but reduced over 20% in Model 4C (IRR=4.75, p > .001) and the full model (IRR=4.79, p < .001) when also controlling for the subject’s aggression. Similar to the models in Table 4.9 and Table 4.10, subject aggression decreases in Table 4.11 58.7% compared to Model 3 (IRR=7.41, p < .001) when controlling for the full complement of PC and subject reports of ETV in the full model and is less significant (IRR=3.06, p < .05), indicating that subject aggression at Wave 2 explains less of the variance in the subject’s rate ratio of community offending at Wave 3 when other ETV factors are included. In fact, the IRR coefficient for subject aggressiveness in the full model (see above) is less than those for subject sex and witnessing violence in the community. The log pseudo-likelihood of the Poisson model for the full model is -984.80 and for the constant only model is -409.33. In comparison, the log pseudo-likelihood for the comparable aggression only model as shown in Table 4.4 is -400.91 (p < .001). Fitting the full model incorporating all reports of Wave 2 ETV and corresponding aggression score is -372.09, indicating the fill] model is the best fit to the data (closest to zero) of all the models in Table 4.11, and like all the models with ETV predictors, is significant at p < .001. This full model supports both SLT and ASBT theory, but with witnessing 104 violence in the community (at Wave 2) more than one and half times larger as a factor than the subject’s aggressiveness (at Wave 2), opposite to the results in the combined model. The Wave 1 Model Table 4.12 examines the effect of Wave 1 family environment, ETV, and subject aggression on Wave 3 community offending. Again, Model 1 shows the relationship between the demographic variables. Model 2 displays the results after the family environment variables at Wave 1 are entered into the model. Model 3 uses the subject aggression factor score from Wave 1 to examine the effect of the subject’s aggression independent from their reported ETV at Wave 1. Model 4A and Model 4C use the PC’s reports of PC—partner violence and child-directed aggression against the subject at Wave 1.. Model 4B and Model 4C include subject reports of each type of ETV from Wave 1 only. Model 4C adds the aggression factor score from Wave 1 to examine the effects of ETV along with the subject’s report of their aggressive behavior from the same period. The full model incorporates all predictors into one model. Comparing across the models in Table 4.11 to Table 4.12, the effect of the sex of the young adult on the rate ratio of YA community offending at Wave 3 is lower and less significant (e.g., the full model, IRR=2.56, p < .05). This factor is highest in Model 3 when subject aggression alone is added in Step 3 (IRR=3.24, p < .01) and lowest and even non-significant (IRR=2.08, ms) in Model 4A when the PC reports alone, especially the significant factor of reports of child-directed aggression, are added in Step 3. Still, compared to females and based on the full model, the expected rate ratio of violent community offending is 2.56 times higher for males, controlling for all other variables. 105 .8. w 9...... .5. w a... .8. w a... 4:82:00 2s 33st? 3:8 2:?» .838 05 E 3855 :5. _ a 8.: 058:0 2.: :0 0:8 88 2: E owner—0 @2898 30% 8:3: 83— E020:— ._o>o_ .926: 2t 26% 86:88: 30:53: 33250595: E 85.8: _ao_mo_oE Son :23 BE on? 8:25.: 833 m_ 9.8m 3:20:3— ”MHOZ .13.. ~92: .13.. mmdm: 3.2.. vam .12.. mNHm 3.3.. modw .w hNH— _.. mvd— UPS—#5450 2:3 mm. _ mm- m _ .mwm- madam- owwam- :_ .wom- no.5? 54:? 255—8563st mod 3.... a: R... a... a: «E .1... m3. mm.» .2222. 35.2%.... 1.... mm: 2..— ..:..... mm: a; 2.... Q: mQN 888 533053? mm: :m.: mm: 3.: mm: 3.: 0:5: 533055, and 3.: we: mm; 3% 5.: .888 £3285me a: a: a: 0:8: 5 voocotomxm atom? Sufism. 1. mm: mm; *1. 3.: SJ qwma 3826-220 wfi: ~:.— 2.: 3.: 8:203 contend: mtommxbk 8.391on 82395 2.5 mm: 3.: am: :1: :m: Eu: .23 8.: :m: we: 9...: 5.: on? 3E5: m:.: mm: m:.: mm: m:.: no: m:.: m:.: v:.: m:.: m:.: ma: :o_m_>.8a:m 3w: ::.— :v: m:._ 3.: m—._ , mm: m:.: 9.: :.— mm: VNA 3:22: 2&3 3:82 um: 2.: cm: 8.: cm: 3.: :m: 3.: mm: Eu: NY: mm: 90.5: 320m 3...: m_._ 3.: 3; mm: 3.: mm: v:.— on: :m.— NY: 3A Ensdgmocéono 3.5.85 EoBEéoZ 2.3m 1.... wmé :2. 2.... KM NY: .1. mmd :m.v _.. wfim $6 1.... vmd 3%: 3: ::.N :4 5:6. 008 550 :5: am; mo: :4 :5: 3..— nw: $4 ... mm; ~:.m m:._ E..— v:.~ E..— xoflm Eu: hm; :m: ::.— 3%: 3w: 5.: 54 me: ha; 3:: :9: _m.: cm: 253$: 88‘ ... 3.: :m.~ .1. co; :b.~ .. 5.: mmd 3w: ::.N .2. mm; 34m a. v: E..~ ... w: Nm.~ 232 How. 2.3m .m.w my: .m.m my: .m.m aw: .m.m ¢~= .m.m ¢~= .m.m my: .m.m My: Ewe: 3E UV 33%: 5.. 333: 3. 3:3: M 3:53 N ~35: ~ Ewe: .252 u 93>? Etc—Z _ 96>» .. big—ESQ 2: E mic—:50 «:29; ..e 8.3— :: 3:295 3 232?”.— ..o 38.5— .~_.v 03:. 106 Although several models show the IRR coefficients for Hispanic subjects relative to White subjects change from negative to positive, none are significant. Only Model 3 where Step 3 included only subject aggressiveness is the difference between Black and White subjects significant (IRR=3.02, p < .05). The effect for other race subjects compared to White subjects is largest and most significant in the models that also included subject aggressiveness (e.g., the full model, IRR=7.16, p < .001). The family environment factors showed little noteworthy fluctuation across the models and none are significant. The PC’s report of PC—partner violence in Wave 1 showed little effect on YA violent community offending, and is not significant. Similar to the models in Table 4.1 l, child-directed aggression in Wave 1 in both models in Table 4.12 is positively related to the rate ratio of YA violent community offending at Wave 3. Although decreasing in magnitude and significance from Model 4A to the full model when subject reports of ETV and aggressiveness are included, this factor remained significant (IRR=1.88, p < .01). For a one unit increase in the amount of child-directed aggression at Wave 1, the expected rate ratio of violent community offending is 1.88 times higher, controlling for all other variables. None of the subjects reported experiencing violence in the home so this factor could not be included in the models in Table 4.12. Experiencing violence in the community at Wave 1 is positively related to community offending at Wave 3, reduces in magnitude in Model 4C and to a negative relationship in the full model when subject aggression is added to the model; yet never significant. Witnessing violence in the home at Wave 1 is, however, only negatively related to community offending at Wave 3 and never significant. Witnessing violence in the community at Wave 1 is positively and 107 significantly related to the YA rate ratio of offending at Wave 3 in Model 4B (IRR=2.35, p < .001), but saw at least a 20% reduction compared to Model 4C (IRR=1.87, p < .001) after subject aggression is added in the full model (IRR=1.70, p < .001) after subject aggression and PC ETV are added. Subject aggression remains highly significant in each model it was present (e.g., Model 3, IRR=7.52, p < .001). Similar to all the previous tables about violent community offending, subject aggression in Table 4.12 decreases 31.1% when controlling for subject reports of ETV in Model 4C (IRR=5.18, p < .001) and 36.6% when controlling for the full complement of PC and subject reports of ETV in the full model (IRR=4.77, p < .001), indicating these significant ETV factors offset aggression in explaining the variance in the subjects’ violent community offending at Wave 3. However, the expected effect of subject aggressiveness at Wave 1 on YA rate ratio of violent community offending at Wave 3 is almost three times higher than the expected effect of witnessing violence in the community at Wave 1. The log pseudo- likelihood of the Poisson model for the full model is -1011.00 and for the constant only model is -409.33. In comparison, the log pseudo-likelihood for the comparable aggression only model as shown in Table 4.4 is -401.84 (p < .001). Fitting the full model incorporating all reports of Wave 1 ETV and corresponding aggression score is -381.57, indicating the full model is the best fit to the data (closest to zero) of all the models in Table 4.12, all of which were significant at p < .001 when subject aggressiveness and/or ETV are included in the model. Like the combined model, this full model supports both SLT and ASBT theory, yet with the subject’s aggressiveness more than two and half times larger a factor than their experiencing violence in the home or witnessing violence in the community. 108 Comparisons Across Time Since the full models for each time period are the best fit to the data, the full models for each dependent variable are organized into one table for each to facilitate comparisons over time. Table 4.13 shows the full models for the five time periods for the effect of exposure to violence on the rate ratio of IPV offending by the young adults at Wave 3. The full model at Wave 2 provides the best model fit for IPV offending according to the log pseudo-likelihood and is significantly different than all the other IPV offending models (cross-sectional model, p < .01; all others, p < .001). The Wave 2 model shows the largest effect of the sex of the subject, with males expected to have a rate ratio of IPV offending that is .12 times (p < .001) that of females — in other words, the expected rate ratio for males is 88% of the expected rate ratio for females, while other variables are held constant. In fact, the sex of the young adult as a predictor of the rate ratio of the IPV offending at Wave 3 is the most stable predictor over time — the smallest gender difference is 78% (IRR=.22, p < .001) at Wave 1. In the Wave 2 model, the effect associated with the subject’s race is significant (IRR=2.67, p < .05) only for Black young adults compared to White young adults. The largest effect associated with the subject’s race occurs during mid-adolescence at Wave 1, with the effect ranging from three times (IRR=3.17, p < .01) the rate ratio of IPV offending for Hispanic adolescents relative to White adolescents to seven times (IRR=7.09, p < .01) for adolescents of other races. Across time, the only family environment factor that is significant to the rate ratio of YA IPV offending at Wave 3 was that of not living with both biological parents at Wave 1 (IRR=2.39, p < .01), with those that did not displaying an expected rate ratio 2.4 times of 109 adolescents that did live with both biological parents at Wave 1, while controlling for all other variables. The Wave 2 model is the only time the PC’s reports of child—directed aggression is significant (IRR=.47, p < .05), which is also unexpectedly negative (rather than positive as predicted by SLT), indicating young adults who received child-directed aggression from a household member at Wave 2 are expected to have a rate ratio of IPV offending at Wave 3 that is about 50% of those that did not. This result is very similar to the PC’s report of PC-partner violence at Wave 2, which is not significant (IRR=.52, p < .10) as well as the subject’s report of experiencing violence in the home at Wave 2, which is significant (IRR=.33, p < .01). This latter comparison is especially interesting in that similar results are shown from both informants for comparable measures despite the binary coding of the PC report and the subject report represented by a count. Only at Wave 1 is the factor of experiencing violence in the community significant (IRR=2.39, p < .05), which indicates for a one unit increase in violence experienced in the community during mid-adolescence the expected rate ratio of IPV offending increases nearly 2.4 times, while other variables are held constant. Conversely, for each unit increase in violence witnessed in the home in Wave 1 as reported by the adolescent, their expected rate ratio of IPV offending at Wave 3 decreases by over 80% (IRR=.18, p < .001). As Table 4.3 indicates, the percent of adolescents reporting either type of violence exposure in Wave 1 was small; however, for those that reported either in Wave 1, the effect is still. a factor six years later. Each unit increase in witnessed violence in the community at Wave 2, however, increases the expected rate ratio of YA IPV offending at Wave 3 by over 1.6 times (IRR=1.67, p < .05), second only to the effect of 2 times (IRR=2. 14, p < 110 .8. w 9...... .5. w a... .8. w a. .Efimcoo 2a 3338.» 850 BE? 35:8 05 E 3385 «E: _ a co: 082:: 05 :0 0:8 82 on. E owSEo @8898 30;: 833: Sum 2522: ._o>o_ €95: 2: 025: $3.630: 809.30: 332820.20: E 8:83 Egg—05 £0: £3, :3: 0:3 83:5: 35>) m_ 95% 3:283: ”MHOZ 3.3 ovd: .32.. MON: .13.. deN— .12.. dem— 3.2.. oeNa wen—5:4:o 235 8.51 :22? 33.. m _ .31 8.24. 82:33:88 :3 .1... R; 3.: I... :3. ::.~_ I... 8.: 3.: r; 2.: :~.m 1... 3.: 3.: 8323 o>_mm8ww< w_.: 3.: * am: Nb.— wm: .3.— r; 9:: 3: 3.: .5.— .EEoo £83253 3.... 8.: 2.: mm: 34 S: v:._ R: ah: mm: 5.: 6:8; 5:88:23 ._. 34 ::.~ 9.: cm: ::.: ::._ S: :2 m:.: :2 .888 Ecoocotoaxm Q: I. 3.: ::.: .. 2.: ::.: 3.: :2 m:.: :3 use: Ewoocotoaxm when? 36.33% 2.: :.~ .. 3.: 5.: 2.: S.— a: 8:22., 36266:;0 4a.: :3 :m.: S: 2.: S; 2. 853355590: highway: hmkzams NUQNNQNA «4.3: 3.: :3 on: E... :3 .5; a. 83:8“: mod 3.: 3.: 3.: 0:: 3.: a: 863.6%»: 1. 5.: ::.N 2:: :mA 5.: EA :m.: m:.— 3:: NM; 853: 2&3 3:82 cm: 3.. :v: :v._ mm: 8.: m: :m.: 8.: mm: €96: 320m 3:: a: 9x: 8..— wm: 3.: mm: 3.: cm: 3.: Eugene: 333820 wwkzgwg NQNNONK—ntoz :3: I we: :3. at: 5.: :N: ~:.v . 5.: m:.v ... 3.: ::.v 88 350 1.... mo: 3.: .. _:._ 3.: .. 5; :9: r. w._ 3.: ._. no: 3.: x85 .2. :3 2.: mm; :2: w: 3.: . 3; 3.: ... mm; :m.~ 2:82: 86% .12.. 8.: «N: 1.... 3.: 2.: 1.... co: :~.: .1. mo: 5.: *1. co: 2.: 232 Row. .98: .m.m my: .m.m 5: .m: 5: .m.m 5: .m: 5: ~33: ~ 3:3. 33%: N 9.3: 3:33 szmefieb 3:5: totooaumeub 38: buxomfiwutob 2252:: £252 u: guinea—:0 - win—SEC :53; 35.3.— 32-55 u: 823— .8 «2.2:; 3 2:89”: :: 38.5— .2... 03:. 111 .001) displayed for their exposure to violence witnessed in the community in the cross- sectional model. Although supporting SLT, the effect of witnessed violence in the community is stronger for the more recent exposures, despite the relative stability of community violence witnessed over time as shown in Table 4.3, and indicates that the effect of this factor can fade over time. Finally, the model for Wave 2 with the best fit to the data supports ASBT theory and also shows the largest effect of subject aggression relative to their YA IPV offending at Wave 3, indicating an over 12 time increase in IPV offending at Wave 3 for each unit increase in the subject’s Wave 2 aggression score (IRR=12.30, p < .001), while controlling for all other variables. Although subject aggression is always significant relative to the rate ratio of IPV offending at Wave 3, its effect is lowest when using the Wave 1 aggression factor score, alone in the Wave 1 model (IRR=3.64, p < .001), as well as apparently reducing the effect of this factor in the first three models in Table 4.13 for which the mean of the Wave 1 and Wave 2 subject aggression scores are used. Overall, the IPV offending models support ASBT theory as a predictor of this type of offending, but not the only factor. Current ETV and ETV three to six years earlier are also significant predictors - in the home negative and in the community positive, lending mixed support for SLT, and indicate that violent models can encourage or discourage future intimate partner violence. Table 4.14 shows the full models for the five time periods for the effect of exposure to violence on the rate ratio of violent offending in the community by the young ’ adults at Wave 3. In contrast to the IPV offending models for which the Wave 2 full model was the best fit to the data, the cross-sectional model using Wave 3 predictors and the mean of the Wave 1 and Wave 2 aggression score provides the best model fit for 112 community violence offending according to the log pseudo-likelihood. The cross- sectional model is not, though, significantly better than the other two Wave 3 predictor models (i.e., the contemporary and combined models -— columns 1 and 3 in Table 4.14). However, similar to the best IPV offending model, the Wave 3 cross-sectional model also shows the largest effect of the sex of the subject, but instead with males expected to have a rate ratio of community violence offending at Wave 3 that is higher rather than lower, which is 3.48 times (p < .001) that of females, while controlling for all other variables. Again, the sex of the young adult as a predictor of their community violence offending is the most stable predictor over time — the largest gender difference compared to the cross- sectional model is again Wave 1, which shows a factor of 2.56 (p < .05) between the sexes at Wave 1. In the cross-sectional model, an effect associated with the young adult’s race is not significant for any of the three races as compared to White young adults; in fact only for young adults of other races are any of the models significant. Whereas not living with both biological parents at Wave 1 is significant for IPV offending, this factor is not significant in any of the community violence offending models. Instead, living below the poverty level at Wave 3 is negative and significant (IRR=.32, p < .01) to the rate ratio of committing violence in the community at Wave 3, indicating those living below the poverty level at Wave 3 have an expected rate ratio of offending at Wave 3 that is 68% less than the young adults that did not. This negative result is also found in the two other models that used living below the poverty level at Wave 3 as a predictor. Across time, the only other family environment factor that is significant to the rate ratio of community violence offending at Wave 3 is that of living in a chemically-dependent household at Wave 1 when included in the Wave 2 model (IRR=2.21, p < .05), which 113 .8. w 9...... .S. w a: .2. w a... €8.38 8: 832:3 :05: 22? 33:8 2: E 83.85 :5: _ a 8.: 080:5 2: :0 0:2 88 2: E owSEo 3:898 26% mega 83: E020:— ._o>o_ 5.55: 2: 30:: mEozowzo: Eugene: 332505.:0: E 3:23 $23.05 53 £3, :3: 0:3 8.9:“: SE? 2 :38» 00:20.3”. ”m #0 Z ......... _o.mo_ .13.. ~od: .12.. dem— .13.. ww._m_ .13.. NbfiO— teamsvmufio 2:3 mm. _ wm- 3&5”- wvdnm- $63.. $65. EOE—8:70:03: wou— ......... .8.— hbé ... Q: 8.: .12.. mod 3%: a ma; mud ... v9— m:.n 5320: o>_mm2ww< .22.. mm: :hé ......... mm; in: 3.... 9»: 3nd 2... mm: and 1.... m5: ~:.n .888 5:38:23 mm: mm: _m.: 3.: mm: 3.: 3.: Eu: .. 2.: 3.: 0:8: 5:88:23 mm: 3.: cm: 5.: mm: ah: 1. 3V.— Nvfi .. mm: m5.— .888 5:00:28qu :: :1: «a: on: me: on: N~.— Ex: VNA 2:0: 5:025:0qu 5.8%.: 8:330 ...... mm: 3.— 3.: ma; 3.: 34 8:22.» 3826-250 w_.: _:._ MN: 3.: 2.: 3.: 8:20? 35.890: 35%.: 9% $2562: 9233.5 2.3m MN: 3%: mm: vm: .3: an; a: a: 0N6 AGE»: 3.: mo: 3.: no: 3.: «a: a: a: :o_w_to::m 3w: ::._ :m: 3.: 3.: an; on: :m.— mm: :m.— 3:83 :5?» 03—82 mm: 3...: 5.: NBA _N.: :m.: t. 2.: NM: ... 2.: on: 35.5: 320m mm: mm; ... mud. :.~ 3.: cu; wm: 9:.— Ex: mvé 30:53: 332825 83:85 3:33:02 2.5 1.... wmd 34. i. :NN 2... ...... flan :m.: vwd Rum ... ~w.m m:.v com: 350 mm: :m.— :9: mm.— :5: 34 am: 3.: we: ::.— xoflm v5: Em.— :N.: 3.: 9.: 3.: .N: on: em: 3.: 25%;: mock .. 3.: :m.~ 1.... mo; ::.m .12.. ::._ ::.m t... v: :v.m ...... cw: :m.~ 2:2 Rom. 2.8m .m.m my: .m.m my: .m.m Kw: .m.m my: .m.m :— 33: ~ .33: ~38: N was: 3:3: 3:35:39 -§§< zfinumhunmeb 3?»: $3335.39: 2252:5— mfiéoz u: 53.39.80 - bias—=80 2: E NEE—«=0 2.2:; ..e 8.3— :: 8:23.» 3 2:8me .3 38...:— .v—.v 035—. 114 displays an expected rate ratio of 2.2 times that of adolescents that did not live in such a household at Wave 2. The cross-sectional community offending model could not include the PC reports of family violence. However, the Wave 1 model shows the PC’s reports of child-directed aggression is positive and significant (IRR=1.88, p < .01) (in contrast to the Wave 2 IPV offending model which is significant and negative), indicating young adults who received child-directed aggression from a household member at Wave 1 is expected to have a rate ratio of community violence offending at Wave 3 that is 1.88 times of those that did not, while controlling for all other variables. However, none of the subject reports of experiencing violence in the home are significant in any of the community violence offending models and subject reports of witnessing violence in the home is significant only in the contemporary model (IRR=.64, p < .05), yet negative as in all the models. The factor of experiencing violence in the community is significant in the cross-sectional (IRR=3.42, p < .01) and contemporary models (IRR=1.86, p < .05), with the largest effect in the cross-sectional model. Since the effect of experienced violence in the community is stronger, and positive rather than negative, for the more recent exposures, despite the higher reports of community violence experienced in Wave 2 (compared to Wave 3) as shown in Table 4.3, indicates that the effect of this factor can fade over time or violence in the community experienced at the younger ages affects individuals differently than when older. Witnessing violence in the community is a consistently significant factor (e. g., in the cross-sectional model, IRR=2.33, p < .001) in the rate ratio of young adult violent community offending at Wave 3, registering an effect consistent with the subject reports in Table 4.3, lowest for Wave 1, highest for Wave 2, and in-between for the two 115 models (contemporary and cross-sectional) incorporating Wave 3 witnessed community violence as a predictor. Although subject aggression is always significant relative to the rate ratio of YA community violence offending at Wave 3, its effect is offset by other factors, showing its strongest effect in the combined model (IRR=7.44, p < .001) with only one non-demographic predictor significant and lowest in the contemporary model (IRR=3.03, p < .05) when four non-demographic predictors are significant. Still, subject aggression is not as strong a factor in the rate ratio of YA community violence offending at Wave 3 as it is with YA IPV offending at Wave 3, both in magnitude, except for the Wave 1 model, as well as significance for three of the five models. For example, the cross-sectional model with the best fit to the data indicates for each unit increase in the subject’s mean aggression score from Wave 1 and Wave 2 the expected rate ratio of community violence offending increases by 3.75 times (p < .05), which all other variables are held constant. In contrast, in the Wave 2 model, which is the best fit to the data for IPV offending, fore each one unit increase in the Wave 2 aggression score, the expected rate ratio of IPV offending at Wave 3 increases by 12.30 times (IRR=12.30, p < .001). Overall, these models support ASBT theory and community ETV at any of the time periods as predictors of violent community offending at Wave 3. Focusing on just the two best models for each dependent variable, the Wave 2 model for IPV offending and the cross-sectional model for community offending, both show strong effects associated with the subject’s sex, but in opposite directions. The expected rate ratio of IPV offending for males is 88% (IRR=.12, p < .001) of females, but their expected rate ratio of community violence offending is expected to be 3.48 times (p < .001) that of the female young adults. Only IPV offending shows any effect associated 116 with race, with Black young adults expected to have a rate ratio of offending that is 2.67 times (p < .05) that of White young adults. No effect associated with the subject’s race is seen in the community violence offending model. However, none of the family environment factors are significant for the IPV offending model, and only living below the poverty level is significant (IRR=.32, p < .01) for the violent community offending model, yet surprisingly showing a negative relationship that indicates those living below the poverty level have an expected rate ratio that is 68% of those living above the poverty level at Wave 3. For the IPV offending model, experiencing violence in either setting is negative but witnessing violence in either setting is positive to the rate ratio of IPV offending at Wave 3. Specifically, PC’s reports of child-directed aggression is negative and significant (IRR=.47, p < .05) as is the subject’s report of experiencing violence in the home at Wave 2, which is also significant (IRR=.33, p < .01). Together they indicate for a unit increase in experiencing violence in the home the rate ratio of IPV offending is expected to decrease by 53% to 67%. As with experiencing in the home, the subject’s reports of experiencing violence in the community are negative to IPV offending, but miss significance (IRR=.56, p < .10). However, subject reports of witnessing violence in the home is positive but not significant, yet witnessing violence in the community is positive to the rate ratio IPV offending, showing a factor of 1.67 times, and as is significant (p < .01). Although neither are significant in the community offending models, the factor of violence in the home is opposite between IPV and community violent offending, for which in the latter experiencing in the home is positive while witnessing in the home is negative. Both experiencing and witnessing violence in the community are positive (3.42 117 times and 2.33 times, respectively) and significant (p < .01 and p < .001, respectively) to the rate ratio of committing violence in the community at Wave 3. The subject’s aggression score at Wave 2 is a large factor (IRR=12.3, p < .001) in the IPV offending at Wave 3; the mean aggression score used in the cross-sectional model at Wave 3 shows a factor to community violence offending at Wave 3 that is 60% less in magnitude as well as less in significance (IRR=3.75, p < .05). Overall, according to the log pseudo- likelihood, the community offending model (-367.54, p < .001) better fits the data than the IPV offending model (-425.3 1 , p < .001). Regarding the social learning theory vs. anti-social behavior trait debate, ASBT appears a larger factor than ETV for IPV offending whereas ETV appears a larger factor in community violent offending. However, this supposition is confounded by the opposite directions of the relationship of experiencing violence (negative) and witnessing violence (positive) in regard to IPV offending at Wave 3. 118 Chapter Five Discussion The analysis of the PHDCN data provides evidence that adolescent exposure to violence is a significant factor of offending as adults. However exposure in some settings produce a larger effect than others and the relationship of some types of exposure were in a different direction than expected based on social learning theory (SLT). Each project question will be addressed individually. 0 Does witnessing intimate partner violence between a primary caregiver and a partner or experiencing child-directed aggression during adolescence correspond to committing intimate partner violence in later romantic relationships? Violent interactions between family members as reported by the PC in Wave 1 and Wave 2 or by the subject for all three waves display significant relationships to the rate ratio of IPV offending as a young adult. However, the relationships are significant only when the subject’s contemporary ETV is not in the model. In other words, in the full contemporary model and the full cross-sectional model (see columns 1 and 2 in Table 4.13), based on the subject reports alone, experiencing violence in the home is positive but not significant to their IPV offending and witnessing violence in the home is negative but not significant. The non-significant results may be influenced by the young adult spending less time in the home during Wave 3. As shown in Table 4.3, subject reports of ETV are largest by percentage in Wave 2 and the PC reports largest in Wave 1. Subject reports of experiencing violence in the home during Wave 2 (since none reported experiencing violence in the home in Wave 1) in the combined model (column 3 in Table 4.13) and the Wave 2 model (column 4 in 119 Table 4.13) are negative and significant (IRR= .39, p < .05, IRR=.33, p < .01, respectively). Yet, they are significant only when aggression is also in the model as shown in Table 4.6 and Table 4.7. Subject reports of witnessing violence in the home in either the combined or Wave 2 models are positive (as expected by SLT), but not significant. However, in the Wave 1 model (column 5 in Table 4.13) their reports of witnessing violence in the home is negative and significant (IRR=. 18, p < .001) and is unaffected by the addition of the PC reports of family violence or by the subject’s aggression score in the models as shown in Table 4.8. The PC reports of child-directed aggression in Wave 2 confirms their youths’ reports (IRR=.47, p < .05), although the combined model, which summed the Wave 1 and Wave 2 PC reports of child-directed aggression as well as the reports in the Wave 1 model alone are positive to the YA IPV offending at Wave 3, but neither are significant. The PC reports of IPV in the combined model and Wave 1 model are positive to the rate ratio of YA IPV offending in Wave 3 and in Wave 2 changes from positive to negative with subject ETV and aggression in the full model (as shown in Table 4.7), but again none are significant. Therefore, based on their own reports as well as the PC, violence experienced or witnessed in the home as an adolescent shows a significant relationship to the rate ratio of YA IPV offending in Wave 3, however, only significant when a negative relationship. The negative relationship indicates the exposure to violence when an adolescent produces a legacy of less future IPV offending, rather than more as expected by social learning theory. This analysis may have been influenced by using data that started when the adolescents averaged 15 years old, at a time when they begin to spend less time in the home. Future research may want to investigate other factors that intersect with exposure 120 to violence in the home that clarify the above negative results, including coping mechanisms or perceptions of violence not leading to rewards or positive outcomes. 0 Does witnessing or experiencing violence in the community during adolescence relate to future incidents of committing community violence? The data demonstrate a clear relationship between exposure to violence in the community during adolescence and the rate ratio of committing violence in the community as a young adult, and therefore, provides support for SLT. Across all models at all time periods, subject reports of witnessing violence in the community is positive and significant to the rate ratio of violent offending in the community at Wave 3. The factor of witnessing violence in the community is always reduced, however, in magnitude but not significance by the addition of subject aggression into the models. The strongest influence of witnessing violence in the community is in the Wave 2 model (column 4 in Table 4.14, IRR=4.79, p < .001), the lowest in the Wave 1 model (column 5 in Table 4.14, IRR=1.70, p < .001), and in-between for the contemporary model (column 1 in Table 4.14, IRR=3.02, p < .001) which used the sum of the Wave 1, 2, and 3 subject ETV reports and the cross-sectional model (column 2 in Table 4.14, IRR=2.33, p < .001), which used the Wave 3 subject ETV alone. The combined model (column 3 in Table 4.14, IRR=2.34, p < .001), which used the summed ETV reports from Wave 1 and Wave 2 shows results most similar to the cross-sectional model; however, this model also shows the largest and most significant effect of the (mean) subject aggression score. These comparisons correspond with the percents of the subject reports of witnessed violence in the community at each wave as shown in Table 4.3. Therefore, although 121 supporting SLT, the models also indicate that reductions in witnessing community violence could also lead to reductions in community violence offending. Subject reports of experiencing violence in the community in both the contemporary (column 1 in Table 4.14, IRR=1.86, p < .05) and cross-sectional (column 2 in Table 4.14, IRR=3.42, p < .01) models are also positive and significant and to the rate ratio of violent offending in the community at Wave 3. However, in the other full models (columns 3, 4, and 5 in Table 4.14) which did not include the Wave 3 ETV predictors experiencing violence in the community is negative and not significant after the addition of the subject aggression score and the PC ETV reports (see Tables 4.6, 4.7, and 4.8). As shown in Table 4.3, subjects reported experiencing more violence in the community at Wave 2 than at Wave 3, but the largest effect of this factor is in the cross-sectional model, which shows for a one unit increase in violence experienced in the community at Wave 3, the expected rate ratio of young adult violent offending in the community increases by nearly 3.5 times. These results also support SLT, and yet, indicate that current, rather than a history of experiencing violence in the community, is most influential as a factor in community violence offending. 0 Do situational crossovers occur, e. g. does exposure to caregiver-partner violence during adolescence or experiencing child-directed aggression during adolescence correspond to future incidents of committing community violence, or vise versa? Some situational crossovers did occur regarding predictors of both young adult IPV offending and violent offending in the community, lending support for SLT across both settings. For the young adult IPV offending at Wave 3, the cross-sectional model (column 2 in Table 4.13) shows that the subject’s witnessing violence in the community 122 at Wave 3 is positive and significant to their rate ratio of IPV offending at Wave 3, even though reduced in magnitude by the addition of aggression (IRR=2.14, p < .001). Witnessing violence in the community is positive and significant to IPV offending at Wave 3 as well for the contemporary model (Model 4A in Table 4.5) and the combined model (Model 4A in Table 4.6), but only before the addition of aggression into the model after which this factor is reduced in magnitude and to non-significance (columns 1 and 3 in Table 4.13, respectively). Aggression lead to a similar result in Model 4C in the Wave 2 model in Table 4.7, however, the addition of the PC ETV predictors in the full model (see column 4 in Table 4.13) returned witnessing violence in the community to significance (IRR=1.67, p < .05). In the Wave 1 model (column 5 in Table 4.13), the addition of aggression changed the positive relationship of witnessing violence in the community on the rate ratio of IPV offending at Wave 3 to a slightly negative relationship. Therefore, witnessing violence in the community in their current situation, rather than a history of ETV in the community, appears the most influential to the young adults’ rate ratio of IPV offending at Wave 3 and without also controlling for the subject’s aggressiveness, could be overestimated in its influence. Experiencing violence in the community is only positive and non-significant in the cross-sectional, contemporary, and combined models (columns 1, 2, and 3 in Table 4.13). However, in the Wave 2 model (column 4 in Table 4.13), this factor is instead negative, yet not significant (IRR=.56, p < .10), to the rate ratio of IPV offending at Wave 3. Although reduced in both magnitude like the other models when subject aggression is added, the Wave 1 model (column 5 in Table 4.13) remains positive and significant (IRR=2.39, p < .05) to the expected rate ratio of IPV offending at Wave 3. 123 The number of subjects who reported experiencing violence in the community at Wave 1 was small, largest at Wave 2, and still substantial at Wave 3. So this factor appears to be of little effect on IPV offending at Wave 3 except for the small number who experienced violence in the community at Wave 1. Regarding the rate ratio of violent offending in the community at Wave 3, the cross-sectional model (column 2 in Table 4.14) found no significant predictors in witnessing or experiencing violence in the home at Wave 3. The contemporary model (column 1 in Table 4.14) which accumulated subject reports across the three waves found, however, the subject reports of witnessing violence in the home to be both negative and significant (IRR=.64, p < .05). None of the subject reports of ETV in the home are significant for the other models. The PC reports of child-directed aggression against the subject are positive and significant to the subject’s violent offending in the community at Wave 3 for the combined model (Model 4A in Table 4.10) and the Wave 2 (Model 4A in Table 4.11), but only before the addition of aggression and subject ETV into the models, after which this factor is reduced in magnitude and to non-significance (columns 3 and 4 in Table 4.14, respectively). Aggression and subject ETV lead to a similar reduction in the Wave 1 model (column 5 in Table 4.14), however, the factor of child-directed aggression (the percent of which was highest at Wave 1 in Table 4.3) to the rate ratio of violent offending in the community at Wave 3 remained significant (IRR=1.88, p < .01). This latter result of experiencing violence in the home as well as the earlier result of experiencing violence in the community when younger relative to future violent offending is congruent with prior research that found victims of violence at 124 increased risk of becoming violent offenders (Widom et al., 2006; Macmillan, 2001; Nofziger & Kurtz, 2005). The Social Learning Theory/Anti-Social Behavior Trait Debate The project investigated whether “violence begets violence” while controlling for family environment as well as the adolescent’s sex, race, and self-reported aggressiveness as a measure of an anti-social behavior trait. The final comparisons describe the similar and differential effects of ETV and aggression on offending at Wave 3 to provide results regarding the social learning theory vs. antisocial trait behavior debate. Results for IPV offending reference models in Table 4.13; results for violent offending in the community reference Table 4.14, unless otherwise specified. Regarding the young adult IPV offending at Wave 3, neither subject ETV at Wave 3 or aggression (the mean of the Wave 1 and Wave 2 aggression scores) in the cross-sectional model show a notable influence on the effect of the subject’s sex, race, or family environment regarding their rate ratio of IPV offending. However, in other models, the two factors work in concert with each other in relation to subject demographics. In the contemporary model, the addition of subject ETV (sum of Waves 1 through 3) or aggression (the mean of the Wave 1 and Wave 2 aggression scores) reduces the effect associated with the subject’s race as did PC ETV (sum of Waves 1 and 2) and/or the aggression score (the mean of the Wave 1 and Wave 2 aggression scores) in the combined model. In the Wave 2 model, aggression reduces the effect associated with subject race in Model 3 and subject ETV with or without aggression increases the effect associated with the subject being Hispanic, but reduces the effect for Black and other race subjects. In the Wave 1 model, the effect associated with subject race is reduced by PC or subject ETV in the 125 model whereas the presence of aggression alone or with subject and PC ETV decreases the effect for Black young adults but increases the effect for Hispanic and other race subjects relative to White subjects. Also in the Wave 1 model, the factor of not living with both biological parents increases with subject ETV and when subject ETV and aggression are in the models. The effect of the subject’s sex is in large part stable and unassociated with any predictors. The only notable exceptions are where subject ETV and aggression produce opposite influences — in the combined model, the effect of the subject’s sex increases (becomes more negative) with the addition of subject ETV in the model; in Wave 1 the effect of subject sex is lowest (least negative) in models with aggression included. Regarding their effect on each other, in the cross-sectional model subject reports of witnessing violence in the community is reduced by aggression in its effect on the rate ratio of IPV offending at Wave 3, but remains positive and significant (IRR=2.14, p < .001). In the contemporary, combined, and Wave 2 models, witnessing in the community remains positive to the rate ratio of IPV offending but was reduced to non-significance when aggression is included in the model. Yet, witnessing in the community is once again significant (IRR=1.67, p < .05) in the full Wave 2 model with the addition of PC ETV. Subject reports of experiencing violence in the home in the combined (IRR=.39, p < .05) and the Wave 2 (IRR=.33, p < .01) models and PC reports of child-directed aggression in the Wave 2 model (IRR=.47, p < .05) increases (more negative) and becomes significant when aggression is included in the model. However, in the Wave 1 model, the effect of subject reports of experiencing violence in the community reduces by the addition of aggression and filrther by the addition of aggression and PC ETV into the 126 models, but remains significant (IRR=2.39, p < .05). Witnessing in the home, though, is negative and significant (IRR=.18, p < .001) and is unaffected by aggression or PC ETV in the models. Overall, aggression appears to influence negative ETV relationships to become more negative relative to the rate ratio of IPV offending at Wave 3, but reduces positive ETV relationships, sometimes to non-significance. Similarly, the presence of subject ETV in the contemporary and cross-sectional models (see Table 4.5) reduces the effect of aggression by 40% or more (IRR=9.98 to IRR=5.99 and to IRR=5.23, respectively; p < .001). In the combined model (see Table 4.6) PC ETV reduces the effect of aggression by around 10% (IRR=10.42 to IRR=9.42, p < .001) as well as a reduction of about 9% in the Wave 1 model (see Table 4.8, IRR=4.01 to IRR=3.64, p < .001). Only in the Wave 2 model (see Table 4.7) did aggression increase, about 25%, as a factor of the rate ratio of IPV offending at Wave 3 with the inclusion of PC ETV in the full model (IRR=9.13 to IRR=12.30, p < .001). Regarding the rate ratio of violent offending in the community at Wave 3, the addition of aggression (the mean of the Wave 1 and Wave 2 aggression scores) in the cross-sectional model (see Table 4.9) slightly increases the effect of the subject’s sex on their rate ratio of offending. However, aggression (see Table 4.14) has no effect on the factor of being under the poverty level at Wave 3 (IRR=.32, p < .01), which is similarly unaffected by aggression or subject ETV in the contemporary model (IRR=.36, p < .05). Aggression, especially in Model 3 of the contemporary model (see Table 4.9) increases the effect of the subject’s sex (IRR=3.52, p < .001), of being Black (IRR=2.91, p < .05) or of an other race (IRR=6.78, p < .001), and increases not living with both biological parents at Wave 3 to significance (IRR=2.05, p < .05) on the young adult’s rate ratio of 127 offending at Wave 3. This effect of aggression is also seen in the combined model (see Table 4.10), again especially in Model 3 on the same four factors as well as family size (IRR=3.64, p < .05), which is the only time this factor is significant in all of the models. In contrast, however, living below the poverty level at Wave 3 (see Table 4.10) is significant only when PC ETV is in Model 4A (IRR=.37, p < .05). In the Wave 2 models (see Table 4.11) the effect of the subject’s sex on their rate ratio of community violent offending is most significant when both aggression and/or subject ETV are in the model as is the subject being from a chemically—dependent household at Wave 1. However, as seen in the combined model (see Table 4.10), being in a household under the poverty level is only significant in Model 3 (IRR=2.22, p < .05) and supervision is only significant (IRR=.89, p < .05) in Model 4B with subject ETV also in the model. In the Wave 1 models (see Table 4.12) the effect of the subject’s sex on their rate ratio of community violent offending at Wave 3 is largest (IRR=3.24, p < .01) in Model 3 when aggression is in the model and smallest and non-significant when PC ETV is in Model 4A. The effect associated with the subject being Black on the young adult’s rate ratio of offending is significant only in Model 3 (IRR=3.02, p < .05) and for subjects of other race is significant only when aggression or any ETV are in the model, and largest and most significant with aggression (e.g., full model, IRR=7.16, p < .001). In contrast, the ETV predictors created from adding the reports from all three waves as used in Model 4A of the contemporary model (see Table 4.9) reduces the effect of demographic variables both in expected magnitude as well as significance, including eliminating any significance to the race variables, reduces the effect of being in a chemically-dependent household and not living with both biological parents, the latter to non-significance, but 128 did not change the effect of living below the poverty level, which remains significant (IRR=.38, p < .05). Regarding their effect on each other, in the cross-sectional model (see Table 4.9) subject reports of experiencing (IRR=3.42, p < .01) or witnessing violence (IRR=2.33, p < .001) in the community is reduced by aggression in its effect on the rate ratio of violent offending in the community at Wave 3. The same is seen in the contemporary model for each; yet witnessing violence in the home, increases (more negative) and becomes significant (IRR=.64, p < .05) in Model 48, with aggression in the model. In the combined, Wave 2, and Wave 1 models, aggression had a similar effect of reducing the magnitude, but not significance, of witnessing violence in the community 40% (Table 4.10, IRR=2.34, p < .001), 20% (Table 4.11, IRR=4.79, p < .001), and 27% (Table 4.12, IRR=1.70, p < .001) on the young adult’s rate ratio of community violent offending at Wave 3. In the same combined, Wave 2, and Wave 1 models, the PC report of child- directed aggression is positive and significant, yet maintains significance only in the Wave 1 (IRR=1.88, p < .01) model after aggression is also in the full model. As seen with the IPV offending models, aggression appears to influence negative ETV relationships to become more negative relative to the rate ratio of violent community offending at Wave 3, but reduces the effect of positive ETV relationships. Aggression as a factor in the rate ratio of the young adult’s violent offending in the community at Wave 3 is much reduced in all models when subject and/or PC ETV factors are included in the models. The reduction ranged from around 80% in the contemporary model (see Table 4.9) to 30% in the Wave 1 model (see Table 4.12). In the contemporary model, aggression and witnessing in the community are almost identical in 129 their individual effects on the rate ratio of violent offending in the community at Wave 3 (IRR=3.03, p < .05 and IRR=3.02, p < .001, respectively); in the Wave 1 full model, aggression is 2.8 times the effect of witnessing in the community, and in the combined full model (see Table 4.10), the difference is over 3 times higher for aggression (IRR=7.44, p < .001). However, at Wave 2, when subject reports of witnessing violence in the community is largest (see Table 4.3), the effect of witnessing in the community is 36% higher (see Table 4.11, IRR=4.79, p < .001) than aggression as a factor in the rate ratio of young adult violent offending in the community at Wave 3. Conclusion Based on the research reviewed for this project, this project is the first to analyze the effect of multiple forms of adolescent exposure to violence on violent offending within one analysis model, incorporating ETV from within their homes and communities, witnessed and experienced, as well as over time. The challenges of multicollinearity were sufficiently addressed. In addition, the project investigated the independent and combined effects of ETV and subject aggression on violent offending using longitudinal, prospective data collected from multiple informants from an ethnically diverse urban sample of both male and female adolescents from the general population rather than a school-based sample from which delinquent adolescents may be absent. Results provided mixed support for social learning theory as an explanation of future violent offending and so the main hypothesis was only partially supported. The negative and significant relationship of ETV in the home on the rate ratio of IPV offending at Wave 3 across the earlier models was opposite to expectations based on SLT and indicates a legacy from ETV in the home predictive of less future relationship 130 violence. However, SLT receives support from the positive and significant ETV in the community measures but indicates that relationship violence may be learned from models outside the home (Williams, 1988). Yet, ASBT is always a larger, positive, and significant factor for IPV offending than for any ETV measure in any time period. Nonetheless, all of the full models which incorporated both ASBT and ETV better fit the data than ASBT as a sole factor of either type of offending. Regarding the rate ratio of violent offending in the community at Wave 3, both SLT and ASBT theory are supported, with ETV even indicated as a larger, positive factor than ASBT in Wave 2 and an equal factor in the contemporary model which incorporates all subject ETV reports. Overall, aggression (i.e., ASBT) appears a larger factor in the rate ratio of young adult IPV offending at Wave 3 than for their community violent offending, for which SLT appears as comparable an explanation as ASBT theory. Similar to IPV offending, both home and community environments show significant effects on violent community offending. In addition, both models show that ETV in adolescence can increase or decrease risk of committing violent offenses years later; that current ETV in the community and violent offending coincide; and that the effect of ETV on violent offending can fade over time. Still, the interplay of exposure to violence and aggression on the rate ratio of either type of young adult offending at Wave 3 cannot be ignored; controlling for one factor reduces the positive effect of the other. In other words, future research that investigates correlates of offending needs to control for both the individual’s aggressive behavior and their ETV or the model will risk being misspecified. Finally, reports from the perspective of the child are needed; however, even with adolescents as seen with these data, the PCs reported more violence in the home than did 131 their child. Results show that the PC reports of child maltreatment at Wave 1 are significant to their violent offending in the community at Wave 3, but this finding would not have been measured if the data obtained child reports alone as they reported less home violence than their PC. Whether a reporting issue or a measurement issue, these data underscore the importance of obtaining data from both the caregiver and the child. Additionally, the effect of community factors indicate that the next step would be to add neighborhood—level factors into the model, particularly to investigate the differences by race. Limitations These data were collected only in Chicago neighborhoods and so the results are not generalizable to other populations. In addition, since over 90% of the primary caregivers were female, sex-specific modeling of learned behavior to the adolescent’s future behavior cannot be examined. The analysis assumed that the PC reports of PC-partner violence represented in- home exposure to violence for the adolescent. The ETV reports by the adolescents of exposure to violence in the home, however, were lower than the reports by the PC to the CTSP. Similarly, the My E T V reports by the adolescents of experiencing violence in the home were lower than child-directed aggression reported by the PC to the CTSS. These discrepancies may have resulted from obtaining the data from the informants using different instruments, although differences by multiple informants to similar measures are known (Kuo et al., 2000; Stemberg et al., 2006). Additionally, the adolescent’s time away from the home during their mid- and later teen years may have affected their opportunity to witness intimate partner violence between the PC and partner. 132 While the CTSP provides symmetry in measurement by recording the respondent and partner actions separately, intentions behind the reported acts cannot be assessed — “in the absence of context. . .a victim’s behaviors can look just like a batterer’s behaviors” (Conte & Savage, 2003). Physical self-protection (e. g., “fighting back”) was reported by a higher proportion of women victimized by an intimate than by a stranger, 40% vs. 20% (BJ S, 1994). Therefore, the difference in the IPV offending reported by the female young adults at Wave 3 compared to their male counterparts does not account for the context of the situation in which the violence reported had occurred. Future research will need to expand the investigation of the impacts of adolescent exposure to violence, especially in the settings which showed significant relationships in these analyses. The PHDCN data provide many other variables that research indicates may also be factors that can moderate or mediate any impacts. These include: 1) adding more risk and protective factors at the individual-level, such as internalizing or externalizing responses (Spilsbury, Kahana, Drotar, Creeden, Flannery, & Friedman, 2008) or self-concept (Youngstrom et al., 2003); social supports at the family level, such as close family relationships, parenting practices, and family beliefs (Youngstrom et al., 2003; Gorman-Smith et al., 2001) and parental attachments (Emery, 2006); and community levels (Patchin et al., 2006; Sampson et al., 2005; Loeber & Hay, 1997); as well as peer influences (Patchin et al., 2006; Nofziger & Kurtz, 2005; Ehrensaft et al., 2003; Salzinger et al., 2002; Williams, 1988; Loeber & Hay, 1997); 2) considering frequency as well as incidence and severity (Fagan & Browne, 1994); 3) investigating exposure to traumatic non-violent events as well as to violent events (Buka et al., 2000) as well as the relationship of verbal and psychological abuse separately or in conjunction 133 with violent behaviors (Gorman-Smith et al., 2001); and, 4) studying the role of resilience, immunities, and other factors that “interrupt aggressive, coercive interaction patterns” (Margolin et al., 2003, p. 437; Stemberg et al., 2006; Fergus & Zimmerman, 2005; Edleson, 1999; Fagan & Wexler, 1987). 134 APPENDIX 135 :5. W 91.... ._o. W 9.2.. .wo. W a... 8. 8. 8. 8.- 1. 8m. 1. 8. 58.8288: .. 2. 8. 8. .. 2. ...... S. 1. cm. .558 5 .53 .2 8.- 8.- 8.- 8.- 2. r 2. 858 5 .53 .8 8. S. 8.- 8. r. a. .. 3. .558 5 gm. .8 8. 2. 8.- 8.- _. 2. 8. 858 5 .3 a 1 om. .. 2. .. 2.- 8. 8. 58 58552.8 8.- 8. 8.- 8. 888 328m .m t. 2.- 8. 8. 8.88.5858 .. :. r. 2.. 2848 .3. 8. to .558 .55 .N 888% >8 ._ e m. 4 m N _ m-o—vc—Z —a:¢38mummchu kc.— xmhua—Z EGSEOLHQU .N.< 03-:- _8. w 9...... .8. w 5...... .8. w a... 8. 8. 8. 8.- i. mm. 3.. 8. 58.3288: .2. 8. 8. ..:. :8. :2. 5585538. 2. 8. S. 8.- .. :. 88. ”58553.8 8. 8. 8. 8. t. on 8. .558 5 Eu .8 8. 8. 8. .2. 3.- 8. r. 2. 8585.868 :8. *2. ..N_.- 8. 8. 58253828 8.- 8. 8.- 8. 888 3288 1. 2.- 8. 8. 55.88.5808 ..:. 2.. M2.- 2828 .1. mm. to .558 .55 .N 868% >8 ._ o m 4 m N _ £252 banana—850 .8.— anE cause-:90 .—.< «3:. 136 :8. w 5......- .8. w 8...... .8. w a... ...... 8. ...... 2. ..- 8. I 8. ...... 8. 8. 8.- ...... m..- 8. 8. 8. 8.- r. 8. 2.. 8. 58.82882 _. 2. ...... 8. *3. t 2. 8. 8.- o..- ..- t. 8.- 2. 8. ...... 8. .. E. .5585 53 .3 r. 8. .1. 8. ...... 8. 8. 8.- 8.- .. 2. 8. 8. 8.- .. :. 8. 058 5 53 .2 .2. 8. .. 2. 8. 8.- .. 2.- 8. 8. E. 8. ...... 8. 8. .558 5 Sm .2 I S. 8.- 8.- 8. 8. 8. 8. .. 2.- 8. 8. 858 5 .98.: r. 3. 8.- 8.- ..... cm. 8. .. :. 8. ...... a. 8. 8358-5592 8.- 8.- 8. 8. 2. .. 2.- 8. 8. >.=858988 8. ...... a; 8. 8.- 8.- 8. 8. ”51:58.8 8.- 8. 8. 8. 8.- 8.- 555388.» r. om. .. 2. ._. 2.- 8. 8. 58 58582 .8 8.- 8. 8.- 8. 888 3588 r. a; 8. 8. 8.88.5258 .. :. r. 2.- 2828 r. 8. 8.558 .35 .N 88.80 >2 ._ E 2 2 : 2 a w h o n v m N _ 20592 8:55ng .3.— xmhua—Z cams—0.950 .M.< 039—. 137 88. w 8.1.... .8. w 8...... .8. w 8... r. 2. 8. 8.- 8.- 2. 8.- 8. 8. 1. 8. ...... 8. .58 .2286. .2 2.. 2. .. 2. 8. 8.- .. :. 8. 8. .. :. r. 8. .. 2. .5585 53 .2 8. 8.- 2.- 8.- .. E. 8. 8. .. 2.- 8. .. 2. 858 5 53 .2 1. 8. 8. 8.- 8.- 8. 8.- 8. 8. a. 8. 8. .5585 .88 .2 .. 2. 8.- .. 8.- 8. 8. 8. 8. .. 2.- 8. 8. 858 5 .88 .: 2. 8.- 8.- ..... 2. 8.- .. 2. 8. _. 2. 8.- 253 58-550 .2 .. 8. 8.- 8. 8. 8. 8. 8. 8. 2858-28 8.- r. 2.- 8. 8.- 8.- 8.- 8. 81:58.8 8.- 8.- 8. 8. 8.- 8.- 5553.88 8 ...... 2. :. .. :.- 8. 8. .58 5&3 82 8 .. :.- 8.- 8. 8. 8883288 1. 2.- 8. 8. 8.88.5894 .. :. 1.2.- 2828 r. 8. 8 .558 .85 8 8888 >2 ._ 3 o w h o n v m N _ £0.52 N 93>? .5. 5.582 noun—9.30 .v.< «Ea-r 138 :5. W at; ._O. W PI. 60. W a... 8. 8.- ...... 8. 1. 2. 58.5588: 8. 8. 1. 2. . 2. .558 5 5B .2 8.- 8. 8. 8.- 0552 5 53 .2 ... 2. 8.- . 2. 8. .5585 .55 .: .. :. 8. 2 8. 8. 2535-2502 8. .. E.- 8.- 2. 22858-2 .8 8.- 8. 8.- 8. 0% £58 .8 8.- 8.- 8.- 8. 555585 .5 8. 8.- 8. 8. 58 55352 .8 8.- 8. 8.- 8.- 828 320m .8 t. 2.- 8. 8. .5. .55 .526 .v .. :. 2.. 2.- 2528 .1. 8. to .558 .55 .N 858% >2 ._ v m N _ 235—2 — Era; kc.— xmhawz Ecmua—GHHOU .m.< flash. 139 REFERENCES 140 REFERENCES Alcéntara, C., & Gone, J. P. (2007, May). Reviewing suicide in Native American communities: Situating risk and protective factors within a transactional- ecological framework. Death Studies, 31(5), 457—477. Anderson, C. A., & Bushman, B. J. (2002). Human aggression. Annual Review of Psychology, 53, 27-51. Annotated Stata Output: Negative Binomial Regression. (2008). Retrieved 14/04/2008, from UCLA Academic Technology Services. Statistical Consulting Group: http://www.ats.ucla.edu/stat/Stata/output/stata_nbreg_output.htm. Babcock, J. C., Canady, B. E., Senior, A., & Eckhardt, C. I. (2005, Apr). Applying the transtheoretical model to female and male perpetrators of intimate partner violencezGender differences in stages and processes of change. Violence and Victims, 20(2), 235-250. Bandura, A., Ross, D., & Ross, S. A. (1961). Transmission of aggression through imitation of aggressive models. Journal of A bnormal and Social Psychology, 68(3), 575-582. Banyard, V. L., Cross, C., & Modecki, K. L. (2006, Oct). Interpersonal violence in adolescence: Ecological correlates of self-reported perpetration. Journal of Interpersonal Violence, 21(10), 1314-1332. Blumstein, A., Cohen, J ., & Farrington, D. P. (1988). Criminal career research: Its value for criminology. Criminology, 26(1), 1-35. Bradshaw, C. P. (2004). Social-cognitive mediators of the link between social- environmental risk factors and aggression in adolescence. Washington, DC: US. Department of Justice, Office of Justice Programs, National Institute of Justice (25). Buka, S. L., Stichick, T. L., Birdthistle, I., & Earls, F. J. (2000, ll/May). The epidemiology of witnessing community violence in childhood and adolescence [Urban Seminar Series on Children's Health and Safety]. In T. R. W. J. F. Jim Knickman (Chair), Session 11: Family and Community Level Predictors of Youth Violence. Youth Violence in Urban Communities, The Robert Wood Johnson Foundation, Cambridge, MA. Buka, S. L., Stichick, T. L., Birdthistle, I., & Earls, F. J. (2001, Jul). Youth exposure to violence: Prevalence, risks, and consequences. American Journal of Orthopsychiatry, 71(3), 298-310. 141 Bureau of Justice Statistics. (1994). Violence between intimates (Bureau of Justice Statistics Selected Findings No. NCJ 149259). Washington, DC: US. Department of Justice. Cappell, C., & Heiner, R. B. (1990). The intergenerational transmission of family aggression. Journal of Family Violence, 5(2), 135-152. Carr, J. L., & VanDeusen Karen M. (2002, Jun). The relationship between family of origin violence and dating violence in college men. Journal of Interpersonal Violence, 17(6), 630-646. Chen, P.-H., & White, H. R. (2004, Nov). Gender differences in adolescent and young adult predictors of later intimate partner violence: A prospective study. Violence Against Women, 10(11), 1283-1301. Conte, J. R., & Savage, S. B. (2003, Apr). Concluding Observations. Journal of Interpersonal Violence, 18(4), 452-468. Crockett, L. J ., & Randall, B. A. (2006). Linking adolescent family and peer relationships to the quality of young adult relationships: The mediating role of conflict tactics. Journal of Social and Personal Relationships, 23(5), 761-780. Dodge, K. A., Bates, J. E., & Petit, G. (1990, Dec). Mechanisms in the cycle of violence. Science, 250, 1678-1683. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2006b). Project on human development in Chicago neighborhoods (PHDCN): Caregiver-subject conflict scale, wave 2, 1997-2000 [MRDF]. ICPSR13612-v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (20070). Project on human development in Chicago neighborhoods (PHDCN): Conflict tactics scale for parent and child, wave 1, [994-1997 [MRDF]. ICPSRl3584-v2. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. 1., Brooks—GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2007a). Project on human development in Chicago neighborhoods (PHDCN): Conflict tactics scale for partner and spouse, wave 1, 1 994-1997 [MRDF]. ICPSRl3583-v2. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. 142 Earls, F. J., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2007b). Project on human development in Chicago neighborhoods (PHDCN): Conflict tactics scale for partner and spouse, wave 3, 2000—2002 [MRDF]. ICPSRl3688-vl. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2006d). Project on human development in Chicago neighborhoods (PHDCN): Demographic file, wave 2, 1997-2000 [MRDF]. ICPSRl3609-v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2006c). Project on human development in Chicago neighborhoods (PHDCN): Demographic file, wave 3, 2000-2002 [MRDF]. ICPSRl3669-vl. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2007d). Project on human development in Chicago neighborhoods (PHDCN): Exposure to violence (subject), wave 1, [994—1997 [MRDF]. ICPSRl3589-v2. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (20070. Project on human development in Chicago neighborhoods (PHDCN): Family mental health and legal history, wave 1, 1994-1997 [MRDF]. ICPSRl359l-v2. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (20051). Project on human development in Chicago neighborhoods (PHDCN): Home observation for measurement of the environment, wave 1, [994-1997 [MRDF]. ICPSRl3594- v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2005g). Project on human development in Chicago neighborhoods (PHDCN): Home observation for measurement of the environment, wave 2, 1997-2000 [MRDF]. ICPSR13630- v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. 143 Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2006i). Project on human development in Chicago neighborhoods (PHDCN): Household composition, wave 2, 1997-2000 [MRDF]. ICPSRl3628-vl. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2007g). Project on human development in Chicago neighborhoods (PHDCN): Household composition, wave 3, 2000-2002 [MRDF]. ICPSRl37l4-vl. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2005d). Project on human development in Chicago neighborhoods (PHDCN): Master file, wave 1, [994-1997 [MRDF]. ICPSRl3580-v3. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2005e). Project on human development in Chicago neighborhoods (PHDCN): Master file, wave 2, 1997-2000 [MRDF]. ICPSR13608-vl. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2006h). Project on human development in Chicago neighborhoods (PHDCN): Master file, wave 3, 2000-2002 [MRDF]. ICPSRl3668-vl. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2005a). Project on human development in Chicago neighborhoods (PHDCN): My exposure to violence (subject), wave 2, 1997-2000 [MRDF]. ICPSRl36l7-v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2007c). Project on human development in Chicago neighborhoods (PHDCN): My exposure to violence (subject and young adult), wave 3, 2000—2002 [MRDF]. ICPSR13697- v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. 144 Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2006a). Project on human development in Chicago neighborhoods (PHDCN): Physical abuse scale, wave 2, 1997-2000 [MRDF]. ICPSR13642-vl. Ann Arbor, MI: Inter- university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2005b). Project on human development in Chicago neighborhoods (PHDCN): Self report of offending, wave 1, [994-1997 [MRDF]. ICPSRl3601-v1. Ann Arbor, MI: Inter- university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2005c). Project on human development in Chicago neighborhoods (PHDCN): Self report of Offending, wave 2, 1997-2000 [MRDF]. ICPSRl3658-vl. Ann Arbor, MI: Inter- university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2006c). Project on human development in Chicago neighborhoods (PHDCN): Self report of offending, wave 3, 2000—2002 [MRDF]. ICPSR13742-v1. Ann Arbor, MI: Inter- university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2006g). Project on human development in Chicago neighborhoods (PHDCN): Young adult self report, wave 2, 1997-2000 [MRDF]. ICPSR13666-vl. Ann Arbor, MI: Inter- university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Earls, F. J ., Brooks-GunnJeanne, Raudenbush, S. W., & Sampson, R. J. (2007h). Project on human development in Chicago neighborhoods (PHDCN): Youth self report, wave 1, [994-1997 [MRDF]. ICPSR13607-v2. Ann Arbor, MI: Inter-university Consortium for Political and Social Research (producer). Boston, MA: Harvard Medical School. Edleson, J. L. (1999, Aug). Children's witnessng of adult domestic violence. Journal of Interpersonal Violence, 14(8), 839-870. Edleson, J. L., Ellerton, A. L., Seagren, E. A., Kirchberg, S. L., Schmidt, S. 0., & Ambrose, A. T. (2007). Assessing child exposure to adult domestic violence. Children and Youth Services Review, 29, 961-971. Ehrensaft, M. K., Cohen, R, Brown, J., Smailes, B, Chen, H., & Johnson, J. G. (2003). Intergenerational transmission of partner violence: A 20-year prospective study. Journal of Consulting and Clinical Psychology, 71(4), 741-753. 145 Emery, C. R. (2006). Consequences of Childhood Exposure to Intimate Partner Violence (Final Technical Report No. NCJ 215347). Washington, DC: US. Department of Justice. National Institute of Justice. Fagan, J. A., & Browne, A. (1994). Violence against spouses and intimates: Physical aggression between women and men in intimate relationships. In A. J. Reiss Jr. & J. A. Roth (Eds), Social Influence. Vol. 3: Understanding and Preventing Violence (pp. 115-292) [Panel on the Understanding and Control of Violent Behavior, Committee on Law and Justice, Commission on Behavioral and Social Science and Education, National Research Council]. Washington, DC: National Academy Press. Fagan, J. A., & Wexler, S. (1987). Crime in the home and crime in the streets: The relationship between family violence and stranger crime. Violence and Victims, 2(1), 5-21. Fantuzzo, J. W., & Fusco, R. A. (2007). Children's direct exposure to types of domestic violence crime: A population-based investigation. Journal of Family Violence, 22(7), 543—552. Fergus, S., & Zimmerman, M. A. (2005). Adolescent resilience: A framework for understanding healthy development in the face of risk. Annual Review of Public Health, 26, 399-419. Finkelhor, D., Ormrod, R., Turner, H., & Hamby, S. (2005, Feb). The victimization of children and youth: A comprehensive, national survey. Child Maltreatment, 10(1), 5-25. Foshee, V. A., Bauman, K. E., & Linder, F. (1999, May). Family violence and the perpetration of adolescent dating violence: Examining social learning and social control processes. Journal of Marriage and the Family, 61, 331-349. Gelles, R. J. (1985). Family violence. Annual Review of Sociology, 11, 347-367. Gorman-Smith, D., Tolan, P. H., Sheidow, A. J., & Henry, D. B. (2001, Sep). Partner violence and street violence among urban adolescents: Do the same family factors relate? Journal of Research on Adolescence, 11(3), 273-295. Hamby, S. L. (2005, Jun). Measuring gender differences in partner violence: Implications from research on other forms of violent and socially undesirable behavior. Sex Roles, 52(1 1/ 12), 725-742. Hanson, R. F., Self-Brown, S., Fricker-Elhai, A. E., Kilpatrick, D. G., Saunders, B. E., & Resnick, H. S. (2006, Feb). The relations between family environment and violence exposure among youth: Findings from the national survey of adolescents. Child Maltreatment, 11(1), 3-15. l46 Heise, L. L. (1998, Jun). Violence against women: An integrated, ecological framework. Violence Against Women, 4(3), 262-290. Hendy, H. M., Weiner, K., Bakerofskie, J ., Eggen, D., Gustitus, C., & McLeod, K. C. (2003, Jun). Comparison of six models for violent romantic relationships in college men and women. Journal of Interpersonal Violence, 18(6), 645-665. Heyman, R. E., & Slep, A. M. S. (2002, Nov). Do child abuse and interparental violence lead to adulthood family violence? Journal of Marriage and the Family, 64(4), 864-870. Hirschi, T., & Gottfredson, M. (1983). Age and the explanation of crime. American Journal of Sociology, 89(3), 552-584. Holtzworth-Munroe, A. (2005, Apr). Female perpetration of physical aggression against an intimate partner: A controversial new topic of study. Violence and Victims, 20(2), 251-259. Huesmann, L. R., Eron, L. D., Lefl(owitz, M. M., & Walder, L. O. (1984b). The stability of aggression over time and generation. Developmental Psychology, 20(6), 1120- 1134. Kalmuss, D. (1984). The intergenerational transmission of marital aggression. Journal of Marriage and the Family, 46(1), 11-19. Minneapolis. Kelleher, K., Gardner, W., Coben, J ., Barth, R., Edleson, J ., & Hazen, A. (2006). C0- occurring intimate partner violence and child maltreatment: Local policies/practices and relationships to child placement, family services and residence (Final Report Submitted to the National Institute of Justice for Grant No. 2002-WG-BX-0014). Washington, DC: US. Department of Justice. National Institute of Justice (31). Kingsfogel, K. M., & Grych, J. H. (2004). Interparental conflict and adolescent dating relationships: Integrating cognitive, emotional, and peer influences. Journal of Family Psychology, 18(3), 505-515. Kitzmann, K. M., Gaylord, N. K., Holt, A. R., & Kenny, E. D. (2003). Child witness to domestic violence: A meta-analytic review. Journal of Consulting and Clinical Psychology, 71(2), 339-352. Knutson, J. F., DeGarmo, D., Koeppl, G., & Reid, J. B. (2005, May). Care neglect, supervisory, neglect, and harsh parenting in the development of children's aggression: A replication and extension. Child Maltreatment, 10(2), 92-107. Knutson, J. F., DeGarmo, D. S., & Reid, 1. B. (2004). Social disadvantage and neglectful parenting as precursors to the development of antisocial and aggressive child behavior: Testing a theoretical model. Aggressive Behavior, 30, 187-205. 147 Kuo, M., Mohler, B., Raudenbush, S. L., & Earls, F. J. (2000). Assessing exposure to violence using multiple informants: Application of hierarchical linear model. Journal of Child Psychology and Psychiatry, 41(8), 1049-1056. Kwong, M. J ., Bartholomew, K., Henderson, A. J ., & Trinke, S. J. (2003). The intergenerational transmission of relationship violence. Journal of Family Psychology, 1 7(3), 288-301. Land, K. C. (2000). Influence of neighborhood, peer, and family context: Trajectories of delinquent/criminal offending across the life course (National Institute of Justice Final Report). Washington, DC: US. Department of Justice. National Institute of Justice. Langhinrichsen-Rohling, J. (2005, Jan). Top 10 greatest "hits": Important findings and future directions for intimate partner violence research. Journal of Interpersonal Violence, 20(1), 108-118. Lewis, D. O., Lovely, R., Yeager, C., & Femina, D. D. (1989). Toward a theory of the genesis of violence: A follow-up study of delinquents. Journal of the American Academy of Child and Adolescent Psychiatry, 28(3), 431-436. Liberman, A. (2007). Adolescents, neighborhoods, and violence: Recent findings from the project on human development in Chicago neighborhoods (National Institute of Justice Research in Brief No. NCJ 217397). Washington, DC: US. Department of Justice. Little, L., & Kantor, G. K. (2002). Using ecological theory to understand intimate partner violence and child maltreatment. Journal of Community Health Nursing, 19(3), 133-145. Loeber, R., & Hay, D. (1997). Key issues in the development of aggression and violence from childhood to early adulthood. Annual Review of Psychology, 48, 371-410. Macmillan, R. (2001). Violence and the life course: The consequences of victimization for personal and social development. Annual Review of Sociologi, 27, 1-22. Maker, A. H., Kemmelmeier, M., & Peterson, C. (1998, Oct). Long-terrn psychological consequences in women of witnessing parental physical conflict and experiencing abuse in childhood. Journal of Interpersonal Violence, 13(5), 574-589. Malik, S., Sorenson, S. B., & Aneshensel, C. S. (1997, Nov). Community and dating violence among adolescents: Perpetration and victimization. The Journal of Adolescent Health, 21(5), 291-302. 148 Margolin, G., Gordis, E. B., Medina, A. M., & Oliver, P. H. (2003, Apr). The co- occurrence of husband-to-wife aggression, family-of-origin aggression, and child abuse potential in a community sample. Journal of Interpersonal Violence, 18(4), 413—440. Margolin, G., & Gordis, E. (2000). The effects of family and community violence on children. Annual Review of Psychology, 51, 445-479. Maxwell, C. D., & Maxwell, S. R. (2003, Dec). Experiencing and witnessing familial aggression and their relationship to aggressive behaviors of Filipino adolescents. Journal of Interpersonal Violence, 18(2), 1432-1451. McCloskey, L. A., & Lichter, E. L. (2003, Apr). The contribution of marital violence to adolescent aggression across different relationships. Journal of Interpersonal Violence, 18(4), 390-412. Mohr, W. K., & Tulman, L. J. (2000, Sep). Children exposed to violence: Measurement considerations within an ecological framework. Advances in Nursing Science, 23(1), 59-68. Molnar, B. E., Buka, S. L., Brennan, R. T., Holton, J. K., & Earls, F. (2003, May). A multilevel study of neighborhoods and parent-to-child physical aggression: Results from the Project on Human Development in Chicago Neighborhoods. Child Maltreatment, 8(2), 84-97. Moretti, M. M., Obsuth, 1., Odgers, C. L., & Reebye, P. (2006). Exposure to maternal vs. paternal partner violence, PTSD, and aggression in adolescent girls and boys. Aggressive Behavior, 32(4), 385—395. Murrell, A. R., Christoff, K. A., & Henning, K. R. (2007, Jul). Characteristics of domestic violence offenders: Associations with childhood exposure to violence. Journal of Family Violence, 22(7), 523-532. Nicholas, K. B., & Rasmussen, E. H. (2006, Jan). Childhood abusive and supportive experiences, inter-parental violence, and parental alcohol use: Prediction of young adult depressive symptoms and aggression. Journal of Family Violence, 21(1), 43-61. Nicholas, K., & Bieber, S. L. (1996). Parental abusive versus supportive behaviors and their relation to hostility and aggression in young adults. Child Abuse and Neglect, 20(12), 1995-1211. Nofziger, S., & Kurtz, D. (2005, Feb). Violent lives: A lifestyle model linking exposure to violence to juvenile violent offending. Journal of Research in Crime and Delinquency, 42(1), 3-26. 149 O'Keefe, M. (2005, Apr). Teen dating violence: A review of risk factors and prevention efforts (VAWnet Applied Research Forum). Retrieved 12112007, fi'om Harrisburg, PA: VAWnet, a project of the National Resource Center on Domestic Violence/Pennsylvania Coalition Against Domestic Violence: http://www.vawnet.org. Park, H. M. (2005, 15/Sep). Regression models for event count data using SAS, STA TA, and LIA/HDEP. Retrieved 18/04/2008, from UITS Center for Statistical and Mathematical Computing: http://www.indiana.edu/~statmath/stat/all/count/count.html. Patchin, J. W., Huebner, B. M., McCluskey, J. D., Varano, S. P., & Bynum, T. S. (2006, Apr). Exposure to community violence and childhood delinquency. Crime and Delinquency, 52(2), 307-332. Peacock, D., & Rothman, E. (2001, Nov). Working with young men who batter: Current strategies and new directions (VAWnet Applied Research Forum). Retrieved 12112007, from Harrisburg, PA: VAWnet, a project of the National Resource Center on Domestic Violence/Pennsylvania Coalition Against Domestic Violence: http://new.vawnet.org. Perry, A. R., & Fromuth, M. E. (2005, Sep). Courtship violence using couple data: Characteristics and perceptions. Journal of Interpersonal Violence, 20(9), 1078- 1095. Raudenbush, S. W., Johnson, C., & Sampson, R. J. (2003). A multivariate, multilevel Rasch model for self-reported criminal behavior. Sociological Methodology, 33(1),169-211. Salzinger, S., F eldman, R. S., Stockhammer, T., & Hood, J. (2002). An ecological framework for understanding risk for exposure to violence and the effects of exposure on children and adolescents. Aggression and Violent Behavior, 7, 423- 451. Sampson, R. J ., & Laub, J. H. (2005). A life—course view of the development of crime [Developmental criminology and its discontents: Trajectories of crime from childhood to old age] (R. J. Sampson, & J. H. Laub eds.) [Special issue]. Annals of the American Academy of Political and Social Science, 602, 12-46. Sage Publication. Sampson, R. J ., & Lauritsen, J. L. (1994). Violent victimization and offending: Individual-,situational, and community-level risk factors. In A. J. Reiss & J. A. Roth (Eds.), Understanding and preventing violence (Vol. 3, pp. 1-115). Washington, DC: National Academy of Science. 150 Sampson, R. J ., Morenoff, J. D., & Raudenbush, S. (2005). Social anatomy of racial and ethnic disparities in violence. American Journal of Public Health, 95(2), 224—32. Saunders, B. E. (2003). Understanding children exposed to violence: Toward an integration of overlapping fields. Journal of Interpersonal Violence, 18, 356-376. Schwartz, J. P., Hage, S. M., Bush, I., & Burns, L. K. (2006, Jul). Unhealthy parenting and potential mediators as contributing factors to future intimate violence: A review of the literature. Trauma, Violence, & Abuse, 7(3), 206-221. Selner-O'Hagan, M. B., Kindlon, D. J ., & Buka, S. L. (1998). Assessing exposure to violence in urban youth. Journal of Child Psychology and Psychiatry, 39(2), 215- 224. Simons, R. L., Lin, K.-H., & Gordon, L. C. (1998, May). Socialization in the family of origin and male dating violence: A prospective study. Journal of Marriage and the Family, 60(2), 467-78. Simons, R. L., Wu, C., Johnson, C., & Conger, R. D. (1995). A test of various perspective on the intergenerational transmission of domestic violence. Criminology, 33(1), 141-171. Spilsbury, J. C., Kahana, S., Drotar, D., Creeden, R., Flannery, D. J ., & Friedman, S. (2008). Profiles of behaviorial problems in children who witness domestic violence. Violence and Victims, 23(10), 3-17. Stemberg, K. J ., Lamb, M. E., Guterman, E., & Abbott, C. B. (2006). Effects of early and later family violence on children's behavior problems and depression: A longitudinal multi-informant perspective. Child Abuse and Neglect, 30, 283-306. Stith, S. M., Rosen, K. H., Middleton, K. A., Busch, A. L., Lundeberg, K., & Carlton, R. P. (2000). The intergenerational transmission of spouse abuse: A Meta-Analysis. Journal of Marriage and Family, 62(3), 640-54. Straus, M. A. (2004, Jul). Prevalence of violence against dating partners by male and female university students. Violence Against Women, 10(7), 790-811. Straus, M. A. (2006, 23/May). Dominance and symmetry in partner violence by male and female university students in 32 nations. Presented at the Trends in Intimate Violence Intervention, New York University: University of Haifa and New York University. Straus, M., Hamby, S., Boney-McCoy, S., & Sugarman, D. (1996). The revised Conflict Tactics Scales (CTSZ): Development and preliminary psychometric data. Journal of Family Issues, 1 7(3), 283-316. 151 Swan, S. C., & Snow, D. L. (2006, Nov). The development of a theory of women's use of violence in intimate relationships. Violence Against Women, 12(11), 1026-1045. Teisl, M., & Cicchetti, D. (2008). Physical abuse, cognitive and emotional processes, and aggressive/disruptive behavior problems. Social Development, 1 7(1), 1-23. Tolan, P., Gorman-Smith, D., & Henry, D. (2006). Family violence. Annual Review of Psychology, 57, 557-583. US. Census Bureau. (2006, 29/Aug). Poverty thresholds: 1995 . Retrieved 31/10/2008, from US. Census Bureau, Housing and Household Economic Statistics Division: http://www.census.gov/hhes/www/poverty/threshld/thresh95.htrnl. Vezina, J ., & Hebert, M. (2007, Aug). Risk factors for victimization in romantic relationships of young women: A review of empirical studies and implication for prevention. Trauma, Violence, & Abuse, 8(1), 33-66. Walsh, J. A., & Krienert, J. L. (2007, Jul). Child-parent violence: An empirical analysis of offender, victim, and event characteristics in a national sample of reported incidents. Journal of Family Violence, 22, 563-574. Wekerle, C., & Avgoustis, E. (2003). Child maltreatment, adolescent dating, and adolescent dating violence. In P. Florsheim (Ed.), Adolescent romantic relations and sexual behavior: Theory, research and practical implications (pp. 213-241). Hillsdale, NJ: Lawrence Erlbaum Associates. White, J. W., & Smith, P. H. (2004). A longitudinal perspective on physical and sexual intimate partner violence against women (From Violence Against Women and Family Violence: Developments in Research, Practice, and Policy, 2004, Bonnie Fisher, Ed. No. NCJ 199708). Washington, DC: US. Department of Justice. National Institute of Justice. Widom, C. S., & Maxfield, M. G. (2001). An update on the "cycle of violence " (National Institute of Justice Research in Brief No. NCJ 184894). Washington, DC: US. Department of Justice. National Institute of Justice. Widom, C. S., Schuck, A. M., & White, H. R. (2006, Dec). An examination of pathways from childhood victimization to violence: The role of early aggression and problematic alcohol use. Violence and Victims, 21(6), 675-690. Widom, C. S., & White, H. R. (1997). Problem behaviours in abused and neglected children grown up; prevalence and co-occurrence of substance abuse, crime and violence. Criminal Behavior and Mental Health, 7, 287-310. Williams, L. M. (2003). Understanding child abuse and violence against women: A life course perspective. Journal of Interpersonal Violence, 18, 441-451. 152 Williams, 0. (Presenter). (1988). Exposure to violence outside the family. 4th National Symposium on Doctoral Research in Social Work (p. 12). The Ohio State University College of Social Work. Wolak, J ., & Finkelhor, D. (1997, 05). Effects of partner violence on children (United States Air Force Domestic Violence Literature Review, Synthesis and Implications for Practice). Retrieved 021 107, from National Network for Family Resiliency: http://www.agnr.umd.edu/nnfr/research/pv/pv_ch4.html. Youngstrom, E., Weist, M. D., & Albus, K. E. (2003, Sep). Exploring violence exposure, stress, protective factors and behavioral problems among inner-city youth. American Journal of Community Psychology, 32(1/2), 115-129. 153 IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII lllllllllllllll|||lllllllllljllIllllllllllllllllllllllllllll 333333 5796