University of Southern California Dissertations and Theses

Aggression and victimization as predictors of drug use in early adolescence

(USC Thesis Other)

Aggression and victimization as predictors of drug use in early adolescence

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content AGGRESSION AND VICTIMIZATION AS PREDICTORS OF DRUG USE IN EARLY ADOLESCENCE by Michelle Diane Weiner A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (PREVENTIVE MEDICINE-HEALTH BEHAVIOR RESEARCH) May 2002 Copyright 2002 Michelle Diane Weiner Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3180785 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3180785 Copyright 2005 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. DEDICATION This dissertation is dedicated to my husband Scott with love. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ACKNOWLEDGMENTS It is humbling to reflect on the number of people who have helped me in completion of my dissertation and doctoral program. Truly, every friend, staff person and faculty member I have known during the past six years at IPR has contributed to this seemingly personal achievement. I would like to acknowledge the support, both in terms of infrastructure and personnel, which the entire Institute of Prevention Research under the leadership of Andy Johnson has provided. It has. been a great environment in which to spend these formative years. Additionally, the National Institute of Drug Abuse supported this research via grant no. 1 F31 DA06055-01 and I would like to acknowledge their assistance-this has been a great training experience which I hope adds to our understanding of drug use among preadolescents. Of course, there are some who have had more “hands on” experience in helping me reach this goal. I would like to thank them individually. First, I am indebted to my advisor and committee chair, Dr. Mary Ann Pentz. She has been a source of confidence, optimism, intelligence and grace throughout this process. She has taught me to trust theory and consider data in its context. Her leadership has taught me not to falter in the face of challenges and to expect a high level of performance from myself. She has pushed and encouraged me. She has been an inspiration to me. Dr. Chih-Ping Chou introduced me to structural equation modeling (SEM) in his course. He has continued to help me master the finer points of this analysis technique Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. as I have applied it in my dissertation research. With his patience and generosity with his time, I have become an SEM enthusiast-a tribute to his excellence as a teacher. Dr. Steve Sussman was responsible for two seminal moments in my doctoral experience. He taught the first course on Health Behavior Research I ever took, giving me an excellent introduction to this discipline— I still find myself referring to the outline he provided in that class. Also, it was under his mentorship that I published my first journal article. His insights have continued to inform my graduate school experience as a member of my dissertation committee. Dr. Penny Trickett has facilitated my development as a researcher in violence prevention though her leadership of the Interdisciplinary Violence Research Initiative. Being a member of this group has broadened my horizons on aggression and violence research and has given me fodder for future research. Additionally, Penny has traveled to foreign lands (the Alhambra Campus) to serve as the outside member of my dissertation committee for which I thank her. Dr. Luanne Rohrbach served on my qualifying exam committee and mentored me in a directed research and apracticum. Her advice in these experiences was valuable and appreciated. However, my gratitude to Luanne extends far beyond this. Looking back, she has helped me at every turn, providing guidance and advice which have calmed, informed and edified, not only when I have been her student but other times when I have needed advice. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. V Mamy Barovich has been a friend and confidante since my arrival. She has provided a sort of one-stop shopping for all questions I have had at USC whether they be related to classes, professors, financial support or USC bureaucracy. She has gone out of her way many times to help me. She has also proved an excellent source of restaurant and movie recommendations. Without her, graduate school would have not been half the fun. I also want to thank graduate school friends Silvana Skara, Cher McMonigle, Cheryl Nordstrom and Darleen Schuster. They have shared this experience with me, providing camaraderie and encouragement. Also, longtime friends Meera Dhanalal, Alecia Domer, Julie Hendin, Andhra Lutz and Julia Robeson have fielded many a desperate phone call and given me the will to write another sentence. Another USC friend, Chaoyang Li, has been both friend and trusty statistical and SAS programming consultant since my first year. Without his advice, I would probably still be working on my analyses. Gencie Turner, although she has long since left USC, was also a great teacher in my first years in SAS programming and data analysis. My sister Jill and brother Jason have provided comic relief, diversion and perspective throughout this experience for which I am grateful. In addition, Jason introduced me to LA, helped me find my apartment here, and bequeathed his Thomas Guide to me, helping me become a true Angeleno. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I would like to thank Pat and Bob Karmozyn for their support and interest in my education. During my Master’s program, their home was my “country house” away from the hubbub of Boston and they have continued to keep apprized of my progress. No acknowledgment would be complete without thanking my parents, Bob and Lois Weiner, for all they have given me throughout my entire education. As a scientist himself, my dad has instilled me a love of investigation and discovery. His support, enthusiasm and confidence in me were pivotal in my decision to return to school. My mom has imbued in me a love of reading (which came in handy in several classes) and has encouraged and supported me in all my educational efforts. Even in graduate school she is always quick to supply a new book or study aid if she thinks it will assist me in my schooling. Without them, I don’t know what I would be doing, but I feel certain it would not be getting a Ph.D. Finally, my greatest thanks go to my husband, Scott Karmozyn. Long before he was my husband, he was a critical influence in my decision to go to graduate school. Over the course of this experience which included a Master’s program at Boston University as well as the doctoral program at USC, he has worn many hats. He has been a proofreader, editor, data entry specialist, crisis manager, cheerleader, moving man, statistics tutor, audience to presentations, sous-chef and computer fix-it guy, to name a few. Most importantly, he has been my best friend. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. vii TABLE OF CONTENTS Dedication .................................................................................................................. ii Acknow]edgments................................................................................................... iii List of Tables........................................................................................................... ix List of Figures ............................................ xii Abstract ................................................................................................................. xiii Introduction................................................................................................................. 1 Relationship between Aggression and Violence .. . . ; . ...................................2 Relational Aggression vs. Overt Aggression...................................................3 Relationship between Drug Use and Aggression.............................................4 Gender Differences: Relationships among Aggression, Victimization and Drug U se ........................................................................ 6 Interaction of Type of Aggression and Gender on Risk for Drug Use............ 7 Gender Differences: Social Competence......................................................... 8 Aims of the Present S tudy............................................................................. 13 M ethods......................................................................................................................14 Overview...................................... 14 Overall Research and Measurement Design ................................................. 16 Study One: The Panel Study ......................................................................... 16 Procedure............................................................................................16 Human Subjects Protection ..................................................... 17 Sam ple.......................... 18 Description of Measures .............. 20 Specific Hypotheses .......... 24 Data Analysis P lan......................................................................... 25 Preliminary Measurement Model................................................ 28 Study Two: The Cross-Sectional Study......................................................... 30 Procedure............................................................................................30 Human Subjects’ Protection ........................................ 31 Sam ple........................ 31 Measures .............................................................. 35 Data Analysis P lan...................... 36 Preliminary Measurement Model.......................................... 37 Specific Hypotheses................ 39 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. viii R esults .....................................................................................................40 Study One: Panel Study ...........................................................................40 Descriptive Analyses......................................................................40 Correlations of Indicators.............................. ..................................43 Effect of Fourth-grade Aggression on Sixth-grade Cigarette U s e .........................................................................45 Effects of Fourth-grade Aggression on Sixth-grade Alcohol U s e ...........................................................................54 Effect of Fourth-grade Aggression on Sixth-grade Marijuana Use .......................................................................62 Effect of Fourth-grade Relational and Overt Aggression on Sixth-grade Cigarette U se .................................................70 Effect of Fourth-grade Relational and Overt Aggression on Sixth-grade Alcohol U se...................................................79 Effect of Fourth-grade Relational and Overt Aggression on Sixth-grade Marijuana U s e ...............................................88 Study Two: The Cross-sectional Study......................................................... 88 Descriptive Analyses......................................................................... 88 Correlations of Indicators.................................................................91 Effect of Aggression and Victimization on Lifetime Cigarette Use Among Sixth-Grade Students .............. 93 Effect of Aggression and Victimization on Lifetime Alcohol Use Among Sixth-Grade Students .......................103 Effect of Aggression and Victimization on Lifetime Marijuana Use Among Sixth-Grade Students.....................112 Discussion ............ 121 Study One: The Panel Study .......... 121 Overview .......................121 Limitations .......... 124 Study Two: The Cross-Sectional Study .............. 127 Limitations ................................................ 129 Overall Implications and Future Directions........................................ 131 Implications and Future Directions ...................................... 133 References ...................... 138 Appendix A. Psychometric Properties of Measures................ 147 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ix LIST OF TABLES Table 1. Comparison between subjects selected for longitudinal analytic sample with those excluded................. 20 Table 2. Characteristics of analytic sample in cross-sectional study........................ 32 Table 3. Correlation of indicators used in the longitudinal study by gender 42 Table 4. Estimates for factor loadings in model showing the effect of fourth-grade aggression on lifetime cigarette use in sixth grade................................46 Table 5. Model development and test of measurement invariance for the effect of fourth-grade aggression on lifetime cigarette use in sixth grade. ....... 48 Table 6. Indirect effect of fourth-grade aggression on lifetime cigarette use in sixth grade mediated by fifth-grade social competence................................... 50 Table 7. Estimates for factor loadings in model showing the effect of fourth-grade aggression on lifetime alcohol use in sixth grade. ...................... 55 Table 8. Model development and test of measurement invariance for the effect of fourth-grade aggression on lifetime alcohol use in sixth grade....................57 Table 9. Indirect effect of fourth-grade aggression on lifetime alcohol use in sixth grade mediated by fifth-grade social competence..........................................59 Table 10. Estimates for factor loadings in model showing the effect of fourth-grade aggression on lifetime marijuana use in sixth grade. .................... 63 Table 11. Model development and test of measurement invariance for the effect of fourth-grade aggression on lifetime marijuana use in sixth grade. ...... 65 Table 12. Indirect effect of fourth-grade aggression on lifetime marijuana use in sixth grade mediated by fifth-grade social competence............................. 67 Table 13. Estimates for factor loadings in model showing the effect of fourth- grade overt and relational aggression on lifetime cigarette use in sixth grade. .... 71 Table 14. Model development and test of measurement invariance for the effect of fourth-grade overt and relational aggression on sixth-grade cigarette use. 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. X LIST OF TABLES (CONTINUED) Table 15. Indirect effect of fourth-grade overt and relational aggression on lifetime cigarette use in sixth grade mediated by fifth-grade social competence. .. 75 Table 16. Estimates for factor loadings in model showing the effect of fourth- grade overt and relational aggression on lifetime alcohol use in sixth grade 80 Table 17. Model development and test of measurement invariance for the effect of fourth-grade overt and relational aggression on lifetime alcohol use in sixth grade...............................................................................................................82 Table 18. Indirect effect of fourth-grade overt and relational aggression on lifetime alcohol use in sixth grade mediated by fifth-grade social competence. . . . 84 Table 19. Correlation of indicators used in the cross-sectional study by gender. .. 90 Table 20. Estimates for factor loadings in model showing the effect of aggression and victimization on lifetime cigarette use among sixth-grade students.................................. 94 Table 21. Model development and test of measurement invariance for the effect of aggression and victimization on lifetime cigarette use among sixth- grade students......................... 97 Table 22. Indirect effect of aggression and victimization on lifetime cigarette use mediated by social competence among sixth-grade .students.............................. 99 Table 23. Estimates for factor loadings in model showing the effect of aggression and victimization on lifetime alcohol use among sixth-grade students....................................... 104 Table 24. Model development and test of measurement invariance for the effect of aggression and victimization on lifetime alcohol use among sixth- grade students. ............... 106 Table 25. Indirect effect of aggression and victimization on lifetime alcohol use mediated by social competence among sixth-grade students 108 Table 26. Estimates for factor loadings in model showing the effect of aggression and victimization on lifetime marijuana use among sixth-grade students. ........................................................................................... 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xi LIST OF TABLES (CONTINUED) Table 27. Model development and test of measurement invariance for the effect of aggression and victimization on lifetime marijuana use among sixth-grade students..................................................................................................115 Table 28. Indirect effect of aggression and victimization on lifetime marijuana use mediated by social competence among sixth-grade students 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xii LIST OF FIGURES Figure 1. Parameter estimates for final model of the effect of fourth-grade aggression on lifetime cigarette use in sixth grade ...................... 53 Figure 2. Parameter estimates for final model of the effect of fourth-grade aggression on lifetime alcohol use in sixth grade .....................................................61 Figure 3. Parameter estimates for final model of the effect of fourth-grade aggression on lifetime marijuana use in sixth grade .................................................69 Figure 4. Parameter estimates for final model of the effect of fourth-grade overt and relational aggression on sixth-grade lifetime cigarette use ...................... 78 Figure 5. Parameters estimates for final model of the effect of fourth-grade overt and relational aggression on sixth-grade lifetime alcohol use ........................ 87 Figure 6. Parameters estimates for final model of the effect of aggression and victimization on lifetime cigarette use among sixth-grade students.................102 Figure 7. Parameter estimates for final model of the effect of aggression and victimization on lifetime alcohol use among sixth-grade students .........................I ll Figure 8. Parameter estimates for final model of the effect of aggression and victimization on lifetime marijuana use among sixth-grade students.....................120 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xiii ABSTRACT A major risk for perpetration of violence is early evidence of aggressive behavior and this may also be an early risk factor for later drug use. A major protective factor is social competence which may protect against aggression as well as drug use. Furthermore, subtypes of aggression, namely relational and overt, may signal differential risk according to gender. The present longitudinal study (Study One) used structural equation modeling (SEM) to investigate the longitudinal relationship between early aggressive behavior and later drug use, mediated by social competence. Self-reported data gathered from 461 subjects annually from fourth to sixth grade were used to test the hypotheses that 1) early aggression would predict later drug use, 2) this relationship would be mediated by social competence, and 3) this relationship would be moderated by gender such that relational aggression would be a stronger drug use predictor for girls and overt aggression a stronger predictor for boys. Results indicated that aggression in fourth grade predicted lifetime cigarette and alcohol use, but not marijuana use, in sixth grade. Social competence did not mediate these relationships, and there was no interaction of gender and aggression subtype. Some research has suggested victimization is associated with aggression while there is little information on its relationship to drug use. Using self-reported data from 824 students surveyed in the Spring of sixth grade and SEM analysis, Study Two tested the hypothesis that higher victimization and aggression independently predict increased Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. xiv drug use and investigated social competence as a mediator and gender as a moderator of these relationships. Results supported aggression as a predictor of lifetime use of cigarettes, alcohol and marijuana, while victimization was inversely associated with use of these substances. Relationships were not mediated by social competence or moderated by gender, as had been hypothesized. Results of these studies argue for universal prevention of aggression as a way of reducing risk for later drug use. They also suggest that victimization may have a unique relationship with drug use that demands further investigation. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 INTRODUCTION Substance use and violence pose serious health and social problems for Americans (Belcher & Shinitzky, 1998; Dahlberg, 1998; Sells & Blum, 1996). In a recent year, over 8 million incidents of violent crime were perpetrated on Americans and more than 15,000 people were murdered. Youth in particular are affected by violence, which is among the leading causes of death among people under 18 (U.S. Department of Justice, 2000). Drug use is also a widespread problem with approximately 13.6 million Americans current users of illicit drugs in a recent year (Dept, of Health and Human Services, 1998). Serious consequences of drug use may include health problems, injury and psychological maladjustment, as well as associated economic costs (Ammerman, Ott, Tarter, & Blackson, 1999). Drug use costs Americans over $2.4 billion annually while the estimated annual cost for people injured in violent criminal acts is over $17 billion (National Institute on Drug Abuse, 1998). Effective prevention efforts, especially with youth, are one promising avenue for reducing the occurrence of these behaviors and mitigating their enormous social consequences. Although the causes ofboth problems are uncertain, substance use and violence are correlated, suggesting they may share a common root (Dukann, Byrd, Auinger, & Weitzman, 1996). Theoretical explanations accounting for this co-occurrence include Jessor’ s problem behavior theory, which suggests that drug use and delinquent or violent behavior in adolescence may represent a proclivity toward unconventionality and/or attempts to achieve a more adult status (Donovan & Jessor, 1985; Jessor, 1984). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2 Delinquency and drug use during late childhood and early adolescence may also reflect the influence of antisocial risk factors, in particular interaction with and bonding to antisocial peers as posited in social development theory (Catalano & Hawkins, 1996). By studying the common predictors and consequences of drug use and violence, it may be possible to identify targets for interventions and inform future prevention efforts. Relationship between Aggression and Violence Aggressive behavior in childhood is among the strongest predictors of later violent behavior (Dahlberg, Toal, & Behrens, 1998). Longitudinal research has consistently shown that aggression in childhood predicts later delinquent or violent behavior (Huesmann, Eron, Lefkowitz, & Walder, 1984; Kellam, 1990; Tremblay et al., 1992). Similarly, other research has demonstrated that aggression is a relatively stable characteristic from as early as first grade to adulthood, with Olweus (1979) reporting a correlation over time of approximately .63. Other research suggests that although early aggression is associated with later aggression, periods of desistance from aggression also occur, with late childhood possibly being one of these periods (Loeber & Stouthamer- Loeber, 1998). Thus, the natural decline in aggression during late childhood may indicate an opportune time to intervene on aggression, as this may be a period when it is vulnerable to influence. Either way, this literature suggests that early intervention with aggressive children and, in particular, in the late childhood years, maybe effective in prevention of violent behavior in adulthood. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3 Relational Aggression vs. Overt Aggression Although the existing literature has demonstrated that early aggression predicts later aggression, these findings are based on studies where aggression is operationalized as overt aggression (Hawkins, Von Cleve, & Catalano, 1991; Tremblay et al., 1992; White, Brick, & Hansell, 1993). Overt aggression is a type of aggression in which harm occurs through direct physical damage or threat of physical damage to another person and includes behaviors such as physical attacks and the threat of physical attacks (Crick, Casas, & Mosher, 1997). Another form, relational aggression, has been identified in the child development literature (Crick, 1996; Crick & Gropeter, 1995). Relational aggression involves inflicting harm through purposeful manipulation or damage to peer relationships, such as purposefully excluding a peer from social plans or telling lies about a peer (Crick, 1996; Galen & Underwood, 1997). Research by Crick, Bigbee and Howes (1996) on third- through sixth-grade students found that children report relationally aggressive behaviors as normative responses to anger and as behaviors they perceive to be as harmful, providing support that this type of behavior is indeed “aggressive.” Existing research suggests that relational aggression is related but distinct from the overt form. Moderate associations between the two types, for example a correlation of .54 as reported by Crick (1995), generally support this distinction. However, other studies reporting even stronger correlations (e.g., r = .77) suggest that, in some cases, the distinction between these two aggression subtypes may be minimal (Crick, 1996). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4 Existing studies do suggest that relationally aggressive children, like their overtly aggressive peers, are at increased risk for a host developmental problems, including peer rejection, depression and loneliness (Crick et al, 1997; Crick & Gropeter, 1995). Additionally, it seems that relational aggression maybe the preferred form of aggression among girls in late childhood and early adolescence, possibly because it attempts to damage one of the most valued commodities of this age-group, intimate peer relationships (Crick & Gropeter, 1995). Thus, without targeting relational aggression alongside overt aggression in prevention programs, it is possible that these harmful consequences of aggression will go unchecked, particularly for girls. Relationship between Drug Use and Aggression Research has consistently demonstrated that aggressive behavior and drug use are associated (Osgood, 1995). This evidence has come from a myriad of sources including correlational studies showing associations between aggression and drug use in general populations of adolescents and criminal offenders (Abram, 1989; Osgood, 1995), prospective studies of childhood aggression leading to adolescent substance use (Kellam, Brown, Rubin, & Ensminger, 1983; White, 1997), and laboratory studies drawing associations between use of some substances, such as alcohol, and short-term increased aggression (Lipsey, Wilson, Cohen, & Derzon, 1997). As mentioned earlier, one explanation for the co-occurrence of various deviant behaviors in adolescence, including drug use and aggression, is problem behavior theory (Donovan & Jessor, 1985; Jessor, 1984). However, as with the research linking early aggression to later Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 5 violence, the research tying drug use to aggression has operationalized aggression only as overt aggression. No research has examined the relationship between relational aggression and drug use. It is believed that aggressive behavior precedes initiation into drug use (White, 1997). Research that has evaluated the temporal relationship between aggression and drug use suggests that aggression is manifested first developmentally. Several studies have documented the emergence of aggressive behavior as early as elementary school in the form of conduct disorders or problems (Lynam, 1996). Other studies have shown that childhood aggression is predictive of drug use in adolescence and adulthood (Kellam, 1990; Lewis, Robins, & Rice, 1985). Moreover, studies have shown that the earlier the onset of aggressive behavior, the greater the risk for more frequent and more severe antisocial behavior late in adolescence and adulthood (Farrington et al, 1990; O'Donnell, Hawkins, & Abbott, 1995). It is important to note that existing studies have used only overt aggression, a style more typical of boys, to predict later drug use (e.g. (Clapper, Buka, Goldfield, Lipsitt, & Tsuang, 1995; White et al., 1993). An aim of this study was to extend the existing body of research by testing not only the temporal relationship of not only overt aggression to drug use but also relational aggression to drug use (Crick & Gropeter, 1995). This work is essential since past research may not have adequately captured the predictive relationship of aggression to drug use, particularly for girls. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 6 Gender Differences: Relationships among Aggression, Victimization and Drug Use Aggressive behavior involves not only the aggressor but also the victim at whom the aggressor’s behavior is aimed, although the literature is not clear on whether aggressors and victims represent separate or overlapping groups. Past research has suggested that a subgroup of victims may contribute to their victimization by aggressive or provocative behaviors (Olweus, 1978). However, at least one prospective study that investigated whether victims are also aggressors found that aggression and victimization were independent of one another (Perry, Kusel, & Perry, 1988). Research that partials out any effects of aggressive behavior is needed to understand the impact of victimization on other problem behaviors. Like aggression, relational and overt subtypes of victimization have been identified in the literature and may be gender-related. Overt victimization refers being a recipient of overt aggression (e.g., hitting, name calling) while being relationally victimized means being the target of damaging social practices, such as being purposely excluded from a peer group or being the subject of malicious gossip (Crick & Gropeter, 1996). Some research suggests that girls experience more relational victimization while boys experience more overt victimization. This is important in that it may explain why studies of victimization, which addressed overt victimization only, have found that girls report a lower frequency of victimization than boys do (Hodges & Perry, 1999; Perry et al., 1988). In fact, one study which examined both subtypes of victimization found that boys reported more overt victimization than girls while girls reported more relational Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 7 victimization (Crick & Bigbee, 1998). Therefore, any study of victimization should include overt and relational subtypes in order to account for the victimization experiences ofboth sexes folly. The relationship between victimization and drug use has received only limited attention in the research literature. Existing research does suggest that substance use and victimization are associated (Windle, 1994; Yudko, Blanchard, Henrie, & Blanchard, 1997). At least two longitudinal studies have found evidence linking early victimization with later drug use, but in both cases the early victimization was parent to child not peer to peer (Ireland & Widom, 1994; Widom, Ireland, & Glynn, 1995). The social development model provides a plausible theoretical foundation for a link between victimization and drug use in that it suggests that children victimized by their peers may be more likely to use drugs because they have limited opportunities for prosocial behavior (Catalano & Hawkins, 1996). Consequent bonding to antisocial others is a precursor to antisocial behaviors, including delinquency and drug use. More research is needed to elucidate the victimization-drug use relationship. Interaction of Type of Aggression and Gender on Risk for Drug Use Relational victimization and aggression maybe more common experiences for girls while boys may more typically experience overt aggression and victimization (Crick & Bigbee, 1998; Crick & Gropeter, 1995). These gender-specific styles of aggression and victimization raise the question of drug use implications of overt aggression or victimization by a girl and relational aggression and victimization by a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 8 boy. Although scant, existing literature suggests that nongender specific aggression or victimization (i.e., relational for boys, overt for girls) may represent a greater risk for drug use than the gender-typical style. Specifically, one study found that overtly aggressive girls and overtly and relationally aggressive boys had the highest level of adjustment problems, although drug use was not examined specifically (Crick, 1997). One reason for elevated risk may be that gender nonnormative behavior may place children at greater risk of social maladjustment because peers view behavior as atypical and, consequently, exclude or reject the child (Crick, 1997). Hence, one aim of this study was to determine whether gender and type of aggression interact to increase the risk of drug use such that girls who are overtly aggressive and boys who are relationally aggressive show the highest occurrence of drug use. Gender Differences: Social Competence Although definitions vary, social competence can be generally understood as a person’s ability to negotiate key developmental tasks. These tasks includes friendly peer relationships, following rules and academic achievement in late childhood (Masten & Coatsworth, 1998). Mastery of these areas continues in adolescence and extends to include development of close friends, romantic relationships and forming a sense of self (Masten & Coatsworth, 1998). Interventions that have increased social competence by fostering the skills necessary to negotiate social interactions have mitigated the effects of some risk behaviors, such as early aggressive behavior and peer acceptance, on later drug use and aggression (Caplan et al., 1992; Hawkins et al., 1991). Skills that have Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 9 been taught to students in competence-building interventions designed to prevent delinquency and drug abuse have included communication, decision-making, negotiation and conflict resolution (Caplan et al., 1992; Hawkins, Catalano, Morrison et al., 1992). Results from these studies suggest that this maybe one effective approach of lowering the risk for later problem behaviors. Theories of aggression, competence, and resiliency provide insight into how competence in social interactions may protect against negative behavior outcomes including drug use. The social development model considers social skills to be mediators of the relationship between risk factors and outcomes of antisocial behavior. In this model, skills change the course of bad developmental outcomes by increasing perceived opportunities and rewards for prosocial behaviors (Catalano & Hawkins, 1996). Similarly, in the coping-competence model, life outcomes are the result of the interaction among surface characteristics, risk-protection variables, competence, challenge exposure and response to challenges, where higher competence leads to better challenge response, and, hence, fewer problem behaviors (Blechman, Prinz, & Dumas, 1995). Consistent with this is resilience theory which identifies parent-child relationships, cognitive development, intellectual functioning, and self-regulation of emotion as key agents which can protect a child even in a high-risk context (Masten & Coatsworth, 1998). The common thrust of these theories is the suggestion that increased social skills may reduce the likelihood of negative outcomes for children by Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 10 helping them minimize the negative impact and maximize the positive impact of life challenges and opportunities. Gender has been found to influence social competence, although its impact is unclear. One study found that boys have less social competence during pre-adolescence, defined as fifth to sixth grade, and higher externalizing and internalizing problems, including depressive mood, anxiety and fighting (Forehand, Neighbors, & Wierson, 1991). This suggests that social competence may be a less effective mediator of risk for boys than for girls during this developmental period. Another study of slightly older subjects (seventh and eighth graders) suggests that social competence has sometimes incongruous relationships with drug use. It found that increased social acceptance from peers, an indicator of social competence, was positively correlated with cigarette use for girls, suggesting that social competence could put girls at greater risk of use. However, this relationship was not found for boys. Furthermore, better behavior, another social competence indicator, was associated with less cigarette and alcohol use for girls but not boys, suggesting, in this case, that social competence would protect girls against drug use (Lifrak, McKay, Rostain, Alterman, & O'Brien, 1997). In general, social competence was not associated with marijuana use for either gender. Clearly, these relationships are complex and, although it seems likely that social competence to some degree is moderated by gender, gender’s impact on social competence as a mediator of problem behavior relationships is uncertain. More research is needed to elucidate these relationships. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 11 Although some interventions which have fostered social competence have been successful in reducing later problem behaviors, more research in this area is needed. In particular, existing social competence interventions seeking to reduce problem behaviors have targeted only overtly aggressive behaviors and later drug use, perhaps explaining why girls have shown less improvement from these programs (O'Donnell et al., 1995). Alternatively, it may be that some types of social competence are liabilities for girls with regard to drug use risk, as discussed previously. No studies to date have examined social competence as a mediator of the effect of early relational aggression on later drug use. One aim of this study was to fill that gap as well as to test social competence as a mediator of the effect of overt aggression on later drug use. Another aim was to explore the interaction of gender with these relationships. Less information is available about the role of social competence in mediating the relationship between victimization and problem behavior outcomes such as drug use. Research does suggest that children who are victimized by their peers show deficits in aspects of social functioning such as social isolation and low assertiveness (Hodges & Perry, 1999) . Additionally, enhancing social skills of children who have low social competence has reduced their rejection among peers (Perry et al., 1988). Together these findings suggest that social competence is associated with victimization by peers. Thus, it seems reasonable that social competence would mediate the victimization-drug use relationship. Testing for social competence as a mediator of this relationship was one aim of the proposed research. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 12 This study proposes that social competence mediates the relationship between aggression and drug use and victimization and drag use. At this point, it is useful to outline the specific reasons that social competence is considered a mediator of the relationships between aggression and victimization and a drag use outcome. A mediator is a causal mechanism, one that “transform[s] the predictor or input variable in some way {p. 1178, (Baron & Kenny, 1986)}.” In the proposed study, it is postulated that social competence reduces the effect of risk (i.e., aggression and victimization) on problem behavior outcome (i.e., drag use) by providing the initially aggressive or victimized child with skills to manage risk situations when they arise, thereby reducing the likelihood of future problem behavior. For example, a child low in anger control may be more aggressive with other children due to an inability to manage this emotion when it arises. Unchecked, the aggressive child may begin to use drags as she gets older as a way to reduce or manage anger. Indeed, one study of early to late adolescents found that alcohol use in early adolescence predicted anger in late adolescence, suggesting that drag use may be one anger coping strategy (Weiner, Pentz, Turner, & Dwyer, 2001). However, by providing the aggressive child with anger management skills, she effectively manages anger when it arises and consequently reduces it. This, in turn, reduces the likelihood that she will seek out anger coping strategies such as drag use as she enters adolescence. Although this is just one hypothetical example, it demonstrates the causal role increased social competence may have in reducing drag use for an aggressive child. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 13 Additionally, past research suggests that social competence meets the empirical requirements for mediating both the aggression-drug use and victimization-drug use relationships. According to Baron and Kenny (1986), in order to demonstrate mediation, there must be a relationship between the predictor and mediating variables and the mediating and outcome variables. Research has established relationships between overt aggression and social competence and, albeit less convincingly, between overt victimization and social competence, the predictor and mediating variables in the proposed study (Hawkins et al., 1991). Given these relationships and research showing that relational aggression and relational victimization are constructs related to their overt counterparts (e.g., Crick (1996); Crick and Gropeter (1996)), it was postulated that relationships exist between the predictor and mediating variable. Other research has demonstrated associations between low social competence and drug use (e.g., Caplan, (1992)), providing the necessary link between the proposed mediator and outcome variables. Collectively, these interrelationships argue for testing social competence as a mediator of aggression to drug use and victimization to drug use relationships. Aims of the Present Study The main purpose of the present study was to explore predictors of drug use related to peer aggression and victimization. The first aim of the study was to test the predictive relationship of aggression and victimization to drug use. Since research has identified specific classes of aggression and victimization (relational and overt), the next goal of this study was to examine these as predictors of drug use. It is possible that Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 14 preference for a particular aggression style may differ by gender. Thus, the second aim of this study was to compare the relative strength of the overt and relational aggress ion to drug use and overt and relational victimization to drug use relationships between boys and girls. One recent study found preliminary evidence that youth at highest risk for problem behaviors may display gender nonnormative forms of aggression and victimization and investigating these relationships was a secondary goal of this study (Crick, 1997). Interventions that have increased social competence and consequently reduced later problem behaviors suggest that the aggression-drug use relationship may be mediated by social competence (Caplan et al., 1992; Hawkins et al., 1991). Therefore, the third aim of this study was to assess social competence as a mediator of the aforementioned predictive relationships. METHODS Overview This study examined potential precursors to drug use including aggression, victimization, and a possible mediating mechanism, low social competence, using data collected as part of a large-scale, school-based intervention known as “Bright Stars” in Indianapolis, Indiana. Bright Stars is a randomized prevention trial that is evaluating the effects of a competence-building program in fourth and fifth grade and risk- reduction efforts of a drug-prevention program in sixth and seventh grade, to prevent adolescent drug use and related problem behaviors, including aggression. The Bright Stars elementary program builds general academic and social competence skills in a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 15 school program with supporting parent and community activities. Beginning in sixth grade, the program targets drug use and violence prevention specifically. The present study utilized data from two samples to examine questions related to aggression as a predictor of drug use mediated by social competence. One sample was a panel followed longitudinally from the Fall of fourth grade to the Spring of sixth grade, and the other was a cross-sectional sixth-grade sample. Both the panel and cross- sectional sixth-grade sample were comprised of subjects from program and control groups of the large Bright Stars prevention study, and the cross-sectional sixth-grade sample included subjects from the panel study as well as new subjects present only in sixth grade. Since Bright Stars was designed primarily as an intervention to protect against drug use, only a limited number of items measured aggressive behavior, and victimization was not assessed in the first two years of the study (grades four and five). When the idea for this study was formulated, additional aggression items and items which assessed victimization were added to the questionnaire for measurement in sixth grade. In order to gain a thorough understanding of aggression and victimization, hypotheses were tested using the longitudinal data, where possible, and this information was supplemented with the cross-sectional data set which contained victimization items and more complete measurement of aggression. Thus, the present study has been divided into two substudies: Study One tested hypotheses related to aggression, drug use and social competence using longitudinal data collected from grades four to six, and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 16 Study Two evaluated additional hypotheses related to aggression, victimization, drug use and social competence among sixth-grade students. Overall Research and Measurement Design The initial forty-eight elementary schools participating in large Bright Stars intervention represented eleven communities in Marion County, Indiana (the Indianapolis Metropolitan area). All forty-eight schools were assigned to treatment condition by middle school feeder patterns, with each of the school districts assigned to a program or control condition. Middle schools were stratified by participation in other intervention programs, particularly DARE, and public/private status, and then randomly assigned to program or control status. Of these, thirty-six schools, comprising one hundred and four classrooms, agreed to participate in a longitudinal study. Data were collected via self-reported survey in the Fall of fourth grade, with a six-month, one-year and two-year follow-up, collected in the Springs of fourth, fifth and sixth grade, respectively. Study One: The Panel Study Procedure Baseline measurements were taken in Fall, 1998, when students were in fourth grade. Three follow-up measurements were taken, one at the end of fourth grade (Spring, 1998, the next at the end of fifth grade (Spring, 1999) and the final measurement at the end of sixth grade (Spring, 2000). The data were collected in classrooms by pairs of trained data collectors following a standardized protocol. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 17 Students absent on the day of data collection were left an absentee package containing the questionnaire for completion upon return to school. Consent was active and the participation rate was approximately 71%. Several attempts were made by proj ect staff to get consent forms from students. Subj ects were given personal identification numbers in order to link repeated measurements on the same individuals. Data management included immediate scanning for completeness of the questionnaire. Classroom statistics were entered on collection envelopes, given a unique batch number, and stored by school. Data were entered and verified with an actual error rate of less than 0.1%. A master file was constructed using SAS statistical software and variables were checked for out of range values (SAS Institute, 1998). Human Subjects Protection This study followed procedures for research with human subjects and was approved by the Institutional Review Board of the University of Southern California. Several steps were taken to insure the protection of participants in this study. First, since this study used subjects ranging in age from 8 to 11 years old, permission for student participation was required from the child’s parents or guardian. Parents were sent, on project letterhead, a description of the program and measurements that were planned for the student. If parents objected to their child’s completion of the questionnaire, they were instructed to so indicate on the forms provided for this purpose. Surveys were administered only to students whose parents provided written consent for their participation in this study. Additionally, student participants were able to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 18 withdraw from the study at any time they chose even if they had parental consent for participation. Third, in order to insure a minimum of anxiety for student participants, data collectors were the only members of the investigation team who actually interacted with students directly. The data collectors were experienced in working with children and sensitive to their needs. Fourth, data were numerically coded and kept locked at USC with only numerical identifiers used for tracking purposes. Once collected, an individual’ s data were not relayed to anyone other than authorized project staff. Finally, in order to further protect the privacy of participants in this research study, subj ects were also protected by DHHS Confidentiality Certificate DA-87-13. Sample Subjects were selected for the analytic sample of Study One from among the 1756 eligible students (i.e., students who had informed consent and were present at baseline) if they met three criteria. First, since this was a longitudinal study, subjects needed to be present at baseline as well as fifth-grade and sixth-grade follow-up measurements. Next, because this study sought to follow the natural history of the aggression-to-drug use relationship, subjects had to be non-drug users at baseline. In this study, non-use was defined as never having used more than one puff of a cigarette, one puff of marijuana and one sip of alcohol. Finally, subjects had to have complete data on variables of interest (i.e., variables measuring aggression, social competence, drug use and demographic information) since EQS, the statistical software used to analyze the data, did not allow for missing data (P. M. Bentler, 1995). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 19 This sample was 59.44% female and 62.26% white, with 31.45% identified as having high socioeconomic status (SES) based on parents’ jobs. Additionally 45.55% of the sample attended private school and 60.52% had received the Bright Stars intervention. Annual prevalence for each of the aggression indicators was calculated by dividing the four-point scale (“NO!,” “no,” “yes,” “YES!”) such that “no” responses were coded as 0 and “yes” responses were coded as 1. Prevalence rates ranged from 10.85% for “destroy[ed] property” to 66.16% for “physical fights with other kids.” A comparison of subjects selected for inclusion in the analytic sample with those excluded revealed that the analytic sample contained significantly higher proportions of female, white and high SES subjects as well as subjects who attended private school. Being in the analytic sample was also significantly associated with lower aggression for two indicators, “destroyfed] property” and “call[ed] others bad names” in the past year. However, there were no significant differences between the analytic sample and others on three other aggression indicators, “physical fights with other kids,” “push[ed] or shove[d] other kids” or “[left] someone out of your group.” See Table 1 for a report of baseline demographic and aggressive behavior information. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 20 Table 1. Comparison between subjects selected for longitudinal analytic sample with those excluded. Characteristic Selected (N = 461) Excluded (N = 1295) % % x2 Demographics Female 59.44 53.17 5.39* White 62.26 56.91 3.99* High SES a 31.45 22.26 14.62*** Private School 45.55 11.04 252.10*** Exposed to intervention 60.52 58.22 .74 Overt Aggressionb Physical fights with other kids 66.16 66.54 .02 Destruction of property 10.85 17.25 10.59** Pushed or shoved other kids 29.93 33.36 1.82 Called others bad names 29.07 35.72 6.69** Relational Aggressionb Purposely left someone out of your group 27.55 28.72 .23 Note. Inclusion in the analytic sample required measurement at baseline, fifth-grade and sixth-grade, no baseline drug use and no missing data on variables of interest. “SES was based on parents’ jobs (technical = high, nontechnical = low) b Percentages reported for overt and relational aggression behaviors were subjects who answered “yes” or “YES!” from response choices of “NO!,” “no,” “yes,” and “YES!” *p<.05, **p<.01, ***p<.001 Description o f Measures The measure used for the panel study (Study One) was a self-report questionnaire consisting of approximately 135 items (135 items in fourth grade, 130 items in 5 th grade, 138 items in 6th grade) related primarily to social and academic competence and drug use, but also including items related to aggression and victimization. Previous research Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 21 has indicated that self-report is the appropriate method for assessment of aggression when there is interest in the child’s perception of the behavior and self-report has been shown to be significantly correlated with other sources, including teacher and peer report (Crick & Gropeter, 1996; Pakaslahti & Keltikangas-Jarvinen, 2000). Self-reported drug use is also believed to produce valid data (Johnston, O'Malley, & Bachman, 1998). Thus, there was confidence that the measurement tool used provided accurate information about the behaviors under study. Aggression items for this study were assessed in the Fall of fourth grade and addressed behaviors in the past year. Overt aggression was measured using a four-item scale which asked students to indicate if they had engaged in aggressive behaviors including physical fights, destruction of property, pushing and shoving, and calling others bad names. Relational aggression was measured with a single item "Did you, on purpose, leave someone out of your group or not invite them to play?”adapted from an existing relational aggression scale (Crick & Gropeter, 1995). Response choices for all aggression items were “NO!,” “no,” “yes,” “YES!”, coded one to four, respectively, and were accompanied by verbal instructions from the teacher explaining the distinction between a strong “no” and “yes” response (“NO!” and “YES!”) and a weak “no” and “yes” response. These responses were adapted from another intervention trial were designed for ease of use with younger subjects (Hawkins, Guo, Hill, Battin-Pearson, & Abbott, 2001). In order to calculate prevalence rates, responses of “NO!” and “no” were Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 22 coded 0 and “YES!” and “yes” responses were coded 1. Original scaling was used for all other analyses including means, correlations and structural equation modeling. Social competence was measured in fifth grade using six items. Items measuring social competence were developed specifically for the large Bright Stars intervention trial and assessed skills that were taught as part Bright Stars curriculum. Since many items addressed social competence, an item-reduction strategy was used to arrive at the six items ultimately used. First, items that were related to social competence with peers were identified based on face validity. Next, since there was a desire to use the same social competence construct in the sixth-grade cross-sectional study (Study 2), any items that were not also assessed in sixth grade were eliminated. Finally, a factor analysis was conducted for fifth and sixth grade and the items that had the best loadings across the two grades were retained (all loadings for items retained were greater than .40). This process resulted in selection of six items. They related to listening to friends, helping classmates with homework, problem solving with friends, mentoring, standing up for personal values and negotiating with friends. Response choices were 1 = “No,” 2 = “Yes, 1 or 2 times,” and 3 = “Yes, more than 2 times.” In order to calculate mean frequency, items were recoded with approximate average scores: “No”=0, “Yes, 1 or 2 times” = 1.5 and “Yes, more than 2 times” = 3. The original scaling was used for all other analyses including correlation matrices and structural equation models. Drug use items were measured in sixth grade and assessed lifetime prevalence of cigarette, alcohol, and marijuana use. They were adapted from an existing measure Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 23 of drug use prevalence (Johnston et al., 1998). Lifetime cigarette use asked subjects how many cigarettes they had smoked in their whole life. Responses were seven choices ranging from “none, I’ve never had one puff of a cigarette,” to “more than 5 packs.” The lifetime alcohol use item asked how many alcoholic drinks the subj ect had ever had, with eight responses ranging from ’’ none, I haven’t ever had one sip of an alcoholic drink,” to “more than 100 drinks.” Finally, marijuana use asked about the number of times a subject had used marijuana, with seven response choices which ranged from no use to “more than 100 times.” Original coding (i.e., scales of 1 to 7 for cigarette and marijuana use and 1 to 8 for alcohol use) were used in structural equation modeling analyses. In order to calculate mean frequency of lifetime drug use, these same responses were recoded with approximate average scores. For cigarette use, “none,’’ was recoded to 0, “one puff’ - .25, “part or all of one cigarette” = 1, “2 to 4 cigarettes” = 3, “5 to 20 cigarettes” = 12.5, “1 to 5 packs” = 60.5, and “more than 5 packs” = 101. Lifetime alcohol use responses were recoded as follows: “none” = 0, “only sips” = .25, “part or all of one drink” = 1, “2 to 4 drinks” = 3, “5 to 10 drinks” = 7.5, “11 to 20 drinks” = 15.5, “21 to 100 drinks” = 60.5 and “more than 100 drinks” = 101. For marijuana, “none” was recoded to 0, “once” = 1, “2 to 4 times” = 3, “5 to 10 times” = 7.5, “11 to 20 times” = 15.5, “21 to 100 times” = 60.5 and “more than 100 times” = 101. Demographic characteristics assessed included gender, ethnicity, and socioeconomic status (SES). SES was based on students’ reports of their parents’ jobs Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 24 where technical jobs were categorized as high SES and nontechnical jobs were categorized as low SES. See Appendix A for all response choices and more information about psychometric properties of each measure. Specific Hypotheses The following specific hypotheses were tested in Study One: Aggression as a predictor of drug use: 1. Higher aggression in fourth grade predicts increased drug use in sixth grade. 2. Social competence mediates this relationship. 3. Gender moderates this relationship. Relational and overt aggression as drug use predictors: 4. Higher relational aggression in fourth grade predicts increased drug use in sixth grade. 5. Higher overt aggression in fourth grade predicts increased drug use in sixth grade. 6. Gender moderates the relational aggression-drug use relationship such that this relationship is stronger for girls than for boys. 7. Gender moderates the overt aggression-drug use relationship such that this relationship is stronger for boys than for girls. 8. Social competence mediates these relationships. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 25 Data Analysis Plan Measurement and structural models Hypotheses for the panel study (Study One) were tested using structural equation modeling (SEM) using EQS software (P. Bentler, 1995). Models were applied to a raw data matrix and the maximum likelihood (ML) procedure was used to obtain parameter estimates. The ML estimation procedure has been shown to be robust to violation of the multivariate normal distribution assumption and thus was believed to provide reliable parameter estimates (Pentz & Chou, 1994). The model testing procedure generally followed several discrete steps. First, a measurement model was specified to verify that measured variables represented underlying latent constructs (Byrne, 1994). The next step was to establish a structural model to test the predictive relationships among latent constructs of interest (e.g., between aggression, social competence and drug use). Results of the structural model provided estimates of these relationships, for example, the regression weight from aggression to social competence, the statistical significance of these paths based on the t test, as well as overall model fit information. When overall model fit was inadequate, modification of the model was undertaken to improve fit. Model modification comprised adding correlated errors or paths from demographic variables to latent constructs. These additional paths were based on the results of the LM test and guided by theoretical considerations. A concern that has been noted about model modification is that this process may substantially alter the originally hypothesized model (Pentz & Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 26 Chou, 1994). Thus, modifications were made conservatively and only when it was believed that they did not alter the structure of the original model. Model fit was assessed using the chi-square test and comparative fit index (CFI). The null hypothesis of the chi-square test is that the model fits the underlying data structure, i.e., that there is no significant difference between the data and the model imposed upon them (Crowley & Fan, 1997). Thus, a p-value greater than .05 is the desirable result when using the chi-square test in SEM. However, since relatively large sample sizes are needed to run an EQS model (typically a minimum of five subjects per free parameter with ten preferred for nonnormal distributions) and the power of the chi- square test increases with sample size, the chi-square test may indicate a poor fit with data even when the differences between the data and the model are minimal (Crowley & Fan, 1997). For this reason, the CFI was also assessed. The CFI indicates the degree of model fit from 0 (no fit) to 1 (perfect fit), with values larger than .90 considered good fit, and has been shown to be less dependent on sample size than other available fit indices (Crowley & Fan, 1997; Tanguma, 2001). Together, these indices provided a good indication of overall model fit. Test of measurement invariance using the multiple group approach Several hypotheses in Study One related to gender interactions in predictive relationships. To test these hypotheses, the multiple group approach for testing measurement invariance was used (Pentz & Chou, 1994). For these analyses, separate models were first established for each group following the preceding steps as Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 27 recommended by Byrne (1994). After a satisfactory structural model had been established separately for boys and girls (this was the final model after modification described previously), the groups were combined and a basic model was established. A test of measurement invariance of factor loadings between girls and boys was conducted by imposing a series of equality constraints upon estimated parameters. First, factor loadings were constrained to be equal between groups. Factor loading constraints were then released based on LM test results if they improved model fit and were theoretically feasible. Improvement of model fit was based on a test of the difference between the chi-square values of the basic model and the constrained or partially constrained model (Pentz & Chou, 1994). Next, parameter estimates between latent factors were constrained to be equal between groups while maintaining any constraints on factor loadings established during the previous step. The model was run and the LM test results used to identify constraints that could be released to improve model fit. Again, improvement was based on the chi-square difference test and constraints were released using theory and LM test results for guidance. The result was the final model showing factor loadings and paths that were invariant across groups. Controlling for demographics All models in Study One controlled for a common set of demographic variables. These were ethnicity (white vs. non-white), SES (parent had a technical vs. non technical job as reported by subject), school type (public vs. private), and exposure to the Bright Stars intervention (exposed vs. not exposed). Within each structural model, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 28 demographic variables were treated as exogenous variables that directly predicted the drug use outcome. Additionally, demographics variables were allowed to covary freely with one another. Gender was also controlled in that all models were applied separately to an all-male and all-female sample. Preliminary Measurement Model When the idea for the panel study (Study One) was formulated, drug use in sixth grade was conceived as one construct representing lifetime use of cigarettes, alcohol and marijuana. This was based on past work which has shown that use of these substances is positively associated and they share similar risk factors (Donovan, lessor, & Costa, 1988; Hawkins, Catalano, & Miller, 1992). However, initial inspection of the a basic measurement model indicated that lifetime cigarette use loaded much more strongly on the drug use factor than did either lifetime alcohol use or lifetime marijuana use. Specifically, for girls, unstandardized factor loadings were 1.11 (SE =.09), .46 (SE = .07) and .30 (SE = .03) for lifetime cigarette use, alcohol use and marijuana use, respectively. For boys, factor loadings were .76 (SE = .12), .42 (SE = .10) and .53(.09) for lifetime cigarette use, alcohol use and marijuana use, respectively. Upon reflection, a decision was made not to combine the three lifetime drug use indicators into one construct, but to evaluate each drug alone in a separate model. This decision was based upon several considerations. First, due to the strength of cigarette use compared to the other two indicators, any results of a model with a combined drug use latent factor would primarily reflect the relationship of other constructs to cigarette Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 29 use. Therefore, any unique relationships between aggression or social competence and alcohol or marijuana use would be obscured. Second, the theories that guided development of the general model for this study, that aggression predicts drug use and this relationship is mediated by social competence, did not preclude the possibility that substances may have some unique properties that could further inform the understanding of the aggression-drug use relationship. Finally, a re-examination of the drug use literature provided some indication that each of these substances may have unique risks and consequences. For example, Kellam (1983) reported that aggression combined with shyness increased the likelihood only of heavy cigarette use among males but not other drugs. These results, based on results of a long-term prospective study provide evidence that there may be specificity as well as commonality in the trajectories of different substances. The decision to test each substance as an outcome separately would result in three models for every model initially described in the hypotheses. However, a reductionist approach was adopted for the model testing such that if a more general model did not show a main effect, no further analysis was conducted for that substance. For example, if the most general model testing fourth-grade aggression as a predictor oflifetime marijuana use found no significant relationship, the more specific model that relational and overt types of aggression predict lifetime marijuana use would not be tested. Similarly, if a main effect of the general model were found, but the subsequent model of overt and relational aggression to lifetime marijuana was not supported, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 30 most specific set of hypotheses related to gender nonnormative styles of aggression would not be explored. Thus, more specific models were tested only when support for a main effect was found in the more general model. Since this study utilized the multiple groups approach, it is important to note that inquiry for a substance was curtailed only if there was a main effect for neither gender. Model testing sequence Models in Study One were tested in progression from general to specific. The first set of models tested addressed the question of fourth-grade aggression as a predictor of sixth-grade lifetime cigarette, alcohol and marijuana use. For these analyses, all five aggression indicators (the four overt aggression items and the one relational aggression item) were allowed to load on the single latent construct representing general aggressive behavior. If a main effect of aggression on the specific substance was found, the substance was investigated in relation to the next area of inquiry which was whether subtypes of aggression, specifically overt and relational aggression, differentially predicted use of this substance. For these models, the four overt aggression indicators loaded on the overt aggression latent construct while relational aggression was represented by the single indicator related to this construct. Study Two: The Cross-Sectional Study Procedure Data for Study Two were collected as part of the Bright Stars intervention trial and were collected in the Spring o f2000. Subjects who had participated in the Bright Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 31 Stars trial in elementary school as well as students in a middle school classroom receiving the STAR drug abuse prevention program were approached for participation in this study. Consent was active, although subjects who had consented to participation in fourth or fifth grade were not required to reconsent. The consent rate for the sixth- grade cross-sectional study was approximately 83%. Subjects in this study comprised thirty-one schools and fifty-three classes and were a population-representative sample. Please refer to the Procedure section under Study One for more information about data collection and management procedures. Human Subjects’ Protection These data were gathered as part of the Bright Stars intervention trial and this study was approved by the Institutional Review Board of the University of Southern California. Please refer to the Human Subjects’ Protection section under Study One for a description of the steps taken to insure protection of human subjects. Sample Subjects in the cross-sectional study (Study Two) were selected for the analytic sample from among the 1008 eligible students (i.e., students who had informed consent and were present at sixth-grade measurement) if they had complete data on variables of interest. Complete data was necessary since analyses could not be conducted in EQS, the statistical software used in this study, with missing data (P. M. Bentler, 1995). This resulted in a total of 824 subjects. Of these, 59.47% were female, 53.64% were white, 36.17% attended private schools and 31.07% were high SES based on parents jobs. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 32 Additionally, 58.25% had been exposed to the Bright Stars intervention mentioned previously in fourth or fifth grade and 57.40% had received a related middle school drug prevention program. Table 2. Characteristics of analytic sample in cross-sectional study. Characteristic Selected Excluded N=824 3 I I % % % x2 Demographics Female 59.47 48.37 7.58** White 53.64 33.70 23.60*** High SES a 31.07 27.88 .44 Private School 36.17 17.93 22.67*** Exposed to elementary competence-building intervention 58.25 48.37 5.98* Exposed to middle school drug prevention program 57.40 59.78 .35 Lifetime Drug Use b Cigarette 16.63 21.91 2.82 Alcohol 14.93 14.20 .06 Marijuana 9.34 15.57 5.78* Overt Aggressionc Destruction of property 5.84 12.03 7.96** Physical fights with other kids 30.54 30.38 .002 Pushed or shoved other kids 20.66 25.63 1.96 Called others bad names 36.28 35.00 .10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 33 Table 2. Continued Relational Aggressiond Purposely left a kid out of your group 31.75 24.48 3.03 Got back at someone by not letting them be in your group 26.59 25.17 .12 Told lies about another kid to make others not like them 18.83 19.31 .02 Told a kid you won’ t like them unless they do what you say 10.51 10.64 .002 Tried to keep others from liking a kid by saying mean things about him/her 14.34 16.08 .30 Overt Victimizationc Got hit or pushed by other kids 21.00 17.11 1.20 Got beat up by other kids 4.87 11.76 10.97*** Called names by other kids 32.97 26.49 2.46 Kids did mean things to you 25.36 21.05 1.28 Relational Victimizationd Someone left you out on purpose 28.35 20.69 3.65 Kid got back at you by not letting you be in his/her group 22.18 16.67 2.22 Kid told lies about you to make others not like you 33.99 29.17 1.28 Kid told you they won’ t like you unless you do what they say 19.26 12.59 3.63 Kid tried to keep others from liking you by saying mean things 28.85 20.83 3.39* Note. Inclusion in the analytic sample required sixth-grade measurement and no missing data on variables of interest. * SES is based on parents’ jobs (technical = high, nontechnical = low) b Drag use is percentage reporting use of at least one cigarette, one alcohol drink and one use of marijuana. c Overt aggression and victimization are percentages reporting three or more incidents of the behavior in the past year. d Relational aggression and victimization prevalence are percentages reporting the behavior "sometimes," "most of the time," and "all the time." *p < .05, **p < .01, ***p<.001 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 34 Prevalence rates for indicators of interest were also calculated. Prevalence of overt aggression and victimization reflected the number of subjects reporting three or more occurrences of a behavior in the past year. Relational aggression and victimization were the percentages of subjects reporting a behavior at least "sometimes" on five-point Likert scale from "never" to "all the time." The manner in which lifetime drug use prevalence rates were calculated is described in the "Sample" section under Study One. Lifetime prevalence of cigarette use was 16.63%, alcohol use 14.93% and marijuana use 9.34%. Annual prevalence of overt aggression ranged from 5.84% of subjects reporting destruction of property to 36.28% reporting "call[ed] others bad names". Overt victimization ranged in annual prevalence from 4.87% of subjects reporting that they got beat by other kids to 32.97% for being called names by other kids. Annual rates of relational aggression ranged from 10.52% of subjects who reported that they had told a peer they would not like him or her unless the peer did what they said to 31.75% who reported purposely excluding someone from their peer group. Finally, annual rates of relational victimization ranged from 19.26% who reported being told by a peer that he or she would not like you unless you did what he or she said to 33.98% who reported that a peer had told lies about them to make others not like them. A comparison of the subjects selected for the analytic sample (i.e., those with complete data) with those excluded was made to assess the extent to which the analytic sample was representative of all subjects surveyed. There was a significantly higher proportion of females, whites and private school students in the analytic sample, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 35 although there was no difference in the proportion of subjects with high SES, based on students’ reports of their parents jobs. For drug use items, being in the analytic sample was associated with a lower lifetime prevalence of marijuana use, but was not associated with lifetime prevalence of either alcohol or marijuana use. For overt aggression, being in the analytic sample was associated with a lower annual prevalence of property destruction but was unrelated for other indicators. Similarly, being in the analytic sample was associated with significantly lower prevalence of getting beat up by other kids, but was independent ofthe other overt victimization indicators. Annual prevalence for none of the relational aggression indicators was associated with analytic group membership and higher annual prevalence of only one relational victimization indicator, "[a] kid tried to keep others from liking you by saying mean things about you," was associated with being in the analytic group. See Table 2 for a complete report of prevalence rates including how they were calculated and comparisons between the analytic sample and excluded subjects. Measures All data collected for the purposes of the cross-sectional study (Study Two) were self-reported. Overt aggression, social competence and lifetime drug use were assessed using the same items as in Study One. However, since subjects were older, response choices were changed for overt aggression and social competence items in order to assess behavior frequencies. For overt aggression, responses were on a four-point scale where 1= "No," 2 = "Yes, 1 or 2 times," 3 = "Yes, 3 or 4 times" and 4="Yes, 5 or more Reproduced with permission ofthe copyright owner. Further reproduction prohibited without permission. 36 times." Social competence response choices were expanded from the three-item scale used previously to the same four-point response choices used to assess overt aggression. New items present only in Study Two measured relational aggression and overt and relational victimization. Relational aggression was assessed using a five-item scale which assessed behaviors including excluding peers, telling lies about peers and other forms of social manipulation (Crick & Gropeter, 1995). Five response choices ranged from 1-'Never" to 5 - 'All the time." Relational victimization was also measured using a five-item scale, which assessed behaviors such as excluding others from social plans, lying, and other forms of social manipulation. For relational victimization, the same behaviors assessed in relational aggression were measured but the items were reworded to indicate victimization rather than perpetration (Crick & Gropeter, 1996). The scale for overt victimization was adapted from an existing peer nomination victimization scale and items asked about victimization experiences including being made fun of, being pushed and shoved, being called names and others doing mean things to you (Hodges & Perry, 1999). See Appendix A for a complete list of indicators, response choices and psychometric properties of measures. Data Analysis Plan Measurement model development, structural model development and tests of measurement invariance in Study Two followed the same procedures articulated in Study One. Demographics controlled in Study Two were similar to those in Study One but reflected students’ status in sixth grade rather than at baseline. Thus, ethnicity Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 37 (white vs. nonwhite) and exposure to the Bright Stars intervention in elementary school (yes vs. no) were the same, but school type (public vs. private) and SES (parent had a technical vs. non-technical job as reported by subject) reflected the status of these conditions for subjects in sixth grade. Another covariate, exposure to a middle school drug prevention program (exposed vs. not exposed), was also controlled. Preliminary Measurement Model Following the reasoning articulated under the “Preliminary Measurement Model” section o f Study One, a decision was made to test models with each substance of interest (lifetime cigarette, alcohol and marijuana use) as the outcome in the cross-sectional study as well. Additionally, the same reductionist approach was used such that more specific models were tested only if support was found for the more general model. See the “Preliminary Measurement Model” section of Study One for more explanation regarding these decisions. The initial design of Study Two was to test hypotheses related to the predictive relationship of relational and overt aggression and victimization to drug use, and to examine gender differences in strength of predictive relationships ofrelational and overt subtypes of aggression and victimization to drug use outcomes. With these goals in mind, confirmatory factor analyses (CFA) were conducted separately by gender to assess relationships between measured variables and latent factors of overt aggression, overt victimization, relational aggression and relational victimization, as well as interrelationships among these latent factors. Based on the convention, mentioned Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 38 earlier, that a CFI of .90 is considered acceptable fit, inspection of the CFA’s for boys and girls showed neither model fit the data particularly well: the boys’ CFA had a %2(242) = 698.30, p < .001 and a CFI = .86, and the girls’ CFA had a y2{242) - 798.32, p < .001 and a CFI = .88, although all indicators loaded significantly on latent factors (Crowley & Fan, 1997). In addition to marginal fit, with twenty-four indicators each, these models were unwieldy and lacked parsimony. Overall dissatisfaction with the results of these CFA’s prompted a search for a more satisfactory model of the constructs of interest. As a solution, a decision was made to construct “parcels” of items by creating four summed scores representing overt aggression, relational aggression, overt victimization and relational victimization, as recommended by McCallum and Austin (2000). Based on the previous CFA’s which showed that, for both genders, the strongest correlations were between the two aggression subtypes (boys r = .58, p < .001; girls r = .53, p < .001) and the two victimization subtypes (boys r = .73, p < .001; girls r = .70, p < .001) and theory suggesting that overt and relational styles are subtypes of aggression and victimization, the models were respecified such that the summed scores for relational and overt aggression were indicators for a latent construct of aggression and the summed scores for relational and overt victimization were indicators for a victimization latent factor. CFA’s indicated that this approach resulted in better fitting models. For boys, the new CFA had a %2(32) = 92.17, p < .001 and a CFI = .94 and for girls, the CFA had a x2(32) = 142.147, p < .001 and a CFI = .91. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 39 The obvious disadvantage of this approach was that it precluded investigation of the differential impact of relational and overt styles of aggression and victimization on drug use outcomes. Further, it was not possible to test gender interactions with aggression and victimization styles on drug use outcome. This limited inquiry in Study Two to overall relationships of aggression, victimization, social competence and drug use. An exploratory test of gender differences in these models was conducted in lieu of investigation of gender x aggression and victimization style interactions. Overall, it was believed that the advantages of the parceling approach, including statistical parsimony, fewer chances for correlated residuals and reduction in sampling error, outweighed the drawbacks of fewer areas of inquiry (Little, Cunningham, Shahar, & Widaman). Specific Hypotheses The following specific hypotheses were tested in the Study Two: Aggression and victimization as drug use predictors: 1. Higher victimization predicts increased drug use among sixth grade subjects. 2. Higher aggression predicts increased drug use among sixth grade subjects. 3. Social competence mediates these relationships. 4. Gender interactions exist among these relationships. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 40 RESULTS Study One: Panel Study Descriptive Analyses The means, standard deviations, and intercorrelations for the fourteen variables measured in the panel study (excluding demographic characteristics) are summarized in Table 3. For fourth-grade aggression indicators, mean scores ranged from 1.52 (SD = .84) for “destroyed] property” to 2.99 (SD = 1.17) for ’’ physical fights” among boys. Among girls the pattern was the same with “destroy[ed] property” (M = 1.36, SD = .77) having the lowest average among aggression indicators and “physical fights” the highest (M = 2.69, SD = 1.20). For fifth-grade social competence indicators, the behavior demonstrated the least and most frequently in the past year were the same for both sexes: “help[ing] someone learn a new skill” was the least frequently reported (boys: M = 1.89, SD = 1.10; girls: M = 1.94, SD = 1.01) and the most frequently reported was “standing] up for your values” (boys: M = 2.50, SD = .86; girls: M = 2.36, SD = .86). Average lifetime drug use frequency in the panel study was also examined. Among girls, mean drug use was lowest for marijuana, used an average o f.19 times (SD = .84), and highest for cigarettes, with an average of 1.85 (SD = 10.19) cigarettes used in their lifetime. For alcohol use, girls reported having had, on average, .68 alcoholic drinks in their lifetime. Among boys, mean lifetime use was lowest for cigarettes (M = 1.08, SD - 7.64), slightly higher for alcohol use (M = 1.22, SD = 7.70) and highest for marijuana use (M = 1.30, SD = 10.49). Although it may seem somewhat surprising that Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 41 boys reported more lifetime marijuana use than cigarette or alcohol use, it is important to consider mean use in relation to the low p revalence of drug use overall in sixth grade, Among girls, 14.24% had ever smoked a cigarettes, 9.86% had ever used alcohol and 7.66% had ever used marijuana. Among boys, 12.30% had ever used cigarettes, 17.10% had ever used alcohol and 6.95% had ever used marijuana. These prevalence rates are generally similar to those reported elsewhere, although prevalence of alcohol and cigarette use is slightly lower and marijuana use slightly higher than other studies have reported (Indiana Prevention Resource Center, 2000; Wills et al., 2001). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 3. Correlation of indicators used in the longitudinal study by gender. M SD A1 A2 A3 A4 AS D1 D2 D3 SI S2 S3 S4 S5 S6 M - - 2.69 1.36 1.67 1.76 1.78 1.85 .68 .19 2.27 2.26 2.36 2.14 2.00 1.94 SD - - 1.20 .77 .95 1.00 1.04 10.19 3.99 .84 1.02 .97 .86 1.05 1.01 1.01 A1 2.99 1.17 - .22 .32 .37 .26 .05 .11 .04 -.03 .01 .001 -.01 -.11 -.10 A2 1.52 .84 .23 - .51 .42 .41 .11 .07 -.05 -.02 -.14 -.02 .01 -.07 -.03 A3 2.20 1.09 .46 .12 - .56 .35 .18 .19 .06 -.04 -.05 -.03 .02 -.12 -.08 A4 2.08 1.12 .36 .19 .51 - .37 .06 .09 -.02 .01 -.01 -.07 .02 -.13 -.14 A5 1.93 1.05 .30 .22 .43 .39 - .06 .10 .004 -.06 .01 -.05 -.03 -.12 -.05 D1 1.08 7.64 .07 .22 .04 -.02 -.01 - .44 .59 -.22 -.01 -.24 -.09 -.16 -.19 D2 1.22 7.70 .06 .06 .03 .07 -.004 .26 - .30 -.01 .10 -.15 -.06 -.20 -.10 D3 1.30 10.49 .002 .17 -.07 -.07 -.05 .49 .25 - -.23 -.06 -.12 -11 -.07 -.21 SI 2.13 1.08 -.07 -.15 -.03 -.20 -.11 -.07 -.02 -.03 - .31 .30 .31 .32 .39 S2 2.07 1.07 .02 -.07 -.02 -.13 -.07 -.06 .07 -.07 .41 - .29 .27 .28 .27 S3 2.50 .86 .07 -.04 -.09 -.11 .01 -.13 .04 -.06 .24 .34 - .43 .38 .36 S4 2.00 1.02 .02 -.09 -.06 -.09 -.13 -.08 -.06 -.08 .24 .30 .41 - .44 .41 S5 2.01 1.07 -.07 -.15 -.12 -.13 -.10 -.03 -.09 .02 .19 .21 .24 .43 - .41 S6 1.89 1.10 .04 -.11 .03 .03 -.08 -.05 .07 -.01 .34 .36 .33 .40 .33 - Notes. Boys (N = 187) below, girls (N = 274) above diagonal. Aggression (A1-A5), social competence (SC1-SC6) lifetime drug use (D1-D3) measured in 4th, 5th, 6th grade. Al=physical fights, A2= destroyed property, A3= pushed/shoved, A4 = bad names, A5= left someone out, D l= cigarette, D2= alcohol, D3=marijuana, Sl=helped, S2=listened, S3=stood for values, S4=talked to find solution, S5=worked out problem, S6=helped learn skill, r’s >= .14 or <= -.14 significant at p<.05. 43 Correlations o f Indicators Table 3 shows correlations among indicators used in Study One by gender. Inspection of correlations revealed significant associations among variables of the same construct. For each gender, aggression indicators were significantly correlated with one another, although associations were slightly stronger for girls. Correlation coefficients for boys ranged from .12 to .51, and were statistically significant with the sole exception of the association between “destroy[ed] property” and “push[ed] or shove[d] other kids” among boys r = . 12, p <. 10), which was marginally significant. For girls, all associations were statistically significant, with the range of correlation coefficients from .22 to .56. Similarly, lifetime cigarette, alcohol and marijuana use were all significantly intercorrelated for both boys and girls, although all associations were slightly weaker for boys (range from r = .25 to r -.49) compared to girls (range from r = .30 to r = .59). Social competence showed the same pattern, with significant correlations among all indicators for both genders. For boys, these correlations ranged from .19 to .43 and for girls they ranged from .27 to .44, again suggesting slightly stronger associations for girls. Correlations between indicators across constructs were weaker than those within constructs. For boys, only two significant associations between fourth-grade aggression and sixth-grade lifetime drug use were found, the associations between “destroyed] property” and two drug use items, lifetime cigarette use r = .22, p < .01) and lifetime marijuana use r = .17, p<.05). For girls, significant associations were found between “pushfed] or shove[d] other kids” and lifetime cigarette use r = . 18, p < .01) and lifetime Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 44 alcohol use r=.19,p<.01). In general, the associations were similar across gender, with boys r’s ranging from r = .002 to .22 and girls r’s ranging from .004 to .19. Relationships between fourth-grade aggression and fifth-grade social competence were also weak. Among girls, “destroy[ed] property” and “listen[ed] to a friend” were inversely associated r =-.14, p< .05). Additionally, having “call[ed] others bad names” was inversely correlated with two social competence variables, “work[ed] out a problem” r = -.13, p < .05) and “help[ed] someone learn a new skill” r = -.14, p < .05). The range of correlation coefficients was from .001 to -.14. For boys, “destroy[ed] property” was inversely associated with two indicators of social competence, “help[ed] others” r = -. 15, p < .05) and “work[ed] out a problem” r = -.15, p < .05). “Call[ed] others bad names” was also inversely correlated with “help[ed] others” among boys r = -.20, p < .01). The range of correlation coefficients for boys was from .01 to -.20. Finally, significant associations between fifth-grade social competence and sixth- grade lifetime drug use differed somewhat by gender. For boys, there were no significant correlations between drug use and social competence indicators and only one marginally significant association (“standing] up for your values” and lifetime cigarette use, r = -. 13, p < .10). The range of correlation coefficients was from -.01 to -.13. In contrast, for girls ten of the eighteen correlations were significant or marginally significant, with coefficients ranging from -.01 to -.24. Significant associations were found between lifetime cigarette use and “help[ed] others” r = -.22, p<.001), “standing] up for your values” r = -.24, p < .001), “work[ed] out a problem” r = -.16, p < .01) and “help[ed] Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 45 someone learn a new skill” r = -.19, p < .01). “Standing] up for your values” r = -.15, p < .05) and “workjed] out a problem” r = -.20, p < .001) were significantly inversely correlated with lifetime cigarette use. Finally, lifetime marijuana use was negatively significantly associated with “help[ed] others” r = -.23, p < .001) and “help[ed] someone learn a new skill” r = -.21, p<.001). Effect o f Fourth-grade Aggression on Sixth-grade Cigarette Use Measurement model and structural model: bovs Results of the measurement model showed that indicators loaded significantly on the two latent factors of aggression and social competence. With one indicator constrained to one (“[got] in physical fights”), other loadings ranged from .48 to 1.31 for the aggression latent factor. For social competence, one indicator was also constrained to one (“work[ed] out a problem”) and other loadings ranged from .80 to 1.23. Although one loading, “destroyed] property”, was considerably weaker than other factor loadings for boys, since it still loaded significantly on the latent factor, it was retained as an aggression indicator. See Table 4 for a summary of factor loadings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 46 Table 4. Estimates for factor loadings in model showing the effect of fourth-grade aggression on lifetime cigarette use in sixth grade. Factor Loading Boys (N = 187) Girls (N = 274) Latent Factor Indicator Unstd. p (SE) Std p Unstd. p (SE) Std p Aggression Got in physical fights constrained to 1 .57 constrained to 1 .43 Destroyed property* .48(.13) .38 .93(.16) .62 Pushed or shoved other kids 1.31 (.20) .81 1.45(.23) .79 Called others bad names 1.03(.17) .62 1.40(.23) .72 Left someone out of your group ,86(.15) .55 ,97(.19) .48 Social competence Listened to a friend ,94(.20) .50 .69(.ll) .45 Helped others ,80(.19) .42 .86(.12) .54 Talked with friends to find a solution to a problem 1.23(.22) .68 1.05(. 13) .64 Helped someone learn a new skill 1.21(.23) .62 1.00(.13) .64 Stood up for your values .86(. 17) .56 ,82(.ll) .61 Worked out a problem constrained to 1 .53 constrained to 1 .64 Note. Unstd. = unstandardized, Std. = standardized. All factor loadings are significant at p<.05. a Parameter differs significantly between groups at p < .05. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 47 The initial structural model of fourth-grade aggression as a predictor of sixth-grade cigarette use for boys fit the data relatively well [yl (96) = 140.03, p = .002, CFI = .90]. However, the results of the LM test suggested that adding correlations between two aggression indicators and between two social competence indicators would improve model fit significantly. Additionally, the LM test suggested adding a path from the school type demographic indicator to the latent aggression factor. This path seemed reasonable given that the distinction between private and public schools may reflect an school economic difference and low school economic status has been related to vulnerability to risk for aggression (Battistich & Horn, 1997). Inclusion of these three paths resulted in a significantly better fitting model that did not differ significantly from the underlying data structure [% 2 (93) = 115.06, p = .06, CFI = .95]. See the top of Table 5 for a summary of model development. The final structural model for boys did not support the basic hypotheses related to aggression predicting cigarette use or social competence as a mediator of this relationship, although results did suggests trends in that direction. Specifically, although the path from fourth-grade aggression to sixth-grade lifetime cigarette use was not significant, it was in the expected direction (p = .17, SE = .13). Similarly, fourth-grade aggression was inversely predictive of fifth-grade social competence (P = -.08, SE = .06) and social competence was inversely predictive of sixth-grade lifetime cigarette use (P = -.24, SE = .24). Given the lack of direct effect of aggression on later cigarette use, it was Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 48 not surprising to find that social competence did not mediate this relationship (indirect effect = .019, SE = .035). Final information on indirect effects is shown in Table 6. Table 5. Model development and test of measurement invariance for the effect of fourth-grade aggression on lifetime cigarette use in sixth grade. Group i df P CFI Boys M ( Bq o 140.03 96 .002 .90 M ( Bc )i 115.06 93 .060 .95 Girls M (GC)0 161.59 96 <001 .93 M (GC)1 135.94 93 .002 .95 Combined M O 251.00 186 .001 .95 Ml 261.49 195 .001 .95 Ml-MO 10.49 9 ns - Ml* 256.46 194 .002 .95 M l-M l* 5.03 1 <.05 - M1*-M0 5.46 8 ns - M2 258.70 197 .002 .95 M2-M1* 7.70 11 ns - Note. A subscript ending in “0" denotes a basic theoretical model and a subscript ending in “1" denotes a model modified to improve fit. An asterisk indicates a model with relevant parameters partially constrained equal across groups. MO= basic model combining groups, Ml = model with all factor loadings constrained equal across groups, M2 = model with all regression weights among factors constrained equal across groups over M l*. Boys N = 187 and girls N = 274. Measurement model and structural model: girls Results of the measurement model for girls confirmed the presence of two latent factors, aggression and social competence. With one aggression indicator constrained to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 49 one (“[got] in physical fights”), other factor loadings ranged from .93 for “destroyed] property” to 1.45 for “push[ed] and shove[d] other kids.” For social competence, one indicator, “work[ed] out a problem,” was constrained to one and other loadings ranged from .69 for “listen[ed] to a friend” to 1.05 for “talk[ed] with friends to find a solution to a problem.” See Table 4 for a summary of factor loadings. The initial structural model of fourth-grade aggression as a predictor of sixth-grade lifetime cigarette use for girls fit the data relatively well based on the CFI (.93) but the results of the chi-square test indicated that the model was significantly different from the underlying data structure [% 2 (96) = 161.59, p < .001]. Thus, two correlated errors, between “destroy[ed] property” and “[left] someone out of your group and also between “destroyed] property” and “listen[ed] to a friend”, were added to the model. Additionally, the LM test suggested adding a path from the demographic indicator for ethnicity to the latent aggression factor, a path which seemed reasonable based on prior research which has shown that nonwhite ethnicity may be correlated with higher aggression rates (Snyder & Sickmund, 1999). Inclusion of these three paths improved the model fit, although the chi-square test revealed that the model still was not statistically acceptable [y2 (93) = 135.94, p = .002, CFI = .95]. See the top of Table 5 for a summary of model development. The final model for girls provided partial support for hypotheses. Specifically, aggression in fourth grade predicted lifetime cigarette use (P = .33, SE = .16, p < .05). Additionally, social competence in fifth grade predicted lifetime cigarette use in sixth Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 50 grade and this path was highly statistically significant ((3 = -.56, SE = .18, p < .01). Although fourth-grade aggression did not significantly predict fifth-grade social competence, the relationship was in the expected direction (P = -.10, SE = .07). There was no indirect effect of fourth-grade aggression on lifetime cigarette use in sixth grade (P = .059, SE = .053). Final information on indirect effects is shown in Table 6. Table 6. Indirect effect of fourth-grade aggression on lifetime cigarette use in sixth grade mediated by fifth-grade social competence. Sample Indirect Effect P(SE) Boys .019(.035) Girls ,059(.053) Groups Combined ,041(.035) Note. No indirect effects are significant at p<.05, two-tailed test. Test of measurement invariance between bovs and girls The next question to be addressed was whether the hypothesized model, i.e., aggression as a predictor of lifetime cigarette use, differed between boys and girls. In order to compare the hypothesized model across groups, the final models for boys and girls were combined and a basic model established (M0). The basic model fit the data relatively well [yl (186) = 251.00 p = .001, CFI = .95]. Next, the nine factor loadings (that is, all factor loadings except those constrained at 1) were constrained equal between groups (Ml). Results indicated that this model did not provide a significantly better fit Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 51 of the data than the basic model did based on the difference of the chi-square values for these two models [Ml-MO % 2 (9) = 10.49, ns]. Results of the LM test suggested that one pair of factor loadings constrained equal were significantly different from one another and, therefore, a decision was made to release this constraint (Ml *). The resulting model fit the data well [% 2 (194) = 256.46 p = .002, CFI = .95] and was a significant improvement over Ml (Ml-Ml* % 2 (1) = 5.03, p<.05) but still not statistically different from MO (M1*-M0 (8) = 5.46, ns). Retaining the equality constraints on factor loadings from Ml*, the next model (M2) added constraints on all regression weights between constructs of interest (i.e., aggression, social competence and cigarette use). The resulting model fit the data well [% 2 (197) = 258.70, p = .002, CFI = .95], although it was not a significant improvement over the Ml* model (M2-M1 * % 2 (11) = 7.70, ns). The test of measurement invariance between boys and girls is shown on the lower half of Table 5 and the final structural model is shown in Figure 1. Overall relationships among aggression, social competence and lifetime cigarette use were supported. However, the final structural model of aggression as a predictor of lifetime cigarette use did not differ significantly across gender. As hypothesized, aggression in fourth grade was inversely predictive of social competence in fifth grade (p = -.10, SE = .04, p < .05) and positively associated with lifetime cigarette use in sixth grade (P = .23, SE =. 10, p < .01). Social competence was also negatively associated with lifetime cigarette use (p = -.42, SE = .14, p < .01), as postulated. Although initial inspection of parameter estimates suggested an interaction of gender on the social Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 52 competence to lifetime cigarette use and aggression to cigarette use relationships (since these individual paths were statistically significant for girls but not for boys), the test of measurement invariance indicated that these differences were not statistically significant. The sole difference between groups was the indicator related to “destroy[ed] property” which loaded significantly more strongly on aggression for boys and girls (.93 for girls vs. .48 for boys). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 1. Parameter estimates for f i n a l model of the effect of fourth-grade aggression on lifetime cigarette use in sixth grade. 4th grade 5th grade 6th grade Social Competence Life Cigarette Use Aggression .17 [.33*] Notes. Results are presented separately for boys and girls with girls’ parameter estimates shown in brackets. Multiple group comparison indicated no statistically significant differences among hypothesized paths between boys and girls. Estimates presented are unstandardized. Baseline demographics included but not shown in this model are ethnicity, socioeconomic status, intervention group exposure and school type. *p<-05. 54 Effects o f Fourth-grade Aggression on Sixth-grade Alcohol Use Measurement model and structural model: bo vs Results of the measurement model confirmed that indicators loaded significantly on the two latent factors of aggression and social competence. With one indicator constrained to one (“[got] in physical fights”), other loadings ranged from .33 to 1.22 for the aggression latent factor. For social competence, one indicator was also constrained to one (“work[ed] out a problem”) and other loadings ranged from .80 to 1.23. Although one loading, “destroyed] property”, was considerably lower than other factor loadings for boys, since it still loaded significantly on the latent factor, it was retained as an aggression indicator. See Table 7 for a summary of factor loadings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 55 Table 7. Estimates for factor loadings in model showing the effect of fourth-grade aggression on lifetime alcohol use in sixth grade. Factor Loading Boys (N = 187) Girls (N = 274) Latent Factor Indicator Unstd. P(SE) Std p Unstd. p (SE) Std p Aggression Got in physical fights constrained to 1 .59 constrained to 1 .43 Destroyed propertya .33(.ll) .27 ,92(.16) .61 Pushed or shoved other kids 1.22(.19) .77 1.47(.24) .80 Called others bad names 1.06(.17) .65 1.39(.23) .71 Left someone out of your group .87(. 15) .57 .96(.19) .48 Social competence Listened to a friend ,94(.20) .50 .64(.ll) .45 Helped others .80(.19) .42 .83(.12) .53 Talked with friends to find a solution to a problem 1.23(.22) .69 1.04(.13) .65 Helped someone learn a new skill 1.21 (.22) .62 ,98(.12) .64 Stood up for your values ,85(.17) .56 .79(.10) .61 Worked out a problem constrained to 1 .53 constrained to 1 .65 Note. Unstd. = unstandardized, Std. = standardized. All factor loadings are significant at p<05. * Parameter differs significantly between groups at p<.05. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 56 The initial structural model of aggression as a predictor oflifetime alcohol use for boys fit the data relatively well [yl (96) = 134.73, p = .006, CFI = .91]. However, the results of the LM test suggested that allowing the errors to correlate between two social competence indicators, “listen[ed] to a friend” and “help[ed] others,” would improve model fit significantly. Additionally, the LM test suggested adding a path from the demographic indicator for school type to the latent aggression factor which was considered reasonable, as mentioned previously (Battistich & Horn, 1997). Inclusion of these paths resulted in a significantly better fitting model that fit the data marginally well [% 2 (94) = 117.84, p = .049, CFI = .95]. See the top of Table 8 for a summary of model development. The final model for boys did not support the basic hypotheses of aggression predicting lifetime alcohol use and mediation of this relationship by social competence in that none of the paths among aggression, social competence and lifetime alcohol use was statistically significant. However, all relationships were in the hypothesized direction. Specifically, although the path from fourth-grade aggression to sixth-grade lifetime alcohol use was not significant, the two constructs were positively related as expected (P = .10, SE = .15). Similarly, fourth-grade aggression was inversely predictive of fifth-grade social competence (p = -.07, SE = .06). Finally, social competence was inversely predictive of sixth-grade lifetime alcohol use (P = -.04, SE = .26), although the relationship was close to zero. Given the lack of direct effect of aggression on later alcohol use, it was not surprising to find that there was no indirect effect of aggression on Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 57 alcohol use through social competence (indirect effect = .003, SE = .029). Final information on indirect effects is shown in Table 9. Table 8. Model development and test of measurement invariance for the effect of fourth-grade aggression on lifetime alcohol use in sixth grade. Group x2 df P CFI Boys M (BC)0 134.73 96 .006 .91 M (BC)1 117.84 94 .049 .95 Girls M (G C)0 162.80 96 <.001 .92 M (GC)1 136.49 93 .002 .95 Combined M O 254.33 187 <.001 .95 M l 270.46 198 <.001 .94 Ml-MO 16.13 11 ns - Ml* 260.47 197 .001 .95 Ml-M l* 9.99 1 <.01 - M1*-M0 6.14 10 ns - M2 262.17 198 .002 .95 M2-M1* 1.70 1 ns - Note. A subscript ending in “0" denotes a basic theoretical model and a subscript ending in “1" denotes a model modified to improve fit. An asterisk indicates a model with relevant parameters partially constrained equal across groups. M O = basic model combining groups, M l = model with all factor loadings constrained equal across groups, M2 = model with all regression weights among factors constrained equal across groups over M l*. Boys N = 187 and girls N = 274. Measurement model and structural model: girls Results of the measurement model for girls confirmed the presence of two latent factors, aggression and social competence. With one aggression indicator constrained to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 58 one (“[got] in physical fights”), other factor loadings ranged from .92 for “destroyed] property” to 1.47 for “push[ed] and shove[d] other kids.” For social competence, one indicator, “work[ed] out a problem,” was constrained to one and other loadings ranged from .45 for “listen[ed] to a friend” to 1.04 for “talkfed] with friends to find a solution to a problem.” See Table 7 for a summary of factor loadings. The initial structural model for girls fit the data relatively well based on the CFI (.93) but the results of the chi-square test indicated that the model was significantly different from the underlying data structure [% 2 (96) = 162.80, p < .001]. Thus, two correlated errors, between “destroyed] property” and “[left] someone out of your group” and also between “destroyed] property” and “listen[ed] to a friend”, were added to the model. Additionally, the LM test suggested adding a path from the demographic indicator for ethnicity to the latent aggression factor, a defensible relationship as previously mentioned (Snyder & Sickmund, 1999). Inclusion of these three paths to the model improved fit as shown by the CFI (.95), although the chi-square test revealed that the model still was not statistically acceptable [% 2 (93) = 136.49, p = .002]. See the top of Table 8 for a summary of model development. The final model of aggression as a predictor of lifetime alcohol use for girls provided partial support for hypotheses. Specifically, aggression in fourth grade predicted lifetime alcohol use (p = .42, SE = .14, p < .01). Additionally, social competence in fifth grade predicted lifetime alcohol use in sixth grade (p = -.30, SE = .15, p < .05). Although fourth-grade aggression did not significantly predict fifth-grade social competence, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 59 relationship was in the expected direction (P = -.11, SE = .07). There was no indirect effect of fourth-grade aggression on lifetime alcohol use in sixth grade (P = .032, SE = .032). Final information on indirect effects is shown in Table 9. Table 9. Indirect effect of fourth-grade aggression on lifetime alcohol use in sixth grade mediated by fifth-grade social competence. Sample Indirect Effect P(SE) Boys .003(.029) Girls ,032(.032) Groups Combined .021 (.024) Note. No indirect effects are statistically significant at p<.05, two-tailed test. Test of measurement invariance between bovs and girls The hypothesized model of aggression to lifetime alcohol use was next compared between boys and girls. In order to compare the hypothesized model across groups, the final models for boys and girls were combined and a basic model established (MO). The basic model fit the data relatively well [yl (187) = 254.33, p < .001, CFI = .95]. Next, the nine factor loadings (that is, all factor loadings except the two constrained to 1) were constrained equal between groups (Ml). The resulting model had a slightly worse fit than the basic model according to the CFI (.94) but was not statistically different [Ml-M0 y l (11) = 16.13, ns]. Results of the LM test at this point suggested that one pair of factor loadings constrained equal were significantly different from one another and, therefore, a decision was made to release this constraint (Ml*). The resulting model fit the well Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 60 \y2 (197) = 260.47, p = .001, CFI = .95] and was a significant improvement over Ml (M l-M l* % 2 (1) = 9.99, p < .01), but was not statistically different from MO (Ml*-M0 % 2 (10) = 6.14, ns). Retaining the equality constraints on factor loadings from Ml*, the next model (M2) added constraints on all regression weights. The resulting model had approximately the same fit as the Ml* model [% 2 (198) = 262.17, p = .002, CFI = .95, M2-Ml* % 2 (1) = 1.70, ns). The test of measurement invariance between boys and girls is shown on the lower half of Table 8 and the final structural model is shown in Figure 2 . The final overall model generally provided support for relationships among aggression, social competence and lifetime alcohol use. However, there was no evidence of interactions of these paths with gender. As hypothesized, aggression in fourth grade was inversely predictive of social competence in fifth grade (p = -.09, SE = .04, p < .05) and positively associated with lifetime alcohol use in sixth grade (p = .29, SE = .10, p < .01). There was marginal support for an inverse predictive relationship between social competence and lifetime alcohol use (P = -.22, SE = .13, p < .10). Although initial inspection of parameter estimates suggested group differences on the aggression and social competence to lifetime alcohol use relationships (since these paths were statistically significant for girls but not for boys), the test of measurement invariance indicated that these differences were not statistically significant. The only difference between the male and female models was the indicator for destruction of property which loaded more strongly on aggression for girls than for boys (.93 for girls vs. .48 for boys). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 2. Parameter estimates for final model of the effect of fourth-grade aggression on lifetime alcohol use in sixth grade. 4th grade 5th grade 6th grade Social C om petence L ife A lcohol U se A ggression .10 [.42*] Notes. Results are presented separately for boys and girls with girls’ parameter estimates shown in brackets. Results of multiple group comparison of boys and girls indicated no differences for hypothesized paths between groups. Estimates presented are unstandardized. Baseline demographics included but not shown in this model are ethnicity, socioeconomic status, intervention group exposure and school type. *p<05 os 62 Effect o f Fourth-grade Aggression on Sixth-grade Marijuana Use Measurement model and structural model: boys The measurement model confirmed that indicators loaded significantly on the two latent factors of aggression and social competence. With “[got] in physical fights” constrained to one, other factor loadings for aggression ranged from .34 to 1.22. For social competence, factor loadings ranged from .81 to 1.24 with the “work[ed] out a problem” indicator constrained to one. Although the “destroy[ed] property” indicator was considerably lower than other factor loadings for boys at .34, since it still loaded significantly on the latent factor, it was retained as an aggression indicator. See Table 10 for a summary of factor loadings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 63 Table 10. Estimates for factor loadings in model showing the effect of fourth- grade aggression on lifetime marijuana use in sixth grade. Factor Loading Boys (N = 187) Girls (N = 274) Latent Factor Indicator Unstd. p(SE) Std p Unstd. p (SE) Std p Aggression Got in physical fights constrained to 1 .57 constrained to 1 .44 Destroyed property •34(.ll) .27 ,94(.16) .64 Pushed or shoved other kids 1.22(.19) .76 1.38(.22) .76 Called others bad names 1.11(. 18) .67 1.37(.22) .72 Left someone out of your group .89(.16) .57 1.03(.19) .52 Social competence Listened to a friend .94(.20) .42 ,68(.ll) .45 Helped others .81(. 19) .50 .87(.12) .54 Talked with friends to find a solution to a problem 1.24(.22) .69 1.06(.13) .65 Helped someone leam a new skill 1.21 (.22) .62 1.02(.13) .65 Stood up for your values .86(.17) .56 •Bl(.ll) .60 Worked out a problem constrained to 1 .53 constrained to 1 .63 Note. Unstd. = unstandardized, Std. = standardized. All factor loadings are significant atp<.05. The initial structural model of aggression predicting lifetime marijuana use for boys fit the data relatively well [% 2 (96) = 135.23, p = .005, CFI = .91]. However, the results of the LM test suggested that allowing the errors to correlate between two social Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 64 competence indicators, “listen[ed] to a friend” and “help[ed] others,” would improve model fit significantly. Inclusion of this correlated error term resulted in a significantly better fitting model that fit the data well [% 2 (95) = 125.50, p = .020, CFI = .93]. See the top of Table 11 for a summary of model development. Despite an acceptable fit with the underlying data, the final model for boys showed neither a main effect of aggression on marijuana use nor a mediated effect through social competence. The path from fourth-grade aggression to sixth-grade lifetime marijuana use was close to zero in magnitude (p = -.04, SE = .10). The paths from fourth-grade aggression to fifth-grade social competence (P = -.08, SE = .06) and from fifth-grade social competence to sixth-grade lifetime marijuana use (P = -. 12, SE =. 18) were stronger and in the hypothesized direction, but still were not statistically significant. There was also no mediational effect of social competence on the aggression-lifetime marijuana use relationship (P = .010, SE = .026). Final information on indirect effects is shown in Table 12. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 65 Table 11. Model development and test of measurement invariance for the effect of fourth-grade aggression on lifetime marijuana use in sixth grade. Group x2 df P CFI Boys M (B C )0 135.23 96 .005 .91 M(B qi 125.50 95 .020 .93 Girls M(G C )o 159.86 96 <001 .92 Combined MO 285.37 191 <001 .93 M l 301.77 200 <001 .92 Ml-MO 16.41 9 <10 - M l* 289.29 199 <001 .93 Ml-M l* 12.48 1 <01 - M1*-M0 3.92 8 ns - M2 289.78 202 <001 .93 M2-M1* .49 3 ns - Note. A subscript ending in “0" denotes a basic theoretical model and a subscript ending in “1" denotes a model modified to improve fit. An asterisk indicates a model with relevant parameters partially constrained equal across groups. MO= basic model combining groups, M l = model with all factor loadings constrained equal across groups, M2 = model with all regression weights among factors constrained equal across groups over Ml*. Boys N = 187 and girls N = 274. Measurement model and structural model: girls Results of the measurement model for girls confirmed the presence of two latent factors, aggression and social competence. With one aggression indicator constrained to one (“[got] in physical fights”), other factor loadings ranged from .94 for “destroyed] property” to 1.38 for “push[ed] and shove[d] other kids.” For social competence, one Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 66 indicator, “work[ed] out a problem,” was constrained to one and other loadings ranged from .68 for “listen[ed] to a friend” to 1.06 for “talk[ed] with friends to find a solution to a problem.” See Table 10 for a summary of factor loadings. The initial structural model for girls of the aggression to lifetime marijuana use relationship fit the data relatively well based on the CFI (.92), but the results of the chi- square test indicated that the model was significantly different from the underlying data structure [% 2 (96) = 159.86, p < .001]. However, the paths that the LM test identified as those which would reduce the chi-square value the most were not theoreticallyreasonable. Therefore, a decision was made to retain this model and make no modifications to it. See the top of Table 11 for a summary of model development. The final model for girls was similar to that of boys in that there was no main effect of fourth-grade aggression on sixth-grade lifetime marijuana use (P = -.02, SE = .06). As in the boys’ sample, the path from fourth-grade aggression to fifth-grade social competence (P = -.10, SE = .07) was not statistically significant, although it was in the hypothesized direction. In contrast with the boys’ sample, fifth-grade social competence did inversely predict sixth-grade lifetime marijuana use for girls and this relationship was highly statistically significant (p = -.21, SE = .08, p<.01). Given the weak relationships between aggression and social competence and aggression and lifetime marijuana use, it was unsurprising that there was no indirect effect of fourth-grade aggression on sixth- grade lifetime marijuana use through fifth-grade social competence. Final information on indirect effects is shown in Table 12. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 67 Table 12. Indirect effect of fourth-grade aggression on lifetime marijuana use in sixth grade mediated by fifth-grade social competence. Sample Indirect Effect P(SE) Boys .010(.026) Girls .021(.020) Groups Combined - Note. No indirect effects are statistically significant at p<.05, two-tailed test. No multiple group comparison of models was conducted. Test of measurement invariance between bovs and girls The hypothesized model of aggression to lifetime marijuana use was next compared between boys and girls. In order to compare the hypothesized model across groups, the final models for boys and girls were combined and a basic model established (MO). The basic model fit the data relatively well [% 2 (191) = 285.37, p < .001, CFI = .93]. Next, the nine factor loadings (that is, all factor loadings except the two constrained to 1) were constrained equal between groups (Ml). The resulting model had a slightly worse fit than the basic model according to the CFI (.92) but was not statistically different [M1 -M O % 2 (9) = 16.41, ns]. Results of the LM test indicated that one equality constraint, on the “destroy[ed] property” factor loading, should be released (Ml*). This was done and the resulting model [% 2 (199) = 289.29, p < .001, CFI = .95] and was a significant improvement over Ml (Ml-Ml* y l (1) = 12.48, p < .01), but was not statistically different from M0 (Ml*-M0 % 2 (8) = 3.92, ns). Retaining the equality constraints on factor loadings from Ml*, the next model (M2) added constraints on all regression Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 68 weights. The resulting model had approximately the same fit as the Ml* model [% 2 (202) = 289.78, p < .001, CFI - .93, M2-M1* % 2 (3) = .49, ns). The test of measurement invariance between boys and girls is shown on the lower half of Table 11 and the final structural model is shown in Figure 3. The overall model did not support the primary hypothesis of fourth-grade aggression as a predictor of sixth-grade marijuana use (P = -.03, SE = .05, ns). Moreover, there was no support for a gender interaction among constructs. The sole interaction was on one factor loading, “destroyfed] property” which loaded more strongly on the aggression construct for girls than for boys. Consistent with expectations, there was a predictive relationship from fourth-grade aggression to fifth-grade social competence (P = -.10, SE = .04, p <.05) and from fifth-grade social competence to sixth-grade lifetime marijuana use (P = -.20, SE = .07, p < .01). However, the hypothesis that social competence would mediate the aggression-lifetime marijuana use relationship was not supported (P — -.019, SE = .017, ns). The final structural model is shown in Figure 3. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 3 . Parameter estimates for final model of the effect of fourth-grade aggression on lifetime marijuana use in sixth grade. 4th grade 5th grade 6th grade Social C om petence Life M ariiuana Use A ggression -,04(-.02) Notes. Estimates are presented separately for boys and girls with girls’ estimates shown in parentheses. Since there was no main effect of aggression on marijuana use for either gender, a multiple group comparison of the model was not performed. Estimates presented are unstandardized. Baseline demographics included but not shown in this model are ethnicity, socioeconomic status, intervention group exposure and school type. * p<.05. Os SO 70 Effect o f Fourth-grade Relational and Overt Aggression on Sixth-grade Cigarette Use Measurement model and structural model: bovs Results of the measurement model showed that indicators loaded significantly on the two latent factors of aggression and social competence. With the variance of the latent factor constrained to one, loadings of the four overt aggression indicators ranged from .23 to .66. For social competence, one indicator was constrained to one (“listen[ed] to a friend”) and other loadings ranged from .86 to 1.32. Although one loading, “destroy[ed] property,” was considerably weaker than others, since it still loaded significantly on the overt aggression latent factor, it was retained as an indicator. SeeT able 13 fora summary of factor loadings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 71 Table 13. Estimates for factor loadings in model showing the effect of fourth- grade overt and relational aggression on lifetime cigarette use in sixth grade. Factor Loading Boys (N = 187) Girls (N = 274) Latent Factor Indicator Unstd. p(SE) Std p Unstd. p (SE) Std p Overt Aggression Got in physical fights .67(.09) .58 .53(.08) .44 Destroyed propertya .23(.07) .28 .50(.05) .65 Pushed or shoved other kids .83(.08) .76 .73(.06) .77 Called others bad names .74(.09) .66 .72(.06) .72 Social competence Listened to a friend constrained to 1 .50 constrained to 1 .45 Helped others .86(.18) .42 1.26(.22) .54 Talked with friends to find a solution to a problem 1.32(.25) .69 1.54(.25) .64 Helped someone learn a new skill 1.29(.25) .62 1.47(.24) .64 Stood up for your values ,91(,19) .56 1.20(.20) .61 Worked out a problem 1.06(.23) .53 1.47(.24) .64 Note. Unstd. = unstandardized, Std. = standardized. All factor loadings are significant atp<.05. a Parameter differs significantly between groups at p<.05. The structural model was specified such that both overt and relational aggression in fourth grade were predictors of fifth-grade social competence and sixth-grade lifetime cigarette use. Relational and overt aggression were included in the model simultaneously as past research has shown that they may co-occur (Crick, 1996). Furthermore, this Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 72 approach permitted examining the effects of one type of aggression on cigarette use, while controlling for effects of the other. In addition, the two aggression factors were allowed to correlate since past research has shown a moderate association between overt and relational aggression (Crick & Gropeter, 1995). As with previous aggression models, fifth-grade social competence was also modeled as a predictor of sixth-grade lifetime cigarette use. The initial structural model for boys fit the data relatively well [x2 (94) - 139.04, p = .001, CFI - .90]. However, based on the LM test results a decision was made to add a correlated measurement error between two social competence indicators, “listenfed] to a friend” and “help[ed] others.” Addition of this path improved model fit significantly although the fit, while good, was still not statistically acceptable {yl (93) = 129.21, p = .01, CFI = .92]. See the top of Table 14 for a summary of model development. The final model for boys did not support hypotheses related to overt and relational aggression differentially predicting cigarette use, and social competence mediating this relationship. In fact, none of the hypothesized paths was statistically significant, although most were in the expected direction. The exception was relational aggression which was inversely predictive of lifetime cigarette use (f3 - -.04, SE = .10), although the magnitude of this relationship was almost zero and the inverse relationship may have been due to chance. The lack of significant indirect effects for the overt aggression-lifetime cigarette use (P = .007, SE = .22) and the relational aggression-lifetime cigarette use relationships (P = .010, SE = .022) is consistent with the finding of no significant associations among Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 73 overt and relational aggression, social competence and lifetime cigarette use. See Table 15 for complete information on indirect effects. Table 14. Model development and test of measurement invariance for the effect of fourth-grade overt and relational aggression on sixth-grade cigarette use. Group x2 df P CFI Boys M (B C )0 139.04 94 .001 .90 M (BC)1 129.21 93 .01 .92 Girls M (G C )0 161.38 94 <.001 .92 M (GC)1 153.57 93 <.001 .93 Combined MO 282.78 186 <.001 .93 Ml 299.06 195 <.001 .92 Ml-MO 16.28 9 < 10 - M l* 288.30 194 <001 .93 M l-M l* 10.76 1 <01 - M1*-M0 5.52 8 ns - M2 291.44 200 <001 .93 M2-M1* 8.66 14 ns - Note. A subscript ending in “0" denotes a basic theoretical model and a subscript ending in “1" denotes a model modified to improve fit. An asterisk indicates a model with relevant parameters partially constrained equal across groups. M O = basic model combining groups, M l = model with all factor loadings constrained equal across groups, M2 = model with all regression weights among factors constrained equal across groups over M l*. Boys N = 187 and girls N = 274. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Measurement model and structural model: girls 74 Results of the measurement model for girls confirmed the presence of two latent factors, aggression and social competence and all factor loadings were statistically significant. With the variance of the latent factor constrained to one for model identification purposes, factor loadings ranged from .50 for “destroy[ed] property” to .73 for “push[ed] and shove[d] other kids.” For social competence, one indicator, “listen[ed] to a friend,” was constrained to one and other loadings ranged from 1.20 for the “stood up for your values” indicator to 1.54 for “talk[ed] with friends to find a solution to a problem.” See Table 13 for a summary of factor loadings. The initial structural model of overt and relational aggression as lifetime cigarette use predictors for girls fit the data adequately based on the CFI (.92) hut the results of the chi-square test indicated that the model was significantly different from the underlying data structure [% 2 (94) = 161.38,p < .001]. Thus, a path allowing correlated measurement error between two indicators, “destroy[ed] property” and “listen[ed] to a friend,” was added to the model to improve fit. Addition of this path significantly improved the model fit, although the model still was not statistically acceptable [% 2 (93) = 153.57, p < .001, CFI = .93]. See the top of Table 14 for a summary of model development. Hypotheses related to fourth-grade relational aggression as a stronger predictor of sixth-grade lifetime cigarette use among girls compared to fourth-grade overt aggression were not supported. In fact, relational aggression had almost no relationship with either social competence (P = -.01, SE = .02) or lifetime cigarette use (P = -.01, SE = .08). In Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 75 contrast, overt aggression did have a marginal predictive effect on lifetime cigarette use (P = .17, SE = .09, p < .10), suggesting that overt aggression may be the more potent precursor to cigarette use for girls. Similar to relational aggression, overt aggression (P = -.03, SE = .03) did not predict social competence. However, social competence was inversely predictive of lifetime cigarette use (P = -.82, SE = .28, p < .01). There were no indirect effects of either relational aggression (P = 009, SE = .026) or overt aggression (p = .023, SE = .031) on lifetime cigarette use through social competence. Final information on indirect effects is shown in Table 15. Table 15. Indirect effect of fourth-grade overt and relational aggression on lifetime cigarette use in sixth grade mediated by fifth-grade social competence. Path Indirect Effect P(SE) Overt aggression->cigarette use Boys .007(.022) Girls ,023(.031) Groups Combined .018(.023) Relational aggression->cigarette use Boys .010(.022) Girls .009(.026) Groups Combined .009(.019) Note. No indirect effects are statistically significant at p<.05, two-tailed test. Test of measurement invariance between bo vs and girls In order to test the hypotheses that overt aggression was a stronger predictor of lifetime cigarette use for boys and relational aggression a stronger predictor for girls, a test Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 76 of measurement invariance of the model across groups (i.e., boys and girls) was conducted (Pentz & Chou, 1994). First, a basic model (MO) which combined both groups was established. This model fit the data relatively well [% 2 (186) = 282.78, p < .001, CFI = .93]. Next, the nine factor loadings (that is, all factor loadings except “listen[ed] to a friend” which was constrained to one for identification purposes) were constrained equal between groups (Ml). The resulting model had a j l = 299.06 with 195 degrees of freedom and a slightly lower CFI (.92) than the MO model. Results of the LM test suggested that one pair of factor loadings constrained equal (“destroyfed] property”) were significantly different from one another and, therefore, a decision was made to release this constraint (Ml*). The resulting model fit was a significant improvement over Ml (Ml- Ml* % 2 (1) = 10.76, p < .01) but still not statistically different from M0 (Ml*-M0 % 2 (8) = 5.52, ns). Retaining the equality constraints on factor loadings from Ml*, the next model (M2) added constraints on all regression weights between constructs of interest (i.e., aggression, social competence and cigarette use) as well as on the covariance between overt and relational aggression. The resulting model fit the data well [yl (200) = 291.44, p < .001, CFI = .93], but it was not a significant improvement over the Ml* model (M2-Ml* % 2 (14) = 8.66, ns). The test of measurement invariance between boys and girls is shown on the lower half of Table 14 and the final structural model is shown in Figure 4. The final overall model of fourth-grade overt and relational aggression as predictors of sixth-grade lifetime cigarette use provided limited support for hypotheses. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 77 There was no effect of relational aggression on either social competence (P = -.02, SE = .02, ns) or lifetime cigarette use (p = -.03, SE = .06, ns), as had been hypothesized. Overt aggression was marginally predictive of lifetime cigarette use (P = .14, SE = .07, p <. 10) but had no relationship on social competence (P = -.03, SE = .03, ns). Social competence was inversely predictive of lifetime cigarette use and this relationship was highly significant (p = -.57, SE = .19, p < .01). Results also provided no support for hypotheses related to differences in the strength of these paths between genders. In fact, the only gender difference between the models was the indicator related to destruction of property which had a stronger loading for girls than for boys on the overt aggression factor (.93 for girls vs. .48 for boys). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 4. Parameter estimates for final model of the effect of fourth-grade overt and relational aggression on sixth-grade lifetime cigarette use. 4th grade 5th grade 6th grade * i n i n * o\ in Social Competence Overt Aggression Relational Life Cigarette Aggression -,04[-.01] Use Notes. Estimates presented are unstandardized. Results of multiple group comparison of boys and girls indicated no differences. Circle refers to construct represented by measured items; square refers to single-item indicator. + p<.10, *p<.05. -j 79 Effect o f Fourth-grade Relational and Overt Aggression on Sixth-grade Alcohol Use Measurement model and structural model: bovs Results of the measurement models showed that indicators loaded significantly on the two latent factors of aggression and social competence. With one indicator (“[got] in physical fights”) constrained to one for model identification purposes, loadings of the other three overt aggression indicators ranged from .34 to 1.24. For social competence, one indicator was constrained to one (“work[ed] out a problem”) and other loadings ranged from .80 to 1.24. Although one loading, “destroyed] property,” was considerably weaker than others, since it still loaded significantly on the latent factor, it was retained as an indicator of overt aggression. See Table 16 for a summary of factor loadings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 80 Table 16. Estimates for factor loadings in model showing the effect of fourth-grade overt and relational aggression on lifetime alcohol use in sixth grade. Factor Loading Boys (N = 187) Girls (N = 274) Latent Factor Indicator Unstd. p(SE) Std p Unstd. p (SE) Std p Overt Aggression Got in physical fights constrained to 1 .58 constrained to 1 .44 Destroyed property8 .34(.ll) .27 .95(.16) .64 Pushed or shoved other kids 1.24(,19) .75 1.39(.22) .77 Called others bad names 1.11(.18) .67 1.36(.22) .71 Social competence Listened to a friend .94(.20) .50 .67(.ll) .45 Helped others ,80(. 19) .42 ,83(.12) .53 Talked with friends to find a solution to a problem 1.24(.22) .69 1.04(. 13) .65 Helped someone leam a new skill 1.21(.22) .62 .98(.12) .64 Stood up for your values .85(.17) .56 .80(. 10) .61 Worked out a problem constrained to 1 .53 constrained to 1 .65 Note. Unstd. = unstandardized, Std. = standardized. All factor loadings are significant at p<.05. a Parameter differs significantly between groups at p<.05. The structural model was specified such that both overt and relational aggression were correlated in fourth grade and were predictors of fifth-grade social competence and sixth-grade lifetime alcohol use. Aggression subtypes were allowed to correlate since both aggression subtypes may occur together (Crick, 1996). As with previous aggression Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 81 models, fifth-grade social competence was also modeled as a predictor of sixth-grade lifetime alcohol use. The initial structural model for boys fit the data relatively well [% 2 (94) = 133.07, p = .005, CFI = .91]. However, based on the LM test results a decision was made to add a correlated measurement error between two social competence indicators, “listen[ed] to a friend” and “help[ed] others.” Addition of this path improved model fit significantly [yl (93) = 123.29, p = .02, CFI = .93]. See the top of Table 17 for a summary of model development. The final model for boys did not support the hypotheses related to overt and relational aggression differentially predicting alcohol use and social competence mediating this relationship. Specifically, although overt aggression in fourth grade positively predicted sixth-grade lifetime alcohol use, as hypothesized, the result was not statistically significant (P = .21, SE =. 19). In contrast, fourth-grade relational aggression was negatively predictive of lifetime alcohol use in sixth grade ((3 = -.10, SE = .10), the opposite of the hypothesized direction, although this relationship was not statistically significant and may have been due to chance. Results further showed no relationship between either overt or relational aggression and social competence ((3 = -.04, SE = .07, P = -.04, SE = .04, respectively) or social competence and lifetime alcohol use (P = -.06, SE = .19), although the regression weights were in the hypothesized inverse direction for all of these associations. Given the lack of significant associations among overt and relational aggression, social competence and lifetime alcohol use, the lack of significant Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 82 indirect effects for the overt aggression-lifetime alcohol use (P = .003, SE = .017 ) and the relational aggression-lifetime alcohol use (P = .002SE = .016) were anticipated. See Table 18 for complete information on indirect effects. Table 17. Model development and test of measurement invariance for the effect of fourth-grade overt and relational aggression on lifetime alcohol use in sixth grade. Group x2 df P CFI Boys M (B C )0 133.07 94 .005 .91 M (BC)1 123.29 93 .02 .93 Girls M (G C )0 162.58 94 <.001 .92 M (gc)i 154.49 93 <.001 .93 Combined M0 277.79 186 <.001 .93 Ml 293.92 194 <.001 .92 Ml-M0 16.13 8 <05 - Ml* 281.74 193 <001 .93 M l-M l* 12.18 1 <001 - M1*-M0 3.95 7 ns - M2 286.31 199 <001 .93 M2-M1* 4.57 6 ns - Note. A subscript ending in “0" denotes a basic theoretical model and a subscript ending in “1" denotes a model modified to improve fit. An asterisk indicates a model with relevant parameters partially constrained equal across groups. M0 = basic model combining groups, Ml = model with all factor loadings constrained equal across groups, M2 = model with all regression weights among factors constrained equal across groups over M l*. Boys N = 187 and girls N = 274. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Measurement model and structural model: girls 83 Results of the measurement model showed that indicators loaded significantly on the two latent factors of aggression and social competence. With one indicator (“[got] in physical fights”) constrained to one for model identification purposes, loadings of the three remaining overt aggression indicators ranged from .95 to 1.39. For social competence, one indicator was constrained to one (“work[ed] out a problem”) and other loadings ranged from .67 to 1.04. See Table 16 for a summary of factor loadings. The initial structural model of relational and overt aggression as predictors of lifetime alcohol use for girls fit the data adequately based on the CFI (.92) but the results of the chi-square test indicated that the model was significantly different from the underlying data structure [yl (94) = 162.58, p < .001]. Thus, a path allowing correlated measurement error between two indicators, “destroyed] property” and “listen[ed] to a friend,” was added to the model to improve fit. Addition of this path improved the model, although the fit was still significantly different from the data based on the chi-square test [x2 (93) = 154.49, p < .001, CFI = .93]. See the top of Table 17 for a summary of model development. Results provided partial support for the main hypotheses among girls. Fourth- grade overt aggression significantly predicted sixth-grade lifetime alcohol use (P = .40, SE = .16, p<.05), as anticipated. However, there was no support for fourth-grade relational aggression as a predictor of sixth-grade lifetime alcohol use, with results showing a zero path between these two constructs (P = -.000, SE = .07). Thus, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 84 hypothesis that relational aggression would be a stronger predictor oflifetime alcohol use for girls than would overt aggression was clearly not supported. Neither overt nor relational aggression predicted social competence (p = -.08, SE = .08 and p = -.02, SE = .04, respectively), although social competence did inversely predict lifetime alcohol use (P = -.30, SE - .15, p < .05), as postulated. There were no indirect effects of either relational aggression (P = .005, SE = .014) or overt aggression (p = .024, SE = .034) on lifetime alcohol use through social competence. Final information on indirect effects is shown in Table 18. Table 18. Indirect effect of fourth-grade overt and relational aggression on lifetime alcohol use in sixth grade mediated by fifth-grade social competence. Path Indirect Effect P(SE) Overt aggression->cigarette use Boys .003(.017) Girls ,024(.034) Groups Combined ,016(.023( Relational aggression->cigarette use Boys .002(.016) Girls .005(.014) Groups Combined ,005(.011) Note. No indirect effects are statistically significant at p<.05, two-tailed test. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 85 Test of measurement invariance between bovs and girls In order to test the hypotheses related to gender differences in the overt and relational aggression-lifetime alcohol use model, the measurement invariance test was conducted. First, a basic model (MO) which combined both groups was established. This model fit the data relatively well [% 2 (186) = 277.79, p < .001, CFI = .93]. Next, the eight factor loadings (all factor loadings except “[got] in physical fights” and “work[ed] out a problem” which were constrained to one) were constrained equal between groups (Ml). The resulting model had a % 2 of293.92 with 194 degrees of freedom and a slightly lower CFI (.92) than the MO model. Results of the LM test suggested that one pair of factor loadings constrained equal (“destroy[ed] property”) were significantly different from one another and, therefore, a decision was made to release this constraint (Ml*). The resulting model fit was a significant improvement over Ml (Ml-Ml* yl (1) = 12.18, p < .01) but still not statistically different from MO (M1*-M0 y l (7) = 3.95, ns). Retaining the equality constraints on factor loadings from Ml*, the next model (M2) added constraints on all regression weights between constructs of interest (i.e., aggression, social competence and alcohol use) as well as on the covariance between overt and relational aggression. The resulting model fit the data well [% 2 (199) = 286.31, p < .001, CFI = .93], although it was not a significant improvement over the Ml * model (M2-Ml* % 2 (6) = 4.57, ns). The test of measurement invariance between boys and girls is shown on the lower half of Table 17 and the final structural model is shown in Figure 5. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 86 The final overall model of fourth-grade overt and relational aggression as predictors of sixth-grade lifetime alcohol use provided partial support for hypotheses. Overt aggression predicted of lifetime alcohol use (p = .32, SE = .12, p < .01) but had no relationship on social competence (P - -.07, SE = .06, ns). Social competence was inversely predictive of lifetime alcohol use but this relationship was only marginally significant (P = -.23, SE = .13, p < .10). In contrast, there was no effect of relational aggression on either social competence (P = -.02, SE = .03, ns) or lifetime alcohol use (P = -.03, SE = .06, ns). Results also provided no support for hypotheses related to differences in the strength of these paths between genders. In fact, the only gender difference between the models was the indicator related to destruction of property which loaded more strongly on overt aggression for girls (.93 for girls vs. .48 for boys). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 5. Parameters estimates for final model of the effect of fourth-grade overt and relational aggression on sixth-grade lifetime alcohol use. 4th grade 5th g r a d e 6th grade * o * O Social Competence Overt Aggression Relational Life alcohol Aggression -,10[-.000] Use Notes. Estimates presented are unstandardized. Results of multiple group comparison of boys and girls indicated no differences. Circle refers to construct represented by measured items; square refers to single-item indicator. + p<.10, *p<.05. oo 88 Effect o f Fourth-grade Relational and Overt Aggression on Sixth-grade Marijuana Use Since there was no main effect of aggression on marijuana use, this model was not estimated. Study Two: The Cross-sectional Study Descriptive Analyses The means, standard deviations, and intercorrelations for the thirteen measured variables in the cross-sectional study (excluding demographic characteristics) are summarized in Table 19 for boys and girls. Means for boys were examined first. Among boys, the relational aggression summed composite score average was 8.81 (SD = 4.24) out of a possible score of 25 while the relational victimization composite, which had the same scale, had a slightly higher mean of 9.21 (SD = 4.62). Boys’ overt aggression and overt victimization summed scores showed the opposite pattem-out of a possible high score of 16, overt aggression was more frequently reported with a mean of 8.04 (SD = 3.11) compared to overt victimization with a mean of 7.73 (SD = 3.16). Highest mean frequency of lifetime drug use for boys was cigarette use, with subjects reporting having used an average of 1.57 (SD = 9.23) cigarettes in their lifetime. Alcohol use had the second highest mean frequency with male subjects reporting having ever had 1.24 (SD = 6.81) alcoholic drinks, and marijuana use had the lowest mean frequency for boys (M = .74, SD = 5.86). “Help[ed] others” was the social competence behavior with the highest frequency in the past year (M = 3.12, SD = 1.83) and “talked with friends to find a solution to a problem” the lowest (M = 2.39, SD = 1.90) among boys. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 89 Patterns of aggression, victimization and drug use among girls were similar to those of boys while social competence behaviors differed slightly. As with the boys, the relational aggression summed composite score was lower (M = 7.99, SD = 3.58 out of a possible score of 25) than relational victimization composite score (M = 9.34, SD = 4.68). Girls’ overt aggression and overt victimization summed scores were similar to those of boys-out of a possible high score of 16, overt aggression was more frequently reported with a mean of 7.16 (SD = 2.81) compared to overt victimization with a mean of 6.85 (SD = 2.67). Mean lifetime drug use frequency among girls was highest for cigarette use (M = 2.52, SD = 13.40), with alcohol use second most frequent (M = 1.23, SD = 7.67) and marijuana use the least frequently used (M = .91, SD = 7.59), the same ordering shown among boys. The most frequently used social competence behaviors in the past year on average were different for girls than for boys. Girls reported having “listen[ed] to a friend” the most frequently (M = 3.31, SD = 1.70) and having “work[ed] out a problem” the least (M = 2.54, SD = 1.68). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Table 19. Correlation of indicators used in the cross-sectional study by gender. M SD A1 A2 VI V2 D1 D2 D3 SI S2 S3 S4 S5 S6 M - - 7.99 7.16 9.34 6.85 2.52 1.23 .91 3.31 3.29 3.32 2.71 2.54 2.62 SD - - 3.58 2.81 4.68 2.67 13.40 7.67 7.59 1.70 1.70 1.68 1.74 1.68 1.72 A1 8.81 4.24 - .45 .38 .26 .23 .22 .06 -.04 -.06 .01 .001 -.05 -.01 A2 8.04 3.11 .46 - .21 .39 .27 .22 .22 .10 .09 .13 .02 -.10 .04 VI 9.21 4.62 .36 .22 - .56 .12 .11 .06 .06 .06 .02 .07 .004 .09 V2 7.73 3.16 .27 .39 .58 - .07 .07 .01 .11 .10+ .07 .08 -.002 .10 D1 1.57 9.23 .23 .27 .08 .14 - .47 .53 -.03 -.13 -.05 -.13 -.11 -.02 D2 1.24 6.81 .29 .32 .14 .16 .43 - .35 0.08 .01 .001 -.08 -.03 .03 D3 .74 5.86 .20 .13 .16 .002 .41 .30 - -.08 -.16 -.12 -.13 -.17 -.11 SI 2.85 1.90 -.04 .0002 .14 .13 -.07 .01 -.08 - .40 .32 .34 .34 .35 S2 3.12 1.83 .11 .12 .10+ .20 -.08 .04 -.15 .32 - .44 .36 .33 .48 S3 3.11 1.84 -.002 .08 .09 .11 -.09 .04 -.06 .40 .46 - .46 .39 .47 S4 2.39 1.90 .02 .03 .08 .20 -.04 .03 -.05 .30 .36 .47 - .56 .46 S5 2.40 1.85 -.001 .03 .05 .14 -.07 -.03 -.04 .31 .38 .47 .63 - .48 S6 2.68 1.88 .08 .13 .14 .15 .01 .03 -.005 .37 .45 .48 .49 .51 - A2 = overt aggression summed composite, VI = relational victimization summed composite, V2 =overt aggression summed composite, D1 - life cigarette use, D2 - life alcohol use, D3 = life marijuana use, SI = listened, S2 = helped, S3 = stood up for values, S4 = talked to find a solution to a problem, S5 = worked out problem, S6 = helped kid learn a new skill. Estimates > =.10 or <= -10 are significant and p < .05. so o 91 Correlations o f Indicators Table 19 shows correlations among measured variables used in Study Two by gender. Inspection of correlations revealed highly significant associations (p < .001) among indicators of the same latent construct. For both genders, overt and relational aggression were correlated (boys’ r = .46; girls’ r = .45) as were overt and relational victimization (boys’ r = .58; girls r = .56), which had slightly higher associations than did aggression indicators. Lifetime cigarette, alcohol and marijuana use were all associated for both sexes, although relationships were slightly weaker in boys (range from r = .30 to r = .43) compared to girls (range from r = .35 to r = .53). Correlations among the six social competence correlations were of similar moderate magnitude with a range for boys from .30 to .63 and for girls from .32 to .56 Correlations among aggression, victimization and drug use indicators were moderate for both sexes. For boys, both overt and relational aggression were significantly correlated with each of the lifetime drug use items, with r’s ranging from .13 to .32. Among girls, overt and relational aggression were also associated with all lifetime drug use indicators, with the sole exception of the relational aggression-lifetime marijuana use relationship r = .06) which was not significant. Correlations among aggression and victimization items were all above .20 and highly statistically significant for both sexes. In general, victimization and drug use were more highly associated for boys than for girls. For boys, four of the six correlations were statistically significant, the exceptions being the relational victimization and lifetime cigarette use r = .08) and overt victimization and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 92 lifetime marijuana use r = .002) associations. In contrast, only two of these same six correlations, the overt victimization and lifetime cigarette use r = .12) and the overt victimization and lifetime alcohol use r = .11) associations were statistically significant for girls. Correlations of social competence to other constructs were of small magnitude and often not in the hypothesized direction. For boys, relational and overt aggression were significantly related to only one social competence behavior, “help[ed] others” (r’s = .11 and. 12, respectively) and these associations were not in the expected direction. For girls, only one social competence indicator, “work[ed] out a problem,” was significantly inversely associated with overt aggression as predicted-the three other significant correlations were not in the hypothesized direction. Overt victimization for boys was significantly positively correlated with each social competence behavior, the opposite of the hypothesized relationship. Relational victimization for boys was also positively associated significantly or marginally with three of six social competence indicators, again not in the postulated direction. For girls, there were no significant associations between relational victimization and social competence indicators. Overt victimization was significantly positively associated with four of the six social competence indicators, the inverse of hypothesized relationships. Correlations of drug use with social competence for boys were generally negative, as hypothesized, although only one of these associations, between “help[ed] others” and lifetime marijuana use r = -.15), was statistically significant. For girls, social competence also had primarily negative Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 93 associations with drag use and over half of these associations were statistically significant, suggesting a stronger relationship between drag use and social competence for girls than for boys. Effect o f Aggression and Victimization on Lifetime Cigarette Use Among Sixth-Grade Students Measurement model and structural model: bo vs Results of the measurement model showed that indicators loaded significantly on the three latent factors of aggression, victimization and social competence. With one factor loading (relational victimization summed score) constrained to one for model identification purposes, the overt victimization summed score indicator loaded significantly ((3 = 1.18, SE = .09) on the victimization factor. Similarly, with the relational aggression summed score factor loading constrained to one, the overt aggression summed score indicator loaded significantly on the latent aggression factor (|3 = .98, SE = .15). For social competence, one loading was constrained to one (“talk[ed] with friends to find a solution to a problem”) and other loadings ranged from .80 to 1.10. See Table 20 for a summary of factor loadings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 94 Table 20. Estimates for factor loadings in model showing the effect of aggression and victimization on lifetime cigarette use among sixth-grade students. Factor Loading Boys (N = 334) Girls (N = 490) Latent Factor Indicator Unstd. p (SE) Std p Unstd. p (SE) Std p Victimization Relational sum score constrained to 1 .59 constrained to 1 .56 Overt sum score 1.18(.09) 1.00 1.01(.07) 1.00 Aggression Relational sum score constrained to 1 .59 constrained to 1 .57 Overt sum score .98(.15) .79 1.10(.17) .79 Social competence Listened to a friend .80(.10) .52 .82(.09) .53 Helped others .92(. 10) .62 1.00(.09) .65 Talked with friends to find a solution to a problem constrained to 1 .65 constrained to 1 .64 Helped someone leam a new skill l.lO (.ll) .72 1.13(.10) .73 Stood up for your values 1.06(.ll) .71 1.00(.09) .66 Worked out a problem ,99(.08) .66 .92(.07) .61 Note. Unstd. = unstandardized, Std. = standardized. All factor loadings are significant at p<05. The structural model was specified such that aggression and victimization were predictors of social competence and lifetime cigarette use. Additionally, aggression and victimization were allowed to correlate since past research has shown an association between these two constructs (Olweus, 1978). Social competence was also modeled as a predictor of lifetime cigarette use. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 95 The initial structural model for boys fit the data relatively well [y2 (89) = 193.71, p < .001, CFI = .91]. However, based on the LM test results a decision was made to add a correlated measurement error between the relational victimization and aggression summed scores, a decision that seemed reasonable since scales were adjacent on the questionnaire and asked about the same behaviors from a perpetration and then a victimization standpoint. Additionally, a correlated measurement error was added between two social competence indicators, “work[ed] out a problem” and “talk[ed] with friends to find a solution to a problem.” Addition of these path improved model fit, although it was still significantly different from the underlying data [% 2 (87) = 141.28, p <.001, CFI-.95], There was some concern that the hypothesized model of aggression and victimization as lifetime cigarette use predictors may have been altered by the addition of these errors due to slight changes in parameter estimates of relationships among some latent constructs. To insure that the structure of the hypothesized model had not been compromised, two tests were conducted to assess any differences between the two models, as recommended by Pentz and Chou (1994). First, correlation coefficients for the common set of parameter estimates between the basic model (MO) and the modified model (Ml) were computed to determine whether estimates changed as a result of model modification. The Spearman correlation coefficient was .99 (p < .0001), suggesting an almost perfect correspondence between the basic and modified models. Additionally a t test was conducted to detect differences on mean parameter estimates between basic and Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 96 modified models. Results indicated no difference of mean parameter estimates between the models (t = .39, p = .70), providing further evidence model modification did not substantially change the structure of the original model. See the top of Table 21 for a summary of model development. The final model showed partial support for hypotheses. As postulated, aggression predicted lifetime cigarette use and this relationship was highly significant (P = .17, SE = .04,p < .001). However, victimization did not predict lifetime cigarette use and this relationship was close to 0 (P = -.01, SE = .03, p < .001). Aggression and victimization were significantly associated as anticipated, with a covariance of 3.29 (SE = .70, p < .001). Relationships of aggression and victimization to social competence differed with aggression showing no predictive relationship (p = .01, SE = .03) while victimization was significantly positively predictive, the opposite direction expected ( P = .06, SE = .02, p < .01). Social competence was negatively predictive of lifetime cigarette use (p = -.10, SE = .10), as expected, although this relationship was not statistically significant. Given the lack of predictive relationships among several latent constructs, the lack of indirect effects for the aggression-lifetime cigarette use ( P = -.001, SE = .004) and the victimization-lifetime cigarette use ( P = -.01, SE = .01) were anticipated. See Table 22 for complete information on indirect effects. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 97 Table 21. Model development and test of measurement invariance for the effect of aggression and victimization on lifetime cigarette use among sixth-grade students. Group x2 df P CFI Boys M(bqo 193.71 89 < .0 0 1 .91 M (bc)i 141.28 87 <.001 .95 Girls M (gqo 265.18 89 <.001 .90 M (gc)1 192.93 86 <.001 .94 Combined M O 334.57 175 <.001 .95 Ml 339.38 182 <.001 .95 Ml-MO 4.81 7 ns - M2 348.86 188 <.001 .95 M2-M1 9.48 6 ns - Note. A subscript ending in “0" denotes a basic theoretical model and a subscript ending in “1" denotes a model modified to improve fit. An asterisk indicates a model with relevant parameters partially constrained equal across groups. MO= basic model combining groups, M1 = model with all factor loadings constrained equal across groups, M2 = model with all regression weights among factors constrained equal across groups over Ml*. Boys N = 334 and girls N = 490. Measurement model and structural model: girls Results of the measurement model showed that indicators loaded significantly on the three latent factors of aggression, victimization and social competence. With one factor loading (relational victimization summed score) constrained to one for model identification purposes, the overt victimization summed score indicator loaded significantly (P = 1.01, SE = .07) on the victimization factor. Similarly, with the relational aggression summed score factor loading constrained to one, the overt Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 98 aggression summed score indicator loaded significantly on the latent aggression factor (P = 1.10, SE = .17). For social competence, one loading was constrained to one (“talk[ed] with friends to find a solution to a problem”) and other loadings ranged from .82 to 1.13. See Table 19 for a summary of factor loadings. The initial structural model of aggression and victimization as predictors of lifetime cigarette use for girls fit the data relatively well [% 2 (89) = 265.18, p < .001, CFI = .90]. However, based on the LM test results a decision was made to add a correlated measurement error between the relational victimization and aggression summed scores and the overt victimization and aggression summed scores. These modifications seemed reasonable from a practical standpoint since scales were adjacent on the questionnaire, items asked about the same behaviors from both a perpetration and a victimization standpoint, and the same response choices were used. Additionally, a correlated measurement error was added between two social competence indicators, “work[ed] out a problem” and “talkfed] with friends to find a solution to a problem.” Addition of these path improved model fit significantly [% 2 (86) = 192.93, p< .001, CFI = .94]. See the top of Table 20 for a summary of model development. The final model showed partial support for hypotheses. Both aggression (P=.26, SE = .04, p < .001) and victimization (p= -.06, SE = .03, p < .05) significantly predicted lifetime cigarette use, although the victimization-cigarette use relationship was in the opposite direction predicted. Victimization and aggression were also significantly associated, as predicted, with a covariance of 2.43 (SE = .47, p < .001). Unanticipated Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 99 were the lack of a relationship between aggression and social competence (P = .003, SE = .02), which had been hypothesized, and the positive relationship between victimization and social competence (P = .03, SE = .02, p< .10), which was marginally statistically significant. Finally, there was a negative relationship from social competence to lifetime cigarette use, as predicted, although this path was not statistically significant (P = -. 15, SE = .09). Neither the indirect effect of victimization on cigarette use (p = -.004, SE = .004) nor that of aggression on cigarette use (P = -.001, SE = .005) was significant. See Table 21 for complete information on indirect effects. Table 22. Indirect effect of aggression and victimization on lifetime cigarette use mediated by social competence among sixth-grade students. Path Indirect Effect P(SE) Victimization->cigarette use Boys -.01(.01) Girls -,004(.004) Groups Combined -.007(.007) Aggression->cigarette use Boys -.001 (.004) Girls -,001(.005) Groups Combined -.000(.003) Note. No indirect effects are statistically significant at p<.05, two-tailed test. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 100 Test of measurement invariance between bovs and girls Testing measurement invariance of the aggression and victimization to cigarette use model across groups was the next step. First, a basic model (MO) which combined both groups was established. This model fit the data well [yl (175) = 334.57, p < .001, CFI = .95]. Next, the seven factor loadings (that is, all factor loadings except the relational victimization and aggression summed scores and “talked with friends to find a solution to a problem” which were constrained to one) were constrained equal between groups (Ml). The resulting model had a % 2 of 339.38 with 182 degrees of freedom and was still significantly different from the underlying data structure although the CFI still indicated good fit (.95). Based on the LM test results, a decision was made to retain all equality constraints on factor loadings. The next model (M2) added constraints on all regression weights between constructs of interest (i.e., aggression, social competence and alcohol use) as well as on the covariance between overt and relational aggression. The resulting model fit the data well [x2 (188) = 348.86, p < .001, CFI = .95], although it was not a significant improvement over the Ml model (M2-M1 y l (6) = 9.48, ns), indicating invariance of the model across gender. The test of measurement invariance between boys and girls is shown on the lower half of Table 21 and the final structural model is shown in Figure 6. The final structural model provided partial support for relationships among victimization and aggression, social competence and lifetime cigarette use. This model did not provide support for gender differences or mediation in these relationships. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1 01 Specifically, aggression predicted lifetime cigarette use significantly for both boys and girls (P = .22, SE = .03, p < .001), as hypothesized. However, there was no relationship between aggression and social competence (P = .004, SE - .02, ns). Victimization also predicted lifetime cigarette use (P = -.04, SE = .02, p < .05) although, contrary to hypothesis, this relationship was negative, suggesting that more victimization predicted lower lifetime cigarette use. Moreover, while victimization significantly predicted social competence, this relationship was positive, the inverse of the anticipated result (p = .04, SE = .01, p < .001). Finally, social competence showed a trend of being inversely predictive of lifetime cigarette use, as postulated (P = -.12, SE = .07, p < .10). It did not mediate either the victimization-lifetime cigarette use (p = -.007, SE = .007, ns) or aggression-lifetime cigarette use (P = -.000, SE = .003, ns) relationships. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 6. Parameters estimates for final model of the effect of aggression and victimization on lifetime cigarette use among sixth-grade students. Social Competence Victimization # <N * Os cs CO Life Cigarette Use Aggression Notes: Results are presented separately by gender with girls’ parameter estimates shown in brackets. Estimates are unstandardized. Circle refers to construct represented by measured items; square refers to single-item indicator. Baseline demographics included but not shown in this model are ethnicity, socioeconomic status, exposure to competence-building intervention, exposure to drug use prevention intervention and school type. + p .10, *p<.05. g tsJ 103 Effect o f Aggression and Victimization on Lifetime Alcohol Use Among Sixth-Grade Students Measurement model and structural model: bovs Results of the measurement model showed that indicators loaded significantly on the three latent factors of aggression, victimization and social competence. With one factor loading (relational victimization summed score) constrained to one for model identification purposes, the overt victimization summed score indicator loaded significantly (P = 1.18, SE = .09) on the victimization factor. Similarly, with the relational aggression summed score factor loading constrained to one, the overt aggression summed score indicator loaded significantly on the latent aggression factor (P = .94, SE = .15). For social competence, one loading was constrained to one (“listenjed] to a friend”) and other loadings ranged from 1.15 to 1.38. See Table 23 for a summary of factor loadings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 104 Table 23. Estimates for factor loadings in model showing the effect of aggression and victimization on lifetime alcohol use among sixth-grade students. Factor Loading Boys Girls (N = 334) (N = 490) Latent Factor Indicator Unstd. P(SE) Std p Unstd. p (SE) Std p Victimization Relational sum score constrained to 1 .59 constrained to 1 .57 Overt sum score 1.18(.09) 1.00 1.00(.07) 1.00 Aggression Relational sum score constrained to 1 .61 constrained to 1 .58 Overt sum score ,94(.13) .77 1.09(. 14) .80 Social competence Listened to a friend constrained to 1 .52 constrained to 1 .53 Helped others 1.15(.15) .62 1.22(.13) .64 Talked with friends to find a solution to a problem 1.26(. 16) .65 1.22(.13) .63 Helped someone learn a new skill 1.38(.17) .72 1.38(. 14) .73 Stood up for your values 1.33(.16) .71 1.23(.13) .66 Worked out a problem 1.25 (.16) .66 1.12(.12) .61 Note. Unstd. = unstandardized, Std. = standardized. All factor loadings are significant atp<05. The structural model was specified such that aggression and victimization were predictors of social competence and lifetime alcohol use. Additionally, aggression and victimization were allowed to correlate since past research has shown an association Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 105 between these two constructs, as previously mentioned (Olweus, 1978). Social competence was also modeled as a predictor of lifetime alcohol use. This model fit the data relatively well [yl (89) = 194.38, p < .001, CFI = .91]. However, based on the LM test results a decision was made to add a correlated measurement error between two the relational victimization summed scores. This decision seemed reasonable in that scales were adjacent on the questionnaire and asked about the same behaviors from a perpetration and then a victimization standpoint. Additionally, a correlated measurement error was added between two social competence indicators, “work[ed] out a problem” and “talk[ed] with friends to find a solution to a problem.” These paths improved the overall model fit significantly [yl (87) = 145.50, p < .001, CFI = .95]. Table 24 provides a summary of model development. The final model of overt and relational aggression as predictors oflifetime alcohol use showed partial support for hypotheses. Aggression did predict lifetime alcohol use, as expected, and this relationship was highly significant (P = .23, SE = .04, p < .001). However, victimization did not predict lifetime alcohol use and this relationship, although not statistically significant, was in the opposite direction expected (P = -.04, SE = .03). Also in the inverse direction expected was the positive relationship between victimization and social competence (P = .05, SE - .02, p < .01) and this relationship was highly statistically significant. In contrast, there was no relationship between aggression and social competence (P = .003, SE = .02). Social competence inversely predicted lifetime alcohol use, as hypothesized, but this relationship did not achieve statistical significance Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 106 (P = -.13, SE = .13). Finally, aggression and victimization were significantly associated as anticipated, with a covariance of 3.52 (SE = .91 ,P < .001). There was no indirect effects from victimization to lifetime alcohol use (P = -.006, SE = .009) or aggression to lifetime alcohol use (P = -.000, SE = .004). See Table 25 for complete information on indirect effects. Table 24. Model development and test of measurement invariance for the effect of aggression and victimization on lifetime alcohol use among sixth-grade students. Group x2 df P CFI Boys M(B c ) o 194.38 89 <.001 .91 M(b c )i 145.50 87 <.001 .95 Girls M (GC)0 270.25 91 <.001 .89 M ( c c ) i 192.47 89 <.001 .94 Combined M0 337.97 176 <.001 .94 Ml 341.18 183 <001 .94 Ml-M0 3.21 7 ns - M2 347.91 189 <001 .94 M2-M1 6.73 6 ns - Note. A subscript ending in “0" denotes a basic theoretical model and a subscript ending in “1" denotes a model modified to improve fit. M 0= basic model combining groups, M l = model with all factor loadings constrained equal across groups, M2 = model with all regression weights among factors constrained equal across groups over M l*. Boys N = 334 and girls N = 490. Measurement model and structural model: girls Results of the measurement model showed that indicators loaded significantly on the three latent factors of aggression, victimization and social competence. With one Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 107 factor loading (relational victimization summed score) constrained to one for model identification purposes, the overt victimization summed score indicator loaded significantly ((3 = 1.00, SE = .07) on the victimization factor. Similarly, with the relational aggression summed score factor loading constrained to one, the overt aggression summed score indicator loaded significantly on the latent aggression factor ((3 = 1.09, SE = .14). For social competence, one loading was constrained to one (“listen[ed] to a friend”) and other loadings ranged from 1.12 to 1.38. See Table 23 for a summary of factor loadings. The initial structural model for girls did not fit the data acceptably, based on both the chi-square test {yl (89) = 192.47, p < .001] and the CFI (.89). Thus the model was modified and using the LM test as guidance such that correlated measurement errors were added between the relational victimization and aggression summed scores as well as between two social competence indicators, “work[ed] out a problem” and “talked with friends to find a solution to a problem.” As mentioned previously, addition of these paths was theoretically, as well as empirically, feasible. These paths improved the overall fit of the model significantly {yl (89) - 192.47, p < .001, CFI = .94]. See the top of Table 24 for a summary of model development. Results for the girls ’ final model mirrored those ofboys, although path magnitudes were generally slightly smaller. Aggression predicted lifetime alcohol use, as hypothesized, and this relationship was highly significant ([3 = .20, SE = .04, p < .001) but it did not predict social competence-indeed this path was close to zero (P = .001, SE = Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 108 .02). There was also no support for the hypothesis that social competence would predict lifetime alcohol use (P = -.03, SE = .11). Victimization was a significant predictor of social competence, as postulated, but this relationship was positive, the opposite direction expected (p = .02, SE = .01, p<.05). Also in the opposite direction hypothesized was the marginally significant relationship between victimization and lifetime alcohol use (P = - .05, SE = .02, p < .10). Finally, there was a significant covariance between victimization and aggression (P = 2.67, SE = .49, p < .001). There was no indirect effects from victimization to lifetime alcohol use (P = -.001, SE = .003) or aggression to lifetime alcohol use (P = -.000, SE = .001). See Table 25 for complete information on indirect effects. Table 25. Indirect effect of aggression and victimization on lifetime alcohol use mediated by social competence among sixth-grade students. Path Indirect Effect P(SE) V ictimization->alcohol use Boys -,006(.009) Girls -.001(.003) Groups Combined -,002(.004) Aggression->alcohol use Boys -.000(.004) Girls -.000(.001) Groups Combined -.000(.001) Note. No indirect effects are statistically significant at p<.05, two-tailed test. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 109 Test of measurement invariance between bovs and girls The test of measurement invariance began by creating a basic model (M O) which combined both groups. This model fit the data well [yl (176) = 337.97, p < .001, CFI = .94]. Next, the seven factor loadings (that is, all factor loadings except the relational victimization and aggression summed scores and “talked with Mends to find a solution to a problem” which were constrained to one) were constrained equal between groups (Ml). The resulting model was still significantly different from the underlying data [yl (182) = 341.18, p < .001] but the CFI indicated good fit (.94). The LM test did not recommend the release of factor loadings so the next set of constraints, regression weights on relationships between latent constructs as well as the covariance between aggression and victimization, was imposed. This model (M2) fit the data well [yl (189) = 347.91, p < .001, CFI = .94], although it was not a significant improvement over the Ml model (M2-M1 y l (6) = 6.73, ns). The test of measurement invariance between boys and girls is shown on the lower half of Table 23 and the final structural model is shown in Figure 7. The final structural model provided limited support for hypothesized relationships among victimization, aggression and lifetime alcohol use. Aggression predicted lifetime alcohol use (p = .22, SE = .03, p < .001) as hypothesized, but was not associated with social competence (P - .003, SE = .001, ns). Victimization also predicted lifetime alcohol use (P = -.04, SE = .02, p < .05) but this relationship was negative, suggesting that more victimization predicted lower lifetime alcohol use. Similarly, victimization showed a Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 110 positive predictive relationship with social competence (P = .03, SE = .01, p < .001) and this relationship was also in the opposite direction postulated. Social competence was not related to lifetime alcohol use (P = -.07, SE = .08, ns) and did not mediate either the victimization-lifetime alcohol use (P = -.002., SE = .004, ns) or aggression-lifetime alcohol use (P = -.000., SE = .001, ns) relationships. Finally, there was no evidence of an interaction of a gender among any of these relationships. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 7. Parameter estimates for final model of the effect of aggression and victimization on lifetime alcohol use among sixth-grade students. Social Com petence V ictim ization ■ O Life alcohol Use A ggression Notes. Results are presented separately for boys and girls with girls’ parameter estimates shown in brackets. Estimates presented are unstandardized.. Circle refers to construct represented by measured items; square refers to single-item indicator. Baseline demographics included but not shown in this model are ethnicity, socioeconomic status, exposure to competence-building intervention, exposure to drug use prevention intervention and school type. *p<.05. 112 Effect o f Aggression and Victimization on Lifetime Marijuana Use Among Sixth-Grade Students Measurement model and structural model: bovs Results of the measurement model showed that indicators loaded significantly on the three latent factors of aggression, victimization and social competence. With one factor loading (relational victimization summed score) constrained to one for model identification purposes, the overt victimization summed score indicator loaded significantly ((3 = 1.18, SE = .09) on the victimization factor. Similarly, with the overt aggression summed score factor loading constrained to one, the relational aggression summed score indicator loaded significantly on the latent aggression factor (p = 1.01, SE = .17). For social competence, one loading was constrained to one (“talk[ed] with friends to find a solution to a problem”) and other loadings ranged from .80 to 1.10. See Table 26 for a summary of factor loadings. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 113 Table 26. Estimates for factor loadings in model showing the effect of aggression and victimization on lifetime marijuana use among sixth-grade students. Factor Loading Boys (N = 334) Girls (N = 490) Latent Factor Indicator Unstd. p (SE) Std p Unstd. p (SE) Std p Victimization Relational sum score constrained to 1 .58 constrained to 1 .55 Overt sum score 1.18(.09) 1.00 1.04(.07) 1.00 Aggression Relational sum score 1.01(. 17) .59 ■67(.ll) .49 Overt sum score constrained to 1 .79 constrained to 1 .93 Social competence Listened to a friend .80(.10) .52 .82(.09) .53 Helped others .92(.10) .62 1.00(.09) .65 Talked with friends to find a solution to a problem constrained to 1 .65 constrained to 1 .63 Helped someone learn a new skill l.lO (.ll) .72 1.13(. 10) .73 Stood up for your values l.Q6(.ll) .71 1.01(.09) .66 Worked out a problem .99(.08) .66 .92(.07) .61 Note. Unstd. = unstandardized, Std. = standardized. All factor loadings are significant atp<05. The structural model was specified such that aggression and victimization were predictors of social competence and lifetime marijuana use. Additionally, aggression and victimization were allowed to correlate since past research has shown an association between these two constructs, and social competence was modeled as a predictor of lifetime alcohol use (Olweus, 1978). This model fit the data relatively well [% 2 (91) = Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 114 210.11,p < .001, CFI = .90]. However, based on the LM test results a decision was made to add a correlated measurement error between two the relational victimization summed scores and between two social competence indicators, “work[ed] out a problem” and “talk[ed] with friends to find a solution to a problem.” These paths improved the overall fit of the model significantly [% 2 (89) = 156.78, p < .001, CFI = .94]. See the top of Table 27 for a summary of model development. The final model showed partial support for hypotheses related to aggression and victimization as predictors of lifetime marijuana use among boys. Aggression predicted lifetime marijuana use (j3 = .07, SE = .03, p < .01). However, victimization did not predict lifetime marijuana use and this relationship, although not statistically significant, was in the opposite direction expected ( { 3 = -.03, SE = .02). Also in the inverse direction expected was the positive relationship between victimization and social competence which was statistically significant (P = .06, SE = .02, p < .01). In contrast, there was no relationship between aggression and social competence (p = .01, SE = .03). Finally, social competence inversely predicted lifetime marijuana use, as hypothesized, but this relationship was not significant (P = -.05, SE = .06). Aggression and victimization were significantly associated as anticipated, with a covariance of 3.24 (SE = .55, p < .001). Finally, there was no support for an indirect effect from either victimization to lifetime marijuana use (P = -.003, SE = .006) or aggression to marijuana use (p = -.000, SE = .002). See Table 28 for complete information on indirect effects. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 115 Table 27. Model development and test of measurement invariance for the effect of aggression and victimization on lifetime marijuana use among sixth-grade students. Group x2 df P CFI Boys M(BQ0 210.11 91 <001 .90 M(B c)i 156.78 89 <001 .94 Girls M (OC)0 271.57 91 <001 .89 M (GC)1 193.05 89 <001 .94 Combined MO 349.83 178 <001 .94 M l 357.44 185 <.001 .94 Ml-MO 7.61 7 ns - M2 365.50 191 <001 .94 M2-M1 8.06 6 ns - Note: A subscript ending in “0" denotes a basic theoretical model and a subscript ending in “ 1" denotes a model modified to improve fit. An asterisk indicates a model with relevant parameters partially constrained equal across groups. MO = basic model combining groups, M l = model with all factor loadings constrained equal across groups, M2 = model with all regression weights among factors constrained equal across groups over M l*. Boys N = 334 and girls N = 490. Measurement model and structural model: girls Results of the measurement model for girls showed that indicators loaded significantly on the three latent factors of aggression, victimization and social competence. With one factor loading (relational victimization summed score) constrained to one for model identification purposes, the overt victimization summed score indicator loaded significantly (p = 1.04, SE = .07) on the victimization factor. Similarly, with the overt aggression summed score factor loading constrained to one, the relational Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 116 aggression summed score indicator loaded significantly on the latent aggression factor (P = .67, SE = .11). For social competence, one loading was constrained to one (“talk[ed] with friends to find a solution to a problem”) and other loadings ranged from .82 to 1.13. See Table 26 for a summary of factor loadings. The structural model for aggression and victimization as predictors of lifetime marijuana use among girls was the same as that for boys described previously. This model did not fit the data acceptably, based on both the chi-square test [% 2 (91) = 271.57, p < .001] and the CFI (.89). Thus the model was modified and, using the LM test as guidance, correlated measurement error was added between the relational victimization and aggression summed scores as well as between two social competence indicators, “work[ed] out a problem” and “talk[ed] with friends to find a solution to a problem.” As mentioned previously, addition of these paths was theoretically and empirically feasible. These paths improved the overall fit of the model significantly [y2 (89) = 193.05, p < .001, CFI = .94], See the top of Table 27 for a summary of model development. Results for the girls’ final model provided partial support for hypotheses related to aggression and victimization as predictors of lifetime marijuana use. As it did among boys, aggression predicted lifetime marijuana use, and this relationship was highly significant (P = .07, SE = .02, p < .001) but did not predict social competence (P = .001, SE = .02). However, there was strong support for the hypothesis that lower social competence would predict lifetime marijuana use (p = -.20, SE = .06, p <.001). Victimization marginally predicted social competence, but this relationship was not the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 117 inverse one that had been hypothesized (P = .03, SE = .01, p < .10). Also in the opposite direction hypothesized was the marginally significant relationship between greater victimization and less lifetime marijuana use (p = -.03, SE = .02, p < .10), although this effect size was small. Finally, there was a significant covariance between victimization and aggression as hypothesized (P = 2.88, SE = .40, p < .001). There was no indirect effects from victimization to lifetime marijuana use (P = -.006, SE = .004) or aggression to lifetime marijuana use (P = -.002, SE = .006). See Table 28 for complete information on indirect effects. Table 28. Indirect effect of aggression and victimization on lifetime marijuana use mediated by social competence among sixth-grade students. Path Indirect Effect P(SE) Victimization->marijuana use Boys -.0Q3(.OO6) Girls -,006(.004) Groups Combined -,006(.004) Aggression->marijuana use Boys -,000(.002) Girls -,002(.006) Groups Combined -,001(.004) Note. No indirect effects are statistically significant at pK.05, two-tailed test. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 118 Test of measurement invariance between bovs and girls In order to test the hypotheses related to gender differences, a test of measurement invariance of the model across groups was conducted. First, a basic model (MO) which combined both groups was established. This model fit the data well based on the CFI (.94), although it was not statistically acceptably [% 2 (178) = 349.83, p < .001]. Next, the seven factor loadings (all factor loadings except the relational victimization and overt aggression summed scores and “talk[ed] with friends to find a solution to a problem” which were constrained to one) were constrained equal between groups (Ml). The resulting model was still significantly different from the underlying data [j(2 (185) = 357.44, p < .001] and the CFI remained the same (.94). Since there was no overall difference between the unconstrained and the constrained models, no factor loadings were released. The next set of constraints on regression weights between latent factors and the covariance between aggression and victimization were imposed. This model (M2) was not a significant improvement over the Ml model (M2-M1 x2 (6) = 8.06, ns) and the fit was the same [% 2 (191) = 365.50, p < .001, CFI = .94], Thus, the omnibus test of measurement invariance suggested there were no significant differences between groups. The test of measurement invariance is shown on the lower half of Table 27 and the final structural model is shown in Figure 8. The final model of aggression and victimization as predictors of lifetime marijuana use showed partial support for hypotheses. Aggression predicted lifetime marijuana use, as postulated (p = .07, SE = .02, p < .001). Victimization also predicted Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 119 lifetime marijuana use but this relationship was in the opposite direction expected ( P = -.02, SE = .01, p < .05). Also in the inverse direction expected was the positive relationship between victimization and social competence (p = .04, SE = .01, p < .001). There was no relationship between aggression and social competence (p = .01, SE = .01, ns). Social competence inversely predicted lifetime marijuana use, as hypothesized, but this relationship was only marginally significant ( P = -.14, SE = .08). Finally, aggression and victimization were significantly associated as anticipated, with a covariance of 2.97 (SE = .32, p < .001). There were no indirect effects from victimization to lifetime marijuana use (P - -.006, SE = .004) or aggression to lifetime marijuana use (p = -.001, SE = .004). See Table 28 for complete information on indirect effects. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Reproduced w ith permission o f th e copyright owner. Further reproduction prohibited without permission. Figure 8. Parameter estimates for final model of the effect of aggression and victimization on lifetime marijuana use among sixth-grade students. Social Competence Victimization * 0 © © 0 C N Life Marijuana Use Aggression Notes. Results are presented separately by gender w ith girls’ param eter estim ates shown in brackets. Estimates are unstandardized. Circle refers to construct represented by m easured items; square refers to single-item indicator. Baseline demographics included but not shown in this m odel are ethnicity, socioeconom ic status, exposure to com petence-building intervention, exposure to drug use prevention intervention and school type. + p .10, *p<05. to o 121 DISCUSSION Study One: The Panel Study Overview The panel study examined the longitudinal relationship of peer aggression to lifetime use of three drags, cigarettes, alcohol and marijuana, among a cohort of fourth- grade students. Where relationships were found to exist, this study further evaluated the aggression-to-drag use relationship in terms of differential importance of two aggression subtypes, overt and relational aggression, in predicting later drug use. Moreover, it examined gender interactions in these relationships, particularly whether there was an interaction between gender and preferred style of aggression in predicting drag use. Social competence in peer interactions was also tested as a mediator of these relationships. Although there had also been hypotheses related to nonnormative aggression style and elevated risk for aggression, lack of main effects for gender x aggression-style interaction precluded testing of these hypotheses. Results provided partial support for the main hypothesis of aggression in fourth grade as a predictor of sixth grade lifetime drag use. Specifically, there was a main effect of aggression on both lifetime cigarette use and lifetime alcohol use. Moreover, these relationships were similar in magnitude (P = .23, SE = .10 for cigarette use outcome, P = .29, SE - .10 for alcohol use outcome), suggesting similar trajectories from aggression to later use of these drags. In contrast, there was no relationship between aggression in fourth grade and sixth-grade lifetime marijuana use. In fact, the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 122 magnitude of this path was close to zero ((3 = -.03, SE = .05), suggesting that early aggression may not be a risk factor for marijuana use and that lifetime marijuana use may have predictors in childhood which are distinct from those for later alcohol or cigarette use. There was no support for the hypothesis of a gender interaction on relationships among fourth-grade aggression, fifth-grade social competence and sixth-grade substance use. It had been anticipated that the aggression-substance use paths would be stronger for boys than for girls in the models where aggression was treated as one overall concept. This assumption was predicated on the fact that the aggression construct in these models was primarily comprised of overt aggression items (four out of five indicators). Since some research has suggested that this style of aggression is more typical of boys, it seemed likely that this construct would be a more powerful predictor of later drug use for boys than for girls (Crick, 1996). However, in the two models which showed significant relationships between aggression and substance use, namely the models with alcohol and cigarette use outcomes, the regression weights were stronger (although not significantly so) for girls than for boys. One interpretation of this is that since overt aggression is more normative a behavior for boys, it does not necessarily signal a high-risk trajectory which may lead to drug use. Among girls, overtly aggressive behavior may be less normative and, therefore, more likely to be associated with later high-risk behaviors such as drug use. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 123 Hypotheses related to subtypes of aggression, that is overt and relational aggression, as drug use predictors had partial support, but there was no evidence of an interaction between aggression type and gender. When aggression was subdivided into overt and relational subtypes, overt aggression significantly predicted lifetime alcohol use and marginally predicted lifetime cigarette use. Relational aggression did not predict either lifetime cigarette or alcohol use. In fact, there was virtually no association between relational aggression and either cigarette or alcohol use, with the magnitude of both relationships almost zero (|3 = -.03, SE = .06 for both relationships). Furthermore, no gender differences were found in the strength of these relationships. These findings contrast somewhat with past literature that found that girls used relational forms of aggression more than overt forms and suggested that behaviors related to aggression may be under-reported when relational aggression is not studied alongside overt aggression (e.g., Crick (1996)). This study found no evidence of greater use of relational aggression among girls. Indeed post-hoc calculation of prevalence of relational aggression by gender indicated no significant difference with 31% of boys versus 25% of girls (% 2 = 1.90, ns) classified as relationally aggressive (based on “yes” = 1 versus “no” = 0 response codings). Similarly, a post-hoc t test indicated no gender difference in mean frequency of relational aggression reported between girls (M = 1.65) and boys (M= 1.78) (t= .154, ns). These results suggest that overt aggression may be an important risk factor of later alcohol and cigarette use for both sexes, rather than Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 124 primarily among boys, and that relational aggression is not a risk factor for substance use for either girls or boys. Hypotheses related to social competence as a mediator of the aggression-drug use relationship were not supported, although results did provide some indication that social competence mitigates against risk behaviors. When aggression was considered as one construct, fourth-grade aggression was negatively predictive of fifth-grade social competence. Additionally, higher social competence in fifth grade predicted lower lifetime use in sixth grade of cigarettes and marijuana and marginally lower lifetime alcohol use in these models. Interestingly, when aggression was divided into overt and relational subtypes, neither relational nor overt aggression predicted social competence, although the relationships were still in the inverse direction. However, inverse associations between social competence and problem behaviors (e.g., aggression and drug use) provide some evidence for the utility of social competence with peers in mitigating against later drug use. Limitations Internal Validity The measures used in this study were limited and this may have affected internal validity. First, the content validity of relational aggression was insufficient due to the availability of only one item in the data set to measure relational aggression at baseline. Past research has found that theoretically expected associations have not been detected when only one item was used to assess a psychological construct (Rushton, Brainerd, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 125 & Pressley, 1983). It is possible that the item used to represent relational aggression did not capture the construct and a more comprehensive measure of relational aggression would have been more effective in detecting a relationship with drug use. Second, social competence was limited to the peer domain. It is certain that incidents of aggression and victimization that occur outside of relationships with peers, particularly those that occur in parent-to-child relationships, have serious implications for children’s behavioral outcomes. For example, parenting has been linked to children’s mastery ofkey developmental tasks, such as self-regulation, which are central to social competence (Masten & Coatsworth, 1998). However, due to the decision to limit the construct of social competence in the present study to peer relationships, any impact of social competence with parents or other adults would not be captured. Third, the six particular items used in this study may represent only one aspect of social competence with peers, and this may have precluded detecting hypothesized mediational relationships of social competence to problem behaviors. It may be there are skills central to children’s successful social interaction that were not represented in this study. For example, assertiveness skills may be important for avoiding victimization by peers. Since assertiveness was not included among the skills which represented social competence, its impact as a mediator would not have been captured. Future research on social competence which tests the differential effectiveness of specific skills in reducing drug use risk is needed. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 126 Another consideration related to internal validity is that other models may also plausibly explain the relationships among aggression, social competence and substance use. Past research has established that aggression emerges developmentally before substance use, supporting its placement in the present study as a precursor to drug use (White, 1997). Moreover, any effects of drug use on aggression were statistically controlled by selecting a no-using sample at baseline. However, it is possible that, by preadolescence, a reciprocal relationship exists between aggression and drug use. Some research supports reciprocality, although this research is based on older adolescent samples (Hudley et al., 1998; White, Loeber, Stouthamer-Loeber, & Farrington, 1999). Future research is needed to examine this possibility. External Validity Although the panel was drawn from a population-based study, it was comprised of significantly more female, white, high SES and private school subjects. Additionally, baseline drug users were excluded from this study since its design was to study the natural history of the aggression-to-drug use relationship. Finally, subjects with missing data were also excluded since the SEM does not permit inclusion of missing data. Since female gender, high SES and late drug use initiation have been associated with less risk for drug use, it is likely that some subjects at highest risk for problem behavior were excluded from the present study (Belcher & Shinitzky, 1998; Hawkins, Catalano, & Miller, 1992). For this reason, the external validity of this study may be limited to Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 127 populations similar to the one studied here and, specifically, non-drug users in late childhood. Study Two: The Cross-Sectional Study The cross-sectional study sought to provide further information about the relationship of peer aggression as a predictor oflifetime cigarette, alcohol and marijuana use, mediated by peer social competence. In addition, it explored peer victimization as a predictor of these substances, a heretofore unexamined relationship, and tested peer social competence as a mediator of this relationship. Finally, it compared these relationships between boys and girls. Results supported the major hypothesis that aggression would predict substance use among sixth-grade subjects. In contrast to the panel study where there was no relationship of aggression to marijuana use, aggression was a significant predictor of all three substances in the cross-sectional study (lifetime cigarette use, alcohol use and marijuana use). However, hypotheses related to gender as a moderator and social competence as a mediator of the aggression-substance use relationship were not supported for any substance. This was a finding which was consistent with the panel study, which also found no evidence of a gender interaction. There was no support for the other major hypothesis of the cross-sectional study that increased victimization would predict greater lifetime substance use. Results indicated that lower victimization predicted higher drug use, an unanticipated inverse relationship. Social competence also did not mediate these relationships as had been Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 128 predicted, but there were significant relationships from victimization to social competence for all three models. Although the magnitude of these relationships was small (p = .03 to .04, SE = .01), they were highly statistically significant (all p’s <.001) and opposite of what had been posited. While results indicating that victimization was positively related to social competence and negatively related to drug use outcomes were initially surprising, some explanations are feasible. Past research has shown that children victimized by peers have fewer friends and often demonstrate highly anxious behavior (Hodges, Malone, & Perry, 1997). Since drag use in sixth grade is not yet a normative behavior, it may occur primarily among high-risk or sensation-seeking type youth, a personality type which has been associated with early use of at least one drag (alcohol) (Hawkins, Catalano, & Miller, 1992). Moreover, it likely occurs in social settings with peers, since research has established that the amount of time spent with peers and peer deviant behavior are strongly associated with drag use (Brook & Cohen, 1992). Thus, insofar as drag use is a high-risk social activity, anxious and friendless victimized children would be less likely to participate. Research has also shown positive associations between victimization and submissive behavior, at least among boys (Schwartz, Dodge, & Coie, 1993). It may be that social competence in this study was comprised of skills that would help a child get along with others, but required acquiescence or delayed gratification of the child’s own desires. Indeed, only one of the six social competence items reflected an assertive behavior, “standing] up for your values” while others related to putting Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 129 others’ need first (e.g., “listen[ed] to a friend). Thus, the positive relationship between victimization and social competence may reflect low assertiveness and more willingness to accommodate others on the part of the victimized child. Limitations Internal Validity The primary limitation of Study Two was the use of cross-sectional data. The specification of models such that victimization and aggression were predictors of substance use was based on past studies which suggested that aggression and victimization within the peer group are apparent as early as first grade, well before the typical initiation of substance use (Kosterman, Hawkins, Guo, Catalano, & Abbot, 2000). However, it is possible that once initiated, aggression and victimization have a dynamic relationship with substance use and aggression contributing to one another. Some research has demonstrated a reciprocal relationship between alcohol use and aggression, and marijuana use and aggression to a lesser extent, among adolescents, although no studies related to reciprocity of substance use and peer victimization could be found (Huang, White, Kosterman, Catalano, & Hawkins, 2001; White et al, 1999). Similarly, social competence was placed after aggression and before substance use outcomes due to interest in its capacity as a mediator of these relationships. However, it is possible, and even likely, that social competence may precede both victimization and drug use. For example, a child with low social competence entering sixth grade may be a target of victimization because he has no close friends, since having few friends Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 130 increases vulnerability to victimization by peers (Schwartz & Proctor, 2000). The victimization experience causes the child distress which he then tries to cope with by drug use. In fact, research has shown that victimization predicts anxiety and depression which, in turn, have been linked to drug use (Amaro, Blake, Schwartz, & Flinchbaugh, 2001; Hodges & Perry, 1999; Mehrabian, 2001). Other research with longitudinal data is needed to test the temporal relationships among these constructs. External Validity The sample for the cross-sectional study was comprised of significantly more female and white subjects, and a greater proportion of subjects were from private schools, than subjects excluded from this sample. Additionally, a larger proportion of children selected for the analytic sample had been exposed to a program designed to build academic and social competence in elementary school (the large Bright Stars intervention program). This raised concern that subjects in this study may have been at lower risk for aggression and drug use behaviors than those excluded and not typical of the general sixth-grade population. Nevertheless, subj ects in the analytic sample did not differ from those excluded on most aggression and victimization items. Furthermore, lifetime drug use did not differ between included and excluded youth, with the exception of lifetime marijuana use, which was higher among excluded subjects. Therefore, it seems that the analytic sample did not differ dramatically on constructs under examination in this study, suggesting that the demographic differences may not have biased study results. However, these demographic differences may still have operated Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 131 in influencing current findings and caution should be used when generalizing them to populations different than the one studied here. Overall Implications and Future Directions Several findings were consistent across both studies. First, peer social competence did not mediate relationships between aggression and substance use in either the panel or the cross-sectional study. This could be interpreted as a lack of support for building social competence as a means to reducing later problem behavior. However, more careful inspection of results showed that both aggression and substance use were negatively related to social competence. Hence, this provides some indication of the potential for increasing social competence to reduce problem behaviors. Nevertheless, given that relationships between aggression and social competence and drug use and social competence were in the expected negative direction, an explanation for the lack of mediational relationship is still wanting. One feasible explanation comes from considering the social development model’s representation of behavior as occurring along two parallel paths, prosocial and antisocial behavior, and bonding to peers as a reinforcement for behavior (that is, prosocial peers bond to one another, antisocial peers bond to one another). It is possible that what these results may represent is simply a glimpse of part of the antisocial trajectory where antisocial behavior leads to more antisocial behavior and social competence is negatively associated with both of these. If this is the case, one explanation for the failure of social competence to mediate aggression-substance use relationships is that social competence Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 132 with peers as represented in this study did not represent the skills most critical to interrupting the strong relationship between these two problem behaviors. In a similar vein, it may be that competence in other domains, such as in relationships with adults, is the most critical to moving a child from the antisocial to prosocial trajectory. Further research on the relative importance of different domains of social competence is needed to resolve this question. Another explanation of the lack of a mediational effect is that the use of self- reported data to assess social competence did not accurately reflect subjects’ levels of competence. A recent meta-analysis of 87 school-based interventions studies intended to reduce problem behavior by Najaka and colleagues (2002) concluded that changes in self-reported measures of social competence were unrelated to changes in problem behaviors. The present study did find that self-reported competence was inversely related to problem behavior, as had been anticipated. However, if the associations between aggression and social competence and social competence and drug use outcomes were attenuated due to the use of self-report data, the effect of social competence as a mediator of these relationships may have been diminished. Future research should compare the mediational effect of social competence as measured by self-report and independent observer report. Another finding common across all models was the absence of an interaction of gender with any of the hypothesized relationships. This was somewhat unexpected given that other studies have noted differences in aggression and, to a lesser extent, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 133 substance use and victimization, by gender (Amaro et al, 2001; Crick & Bigbee, 1998; Tolan & Guerra, 1994). One plausible explanation is that differences found in past studies related to overall levels of behaviors and not the strength of the relationships between them. Results of this study would be somewhat consistent with that since there were significant differences between gender in aggression levels but not in relationships from aggression to drug use. Another is that this study simply did not have adequate power to detect gender differences which actually exist. Finally, aggression predicted cigarette and alcohol use both longitudinally and cross-sectionally. Interestingly, it seems to have been overt aggression which was responsible for the association between aggression and cigarette use. When aggression was divided into overt and relational subtypes, the overt form of aggression was still (marginally) predictive of later cigarette use, but no association between relational aggression and cigarette use was found. This pattern was slightly different for the alcohol use in that there was no decrease in the strength of the relationship for overt aggression to substance use when the relational aggression item was separated out. Implications and Future Directions Several results of this study have important implications for the design of effective interventions to prevent substance use. First, the absence of a gender interaction in relationships among aggression, victimization, social competence and drug use indicates that these relationships may operate in the same way across gender. This implies that aggression and low competence are risks for substance use of similar degree Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 134 for girls and boys. Thus, these findings may suggest that prevention programs which seek to reduce substance use through decreasing aggression or increasing social competence may be universal in nature and may be implemented in whole classrooms and schools. Furthermore, the lack of any effect of relational aggression on substance use provided preliminary evidence that relational aggression need not be included in prevention efforts of substance use for either gender. However, more research is needed on relational aggression and substance use before this can be stated with confidence. Second, findings related to victimization as a predictor of high social competence and lower substance use provide a conundrum for prevention researchers. They suggest that programs that attempt to reduce children’s victimization within the peer group may increase their risk for drug abuse. In particular, this may occur if children go from being relatively friendless to forming friendships with antisocial children. These results also seem to suggest that increasing social competence would not be an effective strategy for reducing victimization, as the two constructs are positively correlated. However, the most promising drug prevention efforts have achieved results through bolstering social competence. More research is needed on specific skill sets to determine if there are skills which have the capacity to decrease risk of victimization but do not increase drug use risk. Longitudinal research on the victimization-drug use relationship, as well as on the effects ofparticular types of social competence, is needed to address these issues. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 135 Third, results suggest that more research on social competence and its relationship to aggression and drug use is warranted. Future research should perhaps explore other factors that may mediate the inverse predictive relationships found between fourth-grade aggression and fifth-grade social competence and fifth- grade social competence and sixth-grade drug use outcomes. For example, perceived self- efficacy, which has been established as a predictor of performance, may be an important mediator of relationship between aggression and social competence (Bandura, Adams, & Beyer, 1977). It maybe that increased efficacy leads to greater ability to exercise the social skills that comprise social competence. Similarly, the predictive relationships between low social competence and drug use outcomes may be explained by other factors not measured in this study. One example would be greater bonding to drug using peers, a known risk factor for increased drug use (Hawkins, Catalano, & Miller, 1992). It is possible that studies that account for factors such as these may be more successful in finding a significant mediational relationship. At the least, future studies should incorporate additional factors in order to explain more completely this complex relationship. Finally, the finding that aggression predicts later substance suggests that programs that reduce aggression may also be effective in preventing later drug use. Since aggressive behavior appears developmentally much earlier than drug use initiation, interventions to prevent aggression and subsequent drug use could begin well before initiation of drug use. This is significant since earlier age of substance use Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 136 initiation is linked to greater risk for drug abuse problems (Kosterman et al., 2000). Moreover, since this study did not find conclusive support for substantially different relationships between aggression and the three substances studies, it seems likely that any effective program would reduce the risk of all three substances studied, cigarettes, alcohol and marijuana. Future research should investigate the impact of preventing aggression on drug use initiation and later drug use. Furthermore, one can speculate that the effects of prevention of one child from drug use may have larger consequences. According to social development theory, bonding to others who engage in problem behaviors promotes adherence to their beliefs and behaviors and increases the likelihood of behaviors consistent with their norm (Catalano & Hawkins, 1996). Thus, it is possible that interventions that reduced earlier aggression through increasing bonding to social units holding prosocial beliefs could expect prosocial behaviors to increase and antisocial behaviors in general, not just aggression or drug use, to decrease. In short, a child could change from an antisocial to a prosocial developmental trajectory. This is important in that it suggests that prevention programs may do more than simply eliminate a noxious behavior such as drug use or aggression-they may leverage risk and protection to achieve diversion from risk to protection and prosocial behavior. Moreover, research has shown that antisocial behavior often is fostered by bonding to antisocial peers (Catalano & Hawkins, 1996). Therefore, it may be that preventing drug use in one child could reduce the risk for other children insofar by reducing the Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 137 number of antisocial “partners” available for participation in problem behaviors including substance use. It may be that if enough antisocial peers are redirected to prosocial paths, and the number of antisocial consequently reduced, the collective risk for antisocial behavior would diminish based on the absence of an important risk factor, bonding to antisocial peers. Future research on aggression, victimization, social competence and drag use must resolve these important questions. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 138 REFERENCES Abram, K. M. (1989). The effect of co-occurring disorders on criminal careers: Interaction of antisocial personality, alcoholism, and drug disorders. International Journal of Law and Psychiatry. 12(133-148). Amaro, H., Blake, S. M., Schwartz, P., & Flinchbaugh, L. J. (2001). Developing theory-based substance abuse preention programs for young adolescent girls. Journal of Early Adolescence. 21(3). 256-293. Ammerman, R. T., Ott, P. J., Tarter, R. E., & Blackson, T. C. (1999). Critical issues in prevention of substance abuse. In R. T. Ammerman & P. J. Ott & R. E. Tarter (Eds.), Prevention and Societal Impact of Drug and Alcohol Abuse (pp. 3-20). Mahwah, NJ: Lawrence Erlbaum Associates. Bandura, A., Adams, N. E., & Beyer, J. (1977). Cognitive processes mediating behavioral change. Journal ofPersonalitv and Social Psychology. 35(3). 125-139. Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal ofPersonalitv and Social Psychology. 51(61. 1173-1182. Battistich, V., & Horn, A. (1997). The relationship between students' sense of their school as a community and their involvement in problem behaviors. American Journal of Public Health. 87(121. 1997-2001. Belcher, H. M. E., & Shinitzky, H. E. (1998). Substance abuse in children. Prediction, protection, and prevention. Archives of Pediatric and Adolescent Medicine. 152. 952-960. Bentler, P. (1995). EQS for Windows (Version 5.7b). Encino, CA: Multivariate Software, Inc. Bentler, P. M. (1995). EOS Structural Equations Program Manual. Encino, CA: Multivariate Software, Inc. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 139 Blechman, E. A., Prinz, R. J., & Dumas, J. E. (1995). Coping, competence, and aggression prevention: Part 1. Developmental model. Applied and Preventive Psychology. 4. 211-232. Brook, J. S., & Cohen, P. (1992). A developmental perspective on drug use and delinquency. In J. McCord (Ed.), Facts, frameworks, and forecasts (pp. 231-251). New Brunswick, NJ: Transaction Publishers. Byrne, B. M. (1994). Structural equation modeling with EOS and EOSAVindows: basic concepts, applications, and programming. Thousand Oaks, CA: Sage Publications, Inc. Caplan, M., Weissberg, R. P., Grober, J. S., Sivo, P. J., Grady, K., & Jacoby, C. (1992). Social competence promotion with inner-city and suburban young adolescents: Effects on social adjustment and alcohol use. Journal of Consulting and Clinical Psychology. 60(1). 56-63. Catalano, R. F., & Hawkins, J. D. (1996). The social development model: A theory of antisocial behavior. In J. D. Hawkins (Ed.), Delinquency and Crime. Current Theories (pp. 149-197). New York: Cambridge University Press. Clapper, R. L., Buka, S. L., Goldfield, E. C., Lipsitt, L. P., & Tsuang, M. T. (1995). Adolescent problem behaviors as predictors of adult alcohol diagnoses. International Journal of the Addictions. 30. 507-523. Crick, N. R. (1996). The role of overt aggression, relational aggression, and prosocial behavior in the prediction of children's future social adjustment. Child Development. 67. 2317-2327. Crick, N. R. (1997). Engagement in gender normative versus nonnormatvie forms of aggression: Links to social— psychological adjustment. Developmental Psychology, 33(4), 610-617. Crick, N. R., & Bighee, M. A. (1998). Relational and overt forms of peer victimization: A multiinformant approach. Journal of Consulting and Clinical Psychology. 66. 337-347. Crick, N. R., Bigbee, M. A., & Howes, C. (1996). Gender differences in children's normative beliefs about aggression: How do I hurt thee? Let me count the ways. Child Development, 67. 1003-1014. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 140 Crick, N. R., Casas, J. F., & Mosher, M. (1997). Relational and overt aggression in preschool. Developmental Psychology. 33. 579-588. Crick, N. R., & Gropeter, J. K. (1995). Relational aggression, gender, and social-psychological adjustment. Child Development. 66. 710-722. Crick, N. R., & Gropeter, J. K. (1996). Children's treatment by peers: Victims of relational and overt aggression. Development and Psychopathology. 8. 367-380. Crowley, S. L., & Fan, X. (1997). Structural equation modeling: Basic concepts and applications in personality assessment research. Journal ofPersonalitv Assessment. 68. 508-531. Dahlberg, L. L. (1998). Youth violence in the United States. Major trends, risk factors, and prevention approaches. American Journal of Preventive Medicine. 14, 259-272. Dahlberg, L. L., Toal, S. B., & Behrens, C. B. (1998). Measuring violence- related attitudes, beliefs, and behaviors among youths: A compendium of assessment tools. Atlanta, GA: Centers for Disease Control and Prevention, National Center for Injury Prevention and Control. Dept, of Health and Human Services. (1998). Summary Findings from the 1997 National Household Survey on Drug Abuse. Rockville, MD: National Clearinghouse for Alcohol and Drug Information. Donovan, J. E., & Jessor, R. (1985). Structure of problem behavior in adolescence and young adulthood. Journal of Consulting and Clinical Psychology. 53, 890-904. Donovan, J. E., Jessor, R., & Costa, F. M. (1988). Syndrome of problem behaviors in adolescent young adulthood. Journal of Consulting and Clinical Psychology, 51, 890-904. Dukarm, C. P., Byrd, R. S., Auinger, P., & Weitzman, M. (1996). Illicit substance use, gender and the risk of violent behavior among adolescents. Archives of Pediatric and Adolescent Medicine. 150. 797-801. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 141 Farrington, D. P., Loeber, R., Elliott, D. S., Hawkins, J. D., Kandel, D. B., Klein, M. W., McCord, J., Rowe, D. C., & Tremblay, R. E. (1990). Advancing knowledge about the onset of delinquency and crime. In B. B. Lahey & A. E. Kadzin (Eds.), Advances in clinical child psychology (Vol. 13, pp. 283-342). New York: Plenum Press. Forehand, R., Neighbors, B., & Wierson, M. (1991). The transition to adolescence: the role of gender and stress in problem behavior and competence. Journal of Child Psychology and Psychiatry and Allied Disciplines. 32(6). 929-937. Galen, B. R., & Underwood, M. K. (1997). A developmental investigation of social aggression among children. Developmental Psychology. 33. 589-600. Graham, J. W., Flay, B. R., Johnson, C. A., Hansen, W. B., Grossman, L., & Sobel, J. L. (1984). Reliability of self-report measures of drug use in prevention research: Evaluation of the Project SMART questionnaire via the test-retest reliability matrix. Journal of Drug Education. 14(21.175-193. Hawkins, J. D., Catalano, R. F., & Miller, J. Y. (1992). Risk and protective factors for alcohol and other drug problems in adolescence and early adulthood: Implications for substance abuse prevention. Psychological Bulletin. 112(1). 64-105. Hawkins, J. D., Catalano, R. F., Morrison, D. M., O'Donnell, J., Abbott, R. D., & Day, L. E. (1992). The Seattle Social Development Project: Effects of the first four years on protective factors and problem behaviors. In J. McCord & R. E. Tremblay (Eds.), Preventing Antisocial Behavior: Interventions from Birth through Adolescence, (pp. 139-161). New York: Guilford Press. Hawkins, J. D., Guo, J., Hill, K. G., Battin-Pearson, S., & Abbott, R. D. (2001). Long term effects of the Seattle Social Development Intervention on school bonding trajectories. Applied Developmental Science. 5(4). 225-236. Hawkins, J. D., Von Cleve, E., & Catalano, R. F. (1991). Reducing early childhood aggression: results of a primary prevention program. Journal of the American Academy of Child and Adolescent Psychiatry. 30(21. 208-217. Hodges, E. V. E., Malone, M. J., & Perry, D. G. (1997). Individual risk and social risk as interacting determinants of victimization in the peer group. Developmental Psychology. 33(6). 1032-1039. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 142 Hodges, E. V. E., & Perry, D. G. (1999). Personal and interpersonal antecedents and consequences of victimization by peers. Journal ofPersonalitv & Social Psychology. 76. 677-685. Huang, B., White, H. R., Kosterman, R., Catalano, R. F., & Hawkins, J. D. (2001). Developmental associations between alcohol and interpersonal aggression during adolescence. Journal of Research in Crime and Delinquency. 38(11 64-83. Hudley, C., Britsch, B., Wakefield, W. D., Smith, T., Demorat, M., & Cho, S.-J. (1998). An attribution retraining program to reduce aggression in elementary school students. Psychology in the Schools. 35(31. 271-282. Huesmann, L. R., Eron, L. D., Lefkowitz, M. M., & Walder, L. O. (1984). Stability of aggression over time and generations. Developmental Psychology. 20(61. 1120-1134. Indiana Prevention Resource Center. (2000, August 8, 2000). Alcohol tobacco and other drug use bv Indiana children and adolescents (Prevention Monograph Series: Monograph No. 00-02), [website]. Indiana Prevention Resource Center, Indiana University. Available: http://www.drugs.indiana.edu/drug_stats/youth2000/home.html [2002, January 10]. Ireland, T., & Widom, C. S. (1994). Childhood victimization and risk for alcohol and drug arrests. International Journal of the Addictions. 29. 235-274. Jessor, R. (1984). Adolescent development and behavioral health. In J. D. Matarazzo & S. M. Weiss & J. A. Herd & N. E. Miller & S. M. Weiss (Eds.), Behavioral Health: A Handbook of Health Enhancement and Disease Prevention (pp. 69-90). New York: John Wiley & Sons. Johnston, L. D., O'Malley, P. M., & Bachman, J. G. (1998). National survey results on drug use from the Monitoring the Future Study. 1975-1997. Volume I. Secondary Students (NIH Publication No. 98-4345). Rockville: National Institutes of Health. Kellam, S. G. (1990). Developmental epidemiological framework for family research on depression and aggression. In G. R. Patterson (Ed.), Depression and Aggression in Family Interaction (pp. 11-48). Englewood Cliffs: Lawrence Erlbaum. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 143 Kellam, S. G., Brown, C. H., Rubin, B. R., & Ensminger, M. E. (1983). Paths leading to teenage psychiatric symptoms and substance use: Developmental epidemiological studies in Woodlawn. In S. B. Guze & F. J. Earls & J. E. Barrett (Eds.), Child Psychopathology and Development (pp. 17-51). New York: Raven Press. Kosterman, R., Hawkins, J. D., Guo, J., Catalano, R. F., & Abbot, R. D. (2000). The dynamics of alcohol and marijuana initiation: Patterns and predictors of first use in adolescence. American Journal of Public Health. 90(31. 360-366. Lewis, C. E., Robins, L. N., & Rice, J. (1985). Associations of alcoholism with antisocial personality in urban men. Journal of Nervous and Mental Disease. 173.166-174. Lifrak, P. D., McKay, J. R., Rostain, A., Alterman, A. I., & O'Brien, C. P. (1997). Relationship of perceived competencies, percieved social support, and gender to substance use in young adolescents. In M. Galanter (Ed.), Recent Developments in Alcoholism. Volume 13: Alcoholism and Violence (pp. 245-282). New York: Plenum Press. Lipsey, M. W., Wilson, D. B., Cohen, M. A., & Derzon, J. H. (1997). Is there a causal relationship between alcohol use and violence. In M. Galanter (Ed.), Recent Developments in Alcoholism. Volume 13: Alcoholism and Violence fVol. 13, pp. 245-282). New York: Plenum Press. Little, T. D., Cunningham, W. A., Shahar, G., & Widaman, K. F. To parcel or not to parcel: Exploring the question, weighing the merits. Structural Equation Modeling,J n pj ess. Loeber, R., & Stouthamer-Loeber, M. (1998). Development of juvenile aggression and violence. American Psychologist. 53(21. 242-259. Lynam, D. R. (1996). Early identification of chronic offenders: Who is the fledgling psychopath? Psychological Bulletin. 120(21 209-234. MacCallum, R. C., & Austin, J. T. (2000). Applications of structural equation modeling in psychological research. Annual Review of Psychology. 51. 201-226. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 144 Masten, A. S., & Coatsworth, J. D. (1998). The development of competence in favorable and unfavorable environments. American Psychologist. 53(21. 205-220. Mehrabian, A. (2001). General relations among drug use, alcohol use, and major indexes of psychopathology. J oumal of Psvcholo ev. 135(11. 71-86. Najaka, S. S., Gottfredson, D. C., & Wilson, D. B. (2002). A meta-analytic inquiry into the relationship between selected risk factors and problem behavior. Prevention Science. 2(41.257-271. National Institute on Drug Abuse. (1998). The Economic Costs of Alcohol and Drug Abuse in the United States. 1992 (NIH Publication No. 98-4327.). Washington, DC: National Institutes of Health. O'Donnell, J., Hawkins, J. D., & Abbott, R. D. (1995). Predicting serious delinquency and substance use among aggressive boys. Journal of Consulting and Clinical Psychology. 63(41. 529-537. Olweus, D. (1978). Aggression in the schools: Bullies and whipping bo vs. Washington, DC: Hemisphere. Olweus, D. (1979). Stability of aggressive reaction patterns in males: A review. Psychological Bulletin.. 86. 852-875. Osgood, W. (1995). Drugs, alcohol, and adolescent violence (Center Paper 002). Boulder, CO: Center for the Study and Prevention of Violence. Pakaslahti, L., & Keltikangas-Jarvinen, L. (2000). Comparison of peer, teacher and self-assessments on adolescent direct and indirect aggression. Educational Psychology. 20(2). 177-190. Pentz, M. A., & Chou, C.-P. (1994). Measurement invariance in longitudinal clinical research assuming change from development and intervention. Journal of Consulting and Clinical Psychology. 62.450-462. Perry, D. G., Kusel, S. J., & Perry, L. C. (1988). Victims of peer aggression. Developmental Psychology. 24. 807-814. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 145 Rushton, J. P., Brainerd, C. J., & Pressley, M. (1983). Behavioral development and construct validity: The principle of aggregation. Psychological Bulletin. 9 4 .18-38. SAS Institute. (1998). SAS Statistical Software (Version 6.12). Cary, NC: SAS Institute, Inc. Schwartz, D., Dodge, K. A., & Coie, J. D. (1993). The emergence of chronic peer victimization in boys' play groups. Child Development. 6 4 .1755-1772. Schwartz, D., & Proctor, L. J. (2000). Community violence exposure and children's social adjustment in the school peer group: The mediating roles of emotion regulation and social cognition. Journal of Consulting and Clinical Psychology. 68. 670-683. Sells, C. W., & Blum, R. W. (1996). Morbidity and mortality among U.S. adolescents: An overview of data and trends. American Journal of Public Health. 86. 513-519. Snyder, H. N., & Sickmund, M. (1999). Juvenile offenders and victims: 1999 national report. Washington, DC: Office of Juvenile Justice and Delinquency Prevention. Tanguma, J. (2001). Effects of sample size on the distribution of selected fit indices: A graphical approach. Educational and Psychological Measurement. 61(5). 759-776. Tolan, P. H., & Guerra, N. G. (1994). What works in reducing adolescent violence: An empirical review of the field (Center Paper 001). Boulder, CO: Center for the Study and Prevention of Violence. Tremblay, R. E., Masse, B., Perron, D., Leblanc, M., Schwartzman, A. E., & Ledingham, J. E. (1992). Early disruptive behavior, poor school achievement, delinquent behavior, and delinquent personality: longitudinal analyses. Journal of Consulting and Clinical Psychology. 60. 64-72. U.S. Department of Justice. (2000). Criminal victimization in the United States. 1997 (NCJ-174446): U. S. Department of Justice, Bureau of Justice Statistics. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 146 Weiner, M. D., Pentz, M. A., Turner, G. M., & Dwyer, J. H. (2001). From early adolescence to young adulthood: alcohol use and anger relationships. Journal of Adolescent Health. 28.450-457. White, H. R. (1997). Longitudinal perspective on alcohol use and aggression during adolescence. Recent Developments in Alcoholism. 13. 81-103. White, H. R., Brick, J., & Hansell, S. (1993). A longitudinal investigation of alcohol use and aggression in adolescence. Journal of Studies on Alcohol. Supplement 11. 62-77. White, H. R., Loeber, R., Stouthamer-Loeber, M., & Farrington, D. P. (1999). Developmental associations between substance use and violence. Development and Psychopathology. 11. 785-803. Widom, C. S., Ireland, T., & Glynn, P. J. (1995). Alcohol abuse in abused and neglected children followed-up: are they at increased risk? Journal of Studies on Alcohol. 56. 207-217. Wills, T. A., Cleary, S., Filer, M., Shinar, O., Mariani, J., & Spera, K. (2001). Temperament related to early-onset substance use: Test of a developmental model. Prevention Science. 2(3). 145-163. Windle, M. (1994). Substance use, risky behaviors, and victimization among a US national adolescent sample. Addiction. 89.175-182. Yudko, E., Blanchard, D. C., Henrie, J. A., & Blanchard, R. J. (1997). Emerging themes in preclinical research on alcohol and aggression. Recent Developments in Alcoholism. 13. 123-138. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 147 APPENDIX A. PSYCHOMETRIC PROPERTIES OF MEASURES Ind Construct icators Response Choices Psychometric Properties source Overt Last year, did you get into aggression physical fights with other kids (hitting, pushing, or shoving other kids)? This can include when you fight with your brothers or sisters. Did you destroy anything on purpose which did not belong to you? Last year, did you push or shove other kids? Grades 4-5: l=NO!, 2=no, 3=yes, 4=YES!; Grade 6: l=No, 2=Yes 1 or 2 times, 3=Yes 3 or 4 times, 4=Yes 5 or more times Cronbach’s a = ■73; Test-retest reliability over 6-month interval r = .56, p < .0001 Items adapted from Orpinas, 1993 and Institute of Behavioral Science, 1987 as cited in (Dahlberg et al., 1998) Did you call others bad names? Relational Did you, on purpose, leave aggression someone out o f your group or not invite them to play? a Did you get back at a kid you are mad at by not letting them be in your group anymore? Have you told lies about another kid to make others not like them anymore? Grade 4: l=NO!, 2=no, 3=yes, 4=YES!; Grade 6 l=Never, 2=Almost never, 3=Sometimes, 4=Most of the time, 5=A11 the time Test-retest reliability over 4 week interval, boys r = 86, girls r = .80; over 6- month interval, boys r = .56, girls r = .68 Cronbach’s a = .94 Have you told another kid you won’t like them unless they do what you say? Have you tried to keep others from liking another by saying mean things about the kid? Overt victimization Did you get hit and pushed by other kids? Did you get beat up by other kids? Grade 6: 1 = No, 2 = Yes 1 or 2 times, 3 = Yes 3 or 4 times, 4 = Yes 5 or more times Test-retest reliability over 1-year period in peer nomination format Adapted from peer nomination scale (Hodges et al., 1997) Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 148 Construct Indicators Response Choices Psychometric Source Properties Did you get called names by r= .84 other kids? Did kids do mean things to you? Relational Has someone left you out on victimization purpose when it’s time to play or do an activity? Did a kid who was mad at you get back at you by not letting you be in their group anymore? Has another kid told lies about you to make others not like you anymore? Grade 6: 1 = Never, 2 = Almost never, 3 = Sometimes, 4 = Most o f the time, 5 = All the time Cronbach’s a = .80 Evidence o f convergent validity using correlation between self and peer report, boys r = .39, girls r = .35 Has another kid told you they won’ t like you unless you do what they say? Has another kid tried to keep others from liking you by saying mean things about you? Drug use Have you ever tried a cigarette?b Have you ever tried alcohol?b Have you ever tried marijuana?b How many cigarettes have you smoked in your whole life? c How many alcoholic drinks have you had in your whole life? d How many times have you used marijuana in your whole life? ' Grade 4-all substances: 1 = NO!, 2 = no, 3 = yes, 4 = YES!; Grade 6 Cigarette: 1 = None, I’ve never had one puff of a cigarette, 2 = One puff, 3 = Part or all o f one cigarette, 4 = 2 to 4 cigarettes, 5 = 5 to 20 cigarettes, 6 = 1 to 5 packs, 7 = more than 5 packs Grade 6 Alcohol: 1 = None, I haven’t ever had one sip o f an alcoholic drink, 2 = Only sips, 3 = Part or all o f one drink, 4 = 2 to 4 drinks, 5 = 5 to 10 drinks, 6 = 11 to cigarette use test-retest reliability r = .64, alcohol use test-retest reliability r = .74, marijuana use test-retest reliability r = .82 (Graham et a l, 1984); Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 149 Construct Indicators Response Choices Psychometric Properties Source 20 drinks, 7 = 21 to 100 drinks, 8 = more than 100 drinks Grade 6 Marijuana: 1 = None, I’ve never used marijuana, 2 = Once, 3 = 2 to 4 times, 4 = 5 to 10 times, 5 = 11 to 20 times, 6 = 21 to 100 times, 7 = More than 100 times Social competence This year, did you stop and really listen to a friend, even though you were busy doing something else? Grade 5: 1 = N o, 2 = Yes, 1 or 2 times, 3 = Yes, more than 2 times Cronbach’s a = .66 and .81 (fifth and sixth grades, respectively) Not applicable- Measure developed for this This year, did you help other kids in your class when they needed it (not to cheat but to help)? Did you stand up for your values (things you believe in)? Grade 6: 1 = N o, 2 = Yes 1 or 2 times, 3 = Yes 3 or 4 times, 4 = Yes 5 or more times study This year, did you talk with your friends to find a solution to a problem that you could all agree on? When you didn’t agree with other kids on something, did you work out the problem? Did you help someone learn a new skill? a This relational aggression item alone was measured at grade 4 b Item measured in grade 4 pretest and post-test and used response choices under “Grade 4" c Item measured in grade 6 post-test and used response choices under “Grade 6 Cigarette" * Item measured in grade 6 post-test and used response choices under “Grade 6 Alcohol" ' Item measured in grade 6 post-test and used response choices under “Grade 6 Marijuana" Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Adolescent cigarette use and psychological distress: relation to adult obesity risk

PDF

Adolescent stress and styles of coping: Predictors and moderators of psychological distress

PDF

Emotional intelligence and smoking risk factors in early adolescents

PDF

Motivation and the meanings of health behavior as factors associated with eating behavior in Latino youth

PDF

Effects of parent stress on weight-related parenting practices and child obesity risk

PDF

Adaptation to a life -threatening diagnosis: Dispositional optimism and pessimism and posttraumatic growth among patients with HIV

PDF

Exploring the role of peer influence, linguistic acculturation, and social networks in substance use

PDF

A network analysis of online and offline social influence processes in relation to adolescent smoking and alcohol use

PDF

Social selection and influence of alcohol & marijuana implicit cognitions and behaviors: a longitudinal investigation of peer social network dynamics

PDF

Psychosocial and behavioral ractors associated with emotional eating in adolescents

PDF

Correlates of compliance with mammography screening guidelines among low-income Latinas: An exploratory study

PDF

Using a structural model of psychopathology to distinguish relations between shared and specific features of psychopathology, smoking, and underlying mechanisms

PDF

Correlates of smoking behaviour among Egyptian adolescents

PDF

Competing risks: the role of the perceived consequences of refusing to share injection equipment among injection drug users

PDF

Implicit cognition and dissociative experiences as predictors of adolescent substance use

PDF

Biopsychosocial factors in major depressive disorder

PDF

Investigating factors that influence peer relationships and obesity during middle childhood

PDF

Interaction of dietary fiber and serum cholesterol on early atherosclerosis

PDF

Objectively measured physical activity and related factors in minority youth

PDF

Non-steroidal anti-inflammatory drugs, anti-acid drugs and passive smoking as impact factors for esophageal and gastric adenocarcinomas

Asset Metadata

Creator Weiner, Michelle Diane (author)

Core Title Aggression and victimization as predictors of drug use in early adolescence

School Graduate School

Degree Doctor of Philosophy

Degree Program Preventive Medicine - Health Behavior Research

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag health sciences, public health,OAI-PMH Harvest,psychology, behavioral,psychology, developmental,sociology, individual and family studies

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Pentz, Mary Ann (committee chair), Chou, Chih-Ping (committee member), Sussman, Steve (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-393276

Unique identifier UC11340728

Identifier 3180785.pdf (filename),usctheses-c16-393276 (legacy record id)

Legacy Identifier 3180785.pdf

Dmrecord 393276

Document Type Dissertation

Rights Weiner, Michelle Diane

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA