Citation
Risk Factors for Violence among Early Adolescents

Material Information

Title:
Risk Factors for Violence among Early Adolescents Explaining Gender and Racial/Ethnic Disparities in Trajectories of Violent Delinquency
Creator:
Reingle, Jennifer M.
Place of Publication:
[Gainesville, Fla.]
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (287 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Epidemiology
Committee Chair:
Maldonado-Molina, Mildred
Committee Members:
Shuster, Jonathan J
Komro, Kelli Ann
McCarty, Christopher
Jennings, Wesley
Graduation Date:
8/6/2011

Subjects

Subjects / Keywords:
Adolescents ( jstor )
African Americans ( jstor )
Alcohols ( jstor )
Delinquency ( jstor )
Escalators ( jstor )
Hispanics ( jstor )
Marijuana use ( jstor )
Predisposing factors ( jstor )
Trajectories ( jstor )
Violence ( jstor )
Epidemiology -- Dissertations, Academic -- UF
adolescents -- aggression -- alcohol -- ethnicity -- longitudinal -- race -- trajectory -- urban -- violence
Genre:
Electronic Thesis or Dissertation
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
Epidemiology thesis, Ph.D.

Notes

Abstract:
Objective. The purpose of this study is twofold: 1) to evaluate the patterns of violence by race/ethnicity and gender using two longitudinal samples of adolescents, and 2) to evaluate the differential multiple domains of predictors of violence by subgroup. Methods. Study #1 a longitudinal, nationally representative sample of adolescents followed from ages 15 to 26. Group-based trajectory modeling was used to estimate groups of participants who have similar characteristics in their levels of violence over time. Multinomial and survey logistic regression were used to evaluate the effects of group fighting and multilevel risk and protective factors on trajectory membership. Study #2 utilized the same national, longitudinal study of adolescents followed from ages 15 to 26. In this study, trajectories were estimated for each racial/ethnic and gender subgroup separately. Multinomial and survey logistic regression were used to test for differential risk factors for violent trajectory membership by subgroup. Finally, Study #3 utilized a longitudinal, high-risk sample of urban adolescents followed from 6th ? 8th grade. Group-based trajectory models were used to create latent trajectory groups of violence over time by racial/ethnic subgroup, and multinomial logistic regression were used to test for differential risk factors for membership in these trajectories. Results. In the nationally representative sample, three groups were found for the overall sample as well as each subgroup (White males, White females, African-American males, African-American females, Hispanic males, Hispanic females, Asians, and Native Americans). In the high-risk, urban sample, four trajectory groups were found for both African-Americans and Hispanics. The predictors of membership in the high-risk trajectory groups varied substantially between samples and within samples by race/ethnicity and gender. Conclusions. This study provides evidence that there is some stability in the pattern of violence across racial/ethnic and gender groups. However, the proportion of adolescents involved in the high-risk violence trajectory groups varies by race/ethnicity and gender. This study also provides additional evidence for a late-onset group of escalators, and identified risk and protective factors for serious violence broken down by demographic subgroup. In addition, the large sample size allowed us to identify key variables that serve as mediators for contextual-, peer- and family-level variables, an important component of violence prevention programming. These findings lead future research to examine and identify the driving influences for the disparities in violence by subgroup, and to further understand the risk factors for late-onset escalation of violent behavior. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2011.
Local:
Adviser: Maldonado-Molina, Mildred.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-08-31
Statement of Responsibility:
by Jenn Reingle.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Embargo Date:
8/31/2013
Classification:
LD1780 2011 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

1 RISK FACTORS FOR VIOLENCE AMONG EARLY ADOLESCENT S: EXPLAINING GENDER AND RACIAL/ETHNIC DISPARITI ES IN TRAJECTORIES OF VIOLENT DELINQUENCY By JENNIFER M. REINGLE A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERISTY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2011

PAGE 2

2 2011 Jennifer M. Reingle

PAGE 3

3 ACKNOWLEDGEMENTS First, I wish to thank the members of my dissertation committee, Drs. Mildred Maldonad o-Molina, Kelli Komro, Wesley Jenni ngs, Jonathan Shuster, and Christopher McCarty, for their coun tless hours spent reviewing and revi sing this work. I am thankful to have stumbled upon such a supportive group of mentors who supported and challenged me along my journey. I am confi dent that working with such a brilliant group of professionals has enhanced my future career and individual character. I especially wish to express my appreciati on to my co-chairs, Mildred MaldonadoMolina and Kelli Komro. Kelli was responsible for a dramatic change in my trajectory as a doctoral student, convincing me to become a student in the new Epidemiology program. This was one of the tu rning points in my student car eer, as I did not realize the magnitude of the benefits t hat would come from this transit ion. Being assigned to work with Mildred was a blessing, and I am thankful everyday t hat I had the opportunity to collaborate with such a highly regarded, successful, patien t, and sincere woman. The time and energy that she spent teaching an d supporting me (especially in the early years!) warrants more thanks than this ack nowledgement alone. M ildred has served as a role model both professiona lly and personally, and she inspires me to be successful in all dimensions of my life. Second, I must thank those who laid the foundation for my success in higher education. At the University of North Caro lina Wilmington, Dr. Mike Adams inspired me to major in Criminal Justice. Two years later, Dr. Mi chael Maume hired me as a research assistant, working on my first gr ant. Those days of laying on the beach, recreationally reading a book on disproportionat e minority contact lead me to pursue a masters degree in Criminal Justice at t he University of Cincinnati. At UC, I was

PAGE 4

4 fortunate enough to be mentored by leading res earchers, Ojmarrh Mi tchell and Michael Benson. The encouragement they provided led me to pursue doctoral work at the University of Florida with Dr. Dennis T hombs. Dr. Thombs unbiased mentoring, guidance, and professional network allowed me to find my home in the Department of Health Outcomes and Policy. Third, I wish to thank my family, w ho has been incredibly supportive of my education from freshman year of my under graduate schooling (back in 2002) through the present. Despite my constant travel s and the growing physical distance between us, their continued interest in my work and dedication to my success has been a driving force in allowing me to attain my educational goals. Without this support, I could not be where I am today. I must especially thank my mother for her inst illing in me the desire (and capacity!) to learn, and I thank my fat her for the determination and persistence in overcoming the many obstacles that pr esented themselves while completing my doctoral degree. Finally, this acknowledgement would be incomplete without genuine thanks to my best friend and partner, Bryon. His tremendous love and support throughout the last few years of my doctoral work has made this journey a more enjoyable experience. His constant questioning of my results and how they apply to real life pushed me outside my comfort zone in explaining what our work r eally means to real people. His ambitious nature has instilled drive and determination in me, and I must attribute a portion of my success to his encouragement. This mile stone is ours together representing a segment of the great things to be accomplished in years to come.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDG EMENTS ............................................................................................... 3LIST OF TABLES ............................................................................................................ 9LIST OF FIGURES ........................................................................................................ 14 CHA PTER 1 BACKGROUND AND S IGNIFICA NCE ................................................................... 17Disparities in the Prevalence of Delin quency and Violence among Youth in the United St ates ....................................................................................................... 18Youth Risk Behavio r Surve illance .................................................................... 19Monitoring t he Future ....................................................................................... 20Gender Differences in De linquency and Vi olence ................................................... 20Racial/Ethnic Differences in Delinquency and Violen ce .......................................... 23The Relationship between Alcohol and Vi olence .................................................... 27Gang Membership vers us Group Fi ghting .............................................................. 30Theoretical Foundatio n ........................................................................................... 32Social Learni ng Theo ry .................................................................................... 33Social Bond Theory .......................................................................................... 35Moffitts Life-Course Perspec tive ...................................................................... 36Social Disorganiza tion T heory .......................................................................... 39Contribution of the Current Study ........................................................................... 42Study Goal, Aims, and Research Q uestions ........................................................... 422 METHODOLOGY ................................................................................................... 45Study #1: Risk Factors Associated with Tr ajectories of Violent Delinquency in a Nationally Representative, Longitudinal Sample ................................................. 45National Longitudinal St udy of Adolescent H ealth (Add Health) ....................... 45Study Design .............................................................................................. 45Data Collection ........................................................................................... 46Participants ................................................................................................ 47Measures ................................................................................................... 47Study # 2: Gender and Racial/Ethnic Di fferences in Risk Factors Associated with Trajectories of Violent Delinquency in a Nationally Representative, Longitudinal Sample ............................................................................................ 51Study #3: Risk Factors Associated with Raci al/Ethnic Differences in Trajectories of Violent Delinquency in Longitudinal Sample of High-Risk, Urban Youth ......... 52Project Northland Chicago ................................................................................ 53Study des ign .............................................................................................. 53Participants ................................................................................................ 54

PAGE 6

6 Measures ................................................................................................... 54Analytical Methods ........................................................................................... 58Group-Based Trajec tory M odeling ............................................................. 58Multinomial Logistic Regression ................................................................. 60Mediation Analyses .................................................................................... 61Summary of Methodology ................................................................................. 623 RISK FACTORS ASSOCIATED WITH TRAJECTORIES OF SERIOUS VIOLENCE IN A NATIONALLY REPR ESENTATIVE, LONGITUDINAL SAMPLE 67Background ............................................................................................................. 68Method .................................................................................................................... 71Design .............................................................................................................. 71Data Coll ection ................................................................................................. 72Measures .......................................................................................................... 73Analytical Methods .................................................................................................. 77Group-Based Trajec tory M odelin g .................................................................... 77Multinomial Logistic Regression ....................................................................... 80Mediation Analyses .......................................................................................... 81Results .................................................................................................................... 82Trajectories of Violen ce .................................................................................... 82Effects of risk and protective fact ors at age 15 on trajectories of violence .................................................................................................. 83Mediated effects of contextual variables on violent trajectory membership ............................................................................................ 85Discuss ion .............................................................................................................. 864 GENDER AND RACIAL/ETHNIC DIFFERENCES IN RISK FACTORS ASSOCIAT ED WITH TRAJECTORIES OF VIOLENT DELINQUENCY IN A NATIONALLY REPRESENTATIVE LONGITUDINA L SAMPLE .......................... 104Background ........................................................................................................... 105Race/Ethnic Differenc es in Vi olence ..................................................................... 106Gender Differences in Vi olence ............................................................................ 109Method .................................................................................................................. 111Design ............................................................................................................ 111Data Coll ection ............................................................................................... 112Measures ........................................................................................................ 113Analytical Methods ................................................................................................ 118Group-Based Trajec tory M odelin g .................................................................. 118Multinomial Logistic Regre ssion ..................................................................... 121Mediation A nalysis ......................................................................................... 122Results .................................................................................................................. 123Trajectories of Violen ce .................................................................................. 123Effects of Risk and Protective Factors at Age 15 on Trajectories of Physical Aggression .................................................................................................. 125White ma les ............................................................................................. 125

PAGE 7

7 White females .......................................................................................... 127African-Amer ican ma les ........................................................................... 130African-Americ an females ........................................................................ 132Hispanic males ......................................................................................... 134Hispanic fe males ...................................................................................... 136Asians ...................................................................................................... 138Native Am ericans ..................................................................................... 139Discuss ion ............................................................................................................ 141Gender Diffe rences ........................................................................................ 143Racial/Ethnic Differenc es ............................................................................... 145Disclos ure ............................................................................................................. 1495 RISK FACTORS ASSOCIATED WITH RACIAL/ETHNIC DIFFERENCES IN TRAJECTORIES OF AGGRESSION IN A LONGITUDINAL SAMP LE OF HIGHRISK, URBAN YOUTH ......................................................................................... 214Background ........................................................................................................... 215Methods ................................................................................................................ 220Design ............................................................................................................ 220Participants ..................................................................................................... 221Measures ........................................................................................................ 221Analytical Methods ................................................................................................ 227Group-Based Trajec tory M odelin g .................................................................. 227Multinomial Logistic Regre ssion ..................................................................... 229Mediation Analyses ........................................................................................ 230Results .................................................................................................................. 231Ethnic Differences in Traj ectories of Aggression ............................................ 231Effects of Risk and Protective Factors at 6th Grade on Trajectories of Aggression .................................................................................................. 232Bivariate results, African-Am ericans ........................................................ 232Bivariate result s, Hispani cs ...................................................................... 233Multivariate community-, parent-, peer-, and individual-level results, African-Am ericans ................................................................................ 235Multivariate parent-, peer-, and indi vidual-level results, Hispanics ........... 236Multivariate Results Adjusted fo r Baseline Aggression, AfricanAmericans ............................................................................................. 237Multivariate Results Adjusted fo r Baseline Aggression, Hispanics ........... 238Mediation, Afri can-Amer icans .................................................................. 238Mediation, Hispanics ................................................................................ 239Characteristics of sample at baseline (post-hoc), African-Americans ...... 240Characteristics of sample at baseline (post-hoc), Hispanics .................... 240Discuss ion ............................................................................................................ 2416 DISCUSSI ON ....................................................................................................... 268LIST OF RE FERENCES ............................................................................................. 276

PAGE 8

8 BIOGRAPHICAL SKETCH .......................................................................................... 286

PAGE 9

9 LIST OF TABLES 2-1 Description of dependent variables, Add Health, Waves I-IV. N = 9421. ........... 63 2-2 Description of covari ates, Add Health, Wave I. ................................................... 64 2-3 Description of dependent variabl es, PNC (ST02-ST05). N= 4188. ..................... 65 2-4 Description of co variates, P NC (ST 02). .............................................................. 66 3-2 Description of violence (mean,SD) over time by trajectory group, Add Health. .. 95 3-3 Bivariate effects between risk/protective factors and trajectories of violence. .... 96 3-4 Community, family, and peer effe cts on trajectories of violence. ........................ 97 3-5 Effects of multiple domain s on trajectories of viol ence. ...................................... 98 3-6 Effects of multiple domains on trajecto ries of violence, adjusted for baseline. ... 99 3-7 Regression models testing the association between community-level and parent and peer level variables on indivi dual-level risk and protective factors. 101 3-8 Mediated effect of parentand peer-lev el variables on violence trajectories. ... 102 3-9 Post-hoc description (means and percentages) of adolescents who were violent at Wave I. .............................................................................................. 103 4-1 Description of samp le, Add Health. N=9421. .................................................... 155 4-2 Model fit statistics by racial/e thnic and gender subgroup, 3-group model. ....... 156 4-3 Racial/ethnic and gender group differ ences in mean violence by trajectory group. ............................................................................................................... 157 4-4 Bivariate effects between risk/protecti ve factors and trajectories of violence, White ma les. ..................................................................................................... 158 4-5 Community, family, and peer effects on trajectories of violence, White males. 159 4-6 Effects of multiple domains of risk factors on trajectories of violence, White males. ............................................................................................................... 160 4-7 Effects of multiple domains of risk fa ctors on trajectories of violence, adjusted for baseline, Wh ite males. ................................................................................ 161

PAGE 10

10 4-8 Regression models testing the association bet ween community-level and peer level variables on individual-level risk and protective factors, White males. ............................................................................................................... 163 4-9 Mediated effect of parentand peer-lev el variables on violence trajectories, White ma les. ..................................................................................................... 164 4-10 Post-hoc description (means and propor tions) of adolescents who are violent at Wave I, Wh ite male s. .................................................................................... 165 4-11 Bivariate effects between risk/protecti ve factors and trajectories of violence, White fema les. .................................................................................................. 166 4-12 Effects of multiple domains of risk factors on trajectories of violence, White females. ............................................................................................................ 167 4-13 Effects of multiple domains of risk fa ctors on trajectories of violence, adjusted for baseline, Wh ite fema les. ............................................................................. 168 4-14 Regression models testing the association between parent and peer level variables and individual-level risk and pr otective factors, White females. ........ 170 4-15 Mediated effect of parentand peer-lev el variables on violence trajectories, White fema les. .................................................................................................. 171 4-16 Post-hoc description (means and proportions) of adolescents who were violent at Wave I, White fe males. ..................................................................... 172 4-17 Bivariate effects between risk/protecti ve factors and trajectories of violence, African-Americ an males. .................................................................................. 173 4-18 Effects of multiple domains of risk factors on trajectories of violence, AfricanAmerican ma les. ............................................................................................... 174 4-19 Effects of multiple domains of risk factors on trajectories of violence adjusted for baseline, Afric an-American males. .............................................................. 175 4-20 Regression models testing the association between parent and peer level variables and individual-level risk and protective factors, African-American males. ............................................................................................................... 177 4-21 Mediated effect of peer marijuana use on violence trajectories, AfricanAmerican ma les. ............................................................................................... 178 4-22 Post-hoc description (means and proportions) of adolescents who were violent at Wave I, African-Americ an male s. ...................................................... 179

PAGE 11

11 4-23 Bivariate effects between risk/protecti ve factors and trajectories of vi olence, African-Americ an female s. ............................................................................... 180 4-24 Effects of multiple domains of risk factors on trajectories of violence, AfricanAmerican fe males. ............................................................................................ 181 4-25 Effects of multiple domains of risk factors on violence, adjusted for baseline, African-Americ an female s. ............................................................................... 182 4-26 Regression models testing the asso ciation between peer level variables on individual-level risk and protective fact ors, African-American females. ............ 184 4-27 Mediated effect of peer marijuana use on violence trajectories, AfricanAmerican fe males. ............................................................................................ 185 4-28 Post-hoc description (means and proportions) of adolescents who were violent at Wave I, Afri can-American females. ................................................... 186 4-29 Bivariate effects between risk/protecti ve factors and trajectories of violence, Hispanic ma les. ................................................................................................ 187 4-30 Effects of multiple domains of ri sk factors on violence, Hispanic males. .......... 188 4-31 Effects of multiple domains of risk fa ctors on trajectories of violence, adjusted for baseline violence, Hispanic males. .............................................................. 189 4-32 Regression models testing the association between parent and peer level variables on individual-level risk and pr otective factors, Hispanic males. ......... 191 4-33 Mediated effect of peer-level vari ables on violence trajectories, Hispanic males. ............................................................................................................... 192 4-34 Post-hoc description (means and proportions) of adolescents who were violent at Wave I, Hispanic ma les. .................................................................... 193 4-35 Bivariate effects between risk/protecti ve factors and trajectories of violence, Hispanic fe males. ............................................................................................. 194 4-36 Effects of multiple domains of risk factors on trajectories of violence, Hispanic females. ............................................................................................................ 195 4-37 Effects of multiple domains of risk fa ctors on trajectories of violence, adjusted for baseline violence, Hispanic fe males. ........................................................... 196 4-38 Regression models testing the a ssociation between communityand peer level variables on individual-level risk and protective factors, Hispanic females. ............................................................................................................ 198

PAGE 12

12 4-39 Mediated effect of parentand peer-lev el variables on violence trajectories, Hispanic fe males. ............................................................................................. 199 4-40 Bivariate effects between risk/protecti ve factors and trajectories of violence, Asians. .............................................................................................................. 200 4-41 Effects of multiple domains of risk fa ctors on trajectories of violence, Asians. 201 4-42 Effects of multiple domains of risk fa ctors on trajectories of violence, adjusted for baseline, Asians. ......................................................................................... 202 4-43 Regression models testing the association between parent and peer level variables and individual-level risk and protective fact ors, Asians. .................... 204 4-44 Mediated effect of contextual vari ables on violence trajectories, Asians. ......... 205 4-45 Post-hoc description (means and proportions) of adolescents who were violent at Wave I, As ians. ................................................................................. 206 4-46 Bivariate effects between risk/protecti ve factors and trajectories of violence, Native Am ericans. ............................................................................................ 207 4-47 Effects of multiple domains of risk factors on trajectories of violence, Native Americ ans......................................................................................................... 208 4-48 Effects of multiple domains of risk fa ctors on trajectories of violence, adjusted for baseline, Nati ve Amer icans. ........................................................................ 209 4-49 Regression models testing the a ssociation between community-level and parent and peer level variables on indivi dual-level risk and protective factors, Native Am ericans. ............................................................................................ 211 4-50 Mediated effect of contextual, parentand peer-level variables on violence trajectories, Nati ve Amer icans. ......................................................................... 212 4-51 Post-hoc description (means and proportions) of adolescents who were violent at Wave I, Native Am ericans. ................................................................ 213 5-1 Description of sa mple, PNC. N=4,188. ............................................................. 249 5-2 Mean (SD) aggression in each trajectory group for African-Americans and Hispanics. ......................................................................................................... 250 5-3 Bivariate effects between risk/prot ective factors and trajectories of aggression, Afric an-Americ ans. ........................................................................ 251 5-4 Community, family and peer effects on trajectories of aggression, AfricanAmeric ans......................................................................................................... 252

PAGE 13

13 5-5 Effects of multiple domains of ri sk factors on trajectories of aggression, African-Am ericans. ........................................................................................... 253 5-6 Effects of multiple domains of ri sk factors on trajectories of aggression, adjusted for baseline, African-Am ericans. ........................................................ 254 5-7 Regression models testing the association between parent and peer level variables and individual-level risk and pr otective factors, African-Americans. .. 256 5-8 Mediated effect of peer marijuana use on aggression trajectories, AfricanAmeric ans......................................................................................................... 257 5-9 Post-hoc description (means and proportions) of adolescents who were aggressive at Wave I, African-Am ericans. ........................................................ 258 5-10 Bivariate effects between risk/prot ective factors and trajectories of aggression, Hispanics ..................................................................................... 259 5-11 Community, family and peer effects on trajectories of aggression, Hispanics. 260 5-12 Effects of multiple domains of ri sk factors on trajectories of aggression, Hispanics. ......................................................................................................... 261 5-13 Effects of multiple domains of ri sk factors on trajectories of aggression, adjusted for baseli ne, Hispa nics. ...................................................................... 262 5-14 Regression models testing the association between parent and peer level variables and individual-level risk and protective factors, Hispanics. ................ 264 5-15 Mediated effect of mult iple domains of risk factor s on aggression trajectories, Hispanics. ......................................................................................................... 265 5-16 Post-hoc description (means and proportions) of adolescents who were aggressive at 6th grade (baseline) Hispani cs. .................................................. 267

PAGE 14

14 LIST OF FIGURES Figure page 3-1 Trajectories of violence over time Add Health full sa mple. ............................... 94 3-2 Test of the m ediation pathway (based on bivariate analysis). .......................... 100 4-1 Trajectories of violence by racial/ethnic and gender subg roups. ...................... 150 4-2 Differences in the prevalence of violence between racial/ethnic and gender subgroups ........................................................................................................ 154 4-3 Test of the mediati on pathway (based on bivariate analysis), White males. ..... 162 4-4 Test of the mediati on pathway (based on bivariate analysis), White females. .. 169 4-5 Test of the mediation pathway (based on bi variate analysis), AfricanAmerican ma les. ............................................................................................... 176 4-6 Test of the mediation pathway (based on bi variate analysis), AfricanAmerican fe males. ............................................................................................ 183 4-7 Test of the mediat ion pathway (based on bivari ate analysis), Hispanic males. 190 4-8 Test of the mediation pathway (based on bi variate analysis), Hispanic females. ............................................................................................................ 197 4-9 Test of the mediation pathway (based on bi variate analysis), Asians. .............. 203 4-10 Test of the medi ation pathway (based on bi variate analysis), Native Americ ans......................................................................................................... 210 5-1 Trajectories of aggr ession over ti me, PNC. ...................................................... 248 5-2 Test of the mediation pathway (based on bi variate analysis), AfricanAmeric ans......................................................................................................... 255 5-3 Test of the mediat ion pathway (based on biva riate analysis), Hispanics. ......... 263

PAGE 15

15 Abstract of Dissertation Pr esented to the Graduate School of the University of Florida in Partial Fulf illment of the Requirements for t he Degree of Doctor of Philosophy RISK FACTORS FOR VIOLENCE AMONG EARLY ADOLESCENT S: EXPLAINING GENDER AND RACIAL/ETHNIC DISPARITI ES IN TRAJECTORIES OF VIOLENT DELINQUENCY By Jennifer M. Reingle August 2011 Chair: Mildred M. Maldonado-Molina Major: Epidemiology The purpose of this study is twofold: 1) to evaluate the patterns of violence by race/ethnicity and gender using two longitud inal samples of adolescents, and 2) to evaluate the differential mult iple domains of predictors of violence by subgroup. Methods. Study #1 a longitudinal, nationally representative sample of adolescents followed from ages 15 to 26. Group-based trajectory model ing was used to estimate groups of participants who have similar characteri stics in their levels of violence over time. Multinomial and survey logistic regr ession were used to evaluate the effects of group fighting and multilevel risk and protective factors on trajectory membership. Study #2 utilized the same national, longitu dinal study of adolescents followed from ages 15 to 26. In this study, trajectories were estimated for each racial/ethnic and gender subgroup separately. Mu ltinomial and survey logistic regression were used to test for differential risk fact ors for violent trajectory mem bership by subgroup. Finally, Study #3 utilized a longitudinal, high-risk samp le of urban adolescents followed from 6th 8th grade. Group-based trajectory models were used to create latent trajectory groups

PAGE 16

16 of violence over time by racial/ethnic subgr oup, and multinomial logi stic regression were used to test for differential risk factors for me mbership in these trajectories. Results. In the nationally representative sample, three groups were found for the overall sample as well as each subgroup (White males, White females, African-American males, AfricanAmerican females, Hispanic males, Hispanic fe males, Asians, and Native Americans). In the high-risk, urban sample, four traj ectory groups were found for both AfricanAmericans and Hispanics. The predictors of membership in the high-risk trajectory groups varied substantially between samples and within samples by race/ethnicity and gender. Conclusions. This study provides evidence that there is some stability in the pattern of violence across racial/ethni c and gender groups. However, the proportion of adolescents involved in the high-risk violence trajectory groups varies by race/ethnicity and gender. This study also provides additio nal evidence for a late-onset group of escalators, and identified risk and protective fa ctors for serious violence broken down by demographic subgroup. In additi on, the large sample size allowed us to identify key variables that serve as mediators for contex tual-, peerand family-level variables, an important component of violenc e prevention programming. These findings lead future research to examine and identif y the driving influences for the disparities in violence by subgroup, and to further understand the risk factors for late-onset escalation of violent behavior.

PAGE 17

17 CHAPTER 1 BACKGROUND AND SIGNIFICANCE Each year, nearly 700, 000 adolescents and young adults (10-24) are treated in the emergency room for injuries related to violent activity (CDC, 2009). Evidence suggests that adolescents w ho engage in delinquent behavior are more likely to engage in other high-risk activities (e.g., alcoho l and other drug use, dr opping out of school, gun ownership, gang membership, risky sexual activity and familial independence) (Thornberry, Huizinga, & Loeber, 1995; CDC, 2009; 2010) and increase their risk of health-related consequences (including serious injury and death) (Conseur, Rivara, & Emanuel, 1997; Farrington & Loeber, 2000). Using the epidemiological perspective, the initial stage of prevention is a complete understanding of the causes and risk factors for delinquent behavior, including violence. The purpose of this study is to examine the differential risk and protective factors for violence by subgroup (boys versus girls; African-Americans versus Whites, etc.) longitudinally, focusing on mu ltiple domains of risk factor s for violence over the life course. These results will inform preventive efforts by identifying the specific risk factors and/or protective factors that should be addressed depending upon the unique exposures of adolescents. Racial/ethnic and gender differences in violence are important because the gender gap in violent delinquency has been narrowing in recent decades (ChesneyLind, 1983; Steffensmeier, Zhong, Ackerma n, Schwartz, & Agha, 2006), and many studies focus on Whites and African-Americans, grouping members of other racial and ethnic groups into a heterogeneo us Other category. Res earchers are now beginning to focus on females in the literature on de linquency, as they have historically been

PAGE 18

18 excluded because of their relative low frequen cy of offending compar ed to men. More recently, there has been evidence that gi rls engage in different types of delinquency compared to boys (Bellair & Mcnulty, 2005; Willi ams, Van Dorn, Ayers, Bright, Abbott, & Hawkins, 2007). Racial/ethnic groups also appear to differ in their levels and characteristics of violent offending (Willia ms, Van Dorn, Ayers, Bright, Abbott, & Hawkins, 2007; Farrington & Loeber, 2000). These disparities may be a function of socioeconomic differences between raci al/ethnic groups, including income, neighborhood structural characteristics, employment, and education; as well as racial discrimination and bias (Centerwall, 1995; Williams, 1999; Peterson & Krivo, 2005). Research is necessary to more clearly delineate the gender and ethnic differences (as a proxy for larger socioeconomic differ ences, discrimination, and cumulative disadvantage) in delinquency an d violence between these groups. Disparities in the Prevalence of Deli nquency and Violence among Youth in the United States A number of studies have been conducted to measure the incidence of violent behavior among young people in the United Stat es. Two distinct approaches are used to obtain incidence measures: arrest statisti cs and official reports (external to the individual, generally based upon law enforcements detection of criminal behavior); and self-report victimization and offending surveys (based upon individual reports of violence and victimization). Overall, offi cial records and self-report surveys agree that youth violence has been on the decline si nce the early 1990s (Snyder & Sickmund, 2006; UCR, 2009). In 2001, arrests for homicide, rape, and robber y were similar to rates in 1980 (Snyder, 2003). Since 2000, all types of violent crime has declined (overall, 15.2%), with the most dramat ic declines observed for aggravated assault (-11.5%),

PAGE 19

19 murder and non-negligent manslaughter (-10.4% ), and forcible rape (-10.4%) (UCR, 2009). The most commonly used arrest data are derived from the FBIs Uniform Crime Reports (UCR). The UCR is a compilation of law enforcement reports of arrests for eight index (e.g., more se vere) offenses, as well as more minor offenses (United States Department of Justice, 2009). The eight index offenses are categorized as violent crime (homicide, forcible rape, r obbery, and assault) and property crime (arson, burglary, motor vehicle theft, and larceny). According to the UCR, arrests for persons under age 18 have decreased for homicide (-0.6%) forcible rape (-31.9%), aggravated assault (-27.2%), arson (-35.3%), burglary (-20.9%), motor vehicle theft (-59.6%), and larceny (-15.1%); but have increased fo r robbery (18%) between 2000 and 2009. Youth Risk Behavior Surveillance A number of surveys are conduc ted at regular intervals to evaluate the prevalence of crime and delinquenc y among youth in the United States. The results of these surveys tend to coincide with arrest and victimiz ation trends. Conduct ed biannually by the Centers for Disease Control and Prevention (CDC, 2009), the Youth Risk Behavior Surveillance System (YRBSS) surveys adolesc ents in grades 9-12 in public and private high schools across the United States regardi ng a variety of health behaviors, including physical fighting and weapon possession. Although these items are not comparable directly to the arrest measur es, identical trends provide assurance that violent crime and property crime have been declining. For exam ple, the number of high school students who reported being in a physical fight in t he past 12 months has decreased from 43% in 1991 to 31.5% in 2009 (CDC, 2009). Similarl y, the number of adolescents who reported

PAGE 20

20 carrying a gun in the past 30 days has decr eased from 7.9% in 1993 to 5.9% in 2009 (CDC, 2009). Monitoring the Future The Monitoring the Future (MTF) study is another national, annually conducted survey of high school s tudents (8th, 10th, and 12th grades) in the United States (Johnston, OMalley, Bachman, & Schulenberg, 2010). Althoug h this survey generally focuses on alcohol and drug use, a number of violent behaviors (including group fighting, serious fights, and weapon carrying and use) are measured. Similar to the YRBSS, the percentage of youths who reported serious fighting at least once in the past year has decreased from 1982 (17.3%) to 2009 (11.8%). The past-year prevalence of group fighting has also decreased from 20. 1% in 1989 to 16.1% in 2009. Weapon use has similarly decreased from 5.3% in 1989 to 3.5% in 2009. These trends correspond with other self-report surveys (including the YRBSS), providing evid ence of the validity of the trends. From these large, national data sources, a trend of decreasing violent behavior among youth is evident. Despit e the decline, the prevalence of some types of violence remains relatively high (robbery, physical figh ting, gang violence). The next sections will detail how rates of violence differ by gender and ethnicity using police records and self-reports of violent behavior. Gender Differences in De linquency and Violence Delinquency research is more recently beginning to focus on girls, as boys have historically higher rates of violent offendi ng compared to girls (McNulty & Bellair, 2003; Williams, Van Dorn, Ayers, Bright, A bbott, & Hawkins, 2007; Farrington & Loeber, 2000). However, there is evidence t hat rates of violence among girls have been

PAGE 21

21 continually increasing sinc e 1980 (Snyder, 2003). During this time, aggravated and simple assaults among girls have increased by 113% and 257%, respectively (Snyder, 2003). Although these increases seem large, the types of assault committed by girls differ from the types of assault perpetrated by boys. For example, girls are more likely to fight with parents and peers, rather t han use weapons or commit sexual assault (Chesney-Lind & Shelden, 1998). There is evidence that these gender differ ences in delinquent activities may be partially attributable to differential risk factor exposure between gender groups. For instance, Moffitt et al. (2001) found that boys are exposed to more risk factors for violent behavior than girls, thus increasing their likelih ood of violent offending. Differences in deviant peer associations may also influence the differences in risk between girls and boys. Piquero, Gover, MacDonald, & Piquero (2005) found that association with deviant peers was significantly related to violenc e and delinquency, but this was only true among boys. This may reflect differences in the social learning of delinquency, as boys learn to participate in violent behavior from their peer group (Akers, 1973). In addition to the effects of multip le risk factors and deviant peer group associations, differences in socializat ion may be an explanation for differential participation in violence between males and females. In a meta-analytic evaluation of differences in parental socia lization of their children, Ly tton & Romney (1991) found that parents were more likely to discourage aggre ssive behavior in their female children compared to male children. This small but insignificant difference may indicate that socialization alone is not the reason for differential aggression between the gender groups. Instead, a small degree of social ization differences may have cumulative

PAGE 22

22 effects when the adolescents peer group and sc hool environment is considered (Lytton & Romney, 1991). Together, these multiple ri sk factors may be partially responsible for the observed gender differences in violent behavior. There is debate in the criminological liter ature as to whether or not the risk and protective factors for delinquent behavior are stable across gender. Moffitt et. al. (2001) found that boys are exposed to more risk fact ors than girls, thus increasing their likelihood of violent offending. In a test of this hypothesis, Daigle, Cullen, & Wright (2007) found that although there were similarities in predi ctors of delinquency across gender groups, differences did exist. For ex ample, violent victimization, peer delinquency, age, and previous violence predicted increased levels of delinquent behavior for both boys and girls. Boys were more likely to be violent if they experienced more negative life events and played sports, and boys were less likely to be violent if they watched more hours of television (Daigle, Cullen & Wright, 2007) Girls were more likely to be violent if they we re depressed or experienced signi ficant strain. A study of institutionalized adolescent boys and girls f ound that girls were more likely to engage in delinquent behavior if they were sexually abused (B elknap & Holsinger, 2006). Other studies provide additional evidence that the predictors of delinquency vary across gender. Piquero, Gover, MacDonald, & Piquero (2005) found that the effect of peer delinquency was significantly related to individual delinquency. Ho wever, the effect of delinquent peer association differed for bo ys and girls. Specifically, association with delinquent peers was more strongly associ ated with individual delinquency among boys compared to girls. Jennings et al. (2010) ev aluated the differences in predictors of delinquency by gender groups in Bronx, NY and San Juan, Puerto Rico. They found

PAGE 23

23 that the predictors of delinquent trajectory membership were similar for boys and girls in NY; however, higher levels of acculturat ion predicted delinquency among girls but not boys. In a multinational st udy of delinquency trajectories Broidy, Tremblay, Brame, Fergusson, Horwood, et al., (2003) found th at early, chronic physical aggression (e.g., elementary school) increases the risk of c ontinual physical aggression and other forms of delinquency throughout adolescence. Ho wever, this finding did not hold among females in the sample, as early physical aggression did not predict membership in stable-violence trajectories. Finally, Willia ms, Van Dorn, Ayers, Bright, Abbott, & Hawkins (2007) found evidence of gender differences in the initiation of delinquency, as poor parental supervision was predictive of violence for girls but not boys. There is also evidence that the life-course perspective may not apply universally across gender groups. To understand the trajec tories of violen ce among females, DUnger, Land, & McCall (2002) used data from the Second Philadelphia Cohort study. They found gender differences in the number of trajectories (5 tr ajectory groups for males, 3 for females), and rates of offending. Specifically, the level of offending among females in the high-rate offending group was comparable to the low-rate male offending group. Overall, there are no clear ri sk factors that have c onsistently emerged from the gender differences in delinquency lite rature; however, there is preliminary evidence that differences do exist. Racial/Ethnic Differences in Delinquency and Violence There is a large body of research depicti ng racial/ethnic differences in delinquency among adolescents. Racial differences in the prevalence of delinquency have been identified. For exampl e, Williams et al. (2007) found that self-reported violence initiation rates were higher for African-Americans co mpared to Whites for major delinquency,

PAGE 24

24 violence, and juvenile justice system involv ement. Additionally, W illiams et al. (2007) reported African-Americans higher rates of major delinquent and violent acts when compared to Whites. Additionally, McNu lty & Bellair (2003) found that AfricanAmericans, Hispanics, and Native Americans have higher involvement in serious physical violence compared to White adolescent s ages 15-16. Further, this study found that after accounting for previous violence, African-Americans and Hispanics still engage in higher levels of violence than Whites (McNulty & Bellair, 2003; Blum, Beuhring, Shew, Bearinger, Sieving, & Re snick, 2000). Barnes, Welte, & Hoffman (2002) reported that Hispanics had more delinquent days during the school year than African-Americans or Whites providing additional evidence for variable levels of delinquency by racial and ethnic group. Overall, the evidence suggests that the racial disparities in offending become more pronounced as the delinquency becomes more serious in nature (Esbensen et al., 2010). Using data from the Gang Resistance E ducation and Training (GREAT) program, Esbensen et al. (2010) estimated that race and ethnic differences in violent offending differ by offense. For exampl e, African-Americans were twice as likely as Whites to report having shot someone, attacked somebody with a weapon, or have participated in serious violent offending. Similarly, Hispanic s were 2.4 times as likely as Whites to having participated in a group fight. However, the racial differences in fighting were relatively small (African-Americans were 26% more likely to have hit someone than Whites, and Hispanics were not more lik ely to have hit someone than Whites) (Esbensen et al., 2010).

PAGE 25

25 There is some evidence that the predict ors of delinquent behavior differ across racial groups (Williams et al., 2007; Fa rrington & Loeber, 2000, Esbensen et al., 2010; Farrington, Loeber, & Stouthamer-Loeber, 2003). Williams et al. (2007) found that African-American boys were more likely than White boys to initiate violence, major delinquency, and involvement wit h the criminal justice system. Studies suggest that these racial differences may be partially attr ibutable to differences in community-level risk factors, such as poverty and exposure to guns, delinquent peers, and violence in the community (Sampson, Morenoff, & Roaudenbush, 2005; Farrington & Loeber, 2000). Family and individual-level variables al so contribute violent behavior, including immigrant status (having immigr ated to the United States more recently is a protective factor from violence), parent al marital status (living with parents who are not married puts adolescents at increased risk for violence) lower verbal reading and writing ability, shorter length of residence in the nei ghborhood, and lower inco me, as free lunch eligibility increased the risk of violence initia tion (Sampson et al., 2005; Williams et al., 2007). These findings support the hypothesis that the racial and ethnic differences in violence may be at least partially attri butable to neighborhood-, familyand individuallevel socioeconomic variables. The research on race and ethnic differenc es on violence has not consistently attributed violent behavior to differences in socioeconomic pos ition. For example, Blum and colleagues (2000) found that gender, race, ethnicity, inco me, and family structure explained no more than ten percent of the variance in violent behavior. This finding indicates that the issue of violence is mo re complex than race or socioeconomic

PAGE 26

26 position. Therefore, multiple domains of risk factors that differ by ra ce/ethnicity must be further studied to better understand how these fa ctors contribute to violent behavior. Criminological theory provides insight as to the rationale behind racial and ethnic differences in the prevalence of violent offendi ng. First, social learni ng theory posits that individuals learn to engage in criminal behavior by observing those around them (Akers, 1973). This theoretical framework is compris ed of four central components: excess of definitions favorable to crim inality, association with devi ant peers, reinforcement of criminal behavior, and imitation. When these elements are combined, an individual is more likely to engage in deviant or criminal behavior (Akers, 1973). Using this theory, violent behavior would be more prevalent among racial/ethnic minoriti es if they have more violent peers who teach them to participat e in violence, and this violent behavior is reinforced. Research supports this hypothesis, as African-Americans and Hispanics are more likely to be involved with gangs (Mc Nulty & Bellair, 2003) and have delinquent peers (Stewart, Simons, & Conger, 2002) compared to Whites. Shaw and McKay maintain that the char acteristics of the community facilitate crime, rather than t he individuals within that communi ty. For example, transient individuals are unlikely to watch over their neighbors property, or even become acquainted with their nei ghbors. Therefore, a neighbor could never know if someone is stealing a car out of the driveway, as they do not know who resides in that particular home. Suburban areas have thes e types of neighborly ties, and would recognize when someone does not belong in the community. Using the framework provided by social disorganization theory, the lack of neighborly recognition is one of the reasons why crime is higher in transient, disorganized communities.

PAGE 27

27 Sampson, Raudenbush, and Earls (1997) furt her elaborate on this aspect of their theory in defining the collective efficacy of communities. Collective efficacy is the neighborhoods ability to come together to reduce criminal or disorderly behavior. Cohesive, suburban neighborhoods are very lik ely to do this: When there is a problem within the community, member s of the neighborhood organize themselves and remedy the issue. Urban, disadvant aged, disorganized neighborhoods are not likely to have such a mechanism in place to filter out cr iminal behavior among residents. This lack of social control in a community allows crime to multiply. This theory has substantial implications for racial/ethnic differences in violence, as minorities are more likely to reside in these urban, at-risk comm unities (Sampson & Wilson, 2005). The Relationship between Alcohol and Violence The association between violence and alcohol consumption has been well studied in the criminological and publ ic health literature. At leas t four explanations hav e been proposed to explain the relationship between alcohol and violence: (a) reciprocal and psychopharmacological, meaning that the intoxicating effe cts of alcohol encourage violence to gain resources to support their drug/ alcohol use (Goldstein, 1985); (b) the relationship is correlational, in that alcohol use increases violence because aggressive people self-select into situations that encourage alcohol consumption (Johnston, OMalley, & Eveland, 1978); (c) the relationshi p is bidirectional, and the arrow between alcohol use and violence may point in ei ther or both directions (White, Loeber, Stouthamer-Loeber, & Farrington, 1999) ; or (d) the relationship is spurious, as problem behaviors cluster as part of a more general problem behavior syndrome (Jessor, Donovan, & Costa, 1991).

PAGE 28

28 Few longitudinal studies have attempted to address the dire ctionality of the alcoholviolence relationship, and t he findings have been mixed. Dembo and colleagues (1991) conducted a c ohort study of detained juvenile delinquents and found that alcohol use predicted violent behavior, and violence was a significant predictor of drug use (e.g., cocaine, marijuana) at the 10to 15-month follow-up. Similarly, Ellickson, Tucker, and Klein (2003) followed a cohort of seventh grader s in California and Oregon through age 23 to evaluate the effects of ear ly alcohol use. They found that early alcohol users were more likely to be delinqu ent and use other drugs in middle and high school compared to nondrinkers in seventh gr ade. At age 23, early alcohol users were at an increased risk for substance misuse, violence, and criminality (Ellickson, Tucker, & Klein, 2003). Another longitu dinal study conducted using t he first two follow-up surveys of the National Longitudinal Study of Adole scent Health (Add Health) found that alcohol use was a significant predictor of physical violence two years later (Resnick, Ireland, & Borowsky, 2004). Despite the well-documented relationship between alcohol consumption as a predictor of violent behavior, another body of literature suggests that the violence predicts alcohol use. For example, Windl e (1990) used National Longitudinal Youth Survey data to assess the impact of vari ous anti-social behaviors at ages 14 on other delinquent behaviors 4 years later. The results suggested that general delinquency (a function of the frequency of non-substance-related delinquent behavior) was a significant predictor of alcohol consum ption. Similarly, a study of adolescents age 12 evaluated the effects of early alcohol use among a sample of 218 males and 213 females (White, Brick, & Hansell, 1993). Whit e et al. (1993) found that early aggression

PAGE 29

29 in males predicted alcohol consumption and alcohol-related aggressive behavior. In this study, more specific levels of alcohol use were not significantly associated with later aggression. In support of thes e findings, data from the same study were reanalyzed to evaluate the complex relationships betw een alcohol use, aggression, and alcoholrelated aggression over time using structur al equation models (White & Hansell, 1996). They found that early initiation of alcoho l use predicted physically aggressive behavior. Several studies provide support for a third theoretical explanation for the association between violence and alcohol use, that a bidirectional relationship exists. First, DAmico, Edelen, Miles, and Morra l (2008) conducted a study of high-risk juveniles in the Los Angeles juvenile probat ion system between the ages of 13 and 17. They found that substance use predicted de linquency (a scale of drug-related crime, property crime, and interpersonal violent crime) and delinquency predicted substance use. This study was not specific to alc ohol use, and adolescents were only followed for a period of 1 year. In an 8-year study of high school aged African Americans in Michigan, Xue, Zimmerman, and Cunningham (2009) tested the bi directionality of alcohol use and violent behavior. Their results indicated that early violence (e.g., group fighting, hitting a teacher or supervisor, us ing a knife or gun to get something from a person, etc.) significantly predicted later al cohol use, and early alcohol use predicted future violent behavior. Among 15to 19year-old urban Mexican Americans and European Americans selected from a lar ge health maintenance organization, Brady, Tschann, Pasch, Flores, and Ozer (2008) no ted reciprocal relationships between alcohol use and violence when adolescents we re older. For example, perpetration of violence at age 18 significantly predicted al cohol use at age 19; however, violence at 15

PAGE 30

30 did not predict alcohol use at age 19. Simila rly, alcohol use at age 15 did not predict violent behavior at age 19, whereas alcoho l use at age 18 significantly predicted perpetration of violence. Alt hough these studies have consistent findings within ethnic groups, the limited external validity from single-site and incarcerated populations of juveniles must be considered when applying findings to these group s at the population level. Gang Membership versus Group Fighting In an attempt to measure the correlat es and effects of gang membership, a number of definitions have been proposed to measure the construct. Gang members have been distinguished from other peer groups based upon t heir 1) participation in criminal behavior; 2) shared ident ity; and 3) naming, symbols, colors, or other means of distinguishing themselves and their territory from other groups (Spergel, 1993; 1994; Esbensen, 2000; White 2002). A commonly us ed definition proposed by Esbensen & Weerman (2005, 2008) describes gang members as, any durable, street-oriented youth group whose involvement in illegal activity is part of their group identit y. However, this type of definition did not exist until 2005, as previous gang researchers have not focused on the illegal and violent behavior t hat is typically associated with gang membership today. For example, Thrasher (1963) proposed a definition that continually influences how gang membership is measured t oday. In this work, Thrasher suggested that in order to be characterized as a gang, a group of adolescents must: 1) have organization and a sense of solidarity distingui shing them from mobs; 2) be responsive to outside threats; 3) create a self-identity; and 4) identify and defend a territory. Later, Klein (1971) argued that illegal activity shoul d be added to this list of requirements to identify adolescents as gang members. Spec ifically, Klein (1971) proposed that gang

PAGE 31

31 members are a group of adolesce nts (generally under age 18) who are perceived to be a distinct aggregation by peers in their co mmunity; self-identify as a group (and often have a name); report involvement in delin quency to the extent that neighborhood residents become unhappy and law enforcement has become involved. Participation in illegal activities as a necessary factor fo r gang membership remains an issue for debate (Spergel, 1993; 1994; Esbensen et al ., 2000; 2005; 2010; White 2002). A number of definitions have been used to measure gang membership. Many surveys will simply ask, Are you a me mber of a named gang? (Thornberry,1998), or Do you belong to a gang?, with follo w-up items including gang names and characteristics (Ball & Curry, 1995). Ot hers have built upon th is definition and categorized gang members based upon self -reported gang member ship and selfproclaimed involvement of the gang in illegal activities (Huizinga et al., 1995). Further, Klein, Weerman, & Thornberry (2006, 418) c onsider gangs to be any durable, streetoriented youth group whose own identity incl udes involvement in illegal activity. Although these questions incorporate self-identifying measures of gang membership, they do not evaluate the violent behavior th at is often closely linked with gang membership. Some surveys of adolescent youth, including Monitoring the Future (Johnston et al., 2010) and the National Longitudinal Survey of Adolescent Health (Harris, Halpern, Whitsel, Hussey, Tabor, Entzel, & Udry, 2009), measure participation in group fights, a measure of assault that has been associat ed with gang membership. Other surveys, including the National Youth Survey (Ellio tt, 1984), use a similar construct of gang fight. There is some evidence that the distinction between group and gang fighting

PAGE 32

32 may produce different results and prevalence estimates, as group fighting and gang membership may be measuring different behaviors (Menard & Elliott, 1993; Esbensen, Winfree, He, & Taylor, 2001). This measure of group fighting may partially overlap with gang membership; however, these constructs are not synonymous. For example, many self-identified gang members are not violent and do not engage in fights, and many non-gang members are violent (Esbensen et al., 2010; Howe ll, 1998; Miller, 1998; Spergel et al., 2005). This leads us to be lieve that the independen t construct of group fighting is problematic above and beyond other fo rms of violence or membership in a gang (Maldonado-Molina, Reingle, & Jennings, 2011). Theoretical Foundation A number of theoretical f oundations provide insight as to why adolescents engage in violent behavior. Four prevailing criminol ogica l theories were used as a guideline, as each are complementary and examine behavior at different context ual levels. First, Social Learning theory, proposed by Akers (1973) as a merger of Sutherland and Banduras learning theories, suggests that individuals engage in deviant behavior because their social group does so, and the devi ant behavior is reinforced. Under this theory, young people do not understand deviance to be overwhelmingly harmful to themselves or others, and they are taught how to engage in illegal activity through their peer group (Akers, 1973). Second, Social Bond Theory, proposed by Hirschi (1969), explains that adolescents are deviant becaus e their bonds to conventional society are weak, and the individual has little motivation to adhere to social norms Third, Moffitt offers a dual-taxonomy of criminal behavior, recognizing life course persistent offenders and adolescence-limited offenders as distinct in their reasons for engaging in crime. She argues that life-course persistent offenders (or t hose who engage in criminal

PAGE 33

33 behavior for the majority of their lifetime) begin life with differences in neurological functioning and have early behavioral problems (2005). Unlike life course persistent offenders, adolescent-limit ed offenders have normal childhoods, begin engaging in criminal behavior later in life (usually between fifteen and seventeen), and stop their illegal behavior soon after beginning it (usually by age nineteen). Finally, Shaw and McKay (1942) proposed a macro-level theory of social disorganization using data from Chicago, Illinois. They found t hat most crime took place in the zone of transition, or an area densely populated with immigrants and trans ients, high levels of poverty, and characterized by a high number of single par ent homes and high ethnic diversity (1942). Each of these theories will be discussed in detail in the following sections. Social Learning Theory Social lear ning theory has been applied to studies of violence and delinquency. Specifically, it has been hypothesized that the constructs of differential association (e.g., having a delinquent peer group, adult approval of delinquency), social reinforcers (e.g., peers and parents reactions to violence), neutralization of the delinquent behavior, and positive attitudes towards violence tend to increase violent behavior (Winfree, Backstrom, & Mays, 1994; Esbensen & Deschenes, 1998). Each of these constructs additively contributes to the lik elihood of delinquent behavior. Social learning theory posits that individua ls learn to engage in criminal behavior by observing those around them (Akers, 1973) The social learning theoretical framework was derived from theories by Su therland (Differential Association theory; Sutherland, 1939) and Bandura (operant conditioning, Bandura, 1971), and is comprised of four central components: excess of definitions favorable to criminality, association with deviant peers, reinforcement of criminal behav ior, and imitation. When

PAGE 34

34 these elements are combined, an individual is more likely to engage in deviant or criminal behavior. For each additional element of social learning that an individual is exposed to, the greater the risk of violence (Akers, 1973). First, an individual will be at increased risk for deviant behavior if they believe that such behavior is not intrinsically harmful to themselves or others, and the deviance will benef it them in some way. This would be considered a definition favorable to crimi nality or deviance. When the definitions favorable to violence outweigh the definitions unfavorable to violence, the likelihood that an individual will engage in vi olence increases. Second, in agreement with Sutherland (1947), Akers suggested that association with deviant peers increases the likelihood of deviance. The theory of differential associ ation suggests that criminal behavior is learned via social interaction with others, as the social environment and values gained from relations with deviant peers and family me mbers would affect individuals criminal behavior (Sutherland, 1947). Sutherland (1947) used nine propositions to explain his theory. He believed that criminal or deviant behavior is learned through interaction with others in personal groups, and that the l earning of criminal behavior requires the learning of criminal techniques, attitudes rationalizations and motives. In addition, Sutherland (1947) also stated that, in order to become a criminal, one must learn to view the law negatively and have a deficiency in definitions favorable to law-abiding behavior. His final propositions state that cr iminal behavior is learned in the same way as any other type of behavior, and criminal behavior is not explained by general needs and values (Sutherland, 1947).

PAGE 35

35 In addition to differential association and def initions favorable to deviance, Akers suggested that reinforcement of deviance and imitation ar e important in predicting deviant behavior. When relatively minor dev iant behaviors are positively reinforced by others, it is likely that more frequent and perhaps, more severe, deviant behaviors will result. It is important to note that all relationships are not created equally under this theory, as longer relationships and those w ho became acquainted early in life are more influential in their reinforcement of deviant behavior. Finally, deviant behavior may be learned through mimicry of social circle mem bers (Akers, 1973). Akers theory highlights the role of the peer group and early relations hips in explaining why individuals engage in deviant activities. Social Bond Theory Hirschis social bond theory has frequently been applied in studies of violent delinquency (Bjerregaard & Smith, 1993; Blum, Ireland, & Blum 2003; Booth, Farrell, & Varano, 2008). Social bond variables including school success (Bjerregaard & Smith, 1993; Blum et al., 2003), high levels of parental attachment (Booth et al., 2008; Blum et al., 2003), belief in traditional norms (Booth et al., 2008), and involvement in pro-social activities (Booth et al., 2008; Blum et al., 2003) have been protective from participation in delinquency. Under this theory, t he absence of any construct (attachment, commitment, involvement, or belief) puts an adolescent at risk for delinquency and violence (Hirschi, 1969). Social Bond Theory is an individual-lev el theory of criminal behavior, first conceptualized by Hirschi (1969). Social Bond theory asserts that criminal behavior results from weak bonds to conventional so ciety. The four main domains include attachment (an individuals ability to ident ify with others and main tain their support),

PAGE 36

36 involvement (participation in conventional acti vities), belief (a per sons acceptance of societys conventional values), and commi tment (dedication to conventional behavior) (Hirschi, 1969). According to Hirschi, persons who are more attached to other members of conventional society, the more they will seek their support and adhere to positive social norms. These four constructs all contribute additive ly to the likelihood of deviant behavior. Persons who are more committed to conventi onal activities are less likely to engage in deviant behavior. For instance, conventional ac tivities may include attending college or having a job in their community. Participation in these activi ties invests the individual in the well-being of the community and decreases the likelihood of deviant behavior. In addition, committed individ uals have more to lose if they are caught engaging in deviant behavior, further reducing their likelihood of deviance. Involved persons participate in a large number of conventional activities, and these persons are generally considered to be too busy to engage in deviant behavior. Finally, persons who believe in the conventional norms and values society are less likely to be delinquent, as they are more likely to adhere to these norms. In comparis on, those who do not believe in the norms of conventional society are more likely to be deviant, since they do not endorse the values and norms of society. Weaknesses in any one of these domains increase the likelihood of deviant and criminal behavior. Moffitts Life-Course Perspective Many studies have supported the existence of the age-crime curve (Hirschi & Gottfredson, 1983; Farrington, 1986), which disp lays the populatio n-level trajectory of crime trends across the life course. Mo ffitts taxonomy of crime applies a developmental perspective to criminal behavi or, focusing on the decrease in criminality

PAGE 37

37 and delinquency as adolescents age (Moffitt, 1993). Moffitts taxonomy suggests that group fighters comprise most of the life course persistent offenders, or the small group who continues to engage in delinquency an d violence throughout adulthood (Moffitt, 1993). This perspective will be de scribed in greater detail below. Moffitt (1993; 2005; 2001) offe rs a dual-taxonomy of criminal behavior, recognizing life-course persistent offender s and adolescence-limited offender s as distinct in their reasons for engaging in crime. She argues that life-course persistent offenders (or those who engage in criminal behavior for the majority of their lif etime) begin life with differences in neurological functioning and have early behavioral problems (Moffitt, 2005). These children may have been abused or neglected, and often exhibit antisocial behavior (Moffitt, 2005). These individuals often come from poorly functioning families, do poorly in school, and associate with other criminal and/or deviant peers (Moffitt, 1993; 2001). They are also more likely to be raised in neglected, low-income neighborhoods and have parents with antisocial tendencies or below-average cognitive abilities. These environments increase t he likelihood that the childs genetic predispositions to violence will surface, and antisocial behavior will prevail (Moffitt, 1993; 2010). Criminal behavior in life-course persist ent offenders begins early and continues throughout life. A cycle of cumulative disa dvantage entrenches them in the criminal world: These individuals have few law-ab iding friends (because of their antisocial behavior) and do poorly in school. Outside of school, they engage in delinquent and later criminal behavior. This lifestyle is very difficult for the indi vidual to break away from, as the individual has very few conventi onal ties (Moffitt, 199 3; 2010). Gradually

PAGE 38

38 throughout childhood and adolescence, the individual begins to identify themselves as criminal (Moffitt, 1993). As a result, life-course persistent offenders exhibit contemporary continuity, meani ng that the offender brings t he same deviant traits that they exhibited as an adolescent to adulthoo d and act in a similar manner. Now, any major turning points in life (or pathways from crime, such as marriage or higher education) are likely to be bypassed (Moffitt 1993; 2010). After all pathways out of delinquency are no longer availa ble, life-course persistent offenders are embedded in their deviant lifestyle for t he majority of their lives. The alternative pathway of offending that Moffitt describes in her taxonomy is the adolescence-limited trajectory (Moffitt, 1993; 2001). Unlike life-course persistent offenders, adolescence-lim ited offenders have normal childhoods, begin engaging in criminal behavior later in life (usually between fifteen and seventeen), and terminate their illegal behavior shortly after initiating it (usually by age nineteen) (Moffitt, 1993). Adolescence-limited offenders ar e often tempted to engage in criminal acts because of the maturity gap: Young people feel biologically and socially mature, but they are not legally allowed to be indepen dent or given adult freedoms (Moffitt, 1993; 2010). They want to engage in deviant, dangerous, or ot her behaviors to defy their parents and appear mature. To this end, adolescentlimited offenders briefly (and sporadically) model the life-course persistent offenders in their criminal behavior (Moffitt, 2010). A short time later, adolescence-limited offenders return to their law-abiding lives, as the risks of criminal behavior are fa r too severe. These individuals are likely to go to college and/or obtain a good job; therefore, the costs of cr iminal behavior are far higher for this group than the life-course per sistent offenders. These adolescence-limited offenders

PAGE 39

39 recognize the potential consequences of getting caught and discontinue any criminal behavior early in life (Moffitt, 1993; 2010). Social Disorganization Theory Shaw and McKays theory of social diso rganization has been applied to violence and delinquency, as many studies have repor ted a clustering of high homic ide rates, assault, and physical violence in general in ur ban centers of cities in the United States (Curry & Spergel, 1988; Kawachi, Kennedy, & Wilkinson, 1999). Spec ifically, Shaw and McKay (1942) believed that residents in cohes ive communities have more control over youth in their neighborhoods, and could theref ore prevent any behavior that would set the context for violence (e.g., supervise leis ure time activities, challenging youth who appear to be behaving inappropriately, etc.) (Kaw achi et al., 1999). This neighborhoodlevel theory of violence highlights the role of the community in violence and delinquency prevention. Social disorganization theory is primarily based in the work of Shaw and McKay (1942) in Chicago, Illinois. In their cla ssic study, Shaw and McKay analyzed concentric zones radiating from the city center of Chicago to deter mine how crime is distributed and identify the characteristi cs of high-crime nei ghborhoods. They found that most crime took place in the zone of transition, or the impover ished area densely populated with immigrants and transients, and characterized by a high number of single parent homes and high ethnic diversity. It was also noted that t he crime rate stayed the same over time, despite the constant flow of people to and from the ci ty. These zones of transition harbor social diso rganization, and low levels of social organization in these communities facilitates crime. The major factors in defining an area as socially

PAGE 40

40 disorganized include: conc entration of poverty, population density, high mobility rates, and ethnic heter ogeneity. Shaw and McKay maintain that the char acteristics of the community facilitate crime, rather than t he individuals within that communi ty. For example, transient individuals are unlikely to watch over their neighbors property, or even become acquainted with their nei ghbors. Therefore, a neighbor could never know if someone is stealing a car out of the driveway, as they do not know who resides in that particular home. Suburban areas have these types of neighborly ties, and they would recognize when someone does not belong. Using the framework provided by social disorganization theory, the lack of neighborly recognition is one of the reasons why crime is higher in transient, disorganized communities. Sampson, Raudenbush, and Earls (1997) furt her elaborate on this aspect of their theory in defining the collective efficacy of communities. Collective efficacy is the ability of a neighborhood to co me together to reduce criminal or disorderly behavior. Cohesive, suburban neighborhoods are very likel y to do this: When there is some sort of problem within the comm unity, community members organize themselves and remedy the issue. Alternatively, urban, disadvantaged, disorgani zed neighborhoods are not likely to have such a mechanism in pl ace to filter out cr iminal behavior among residents. This lack of social control in a community allows crime to multiply. Generally, crime flourishes in thes e inner-city communities because the community cannot control it (Shaw and McKay, 1942). Disor ganized neighborhoods cannot control children or their behavior, as familial control is minimal (both parents/single parents are constantly working to pay bills or may be involved in illegal

PAGE 41

41 behaviors themselves) and children are frequently left to care for themselves. This may allow them to engage in illegal activities with other children in the same situation. Residents have little stake in their community, as they will be relocating as soon as they have the means to do so. The culmination of all of these factors allows certain neighborhoods to remain disorganized for generation after generation. To summarize, the structural characteristics of the community (poverty, population density, mobility and ethnic heterogeneity) leads to constant so cial disorganization and a persistent delinquent subculture that is consis tently engaged in criminal behavior. All of these theoretical perspectives co me together to bette r explain delinquency among adolescents. Social Learning Theory em phasizes the role of the adolescents social surroundings (peers, adults, sibli ngs, etc.) and their influence on violent and delinquent behavior. Social Bond Theory descri bes risk factors at the individual level (e.g., commitment to school, education, family and peers; attachment to parents, involvement in pro-social extracurricular ac tivities) for violent delinquency. Social Disorganization explains the external in fluence of the communities in which adolescents reside, and how the community influences delinquent behavior. Moffitts life-course perspective contributes multilevel risk factors that are highly influential early in life (e.g., genetic composition, charac teristics of the fa mily, neighborhood, peer delinquency, etc.), and how the presence of these characteristi cs will play a large role in whether an adolescent becomes a life-cour se persistent offender or an adolescencelimited, low-level offender. These pers pectives should converge to provide the foundation for delinquency and viol ence prevention efforts.

PAGE 42

42 Contribution of th e Current Study The majority of the liter ature violence and delinquen cy has been cross-sectional and has been conducted within a single city or region (for notable exceptions, see Esbensen et al., 2010; McNulty & Bellair, 2005), and the research on racial/ethnic and gender differences in violent trajectories is especially sparse. Longitudinal designs are relatively rare in the criminol ogical literatur e, and most studies are not able to evaluate data at the neighborho od, school, family, and individual level simultaneously (Agnew, 1994; Adams, 2009; Bendixen, Endresen, & Olweus, 2006). This study allows for evaluation of change in violence over time, and permits the investigation of multiple domains of predictors to i dentify the risk and protective factors for violence by racial/ethnic and gender subgroups in two contexts (national and high-risk, urban context). Study Goal, Aims, an d Research Questions The current study utilized two long itudina l datasets: 1) the National Longitudinal Study of Adolescent Health (Add Health); and 2) Project Northland Chicago (PNC). The Add Health data include a nat ionally representative sample of adolescents followed from ages 15-26. This sample was used to evaluate overall differences in risk and protective factors for delin quency (Study #1) and differences in risk and protective factors by gender and racial/ethnic subgroup (Study #2). P NC is a high-risk sample of urban adolescents in grades 6-8, and this sample was used in Study #3 to replicate the results of Study #2, and assess the impact of multiple domains of predictors of violent trajectories by racial/ethnic subgroup in an urban, high-risk sample of adolescents. This study is unique in that the anal ysis incorporates longitudinal data on frequently excluded subgroups of adolescents (specifically, females and Hispanics).

PAGE 43

43 Additionally, the longitudinal nature of the current study included data from two large, longitudinal studies, which allows the examination of differences in multiple domains of predictors (including community context, family, peers, and individual-l evel) of violence. The first dataset included a nationally representative sample of adolescents, followed from age 15 to 26 (Add Health; Harris et al., 2009). The second dataset included a sample of high-risk urban youth in grades 6-8 in Chicago, IL (Komro et al., 2008; 2004). An understanding of the differences in the mu ltiple domains of pr edictors of delinquency between these samples and between ethnic and gender groups within each sample is informative for violence preventive efforts. Specifically Study 1 addressed the resear ch question: Does group fighting during early adolescence (before age 15) increase violent behavior in late adolescence and adulthood (ages 16-26)? I hypothesized t hat: 1) Adolescents who engage in group fights early in life are more likely to engage in high levels serious violence during late adolescence and adulthood; 2) Group fighting is a predictor of future violence independent of baseline violent behavior; and 3) Exposure to risk factors at age 15 (individual, peer, family, community) increases violent behavior into adulthood. Study 2 addressed the research question, how does gender and race/ethnicity influence violent behavior? I hypothesized that 1) Males are at increased risk for violence over time, and these differences can be at least partially attributed to group fight ing; 2) There are racial/ethnic differences in the trajectories of violent behavior over time in a national sample of adolescents (African-American and Hispanic adolescents are more violent when compared to Whites), and these differenc es can be attributed to early-onset group fighting; and 3) Multiple domains of viol ence predictors in the general population of

PAGE 44

44 adolescents (e.g., community-level violence, peer alcohol use, and exposure to alcohol) differ by race/ethnicity. Finally, Study 3 addressed the research question, how does race/ethnicity influence trajectories of violent behavior? I hypothesized that: 1) There are racial/ethnic differences in the trajectories of violent behavior in an urban sample of adolescents (African-American and Hispanic adolescents ar e more violent wh en compared to Whites), and these differences can be attributed to early-onset group fighting; and 2) The multiple domains of predictors of violence among urban adolescents (e.g., community-level violence, peer alcohol use, and exposure to alcohol) differ by race/ethnicity.

PAGE 45

45 CHAPTER 2 METHODOLOGY Study #1: Risk Factors Associated with Trajectories of Viol ent Delinquency in a Nationally Representative, Longitudinal Sample The first study estimated group-based trajec tory models of violent behavior to test whether early group fighting (age 15) is a risk factor for mem bership in the most violent trajectories over time using data from the National Longit udinal Study of Adolescent Health (Harris et al., 2009). Ne xt, this study evaluated multiple domains of predictors of violence. The Add Health data are appropriate fo r this analysis because the data follow adolescents over the life course, often before they initiate violence. The study design also followed participants into their late 20s and early 30s, allowing us to identify the life-course persistent offenders, whether they were group fi ghters at age 15, and identify other risk and protective factors that may have been present at age 15. Finally, a multitude of risk and protective factors at multiple levels of influence were measured in this dataset, both via self-repor t and other data sources. National Longitudinal Study of Adolescent Health (Add Health) Study Design The National Longitudinal Study of Adole scent Health (Add Health) is a schoolbased panel study conducted from 1994 (W ave I) through 2008 (Wave IV), when participant ages ranged from 11-32 (Chantal a & Tabor, 1999). The data collect ion for this survey was designed to explore multip le domains of influence on adolescents health behaviors. In Wave I, 80 communi ties were selected to ensure demographic representativeness (ethnic composition, region of the country, urbanicity, school size, and school type) of students in t he United States. Schools (n = 132) were eligible if they

PAGE 46

46 enrolled more than thirty students and had an eleventh grade. All students who were enrolled in the school and were present on the survey day were eligible for participation in the study. Approximatel y 200 students were randomly sele cted from strata of grade and sex, resulting in a final sample of 9421 adolescents who partici pated in all four waves of data collection. Data Collection At Wave I, the baseline sa mple of 20,745 participated in a oneto two-hour inhome interview. This interview c ollect ed data on a variety of topics, including demographics, peer networks, health, employment, substance use, criminal and delinquent behavior, sexual behavior, and ro mantic partnerships. All data were recorded on laptop computers, and the participant listened to the questions and responded on the laptop themselves to maximize validity for sensitive items. At this time, parents were also surveyed to ev aluate their health behaviors, education, neighborhood characteristi cs, and heritable conditions (Chantala & Tabor, 1999). Wave II data collection included another in -home interview one year later, between April and August, 1996. Interview items were similar, but additional nutrition and sun exposure items were added. The response rate for eligible participants in Wave II was 88.2%. All participants who were interviewed in Wave I were elig ible, excluding those who moved out of the country. Jailed adole scents were interviewed when possible. Wave III included data collected between 2001-2002, when participants were ages 1826. The survey instrument changed to reflect the more influential romantic relationships, criminal history, and joband college-related influences rather than school-based influences on health behaviors. A final sample of 15,170 respondents participated in Wave III.

PAGE 47

47 In 2007-2008, Wave IV of data collection co nsisted of a 90-minute questionnaire to be completed by respondents via laptop. Physical and biological markers were collected, and questionnaires included a greater number of financial, marital/cohabitation, medications, physical health, and childhood maltreatment items. The response rate for this sample was 80. 3% of eligible participants (n = 15,701). Participants were eligible for all waves of data collection if they partic ipated in Wave I. Participants The sample used in this study includes participants wh o were pres ent at any of the four waves of data collection. This sa mple was 56.3% male, 63.6% White, 24.1% African-American, 15.0% Hispanic, 5.9% As ian or Pacific Islander, 4.1% American Indian, and 8.6% Other. The average age at Wave I was 15.1 (sd = 1.2), 16.3 (sd = 1.6) at Wave II, 21.7 (sd = 1. 6) at Wave III and 26.5 (sd = 1.8) at Wave IV. Nearly twenty percent of the sample r eported participating in a group fight at Wave I (n = 4,737; 19.96%), 19.5% reported group fighting at Wave II, 9.5% at Wave III, and 4.0% at Wave IV. The prevalence of the independent and dependent variables are detailed in Tables 2-1 and 2-2. Measures Violent Delinquency Violenc e was measured using three items that were measured across each of the four waves of data collection: In the past 12 months, have you 1) hurt someone badly enough that he or she needed care from a doctor or nurse?; 2) pulled a knife or gun on someone?; and 3) shot or stabbed someone? At each wave, a value from 0-12 was assigned to each participant, where a value of , , or was assigned for each of these violent acts in which the individual has participated in during the past year. A zero was assigned for each item if the participant did not report the

PAGE 48

48 behavior. A two was assigned if the adol escent reported hurting someone badly enough to need care from a doctor or nurse one to three times in the past year. A four was assigned for each of the following occu rrences: 1) shooting or stabbing someone; 2) pulling a knife or gun on someone; or 3) hurting someone badly enough to need care from a doctor or nurse four or more times in the past year. These values were used to create trajectories of delinquency over Waves II-IV. Group Fighting Group fighting was measured using the variable, In the past 12 months, how often did you take place in a physical fight where a group of your friends was against another group?. Responses to this item include: =Never, =One or two times, =Three to four times, and =5 or more times. These responses were dichotomized into =never group fighting and =group fighting in the past year. Poverty Poverty was measured using the percentage of families in the respondents census tract whose income was at or below the poverty level. Racial dispersion was measured using the concent ration of racial/et hnic homogeneity within the neighborhood, so higher proportions woul d indicate greater concentration of one race/ethnicity within a neighborhood. These items were incorporated into the analysis in accordance with Shaw and McKays (19 42) theory of Social Disorganization. Risk factors for violence Alcohol Use Lifetime alcohol use was evaluat ed using the item, Have you had a drink of beer, wine, or liquornot just a si p or a taste of someone elses drinkmore than 2 or 3 times in your life?. Those who responded affirmatively to this item were categorized as Alcohol Users.

PAGE 49

49 Depression. This mental health status variable was measured with one item, How often in the past week have you felt depressed?. Values for this variable were dichotomized so that 1 = One or more times and 0 = in stances of depression in the past week. Depression was included as a covariate because higher levels of depression have been associated with violence (Elbogen & Johnson, 2009; Senn, Carey, & Vanable, 2010; Thurnherr, Bechtold Michaud, Akre, & Suris, 2008) and other risk behaviors (Latzman & Swisher, 2005; Senn, Carey, & Vanable, 2010). Academic Achievement Academic performance was measured using the variable, On a scale of 1 to 5, where 1 is low and 5 is high, how likely is it that you will go to college?. This item was included as a covariate because academic achievement and IQ have been associated with increased ri sk of violence (Herrenkohl, McMorris, Catalano, Abbott, Hemphill, & Toumbour ou, 2007; Leech, Day, Richardson, & Goldschmidt, 2003). Parental Involvement Parental influence and invo lvement was measured using a scale of twenty items (10 for maternal involvement, 10 meas uring paternal involvement) (Prado et al., 2009). Each indi vidual item was dichotomized, and the scale is the sum of all twenty items (range: 0-20). The ten it ems which comprised the scale included whether or not the respondent r eported participating in the follo wing activities with their mother and/or father in the pas t four weeks: 1) going shoppi ng; 2) playing a sport; 3) attending a religious or church-related even t; 4) talking about someone they are dating or a party they attended; 5) attending a movie, play, concert, or sporting event; 6) talked about a personal problem they were havi ng; 7) had a serious argument about their behavior; 8) talked about work or grades; 9) worked on a project for school; and 10)

PAGE 50

50 talked about other things they are doing in school. Cronbachs coefficient alpha for this scale was 0.74. This scale was included as a covariate because evidence suggests that parenting variables (e.g., monitoring, invo lvement) are related to violence (Park, Morash, & Stevens, 2010). Marijuana and other drug use Marijuana use was measured using the item, During your life, how many times have you used marijuana? Responses were categorized into users and non-users. Other drug use was created using the selfreported number of times the re spondent used cocaine, inhalants or other drugs in their lifetime. If any of these drugs were used, respondents were categorized as users. These items were included because evidence suggests that the use of marijuana and other drugs (Boles & Miotto, 2003; Dhungana, 2009; Herrenkohl et al., 2007) increases the risk of violent behavior. Desire to leave home This variable was measur ed using the following item: How much do you feel that you want to leave home?. Respondents who reported very much or quite a bit were categorized as , others were categorized as . This variable was included because some ev idence suggests that a negative home environment increases the likelihood of viol ent delinquency (Ou & Reynolds, 2010). Peer Alcohol Use Peer alcohol use was measured using one item: Of your three best friends, how many drink alcohol at least once a month? Respondents who reported having one or more friends who use al cohol monthly were coded as . These items were included because literature suggests that indi viduals who have peers who use alcohol (Herrenkohl et al., 2007; Kunt sche, Gossrau-Breen, & Gmel, 2009; Leech, Day, Richardson, & Goldschmidt, 2003) are mo re likely to engage in violent behavior.

PAGE 51

51 Peer Marijuana Use. Respondents were asked, Of your three best friends, how many use marijuana at least once a month? Respondents who reported having one or more friends who use marijuana monthly were coded as . These items were included because literature suggests that indivi duals who have peers who use marijuana (Herrenkohl et al., 2007; Leech, Day, Richar dson, & Goldschmidt, 2003) are more likely to engage in violent behavior. Study # 2: Gender and Racial/Ethnic Diff erences in Risk Factors Associ ated with Trajectories of Violent Delinquency in a Nationally Representative, Longitudinal Sample The second study evaluated gender and racial/e thnic differences in trajectories of violent behavior, and how differences in group fighting and other domains of risk factors for violence influence trajectories of vi olence using data from the Add Health. Specifically, this study test ed the hypotheses that males ar e at higher risk for violence over time (ages 16-26), and these differenc es may be attributable to differences in group fighting at baseline (Age 15). Second, I tested for t he influence of race/ethnic group membership on violence over time. I hypothesized that racial/ethnic minorities (African-Americans and Hispanics) are more likely to be involved in high levels of serious violence compared to Whites. Finally, I hypothesized that differences in violence by racial/ethnic and gender subgroups are a ttributable to differences in multiple domains of risk factors. Data from Add Health is an ideal dataset to address these hypotheses. First, adolescents were followed over time, allowing fo r an evaluation of trajectories, or intraindividual change over time in violent behavior. Second, this dataset features a large sample size (e.g., approximat ely 10,000), allowing stratifica tion by multiple racial and ethnic subgroups. Finally, a multitude of risk and protective fa ctors at multiple levels of

PAGE 52

52 influence are measured in this dataset, both via self-report and other data sources. Study design, data collection, and measures for the Add Health were previously described. Study #3: Risk Factors Associated with R acial/Ethnic Differences in Trajectories of Violent Delinquency in Longitudinal Sample of High-Risk, Urban Youth The third study evaluated gender and racial/et hnic differences in trajectories of violent behavior, and how differences in group fi ghting influences trajectories of violence using a high-risk sample from Project Northland Chicago. Specifically, this study tested the hypothesis that racial/e thnic minorities (African-Am ericans and Hispanics) are involved in high levels of aggressive behav iors. In addition, I hypothesized that differences in violence between African-Am ericans and Hispanics are attributable to differences in multiple domains of risk factors. Data from Project Nort hland Chicago was an ideal dataset to address these hypotheses. First, adolescents were followed over time, allowing for an evaluation of trajectories, or intra-indi vidual change over time in violent behavior. Second, a multitude of risk and protective factors at multiple levels of influence were measured in this dataset, both via self-report and other dat a sources. Third, t he prevalence of group fighting and delinquency is relatively high in this sample of high-risk adolescents. Finally, the range of ages for youth in this study allows us to identify distinct groups of desistors, initiators, and those who remain consistently low or high in their levels of delinquency throughout adolescence.

PAGE 53

53 Project Northland Chicago Study design Project Northland Chic ago (PNC) was a group-randomized controlled alcohol prevention intervention implemented during 6th through 8th grades in selected Chicago, Illinois schools (Komro et al., 2008). All schools in Chicago were eligible for inclusion if students were enrolled in 5th through 8th grade, had a mobility rate of less than 25%, and had thirty or more students per grade. All magnet schools were ineligible, as they were less likely to include students from a specific neighborhood, and the intervention included a community intervent ion component. All adolescents were included in this analysis, and inclusion in the intervention condition served as a covariate. A total of sixty-six schools participated in the study, and these schools were combined into study units with approximatel y two hundred students per unit. Units were grouped in correspondence with census tracts and matched on ethnicity, mobility rate, reading and math scores, and poverty levels within each unit. After matching, units were randomized to treatment condition ( 30 schools to the intervention, 36 schools served as controls). Five schools withdrew from the study prior to the intervention, leaving 29 intervention and 32 control schools. Baseline surveys were administered in cla ss during the Fall semester of 2002, and three follow-up surveys were conducted dur ing the intervention period (Spring 2003, Spring 2004, Spring 2005). All students who were enrolled in the school during the intervention year were eligible to complete the surveys. During the Fall of 2002, 91% (n=4,259) of eligible students participated in the baseline survey; 94% (n = 4,240) participated in the first follow-up (Spring, 2003); 93% (n = 3,778) completed the second follow-up (Spring, 2004); and 95% (n = 3,802) of students completed the final middle-

PAGE 54

54 school survey (Spring, 2005). Of those pres ent at baseline, the cohort follow-up rate was 61% from Fall 2002 through Spring, 2005. At baseline, parents were surveyed usi ng a sample obtained from the Chicago Public Schools (CPS) address list. When surv eys were administered to students, each student was asked to deliver a modified surv ey to their parent, and parents were given a $25 incentive for returning the survey. Student s were given a $5 gift card to deliver the survey to their parents. Two weeks later, another survey was sent home, and teachers reminded students to prompt thei r parents to return the surv ey. At the end of baseline data collection, a total of 3,250 parents (70% of eligible parents) responded. Participants The sample used in this study includes participants who were present at baseline (6th Grade, Fall) and completed at leas t one additional survey throughout 6th through 8th grades. This cohort consisted 5,815 adolescent s, who were 50.6% male, 12.7% White, 43.1% African-American, 28.5% Hispanic, and 14.7% other races. The average age in 6th grade was 11.84 (sd = 0.58). Nearly th irty percent of the sample reported participating in a group fight in 6th grade (27.5). The prevalenc e of the independent and dependent variables are detail ed in Tables 2-3 and 2-4. Measures The current study used baseline measures as covariates to examine trajectories of violence in the cohort of st udents who participated in the baseline survey and at least one addition follow-up during the P NC data c ollection period. Group fighting Participants were asked, During the last month, how many times have you taken part in a fight where a gr oup of your friends were against another group?. Response options included, never, -3 times, and or more times.

PAGE 55

55 Participants were dichotomized into tw o groups (e.g., group fighters and non-group fighters) for analytical purposes. Aggression. All participants were asked three item s to evaluate past month levels of aggression. These three items included, how many times have you pushed, shoved, pulled someones hair, or gr abbed someone?; how many times have you kicked, hit, or beat up another person?; and how many times have you told someone you were going to hit or beat them up?. All items included the following response options, Never, -3 times, or more times. These items we re coded as , , and for each item at each follow-up, and this scale was used to create the trajectories of aggression. Peer alcohol use Students were asked, How many of your friends drink alcohol?. Response options ranged from None to Almost All. These were recoded as None, a few, or more than a few. These items were included because literature suggests that individuals who have peers who use alcohol (Guo, Elder Cai, & Hamilton, 2009; Herrenkohl et al., 2007; Kuntsche, Go ssrau-Breen, & Gmel, 2009; Leech, Day, Richardson, & Goldschmidt, 2003) are more likely to engage in violent behavior. Risk Factors for Violence Depression. Students were asked, During the la st month, how often have you felt sad or depressed?. Responses were coded as, Never, or One or more times. Depression was included as a covariate becau se higher levels of depression have been associated with violence (E lbogen & Johnson, 2009; Senn, Carey, & Vanable, 2010; Thurnherr, Bechtold, Michaud, Akre, & Suri s, 2008) and other risk behaviors (Latzman & Swisher, 2005; Senn, Carey, & Vanable, 2010).

PAGE 56

56 Parental Involvement The parental involvement scale included ten items measuring parental communication and involv ement. These items included frequency of parental praise and general talking, aski ng about school and where the adolescent was going, discussing problems at school, influenc es of media and other alcohol advertising, problems with alcohol, alcohol rules, and alcohol consequences, dining habits, and music restrictions. Responses included Never, Hardly Ever, Sometimes, A lot, and All the time. Values for each item ranged from 1 to 5, with high er scores indicating greater parental involvement. The standardized Cronbach coefficient alpha for this scale was 0.74. This scale was included as a covari ate because evidence suggests that parenting variables (e.g., monitoring, invo lvement) are related to violence (Park, Morash, & Stevens, 2010). Baseline alcohol use Students who have used alcohol in the past year were measured using the item, Durin g the last 12 months, on how many occasions, or times, have you had alcoholic beverages to drink?. Options included, times, -2 occasions, -5 occasions, -9 occasi ons, -19 occasions, 20-39 occasions, and or more occasions. Responses were dichotomized into drinkers and nondrinkers. This item is included as a co variate because alcohol use has been associated with increased levels of violent behavior (Maldonado-Molina, Reingle, & Jennings, 2010). Marijuana use. Marijuana use in the past year was captured using the item, During the last 12 months, on how many o ccasions, or times, have you used marijuana (other names for marijuana are: pot, grass, weed, reefer, blunt, or hashish)?. Response options ranged from occasions to or more occasions. Responses

PAGE 57

57 were dichotomized to include marijuana user s and marijuana non-users. These items were included because evidence suggests that the use of marijuana (Boles & Miotto, 2003; Dhungana, 2009; Herrenkohl et al., 2007) increases the risk of violent behavior. Unsupervised time Students were asked, About how many hours a day do you usually spend without an adult around?. Pa rticipants responded, None, less than one hour, -2 hours, -4 hours, and or more hours. This item was included as a covariate because evidence sug gests that adolescents are mo re likely to be delinquent when they have more leisure time unsuper vised by a parent or guardian (MaldonadoMolina, Jennings, & Komro, 2009). Race/Ethnicity Participants race and ethnicity was measured at baseline using the item, How do you describe y ourself? Mark all that describe you. If you are not sure, mark other and write in how you describe your self. Response options included, Asian American or Asian Indian, Afr ican-American or Black, Lat ino, Hispanic, or MexicanAmerican, Native American or American Indian, White, Caucasian, or European American, and Other. Partic ipants were coded as Hispani c if they identified as Hispanic, regardless of the other options selected. Natural Parent Household To evaluate students living arrangements, participants were asked, Who do you live with most of the time? Ma rk all that apply. Response options included, Mother and father together, and other combinations of parents and grandparents. Students were c oded as living with both parents (e.g., natural parent household), or other. For a review of the evidence suggesting that single-parent households are a risk factor for violence, see Anderson (2010).

PAGE 58

58 Free or Reduced Lunch An indicator of family socioeconomic position, participants were asked, Do you receive fr ee or reduced price lunches at school? Free or reduced-price lunch means t hat lunch at school is provided for free or you pay less for it. Responses were coded as Yes or No. Analytical Methods Trajectory groups were fitted to the dat a using group-based trajectory modeling (Nagin & Land, 1993; Nagin, 2005). This method of analysis grouped individuals together based upon common attributes (e.g., le vels of violence over time). This approach is appropriate in this situation bec aus e violence varies ov er time (Farrington, 1986), and individuals with different levels of violence may be substantially different from each other. Grouping participants with heterogeneous le vels of violent behavior together and then attempting to predict violence may dilute the effect of risk or protective factors. Group-Based Trajectory Modeling Group-based trajectory models are finite mixture models, which use singleand multiple-gr oup models structures (Nagin, 2005) Finite mixture models (also known as latent class models) represent the heter ogeneity in a finite number on unmeasured (latent) classes. The trajectory groups t hat were created using these analyses were derived from maximum likelihood estimation. In this case, violence data follow a Poisson distribution with a large number of non-violent events (zer o violent events). Therefore, a zero-inflated poisson (ZIP) dist ribution was specified in the model (Jones, Nagin, & Roeder, 2001). Models were tested until the most parsi monious number of trajectory groups maximizes the Bayesian Informati on Criterion (BIC). The BIC refers to: BIC = log(L)

PAGE 59

59 0.5 klog(N), where log likelihood at the maximum likelihood estimate is subtracted from half the number of parameters mult iplied by the log of the sample size. The trajectories are descriptive in nature, and quadratic, cubic and linear models were tested to correctly depict the slopes represented in the data. SAS PROC TRAJ was used to estimate the trajectories (SAS Institute, Cary NC; Jones, Nagin, & Roeder, 2001). Individuals were sorted into trajecto ry groups using the maximum probability procedure (Nagin, 2005, 2001). In other words, participants were assigned to groups in which they have the greatest probability of me mbership (e.g., greater than .80). The modeling strategy estimated posterior probabi lities of assignment to each group, and individuals were assigned to the group wit h the highest probability. This does not guarantee that all individuals had a perfect probabi lity of for membership in a latent group, but the mean assignment probability for each group is expected to be high (>.80; Nagin, 2005; 2001). These high assignment pr obabilities increase confidence in the validity of latent groups. This modeling approach has a number of strengths and weaknesses. Latent group-based modeling allows us to estimate patterns of violent behavior over the lifecourse. These summaries of violence over time provide more information than a traditional dichotomous, violent or non-violent outcome vari able. However, there is a possibility that groups that ar e not meaningful could emerge from the data (e.g., a latent class with 2% or less of observa tions categorized in that group) To avoid this situation, the BIC as well as a judgment of parsimony was used in t he modeling procedure. In addition, the shape of each trajec tory and the number of latent groups is sensitive to the size of the dataset, characteri stics of the sample, or lengt h of the follow-up (Eggleston,

PAGE 60

60 Laub, & Sampson, 2004). Specifically, Eggles ton et al. (2004) fou nd that the shape of each trajectory group remained relatively constant despite changes in the length of follow-up; a doubling of the follow-up time changed their conclusion in identifying the time of peak criminality and estimating in dividual-level group membership. These limitations must be considered when interpreting the results of these analyses. Latent group-based trajectory modeling has been used in several studies in estimating trajectories of violence among adolescents and young adults. Piquero (2008) reviewed 80 studies on trajectories of delinquency over the life-course. This study found evidence of the age-crime curve, as criminal behavior decreased over time. The majority of studies found betwe en three and five classes of delinquency, regardless of methodology and the sample (Piquero, 2008; Maldonado-Molina et al., 2009). Overall, these trajecto ry models have been used in pr evious studies to examine delinquency and violence over time. Multinomial Logistic Regression Once trajectory groups were specified, bivariate and multivariate multinomial logistic regr ession procedures were used to estimate odds-ratios for risk and protective factors on membership in each trajectory. This model is an extension of multiple logistic regressions; however, the model is more appropriate in this situation because trajectory group membership is a nominal variable, and this procedure compares membership in each trajectory group to a reference category (e.g., low-level violence). Under this model, each ( g 1) odds-ratios were generated (Hedeker, 2003). Multinomial logistic regression proc edures can be adapted to account for the multilevel nature of the data (Hedecker, 2003). The Add Health sampling design selected schools as the primary sampli ng unit, and individuals are nested within

PAGE 61

61 schools, and the PNC sampling design f eatures nested student s within schools. Clustered robust standard errors were estimat ed to produce error estimates that take into account the autocorrelation due to the sa mpling design. Failure to account for the sampling deign would result in an inflated Ty pe 1 error rate, artificially increasing the precision of the effects (Twisk, 2006). T he adapted multinomial model does not assume that observations are independent; therefore, it is appropr iate for longitudinal and clustered designs (Hedecker, 2003). STATA 11 software (College Station, TX) was used to conduct all multinomial logistic regression analyses. Mediation Analyses Mediation analyses were conduc ted to evaluat e the indirect effect of community-, parentand peer-level variables on violent traj ectory membership for all three studies. Trajectory groups were dichotomized into violent trajectory group me mber (if classified as a desistor or an escalator) and non-violent trajectory group member (if non-violent) groups for logistic regression modeling (Ma cKinnon, 2008). For each mediator and contextual variable, four logistic regressi ons were examined: 1) the effect of the contextual variable on the mediator (slope a) ; 2) the effect of the mediator on the outcome (violent trajec tory membership, slope b ); 3) the direct effe ct of the contextual variable on the outcome (slope c); and 4) the adjusted effect of both the contextual variable and the mediator on the outcome. All regressi on models were adjusted for other risk factors, demographics, and sampling design. These effect sizes were standardized and adjusted using the covariance matrix and variance fo r each variable in the model (MacKinnon, 2008). To test the significance of the mediator, the Sobel test was used to generate a z statistic and standard error (Baron & Kenny, 1986; Sobel, 1982; MacKinnon, Warsi, &

PAGE 62

62 Dwyer, 1995). The percent mediation for each mediator was calculated using the formula: ab/a1b1axbx +c In this formula, a represents the effect of the contextual variable on the hypothesized mediator, and b represents the effect of the mediator on the outcome variable (in this case, violence). C represents the direct effect of the contextual variable on the outcome. All of these standardized estimates (including all other variables in the model) were used to calculate the proportion of the variance in each contextual variable that is mediat ed by each proximal variable. These percentages were summed by contextual vari able to estimate the proportion of the contextual variable that is mediated by more proximal variables. Summary of Methodology Data sources for this study included a l ongit udinal, nationally representative study (Add Health, Studies 1 and 2), and a longitudina l sample of high-risk, ethnically diverse youth (PNC, Study 3). To test the study hypotheses, gr oup-based trajectory models were estimated, and then prediction of traj ectory group membership was conducted using multinomial logistic r egression. These methods te sted the hypothesis that gender and ethnic differences in violence will be at leas t partially attributab le to group fighting and multiple domains of risk factors extending beyond the individual-level.

PAGE 63

63 Table 2-1. Description of dependent variabl es, Add Health, Wave s I-IV. N = 9421. Violence Wave I (%) Wave II (%) Wave III (%) Wave IV (%) Shot or stabbed some one 2.07 2.01 0.58 9.54 Used knife or gun 5.32 5.53 2.04 11.65 Treatment by docto r 20.20 9.2 7.55 3.00

PAGE 64

64 Table 2-2. Description of cova riates, Add Health, Wave I. Covariate % Community-level Racial Dispersion a 0.31(0.27) % Poverty a 0.13(0.03) Urban area 61.8 Parental and Peer Influences Parental Involvement a 5.81(3.4) Parental alcohol use 65.7 One or more peers use alcohol 57.6 One or more peers use marijuana 36.4 Individual-level Risk Factors Ever used alcohol 58.2 Lifetime marijuana use 30.5 Lifetime use of other drugs 17.7 Past week depression b 42.3 Intend to go to college 72.3 Desire to leave home 38.2 Speaking Spanish at home 6.14 Violence Group fighting in Past Year 22.1 Baseline violence 22.0 Demographics Gender (Male) 42.8 Age at Baseline a 15.4(1.60) White 64.5 African-American or Black 23.6 Hispanic or Latino 14.8 Asian or Pacific Islander 5.8 Native American 4.1 Other Race 1.1 a Mean(SD) are reported. b Depression was measured as feeling sad or depressed one or more times in the past month.

PAGE 65

65 Table 2-3. Description of dependent va riables, PNC (ST02-ST05). N= 4188. Violence T1 T2 T3 T4 Threaten to hit or beat up someone 52.8%61.2% 67.0%68.6% Pushed, shoved, pulled so meones hair, or grabbed someone 63.2%63.3% 73.0%70.6% Kicked, hit, or beat up another person 47.6%54.0% 58.1%56.1%

PAGE 66

66 Table 2-4. Description of covariates, PNC (ST02). Covariate % Community-level Neighborhood exposure to alcohola 0.23(0.16) Area deprivation a -32.90(29.50) Adults in neighborhood drink g 44.0 Parental and Peer Influences Parental Involvement a 35.89(7.61) Home access to alcohol 16.6 Peer alcohol use f 10.9 Individual-level Risk Factors Past year alcohol use 19.3 Past year marijuana use 4.9 Depression c 72.5 Unsupervised time d 50.3 Natural parent household 52.7 Free or Reduced Price Lunch 69.1 Spanish at Home (Hispanics only) 66.2 Low Academic Achievement e 67.9 Violence Past month group fighting 27.5 Demographics Male 50.6 Age at Baseline a 11.84(0.58) White 12.7 African-American or Black 43.1 Hispanic or Latino 28.5 Other Race 14.7 a Mean(SD) are reported. b Home access to alcohol was measured as last obtaining alcohol from either the home or the adolescents parent. c Depression was measured as feeling sad or depressed one or more times in the past month. d Unsupervised time was measured as having one or more hours each day without being supervised by an adult. eLow academic achievement was defined as having reported poor performance on a test or project in the past month. f Some, many, or almost all peers use alcohol. g Many or almost all parent s in the neighborhood use alcohol.

PAGE 67

67 CHAPTER 3 RISK FACTORS ASSOCIATED WITH TRAJEC TORIES OF SERIOUS VIOLENCE IN A NATIONAL LY REPRESENTATIVE, LONGITUDINAL SAMPLE The purpose of this study is to estimate trajectories of serious violence using a longitudinal sample of adoles cents, considering group fight ing (independent of baseline violence) and other multiple domains of ri sk factors as a key differentiator between profiles of violent behaviors. Evidence s uggests that adolescents who engage in violence are more likely to participate in other high-risk activities (e.g., alcohol and other drug use, dropping out of school, gun ow nership, gang membership, risky sexual activity and familial independence), and increase their risk of health-related consequences (including serious injury and death) In order to prevent violent behavior, a more complete understanding of the etiology of violence is necessary. Methods. Participants included a nationally represent ative sample of 9421 adolescents followed from ages 15 through 26. Traj ectories of violence were estimated, and participants were assigned to trajectory groups using latent trajectory modeling. Multinomial logistic regression procedures were used to evaluate t he effect of multiple domains of risk and protective factors in stages (e.g., community-level, parentand peer-level, and individual-level) to understand the predictors of membership in high-violence trajectory groups. Mediation analyses were conducted to further evaluate the direct and indirect effect of community-, parental and peer-level variables on violent trajectories. Results. Three groups of violence trajectories were identified: 1) NonViolent (73.1%); 2) Escalators (14.6%); and 3) Desistors (12.3% ). Group fighting sign ificantly predicted violence above and beyond baseline violent behavior for desistors only. Racial dispersion at the neighborhood level predict ed both escalation and desistance before individual-level characteristics were entered into the model. Peer alcohol use predicted

PAGE 68

68 both escalation and desistance; however, these effects were mediated by individuallevel variables. Aside from baseline violence no other contextual, peer, or individuallevel risk factor predicted membership in t he escalator trajectory group. Conclusion. The lack of significance in predicting escala tion highlights the need for further study on the etiology of late onset violence. Group fi ghting is a significant predictor of violence beyond baseline violence, but only for those who desist from violent ac tivity. Contextual variables (such as peer substance use and racial composition of the neighborhood) appear to increase risk for violence. Background Each year, nearly 700, 000 adolescents and y oung adults (10-24) are treated in the emergency room for injuries re lated to violent activity (CDC, 2009). Evidence suggests that adolescents who engage in delinquent behavior are more likely to engage in other high-risk activities (e.g., alcohol and ot her drug use, dropping out of school, gun ownership, gang membership, risky sexual activity and familial independence) (Thornberry, Huizinga, & Loeber, 1995; CDC, 2009; 2010) and increase their risk of health-related consequences (including serious injury and death) (Conseur, Rivara, & Emanuel, 1997; Farrington & Loeber, 2000). The evidence is clear that individualand family-level factors increase the risk for violent behavior. For exampl e, neurological deficiencie s and cognitive impairments (Moffitt et al., 2001), low IQ, hyperactivity, di fficulty concentrating at school, beliefs and attitudes favorable to violence, antisocial behavior, and impulsivity have been consistently associated with violent behavior (H awkins et al., 2000; Howell, 2009). At the family level, parental criminal behavior, child maltreatment, low levels of parental involvement, parental attitudes favorable to violence and drug/alcohol use, and

PAGE 69

69 separation of the parent and child have been identified as risk factors in a recent metaanalysis of longitudinal studies of risk factors for violence (Hawkins et al., 2000). Each of these factors has been consistently associated with violence and delinquency. Despite the strong evidence in support of some risk factors for violence, other behavioral risk factors within the family and p eer group are less studied. Hawkins et al. (2000) found that delinquent peers and gang membership have been predictive of violent behavior; however, the effect of peer and parental substance use is unclear. Academic failure and dropout has also been a ssociated with violence, but less drastic measures of academic success have not been ev aluated in the empirical literature on violence (Hawkins et al., 2000) Community-level influences such as availability of firearms, exposure to viol ence, and exposure to racism in the neighborhood have consistently been linked to violent behavio r (Kaufman, 2005; Reingle, Jennings, Maldonado-Molina, & Canino, 2010). Finally, although many studies have analyzed the multiple domains of risk and protective fa ctors for violent behavior, few have assessed the degree to which contextual variables have indirect effects through more proximal variables at the individual-level. Although many of these studies provide insight as to t he longitudinal predictors of violent behavior at multiple levels of infl uence, no studies to our knowledge have used multiple domains of predictors while differ entiating patterns of serious violence among adolescents over time. Consistent with the lif e-course perspective (Moffitt, 2001), not all offenders are the same in their patterns of vi olent behavior. Accordi ng to Moffitt (2001), most offenders will desist after participating in violence during their adolescent years, while a small proportion will cont inue offending over the life-c ourse. These life-course

PAGE 70

70 persistent offenders will commit more serious crimes, and will do so over a longer time period than adolescent -limited offenders. Th is study will estimate trajectories of violence to differentiate these two groups of offenders, and assess the predictors for each group indepen dently. Based on the extant literatur e on trajectory modeling of violence and delinquency, most studies of violence over time repor t between three and five trajectory groups, regardless of methodology or sample (Pique ro, 2008). In a review of 80 studies on trajectories of violence, Piquero (2008) f ound that support for t he age-crime curve (e.g., the aging out of crim inal activity over the life-course) (Farrington, 1986), and varying profiles of violent behavior in a va riety of populations. These findings have been replicated in subpopulations of ra ce/ethnicity, and across gender groups (Maldonado-Molina et al., 2010; Maldonado-Molina et al., 2009; Jennings et al., 2010). There are several theoretical perspectives that will be used to explain the role of multiple risk and protective fa ctors of trajectories of viol ent behavior. First, Social Learning Theory (Akers, 1973) emphasizes t he role of the adolescents social surroundings (peers, adults, sibl ings, etc.) and their infl uence on violent and delinquent behavior. Second, Social Bond Theory (Hir schi, 1969) describes risk factors at the individual level (e.g., commitm ent to school, education, family and peers; attachment to parents, involvement in pro-social extracurri cular activities) for violent delinquency. Finally, Social Disorganization (Shaw & McKa y, 1942) explains the external influence of the communities in which adolescents reside, and how the community influences delinquent behavior. Taken toget her, these perspectives info rmed the selection of the multiple domains of risk and protective factors that were selected fo r analysis. Multiple

PAGE 71

71 levels of influence informed by theory shoul d converge to provide the foundation for delinquency and violence prevention efforts. Using an epidemiological approach, the init ial stage of prevention is a thorough understanding of the causes and risk factor s for violence. To our knowledge, no longitudinal studies have evaluated the effect of group fighting specifically (independent of baseline violence) on violent behavior am ong adolescents. The purpose of this study is to examine the differential risk and protec tive factors for violence longitudinally, focusing on group fighting as a risk factor for vi olence over the life-course. Specifically, I hypothesize that 1) adolescents who partici pate in group fights are more likely to engage in high levels of serious violence dur ing late adolescence and adulthood; 2) group fighting is a predictor of future vi olence independent of ot her baseline violent behavior; and 3) exposure to multiple domai ns of risk factors at age 15 (community, family, peer, and individual-level) increases violent behavior into adulthood. Method Design The National Longitudinal Study of Adole scent Health (Add Health) is a schoolbased panel study conducted from 1994 (Wav e I) through 2008 (Wave IV), when participant ages ranged from 11-32 (Chantal a & Tabor, 1999). The data collect ion for this survey was designed to explore effects of multiple domains on adolescents health behaviors. In Wave I, 80 communities were selected to ensure demographic representativeness (ethnic composition, region of the country, urba nicity, school size, and school type) of students in t he United States. Schools (n = 132) were eligible if they enrolled more than thirty students and had an eleventh grade. All students who were enrolled in the school and were present on the survey day were eligible for participation

PAGE 72

72 in the study. Approximatel y 200 students were randomly sele cted from strata of grade and sex, resulting in a final cohor t sample of 9421 adolescents. Data Collection At Wave I, the baseline sa mple of 20,745 participated in a oneto two-hour inhome interview. This interview c ollect ed data on a variety of topics, including demographics, peer networks, health, employment, substance use, criminal and delinquent behavior, sexual behavior, and ro mantic partnerships. All data were recorded on laptop computers, and the participant listened to the questions and responded on the laptop themselves to maximize validity for sensitive items. At this time, parents were also surveyed to ev aluate their health behaviors, education, neighborhood characteristi cs, and heritable conditions (Chantala & Tabor, 1999). Wave II data collection included another in -home interview one year later, between April and August, 1996. Interview items were similar, but additio nal nutrition and sun exposure items were added. The response rate for eligible participants in Wave II was 88.2%. All participants who were interviewed in Wave I were elig ible, excluding those who moved out of the country. Jailed adole scents were interviewed when possible. Wave III included data collected between 2001-2002, when participants were ages 1826. The survey instrument changed to reflect the more influential romantic relationships, criminal history, and joband college-related influences rather than school-based influences on health behaviors. In 2007-2008, Wave IV of data collection co nsisted of a 90-minute questionnaire to be completed by respondents via laptop. Physical and biological markers were collected, and questionnaires included a greater number of financial, marital/cohabitation, medications, physical health, and childhood maltreatment items.

PAGE 73

73 The response rate for this sa mple was 80.3% of eligible participants. Participants were eligible for all waves of data collecti on if they participated in Wave I. The sample used in this study includes par ticipants who were present at all four waves of data collection. This cohort wa s 42.8% male, 64.5% Wh ite, 23.6% AfricanAmerican, 14.8% Hispanic, 5.8% Asian or Pacific Islander 4.1% American Indian, and 1.1% Other. The average age at Wave I was 15.4 (sd = 1.6), 16.3 (sd = 1.6) at Wave II, 21.7 (sd = 1.6) at Wave III and 26.5 (sd = 1.8) at Wave IV. Mo re than twenty percent of the sample reported participat ing in a group fight at Wave I (22.1%). Mean violence at Wave II was 0.52(SE = 0. 04), 0.28(SE=0.03) at Wave II I, and 0.91(SE=0.04) at Wave IV. The prevalence of the independent and dependent variables are detailed in Table 31. Measures Violent Delinquency Violenc e was measured using three items that were measured across each of the f our waves of data collection: In the past 12 months, have you 1) hurt someone badly enough that he or she needed care from a doctor or nurse?; 2) pulled a knife or gun on someone?; and 3) shot or stabbed someone? At each wave, a value from 0-12 was assigned to each participant, where a value of , , or was assigned for each of these violent acts in which the individual has participated in during the past year. A zero was assigned for each item if the participa nt did not report the behavior. A two was assigned if the adol escent reported hurting someone badly enough to need care from a doctor or nurse one to three times in the past year. A four was assigned for each of the following occu rrences: 1) shooting or stabbing someone; 2) pulling a knife or gun on someone; or 3) hurting someone badly enough to need care

PAGE 74

74 from a doctor or nurse four or more times in the past year. These values were used to create trajectories of delin quency across Waves II-IV. Group Fighting Group fighting was measured using the variable, In the past 12 months, how often did you take place in a physical fight where a group of your friends was against another group?. Responses to this item include: =Never, =One or two times, =Three to four times, and =5 or more times. These responses were dichotomized into =never group fighting and =group fighting in the past year. Racial Dispersion Racial dispersion is a measure (ranging from 0 to 1) of the racial heterogeneity in a neighborhood. Dispers ion is equal to zero when all census tract members are members of the same racial group, a nd equal to one when residents are equally distributed among White, AfricanAmerican, Asian, Native American, and Other races. Poverty Poverty was measured using the percentage of families in the respondents census tract whose income was at or below the poverty level. Racial dispersion was measured using the concent ration of racial/eth nic homogeneity within the neighborhood, so higher proportions woul d indicate greater concentration of one race/ethnicity within a neighborhood. These items were incorporated into the analysis in accordance with Shaw and McKays (19 42) theory of Social Disorganization. Urban Neighborhood. All addresses were geocoded at t he time of the interview, and these addresses were linked to U.S. Census data (1990) to determine the urbanicity of the residence. Addresses were considered complet ely urban or not completely urban. Risk factors for violence

PAGE 75

75 Parental Involvement Parental influence and invo lvement was measured using a scale of twenty items (10 for maternal involvement, 10 meas uring paternal involvement) (Prado et al., 2009). Each indi vidual item was dichotomized, and the scale is the sum of all twenty items (range: 0-20). The ten it ems which comprised the scale included whether or not the respondent r eported participating in the follo wing activities with their mother and/or father in the pas t four weeks: 1) going shoppi ng; 2) playing a sport; 3) attending a religious or church-related even t; 4) talking about someone they are dating or a party they attended; 5) attending a movie, play, concert, or sporting event; 6) talked about a personal problem they were havi ng; 7) had a serious argument about their behavior; 8) talked about work or grades; 9) worked on a project for school; and 10) talked about other things they are doing in school. Cronbachs coefficient alpha for this scale was 0.74. This scale was included as a covariate because evidence suggests that parenting variables (e.g., monitoring, invo lvement) are related to violence (Park, Morash, & Stevens, 2010). Parental Alcohol Use. At the Wave I survey, paren ts of surveyed adolescents were asked, How often do you drink alcoho l?. Response options included, Never, Once a month or less, Two or three days a month Once or twice a week, Three to five days a week, and Nearly every day. Responses were dichotomized into parents use alcohol and parents do not use alcohol based upo n the distribution of the responses. Peer Alcohol Use Peer alcohol use was measured using one item: Of your three best friends, how many drink alcohol at least once a month? Respondents who reported having one or more friends who use al cohol monthly were coded as . These

PAGE 76

76 items were included because literature suggests that indi viduals who have peers who use alcohol (Herrenkohl et al., 2007; Kunt sche, Gossrau-Breen, & Gmel, 2009; Leech, Day, Richardson, & Goldschmidt, 2003) are mo re likely to engage in violent behavior. Peer Marijuana Use. Respondents were asked, Of your three best friends, how many use marijuana at least once a mont h? Respondents who reported having one or more friends who use marijuana monthly were coded as . These items were included because literature suggests that indivi duals who have peers who use marijuana (Herrenkohl et al., 2007; Leech, Day, Richar dson, & Goldschmidt, 2003) are more likely to engage in violent behavior. Depression. This mental health status variable was measured with one item, How often in the past week have you felt depressed?. Values for this variable were dichotomized so that 1 = One or more times and 0 = in stances of depression in the past week. Depression was included as a covariate because higher levels of depression have been associated with violence (Elbogen & Johnson, 2009; Senn, Carey, & Vanable, 2010; Thurnherr, Bechtold Michaud, Akre, & Suris, 2008) and other risk behaviors (Latzman & Swisher, 2005; Senn, Carey, & Vanable, 2010). Academic Achievement Academic performance was measured using the variable, On a scale of 1 to 5, where 1 is low and 5 is high, how likely is it that you will go to college?. This item was included as a covariate because academic achievement and IQ have been associated with increased risk of violence (Herrenkohl, McMorris, Catalano, Abbott, Hemphill, & Toumbour ou, 2007; Leech, Day, Richardson, & Goldschmidt, 2003).

PAGE 77

77 Alcohol Use Lifetime alcohol use was evaluat ed using the item, Have you had a drink of beer, wine, or liquornot just a si p or a taste of someone elses drinkmore than 2 or 3 times in your life?. Those who responded affirmatively to this item were categorized as Alcohol Users. Marijuana and other drug use Marijuana use was measured using the item, During your life, how many times hav e you used marijuana? Responses were categorized into users and non-users. Other drug use was created using the selfreported number of times the re spondent used cocaine, inhalants, or other drugs in their lifetime. If any of these drugs were used, respondents were categorized as users. These items were included because evidence suggests that the use of marijuana and other drugs (Boles & Miotto, 2003; Dhungana, 2009; Herrenkohl et al., 2007) increases the risk of violent behavior. Desire to leave home This variable was measur ed using the following item: How much do you feel that you want to leave home?. Respondents who reported very much or quite a bit were categorized as , others were categorized as . This variable was included because some ev idence suggests that a negative home environment increases the likelihood of violen t delinquency (Ou & Reynolds, 2010). Analytical Methods Group-Based Trajectory Modeling To examine the first research questi on, that adolescent s who engage in group fights early in life (age 15) are more likel y to engage in high levels serious violence during late adolescence and adulthood, trajecto ry groups were fitted to the data using group-based trajectory modeling (Nagin & La nd, 1993; Nagin, 2005). This method of analysis grouped individuals together based upon common attributes (e.g., levels of

PAGE 78

78 violence over time). This approach is appr opriate because violence varies over time (Farrington, 1986), and individuals with different levels of violence may be substantially different from each other. Grouping participant s with heterogeneous levels of violent behavior together and then attempting to predict vi olence may dilute the effect of risk or protective factors. Group-based trajectory models are finite mixture models, which use singleand multiple-group models structures (Nagin, 2005) Finite mixture models (also known as latent class models) represent the heter ogeneity in a finite number on unmeasured (latent) classes. The trajectory groups that are created usi ng these analyses are derived from maximum likelihood estimation. In this case, violence data follow a Poisson distribution with a large number of non-violent events (zer o violent events). Therefore, a zero-inflated poisson (ZIP) dist ribution was specified in the model (Jones, Nagin, & Roeder, 2001). Models were tested until the most parsi monious number of trajectory groups maximizes the Bayesian Informati on Criterion (BIC). The BIC refers to: BIC = log(L) 0.5 klog(N), where log likelihood at the maximum likelihood estimate is subtracted from half the number of parameters mult iplied by the log of the sample size. The trajectories are descriptive in nature, and quadratic, cubic and linear models were tested to correctly depict the slopes represented in the data. SAS PROC TRAJ was used to estimate the trajectories (SAS Institute, Cary NC; Jones, Nagin, & Roeder, 2001). Individuals were classified into mutually -exclusive trajectory groups using the maximum probability procedure (Nagin, 2005, 2001). In other words, participants were be assigned to groups in which they have the greatest probability of membership

PAGE 79

79 (e.g., greater than .80). The modeling st rategy estimated poste rior probabilities of assignment to each group, and individuals were assigned to the group with the highest probability. This does not guarantee that a ll individuals have a pr obability of or membership in a latent group, but the mean assignment probability for each group is expected to be high (>.80; Nagin, 2005; 2001) These high assignment probabilities increase confidence in the validity of latent groups. This modeling approach has a number of strengths and weaknesses. Latent group-based modeling allows us to estimate patterns of violent behavior over the lifecourse. These summaries of violence over time provide more information than a traditional dichotomous, violent or non-violent outcome vari able. However, there is a possibility that groups that ar e not meaningful could emerge from the data (e.g., a latent class with 2% or less of observa tions categorized in that group) To avoid this situation, the BIC as well as a judgment of parsimony was used in t he modeling procedure. In addition, the shape of each trajec tory and the number of latent groups is sensitive to the size of the dataset, characteri stics of the sample, or lengt h of the follow-up (Eggleston, Laub, & Sampson, 2004). Specifically, Eggles ton et al. (2004) fou nd that the shape of each trajectory group remained relatively constant despite changes in the length of follow-up; a doubling of the follow-up time changed their conclusion in identifying the time of peak criminality and estimating in dividual-level group membership. These limitations must be considered when interpreting the results of these analyses. Latent group-based trajectory modeling has been used in several studies in estimating trajectories of violence among adolescents and young adults. Piquero (2008) reviewed 80 studies on trajectories of delinquency over the life-course. This

PAGE 80

80 study found evidence of the age-crime curve, as criminal behavior decreased over time. The majority of studies found betwe en three and five classes of delinquency, regardless of methodology and the sample (Piquero, 2008; Maldonado-Molina et al., 2009). Overall, these trajecto ry models have been used in pr evious studies to examine delinquency and violence over time. All baseline covariates in the model were recoded as dichotomous (with the exception of parental involvem ent, poverty, racial dispersion, and age). The distribution of the parental involvement measure is normal, with a relatively high internal consistency ( = 0.74). A description of all covariat es in the model is displayed in Table 4-1. Trajectory estimation was conducted usi ng the recoded violence measures from Waves II-IV. Because there are a large number of non-violent parti cipants in the data, the violence measure follows a Poisson distribution, with a small num ber of adolescents reporting very high levels of violence. SAS PROC TRAJ was used to extract latent groups from the data. Both the Bayesian In formation Criterion (BIC) and the proportion of the sample in each latent group were used to select the appropriate number of groupings without over-extracting groups from the data. Multinomial Logistic Regression Once trajectory groups have been specified, bivariate and multivariate multinomial logistic regr ession were used to estimate odds-ratios for risk and protective factors on membership in each trajectory. This model is an extension of multiple logistic regressions; however, the model is more appropriate in this situation because trajectory group membership is a nominal variable, and this procedure compares membership in

PAGE 81

81 each trajectory group to a reference category (e.g., low-level violence). Under this model, each ( g 1) odds-ratios were generated (Hedeker, 2003). Multinomial logistic regression proc edures can be adapted to account for the multilevel nature of the data (Hedecker, 2003). The Add Health sampling design selected schools as the primary sampling uni t. Clustered robust standard errors were estimated to produce error estimates that take into account the autocorrelation due to the sampling design. Failure to account for the sampling deign would result in an inflated Type 1 error rate, artificially increasi ng the precision of the effects (Twisk, 2006). The adapted multinomial model does not a ssume that observations are independent; therefore, it is appropriate for longitudi nal and clustered designs (Hedecker, 2003). STATA 11 software (College Station, TX) was used to conduct all multinomial logistic regression analyses. The first stage of model selection involved a bivariate test of the association of each predictor variable with the trajectory groups. All variabl es that are not marginally predictive of any dependent variable (traje ctory group) in bivariate analyses were removed from the multivariate model. After this initial model selection, distal variables (e.g., community-level) were added to the model first, followed by parental and peerlevel variables, and then individual-level characteristics. The final model assessed the influence of all risk and protective fact ors, accounting for baseline violence. Mediation Analyses Mediation analyses were conduc ted with any variables that are significantly associated with both the contextual variables and the outcome (violence). Covariance matrices were generated and regression es timates were standardized to obtain an overall mediated effect of each individual-lev el variable for each c ontextual variable.

PAGE 82

82 To test the significance of the mediator, the Sobel test was used to generate a z statistic and standard error (Baron & Kenny, 1986; Sobel, 1982; MacKinnon, Warsi, & Dwyer, 1995). The percent mediation for each mediator was calculated using the formula: ab/a1b1axbx +c In this formula, a represents the effect of the contextual variable on the hypothesized mediator, and b represents the effect of the mediator on the outcome variable (in this case, violence). C represents the direct effect of the contextual variable on the outcome. All of these standardized estimates (including all other variables in the model) were used to calculate the proportion of the variance in each contextual variable on violence that is mediated by each proximal variable. These percentages were summed by contextual vari able to estimate the proportion of the contextual variables effect on violence that is mediated by more proximal variables. Finally, when baseline violence was a signifi cant predictor of violent trajectory membership in the bivariate analyses, pos t hoc analysis was conducted to understand the risk factors associated with baseline violence. For this analysis, weighted proportions and means are calculated for each category of baseline violence (violent or non-violent) to understand the differences in risk factors for violence at age 15. Results Trajectories of Violence To determine the number of trajectories of vi olence, latent group-based trajectory modeling was used. Three distinct classes were identified: Non-Aggressive (73.1%), Desistors (14.6%), and Escalators ( 12.3%). Non-aggre ssive adolescents had trajectories of violence that averaged zero at each of the three waves. Desistors participated in violence at ear lier waves (II and III), but this violent behavior declined by Wave IV. Escalators had lower levels of viol ence at waves II and I II, but their violent

PAGE 83

83 behavior increased drastically at Wave IV. Th is three-group trajectory model showed the lowest AIC and BIC (AIC = -23272, BIC = -17354) when co mpared to a 4(did not converge), and 2(AIC = 70235, BIC = -23290) class model. The mean posterior probabilities ranged from (0.73-0. 90). Figure 4-1 displays the trajectories of violence from ages 16-26. The average violence values for each traj ectory group at each wave are detailed in Table 4-2. Mean violence for those in th e non-violent trajectory group was zero across all waves. The average vi olence for desistors was 3.18 (SD=3.29) at Wave II, 1.53(SD=2.28) at Wave III, and 0.35 (SD=1.20) at Wave IV. Mean violence for escalators was 0.20(SD=0.82) at Wave II, 0.28(SD=1.11) at Wave III, and 7.15(SD=1.90) at Wave IV. Effects of risk and protect ive factors at age 15 on tr ajectories of violence Table 3-3 shows the bivariate relationship bet ween each risk or protective factor and violence trajectories. For desistors, ra cial heterogeneity in the neighborhood (OR = 2.18; 95% CI 1.42-3.36), peer alcohol use (OR = 1.93; 95% CI 1.52-2.44), peer marijuana use (OR = 2.21; 95% CI 1.78-2.73), alcohol use (2.33; 95% CI 1.92-2.83), marijuana use (OR = 2.52; 95% CI 2.07-3.06), other drug us e (OR = 2.27; 95% CI 1.762.94), desire to leave home (O R = 1.33; 95% CI 1.16-1.53), group fighting (OR = 2.54; 95% CI 2.22-2.92), and baseline violence (OR = 11.96; 95% CI 4.93-7.90) were identified as risk factors for being in the desistor trajectory group compared to the nonaggressive group. Protective factors for des istors included higher levels of parental involvement (OR = 0.96; 95% CI 0.93-1.00; p <0.10), parental alcohol use (OR = 0.84; 95% CI 0.68-1.03; p <0.10), and academic achievement (O R = 0.80; 95% CI 0.74-0.86). For escalators, racial dispersion (OR = 2.01; 95% CI 1.32-3.07), peer alcohol use (OR = 1.34; 95% CI 1.081.65), peer marijuana use (OR = 1. 28; 95% CI 1.01-1.62), marijuana

PAGE 84

84 use (OR = 1.28; 95% CI 0.99-1.67; p <0.10), desire to leave home (OR = 1.16; 95% CI 0.99-1.34; p <0.10), group fighting (OR = 1.32; 95% CI 1.07-1.63), and baseline violence (OR = 2.74; 95% CI 1.30-2.32) were identified as risk fact ors. Academic achievement (OR = 0.91; 95% CI 0.83-1.01; p <0.10) was identified as a protective factor for membership in the escalator group compar ed to the non-aggressi ve group. Because poverty, urban neighborhood, and depression were not significant in predicting violence for either group, they were dropped from further multivariate analyses. Table 3-4 shows the multivariate effect of community-, parentand peer-level variables on the desistor and escalator trajecto ry groups. Racial dispersion (OR = 1.60, 95% CI 0.94-2.59, p <0.10 for desistors; OR = 1.73; 95% CI 1.04-2.91 for escalators) and peer alcohol use (OR = 1.89; 95% CI 1. 50-2.64 for desistors; OR = 1.31; 95% CI 1.02-1.65 for escalators) were significant ri sk factors for both groups. Peer marijuana use remained a risk factor for desistors only (OR = 1.89; 95% CI 1.50-2.64). When individual-level variables were added to the multivariate model (Table 3-5), racial dispersion among escalators was the only contextual variable that remained marginally significant (OR = 1.70; 95% CI 0.99-2.93; p <0.10). A number of individuallevel variables were significant for desistors only. Specifically, alcohol use (OR = 1.68; 95% CI 1.29-2.33), marijuana use (OR = 1.25; 95% CI 1.031.72), other drug use (OR = 1.27; 95% CI 1.09-1.47), and group fighting (OR = 2.02; 95% CI 1.64-2.31) were identified as risk factors for memb ership in the desistors trajectory group. No individuallevel variables significantly predicted member ship in the escalator trajectory group. The full model, adjusted for baseline viol ence and all other risk and protective factors, is presented in Table 3-6. For escala tors, racial dispersion remained marginally

PAGE 85

85 significant (OR = 1.68; 95% CI 0.98-2.88; p <0.10), and baseline violence was also identified as a risk factor (O R = 1.39; 95% CI 1.05-1.86). For desistors, marijuana use was no longer significant. However, alcohol use (OR = 1.55; 95% CI 1.21-2.91), other drug use (OR = 1.21; 95% CI 1.02-1.41), group fighting (OR = 1.69; 95% CI 1.36-1.95) and baseline violence (OR = 3.08; 95% CI 1.27-4.09) predicted membership in the desistor group. Mediated effects of cont extual variables on violent trajectory membership The test of the mediation pathway (based on bivariate analysis) is presented in Figure 3-2. The significance of pathway a indicated that all variables except racial dispersion met the criteria for inclusion in the mediation analysis (e.g., contextual variables must be significantly associated wit h at least one mediator (Table 3-7). The indirect effects of each contex tual variable by each individu al-level variable are detailed in Table 3-8. For the effect of parental involvement on viol ence, 55.1% of the effect was mediated through the individual-le vel variables. More than th ree-quarters of the effect of peer alcohol (76.1%) and peer marij uana use (75.6%) on violence was mediated through proximal variables (such as individual-level alcohol and marijuana use, other drug use, academic achievement, desire to leave home, group fighting, and baseline violence). For parental involvement, group fi ghting was the principal mediator (11.9%), followed by desire to leave home (11.0%). For the effect of peer alcohol use on violence, the greatest mediator was alcoho l use (16.2%), followed by marijuana use (15.7%). For peer marijuana use, the mo st pronounced mediator was marijuana use (18.3%), followed by other drug use (11.9%). Because baseline violence was a significant predictor of both violent trajectory groups, a post-hoc analysis was conducted to understand the characteristics associated

PAGE 86

86 with baseline violence (Table 3-9). Higher levels of racial heterogeneity in the neighborhood; peer alcohol use and marijuana use; alcohol, marijuana, and other drug use; lower academic achievement; greater desire to leave home; depression; group fighting; and various demographic variables (males, African-Americans, and Asians) were all significantly higher among thos e who were violent at baseline. Discussion The present study examined the number and s hape of trajectories of violence, as well as the direct and indirect effects of multiple domains of risk and protective factors for membership in eac h trajectory group. The latent-group based trajectory models fit a three-class model: a non-aggressive group, a group who desisted violence, and a group of escalators whose severity of viol ence increased over time. Group fighting significantly predicted membership in the desistor group independent of other baseline violence, but group fighting was not significa nt in predicting escalation. Racial dispersion (i.e., racial heter ogeneity) at the cens us tract level had a marginal, direct effect on escalation, independent of demogr aphics and baseline violence. Effects of peer alcohol and marijuana use were mediat ed through individual-level alcohol and marijuana use, and the effect of parental involvement was mediated through multiple individual-level variables, including the adole scents desire to leave home, as well as their drug and alcohol use. Baseline violence was significantly associated with both desistance and escalation. These results are consistent with previous research on trajectories of violence, and risk factors for violent behavior among adolesce nts. Three trajectory groups were extracted from the data in this study, and this is consistent with the extant literature that suggests there are between three and five unique groups of adolescents who

PAGE 87

87 participate in violent behavior (Piquero, 2008; Maldonado-Molina, Reingle, Tobler, Jennings, & Komro, 2010; Maldonado-Molina, Piquero, Jennings, Bird & Canino, 2009). The findings from this study are unique in that a late-onset group of aggressive adolescents was identified. Although some studies have found support for the existence of this group (DUnger et al., 1998; Zara & Farrington, 2009), the majority of the literature on trajectories of delinquency supports the age-crime curve, in which adolescents age out of delinquent behaviors before age 20 (Piquero, 2008; Farrington, 1986). The sharp increase found among the lateonset escalators in this study may be due to the weighting of more severe forms of violence used in this study (e.g., using a knife or gun in a fighting, s hooting someone). If this is the case, participants assigned to the escalator group initiate late in life, and with more severe forms of violent behavior. This study identified a variety of risk and protective factors that significantly predicted violent trajectory group membership. These findings are consistent with prior literature on community-, familyand peer-level risk factors for violence. Specifically, parental involvement has been ident ified as a protective factor from violence (Hawkins et al., 2000), and this study found evidence of direct and indirect effects for parental involvement. Additionally, exposure to ra cism in the neighborhood have consistently been linked to violent behavior (Kaufman, 2005). This study found that racial heterogeneity had a direct effect on escalation of violence. This finding is consistent with the racism finding reported in previ ous literature, as racially homogeneous neighborhoods are likely to have less raci sm, which heterogeneous neighborhoods may foster more racial t ension (Kaufman, 2005).

PAGE 88

88 There has been disagreement as to the role of peer substance use on adolescent violence (Hawkins et al., 2000; Mattila, Par kkari, Rimpela, 2006). This study provides support for the argument that peer substance use indirectly effects violence, which is largely mediated through individual-level drug and alcohol use. However, this study is unable to account for the effect of peer delin quency, a historically potent predictor of individual violence (Hawkins et al., 2000; Hu izinga et al., 2003; Loeber et al., 2003). It is possible that the effect of peer delinquency is partially controlled under the group fighting construct, however, this relations hip provides a venue for future research on the peerlevel effects on individual delinquency. At the individual-level, this study found support for many risk and protective factors that have previously been identified in t he literature on violence. For instance, academic failure and dropout from school has been associated with violence (Hawkins et al., 2000). This study also found a signifi cant relationship between higher intentions of college attendance and lower odds of member ship in a violent trajectory group. However, although the effect remained in t he expected direction, this effect was no longer significant once baseline violence was added to the model. Also, Hawkins et al. (2000) found that delinquent peers and gang membership have been predictive of violent behavior. The results of this study did not support this fi nding, as group fighting was not a risk factor for membership in the escalator group. This may be because those in the escalator group have yet to be ex posed to this type of violent behavior at baseline. Instead, a time-varying effect may be present, in which group fighting at Wave II or Wave III would si gnificantly predict violence for the escalator group. This implies that adolescents who are categoriz ed as escalators may not be exposed to

PAGE 89

89 group fighting until a different developmental stage, while desistors are exposed to group fighting earlier in life. The results from this study did not ident ify patterns of predictors among escalators, a high-risk late-onset trajectory group. In the fully adjusted model, only baseline violence predicted membership in this high-ri sk group. The effect of peer alcohol use on violence was mediated through individual-level variables; specifically, individual-level substance use. This finding highlights the need for future research on this group of escalators, as unique and early risk factors ma y be present. In one study that identified this late-onset escalator group (Zara & Farrington, 2009), a variety of psychological predictors were identified, including high an xiety, low IQ, delinquent friends, having few friends early in life, and late onset of sexual intercourse. These results indicate that childhood risk factors may predict this late-onset group of violent young adults, and more research on this unique group is neces sary to further understand the etiology of late-onset escalation. The number of risk factors for violent behavior that are pr esent at baseline highlights the need for early violence preven tion programming. Specifically, higher levels of racial heterogeneity in the nei ghborhood, peer alcohol use, peer marijuana use, individual-level alcohol and marijuana use, other drug use, lower academic achievement, greater desire to leave home, depression, and group fighting were significantly more prevalent among those who were violent at baseline compared to those were not violent at baseline. These findings highlight the early risk factors that are present prior to age 15 that may serve as targets for large-scale violence prevention programming.

PAGE 90

90 This study had several limitations. Firs t, this study was unable to account for some of the variables that ar e important in predicting violence, such as peer violence, IQ, and psychological disorders. Second, latent-group based trajectory modeling provides an estimation of the type and number of groups in the data, and this process is exploratory in nature. Despit e the exploratory natur e of trajectory estimation, the results of this study were consistent with the number and shape of trajectory groups from other studies (Piquero, 2008; Zara & Farrington, 2009) Finally, risk factors were analyzed at multiple levels; however, hierarchical linear modeling (HLM) was not used due to the small sample sizes available in some of t he trajectory groups. However, because of the sampling design, all analyses accounted for the nesting of adolescents within schools, which may account for a portion of the variab ility in Census tract measures. Use of HLM to account for the nesting within Census bl ocks is a direction for future research. Despite these weaknesses, the current study had a number of strengths. First, data were derived from a longi tudinal, nationally representative sample of adolescents followed into young adulthood. This samplin g design allows generalization to a national sample of adolescents across the United States. Second, although many studies have analyzed the multiple domains of risk and prot ective factors for violent behavior, few have assessed the degree to which the effect s of contextual variables are mediated through more proximal variables at the indivi dual-level. The mediat ed effects allow this study to acknowledge that cont extual variables are important in predicting violence even though their effects are mitigated using multiv ariate regression models. Finally, the trajectories estimated in this study are especially appropriate for studies of delinquency and violence, as patterns tend to change over time (Farrington, 1986; Piquero, 2008).

PAGE 91

91 In conclusion, the findings from this st udy indicate that the risk factors for membership in each of the three violence trajectory groups diffe r. Taken together, these findings have significant implications for violence prevention. First, social influences, such as exposure to peers who use alcohol or marijuana, and communitylevel exposure to alcohol influence adole scents risk for violent behavior. Second, violent behavior begins even before age 15 in t he general population, indicating that the current prevention programming occurs too late. Prevention programming should begin early in elementary school settings to pr event initiation of aggressive behavior. Disclosure This study was supported by Awar d Numbers K01 AA017480 (PI: Mildred Maldonado-Molina), R01 AA013458 (PI: Kelli A. Komro) from the Na tional Institute on Alcohol Abuse and Alcoholism and the National In stitute for Minority Health and Health Disparities, and from the Institute for Child Heal th Policy at the University of Florida. The content is solely the responsibilit y of the authors and does not necessarily represent the official views of the National Institute on Alcohol Abuse and Alcoholism of the National Institute of Health. This research uses data from Add Health a program project directed by Kathleen Mullan Harris and designed by J. Richard Udry Peter S. Bearman, and Kathleen Mullan Harris at the University of North Caro lina at Chapel Hill, and funded by grant P01HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative f unding from 23 other federal agencies and foundations. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Info rmation on how to obtain the Add Health data

PAGE 92

92 files is available on the Add Health w ebsite (http://www.cpc.unc.edu/addhealth). No direct support was received from gr ant P01-HD31921 for this analysis.

PAGE 93

93 Table 3-1. Description of sample, Add Health. N=9421. Variable % Trajectory Groups Violence, Wave II (Mean, SE) 0.52(0.04) Violence, Wave III (Mean, SE) 0.28(0.03) Violence, Wave IV (Mean, SE) 0.91(0.04) Community-level Racial Dispersion a 0.31(0.27) % Poverty a 0.13(0.03) Urban area 61.8 Parental and Peer Influences Parental Involvement a 5.81(3.4) Parental alcohol use 65.7 One or more peers use alcohol 57.6 One or more peers use marijuana 36.4 Individual-level Risk Factors Ever alcohol use 58.2 Ever marijuana use 30.5 Ever used other drugs 17.7 Depression b 42.3 Intend to go to college 72.3 Desire to leave home 38.2 Speaking Spanish at home 6.14 Violence Group fighting in past year 22.1 Baseline violence 22.0 Demographics Gender (Male) 42.8 Age at Baseline 15.4(1.60) White 64.5 African-American or Black 23.6 Hispanic or Latino 14.8 Asian or Pacific Islander 5.8 Native American 4.1 Other Race 1.1 a Mean(SE) are reported. b Depression was measured as feeling sad or depressed one or more times in the past month.

PAGE 94

94 Figure 3-1. Trajectories of violence over time, Add Health, full sample. # of EventsMean Age

PAGE 95

95 Table 3-2. Description of violence (mean,SD) ov er time by trajectory group, Add Health. Trajectory Group NonViolent N=7193 Desistors N=1186 Escalators N=1042 Mean Violence (Age 16) 0 3.18(3.29)*** 0.20(0.82)** Mean Violence (Age 21) 0 1.53(2.28)*** 0.28(1.11)*** Mean Violence (Age 26) 0 0.35(1.20)*** 7.15(1.90)*** Note: Mean violence for desistors and escalators is being compared to the mean of the non-violent group. **p<0.01 ***p<0.001

PAGE 96

96 Table 3-3. Bivariate effects between risk/protec tive factors and trajectories of violence. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 2.18*** 1.42-3.36 2.01** 1.32-3.07 Poverty 16.09 0.47-551. 6014.44 0.57-366.05 Urban Area 1.03 0.791.33 0.87 0.70-1.10 Parental and Peer Influences Parental Involvement 0. 96 0.93-1.00 0.98 0.95-1.01 Parental alcohol use 0. 84 0.68-1.03 1.02 0.81-1.28 Peer alcohol use 1.93*** 1.52-2.44 1.34** 1.08-1.65 Peer marijuana use 2.21*** 1.78-2.73 1.28* 1.01-1.62 Individual-level Risk Factors Alcohol use 2.33*** 1. 92-2.83 1.11 0.91-1.36 Marijuana use 2.52*** 2. 07-3.06 1.28 0.99-1.67 Other drug use 2.27*** 1.76-2.94 0.95 0.61-1.46 Academic achievement 0.80 *** 0.74-0.86 0.91 0.83-1.01 Desire to leave home 1. 33*** 1.16-1.53 1.16 0.99-1.34 Depression 1.10 0.871.39 1.02 0.83-1.26 Violence Group fighting 2.54*** 2. 22-2.92 1.32** 1.07-1.63 Baseline violence 11.96*** 4.93-7.90 2.74** 1.30-2.32 Demographics Male 4.03*** 3.30-4.90 1.42*** 1.14-1.78 Age 0.99 0.93-1.05 1.05 0.97-1.13 White 0.57*** 0.47-0. 73 0.70** 0.56-0.88 African-American or Black 1. 84*** 1.40-2.41 1.63*** 1.28-2.07 Hispanic or Latino 1.39 ** 1.15-1.67 1.02 0.75-1.38 Asian or Pacific Islander 0.56* 0.32-0.98 1.19 0.79-1.78 Native American 2.09 ** 1.23-3.55 0.87 0.46-1.64 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 **p<0.01 ***p<0.001

PAGE 97

97 Table 3-4. Community, family, and peer effects on trajectories of violence. Trajectory Group Desistors Escalators OR 95% CI OR Community-level Racial Dispersion 1.60 0.94-2.59 1.73* 1.04-2.91 Parental and Peer Influences Parental Involvement 1. 00 0.96-1.04 0.99 0.97-1.03 Parental alcohol use Peer alcohol use 1.50**1.26-2.24 1.31* 1.02-1.65 Peer marijuana use 1.89*** 1.50-2.64 1.10 0.94-1.42 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for demographic variables. *p<0.05 **p<0.01 ***p<0.001

PAGE 98

98 Table 3-5. Effects of multiple doma ins on trajectories of violence. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.39 0.78-2.30 1.70 0.99-2.93 Parental and Peer Influences Parental Involvement 1. 01 0.98-1.06 1.00 0.97-1.04 Peer alcohol use 1.13 0.83-1.54 1.23 0.94-1.59 Peer marijuana use 1.24 0.91-1.74 1.05 0.79-1.38 Individual-level Risk Factors Alcohol use 1.68***1.29-2.33 0.97 0.76-1.24 Marijuana use 1.25 1.03-1.72 1.10 0.79-1.51 Other drug use 1.27**1. 09-1.47 0.93 0.71-1.22 Depression Academic achievement 0.93 0.84-1.02 0.93 0.84-1.04 Desire to leave home 1.12 0.99-1.39 1.10 0.92-1.31 Violence Group fighting 2.02***1.64-2.31 1.17 0.95-1.45 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for demographic variables. **p<0.01 ***p<0.001

PAGE 99

99 Table 3-6. Effects of multiple domains on traj ectories of violence, adjusted for baseline. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.31 0.72-2.231.68 0.98-2.88 Parental and Peer Influences Parental Involvement 1. 01 0.97-1.051.00 0.89-1.04 Peer alcohol use 0.97 0.76-1.481.22 0.93-1.58 Peer marijuana use 1.26 0.91-1.821.05 0.79-1.38 Individual-level Risk Factors Alcohol use 1.55** 1. 21-2.190.97 0.75-1.23 Marijuana use 1.12 0. 92-1.551.07 0.78-1.48 Other drug use 1.21* 1.02-1.410.92 0.71-1.20 Academic achievement 0. 95 0.86-1.050.93 0.85-1.04 Desire to leave home 1. 09 0.96-1.361.10 0.91-1.31 Violence Group fighting 1.69*** 1. 36-1.951.10 0.89-1.37 Baseline violence 3.08*** 1.27-4.091.39* 1.05-1.86 Note: The Non-Violent trajec tory group serves as the reference category. All analyses are controlling for demographic variables. *p<0.05 **p<0.01 ***p<0.001

PAGE 100

100 Figure 3-2. Test of the mediation pathway (bas ed on bivariate analysis). Alcohol, Marijuana, and other drug use Desire to leave home Academic Achievement Group fighting Baseline Violence Individual-Level Risks Social/Contextual Risks Racial dispersion Parental Involvement Peer alcohol and marijuana use Violent Trajectory Membership a b c

PAGE 101

101 Table 3-7. Regression models testing t he association between community-level and parent and peer level variables on indivi dual-level risk and protective factors. Individual-level Variables Alcohol Use Marijuana Use Other Drug Use Academic Achievement Desire to leave home Group Fighting Baseline Violence OR OR OR OR OR OR OR Community-level variables Racial Dispersion 1.02 1.19 1.15 0.83 0.81 1.36 1.41 Parental and Peer Influences Parental Involvement 0.97** 0.94*** 0.94 1.15*** 0.92*** 0.98 0.99 Peer alcohol use 6.77*** 5.45*** 5.28 0.74** 2.05*** 2.96***2.53*** Peer marijuana use 4.43*** 11.60*** 8.07 0.65*** 2.04*** 2.81***2.45*** Note: All analyses are controlling for demographic variables. **p<0.01 ***p<0.001

PAGE 102

102Table 3-8. Mediated effect of parentand peer-l evel variables on violence trajectories. Mediator Indirect Effect ( ab) z SE Percent Mediated Parental Involvement Alcohol Use 0.19913.01*** 0.01 10.2 Marijuana Use 0.20164.02*** 0.003 10.2 Other Drug Use 0.1418.21*** 0.02 7.2 Academic Achievement 0.79810.93*** 0.007 4.1 Desire to Leave Home 0.21715.98*** 0.01 11.0 Group Fighting 0.23311.36*** 0.02 11.9 Baseline Violence 0.0109.25*** 0.03 0.5 Total 55.11 Peer alcohol use Alcohol Use 0.3018.57*** 0.04 16.2 Marijuana Use 0.2917.41*** 0.04 15.7 Other Drug Use 0.2095.41*** 0.04 11.3 Academic Achievement 0.0367.95*** 0.005 1.9 Desire to Leave Home 0.20 9.60*** 0.02 1.1 Group Fighting 0.2667.36*** 0.04 14.3 Baseline Violence 0.2907.28*** 0.04 15.6 Total 76.1 Peer marijuana use Alcohol Use 0.0547.89*** 0.03 3.1 Marijuana Use 0.3218.03*** 0.04 18.3 Other Drug Use 0.2088.44*** 0.02 11.9 Academic Achievement 0.0327.89*** 0.004 1.8 Desire to Leave Home 0.19410.60*** 0.02 11.0 Group Fighting 0.2487.77*** 0.03 14.1 Baseline Violence 0.2716.96*** 0.04 15.5 Total 75.6 Notes: All models are adjusted for demographic variables. (a): These mediated effects were generated in acco rdance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhsh, & Veblen-Mortenson (2001). The perce nt mediation was generated using the formula: [(a*b/(a*b + c)] (MacKinnon, 2008). (b): Indirect effects are not direct ly comparable across variables. Percent mediation is comparable across variables and groups of variables. ***p<0.001

PAGE 103

103 Table 3-9. Post-hoc description (means and percentages) of adolescents who were violent at Wave I. Violence at Baseline Violent Non-Violent p Community-level Racial Dispersion (Mean) 0.29** 0.25 0.009 Poverty (Mean) 0.13 0.13 0.683 Urban Area 0.12 0.39 0.318 Parental and Peer Influences Parental Involvement (Mean) 5.57 5.85 0.056 Parental alcohol use 0.59 0.58 0.377 Peer alcohol use 0.71*** 0.53 <0.001 Peer marijuana use 0.52*** 0.31 <0.001 Individual-level Risk Factors Alcohol use 0.74*** 0.53 <0.001 Marijuana use 0.49*** 0.25 <0.001 Other drug use 0.24*** 0.10 <0.001 Academic achievement 0.60*** 0.74 <0.001 Desire to leave home 0.46*** 0.33 <0.001 Depression 0.47*** 0.40 <0.001 Violence Group fighting 0.50*** 0.15 <0.001 Demographics Male 0.68*** 0.42 <0.001 Age (Mean) 15.27 15.16 0.117 White 0.66*** 0.75 <0.001 African-American or Black 0.25*** 0.17 <0.001 Hispanic or Latino 0.14 0.11 0.025 Asian or Pacific Islander 0.06** 0.03 0.003 Native American 0.02** 0.03 0.005 Notes: Participants were considered violen t at baseline if they reported any of the violence items that were used to estima te violence trajectories: shot or stabbed someone, used knife or gun in a fight, or hurt someone badly enough to need care from a doctor or nurse. The Wave IV weight ing variable used for these analysis. **p<0.01 ***p<0.001

PAGE 104

104 CHAPTER 4 GENDER AND RACIAL/ETHNIC DIFFERENCES IN RISK FACTORS ASSOCIATED WITH TRAJECTORIES OF VIOL ENT DELINQUENCY IN A NATIONALLY REPRESENTATIVE, LONGITUDINAL SAMPLE Racial/ethnic and gender disparities are present in violent offending, and evidence suggests that certain groups (e.g., Afric an-Americans, Hispanics, and males) are at greater risk for violence compared to Whites and females. The purpose of this study is to estimate trajectories of serious violence using a longitudinal sample of adolescents by race/ethnic and gender subgroups, consi dering multiple domains of risk and protective factors in differentiating pr ofiles of delinquent and vi olent behaviors. Methods. Participants included a nationally representative sample of 9421 adolescents followed from ages 15 through 26. Trajectori es of violence were estimated for each subgroup (e.g., White males and females, African-American males and females, Hispanic males and females, Native Americans, and Asians), and participants were assigned to each trajectory group using latent trajectory modeling. Multinomial logistic regression procedures were used to evaluate t he effect of multiple domains of risk and protective factors in stages (e.g., community-level, parentand peer-level, and individual-level) to understand the predictors of membership in high-violence trajectory groups. Mediation analyses were conducted to further evaluate the direct and indirect effect of community-, parental and peer-level variables on violent trajectories. Results. Three groups of violence trajectories were identified for all subgroups : 1) Non-Violent; 2) Escalators; and 3) Desistors. Group fighting significantly predicted violence above and beyond baseline violent behavior for Asian desistors, White male and female desistors, African-American male and female desistors, and African-American male escalators. A number of racial/ethnic and gender differences in the multiple domains of risk and

PAGE 105

105 protective factors emerged. Conclusion. Group fighting is a significant predictor of violence beyond baseline violen ce for certain groups of viol ent adolescents. Contextual variables (urban neighborhood residence, racial dispersion, peer alcohol and marijuana use, and parental alcohol use) appear to in crease risk for violence differentially by subgroup. Background Each year, nearly 700, 000 adolescents and y oung adults (10-24) are treated in the emergency room for injuries re lated to violent activity (CDC, 2009). Evidence suggests that adolescents who engage in delinquent behavior are more likely to engage in other high-risk activities (e.g., alcohol and ot her drug use, dropping out of school, gun ownership, gang membership, risky sexual activity and familial independence) (Thornberry, Huizinga, & Loeber, 1995; CDC, 2009; 2010) and increase their risk of health-related consequences (including serious injury and death) (Conseur, Rivara, & Emanuel, 1997; Farrington & Loeber, 2000). The evidence is clear that individualand family-level characteristics increase the risk for violent behavior. For example, neurological deficiencies and cognitive impairments (Moffitt et al., 2001), low IQ, hyperact ivity, difficulty concentrating at school, beliefs and attitudes favorable to violence antisocial behavior, and impulsivity have been consistently associated with violent behavio r at the individual-level (Hawkins et al., 2000; Howell, 2009). At the fam ily level, parental criminal behavior, child maltreatment, low levels of parental invo lvement, parental attitudes fa vorable to violence and drug/alcohol use, and separatio n of the parent and child have been identified as risk factors in a recent meta-analysis of longit udinal studies of risk factors for violence

PAGE 106

106 (Hawkins et al., 2000). Each of these factors has been consistently associated with increases in violent activity. Despite the strong evidence in support of some risk factors for violence, other behavioral risk factors within the family and p eer group are less studied. Hawkins et al. (2000) found that delinquent peers and gang membership have been predictive of violent behavior; however, the effect of peer and parental substance use is unclear. Academic failure and dropout has also been a ssociated with violence, but less drastic measures of academic success have not been ev aluated in the empirical literature on violence (Hawkins et al., 2000) Community-level influences such as availability of firearms, exposure to viol ence, and exposure to raci sm in the neighborhood have consistently been linked to violent behavio r (Kaufman, 2005; Reingle, Jennings, Maldonado-Molina, & Canino, 2010). Finally, although many studies have analyzed the multiple domains of risk and protective fa ctors for violent behavior, few have assessed the degree to which contextual variables are m ediated by more proximal variables at the individual-level. Race/Ethnic Differences in Violence Racial differences in the prevalence of violence have been identified. For examp le, Williams et al. ( 2007) found that self-reported violence initiation rates were higher for African-Americans compared to Whites for majo r delinquency, violence, and juvenile justice system involvement. Additi onally, Williams et al. (2007) reported African-Americans higher rates of major del inquent and violent acts when compared to Whites. Additionally, McNulty & Bellair (2003) found that AfricanAmericans, Hispanics, and Native Americans have higher involvemen t in serious physical violence compared to White adolescents at ages 15 to 16.

PAGE 107

107 Evidence suggest that these racial differ ences may be partially attributable to differences in the presence of community-l evel risk factors, such as neighborhood poverty and exposure to guns, delinquent peer s, and violence (Sampson, Morenoff, & Roaudenbush, 2005; Farrington & Loeber, 2000). Fa mily and individual-level variables also contribute to violent behavior, incl uding acculturation level among Hispanics (having immigrated to the Unit ed States more recently is a protective factor from violence), parental marital status (liv ing with parents who are not married puts adolescents at risk for violence), lower verbal reading and writing ability, shorter length of residence in the neighborhood, and lower income, as low socioeconomic position has been shown to increase the risk of violence in itiation (Sampson et al ., 2005; Williams et al., 2007). These disparities may be a function of socioeconomic differences between racial/ethnic groups, including income, neighborhood structural characteristics, oppression, cumulative disadvantage within t he family, employment, and education; as well as racial discrimination and bias (Centerwall, 1995; Williams, 1999; Peterson & Krivo, 2005). These findings support t he hypothesis that the racial and ethnic differences in violence may be at least parti ally attributable to neighborhood-, family-, and individual-level socioeconomic variables. Although the risk factors above have expl ained some of the racial/ethnic differences in violent behavior, further resear ch is necessary to more clearly delineate racial and ethnic differences (which hav e been accepted as a proxy for larger socioeconomic differences, discrimination, and cumulative disadvantage) in delinquency and violence between these groups. The research on race and ethnic differences on violence has not consistently attributed violent behavior to differences in race/ethnicity

PAGE 108

108 or socioeconomic position. For example, Blum and colleagues (2000) found that gender, race, ethnicity, income, and family stru cture explained no more than ten percent of the variance in violent behavior. This fi nding indicates that the differential risk factors for violence by racial/ethnic gr oup is more complex than race or socioeconomic position. Therefore, multiple domains of risk factors that differ by race/ethnicity must be further studied to better understand how these factor s contribute to violent behavior. Criminological theory provides insight as to the rationale behind racial and ethnic differences in the prevalence of violent offendi ng. First, social learning theory posits that individuals learn to engage in criminal behavior by observing those around them (Akers, 1973). This theoretical framework is compris ed of four central components: excess of definitions favorable to crim inality, association with devi ant peers, reinforcement of criminal behavior, and imitation. When these elements are combined, an individual is more likely to engage in deviant or criminal behavior (Akers, 1973). Using this theory, violent behavior would be more prevalent among racial/ethnic minoriti es if they have more violent peers who teach them to participat e in violence, and this violent behavior is reinforced. Research supports this hypothesis, as African-Americans and Hispanics are more likely to be involved with gangs (Mc Nulty & Bellair, 2003) and have delinquent peers (Stewart, Simons, & Conger, 2002) compared to Whites. Shaw and McKay maintain that the charac teristics of the co mmunity (rather than the individuals within the comm unity) facilitate crime. For example, transient individuals are unlikely to watch over their neighbors property, or even become acquainted with their neighbors. Therefore, a neighbor could never know if someone is stealing a car out of the driveway, as they do not know w ho resides in that parti cular home. Suburban

PAGE 109

109 areas have these types of neighborly ties and they would recognize when someone does not belong. Using the framework provided by Social Disorganization theory, the lack of neighborly recognition is one of the reasons why crime is higher in transient, disorganized communities. Sampson, Raudenbush, and Earls (1997) furt her elaborate on this aspect of their theory in defining the collective efficacy of communities. Collective efficacy is the neighborhoods ability to come together to reduce criminal or disorderly behavior. Cohesive, suburban neighborhoods are very lik ely to do this: When there is a problem within the community, member s of the neighborhood organize themselves and remedy the issue. Urban, disadvant aged, disorganized neighborhoods are not likely to have such a mechanism in place to filter out cr iminal behavior among residents. This lack of social control in a community allows crime to multiply. This theory has substantial implications for racial/ethnic differences in violence, as minorities are more likely to reside in these urban, at-risk comm unities (Sampson & Wilson, 2005). Gender Differences in Violence There are also substantial gender differences in the prevalence of violent behavior. There is evidence that rates of violence among girls have been continually increasing since 1980 (Snyder, 2003). During this time, aggravated and simple assaults among girls have increased by 113% and 257%, re spectively (Snyder, 2003). Although these increases seem large, the types of a ssault committed by girls differ from the types of assault perpetrated by boys. For example, girls are more likely to fight with parents and peers, rather than use w eapons or commit sexual assault (Chesney-Lind & Shelden, 1998).

PAGE 110

110 Gender differences may be partially attributable to differential risk factor exposure by gender group. For instance, Moffitt et al. (2001) found that boys are exposed to more risk factors for violent beha vior than girls, thus increasi ng their likelihood of violent offending among boys compared to girls. Diffe rences in deviant peer associations may also influence the differences in risk between girls and boys. Piquero, Gover, MacDonald, & Piquero (2005) found that association with deviant peers was significantly related to violence and delinqu ency, but this was only true among boys. This may reflect differences in the social learning effect discussed above, as boys learn to participate in violent behavior from their peer group (Akers, 1973). In addition to the effects of multip le risk factors and deviant peer group associations, differences in socializat ion may be an explanation for differential participation in violence between males and females. In a meta-analytic evaluation of differences in parental socialization of their children, Lytton & Romney (1991) found that parents were more likely to discourage aggre ssive behavior in their female children compared to male children. This small but insignificant difference may indicate that socialization alone is not the reason for differential aggression between the gender groups. Instead, a small degree of social ization differences may have cumulative effects when the adolescents peer group and sc hool environment is considered (Lytton & Romney, 1991). Together, these multiple ri sk factors may be partially responsible for observed gender differences in violent behavior. To further investigate the factors driving gender and raci al/ethnic differences in delinquency over the life-course (and the risk fa ctors for violence over time), this study stratified a longitudinal, nationally representative sample of adolescents and young

PAGE 111

111 adults by genderand race/ethnicity. Trajectori es were estimated by racial/ethnic and gender subgroups, and multiple domains of risk and protective factors were identified for each racial/ethnic and gender group individually This will contribute substantially to the literature on disparities and etiology of violence in the general population. Specifically, I hypothesize that: 1) males are at higher risk for violence between late adolescence and early adulthood (ages 1626), and these differences may be attributable to differences in group fighti ng at baseline (age 15); 2) racial/ethnic minorities (African-Americans and Hispanics) are more likel y to be involved in high levels of serious violence compared to Whites; and 3) differences in violence by racial/ethnic and gender subgroups are attributable to differences in multiple domains of risk factors. Method Design The National Longitudinal Study of Adole scent Health (Add Health) is a schoolbased panel study conducted from 1994 (Wav e I) through 2008 (Wave IV), when participant ages ranged from 11-32 (Chantal a & Tabor, 1999). The data collect ion for this survey was designed to explore the effe cts of multiple domains of risk factors on adolescents health behaviors. In Wave I, 80 communities were selected to ensure demographic representativeness (e thnic composition, region of the country, urbanicity, school size, and school type) of students in t he United States. Schools (n = 132) were eligible if they enrolled more than thir ty students and had an eleventh grade. All students who were enrolled in the school an d were present on the survey day were eligible for participation in the study. Approximat ely 200 students were randomly

PAGE 112

112 selected from strata of grade and sex, re sulting in a final cohort sample of 9421 adolescents. Data Collection At Wave I, the baseline sa mple of 20,745 participated in a oneto two-hour inhome interview. This interview c ollect ed data on a variety of topics, including demographics, peer networks, health, employment, substance use, criminal and delinquent behavior, sexual behavior, and ro mantic partnerships. All data were recorded on laptop computers, and the participant listened to the questions and responded on the laptop themselves to maximize validity for sensitive items. At this time, parents were also surveyed to evaluate their health behaviors, education, neighborhood characteristi cs, and heritable conditions (Chantala & Tabor, 1999). Wave II data collection included another in -home interview one year later, between April and August, 1996. Interview items were similar, but additional nutrition and sun exposure items were added. The response rate for eligible participants in Wave II was 88.2%. All participants who were interviewed in Wave I were elig ible, excluding those who moved out of the country. Jailed adole scents were interviewed when possible. Wave III included data collected between 2001-2002, when participants were ages 1826. The survey instrument changed to reflect the more influential romantic relationships, criminal history, and joband college-related influences rather than school-based influences on health behaviors. In 2007-2008, Wave IV of data collection co nsisted of a 90-minute questionnaire to be completed by respondents via laptop. Physical and biological markers were collected, and questionnaires included a greater number of financial, marital/cohabitation, medications, physical health, and childhood maltreatment items.

PAGE 113

113 The response rate for this sample was 80.3% of eligible participants. Participants were eligible for all waves of data collecti on if they participated in Wave I. The sample used in this study includes par ticipants who were present at all four waves of data collection. The sample used in this study includes participants who were present at all four waves of data collection. This cohort was 42.8% male, 64.5% White, 23.6% African-American, 14.8% Hispanic, 5.8% Asian or Pacific Islander, 4.1% American Indian, and 1.1% Ot her. The average age at Wave I was 15.4 (sd = 1.6), 16.3 (sd = 1.6) at Wave II, 21.7 (sd = 1.6) at Wave III and 26.5 (sd = 1.8) at Wave IV. More than twenty percent of t he sample reported participating in a group fight at Wave I (22.1%). Mean violence at Wave II wa s 0.52(SE = 0.04), 0.28(SE= 0.03) at Wave III, and 0.91(SE=0.04) at Wave IV The prevalence of the independent and dependent variables are detailed in Table 4-1. Measures Violent Delinquency Violenc e was measured using three items that were measured across each of the f our waves of data collection: In the past 12 months, have you 1) hurt someone badly enough that he or she needed care from a doctor or nurse?; 2) pulled a knife or gun on someone?; and 3) shot or stabbed someone? At each wave, a value from 0-12 was assigned to each participant, where a value of , , or was assigned for each of these violent acts in which the individual has participated in during the past year. A zero was assigned for each item if the participa nt did not report the behavior. A two was assigned if the adol escent reported hurting someone badly enough to need care from a doctor or nurse one to three times in the past year. A four was assigned for each of the following occu rrences: 1) shooting or stabbing someone; 2) pulling a knife or gun on someone; or 3) hurting someone badly enough to need care

PAGE 114

114 from a doctor or nurse four or more times in the past year. These values were used to create trajectories of delin quency over Waves II-IV. Group Fighting Group fighting was measured using the variable, In the past 12 months, how often did you take place in a physical fight where a group of your friends was against another group?. Responses to this item include: =Never, =One or two times, =Three to four times, and =5 or more times. These responses were dichotomized into =never group fighting and =group fighting in the past year. Racial Dispersion Racial dispersion is a measure (ranging from 0 to 1) of the racial heterogeneity in a neighborhood. Dispers ion is equal to zero when all census tract members are members of the same racial group, a nd equal to one when residents are equally distributed among White, AfricanAmerican, Asian, Native American, and Other races. Poverty Poverty was measured using the percentage of families in the respondents census tract whose income was at or below the poverty level. Racial dispersion was measured using the concent ration of racial/et hnic homogeneity within the neighborhood, so higher prop ortions would indicate greater concentration of one race/ethnicity within a neighborhood. These items were incorporated into the analysis in accordance with Shaw and McKays (19 42) theory of Social Disorganization. Urban Neighborhood. All addresses were geocoded at the time of the interview, and these addresses were linked to U.S. Census data (1990) to determine the urbanicity of the residence. Addresses were considered complet ely urban or not completely urban.

PAGE 115

115 Risk factors for violence Parental Involvement Parental influence and invo lvement was measured using a scale of twenty items (10 for maternal involvement, 10 meas uring paternal involvement) (Prado et al., 2009). Each indi vidual item was dichotomized, and the scale is the sum of all twenty items (range: 0-20). The ten it ems which comprised the scale included whether or not the respondent r eported participating in the follo wing activities with their mother and/or father in the pas t four weeks: 1) going shoppi ng; 2) playing a sport; 3) attending a religious or church-related even t; 4) talking about someone they are dating or a party they attended; 5) attending a movie, play, concert, or sporting event; 6) talked about a personal problem they were havi ng; 7) had a serious argument about their behavior; 8) talked about work or grades; 9) worked on a project for school; and 10) talked about other things they are doing in school. Cronbachs coefficient alpha for this scale was 0.74. This scale was included as a covariate because evidence suggests that parenting variables (e.g., monitoring, invo lvement) are related to violence (Park, Morash, & Stevens, 2010). Parental Alcohol Use. At the Wave I survey, paren ts of surveyed adolescents were asked, How often do you drink alcoho l?. Response options included, Never, Once a month or less, Two or three days a month Once or twice a week, Three to five days a week, and Nearly every day. Responses were dichotomized into parents use alcohol and parents do not use alcohol based upo n the distribution of the responses. Peer Alcohol Use Peer alcohol use was measured using one item: Of your three best friends, how many drink alcohol at least once a month? Respondents who

PAGE 116

116 reported having one or more friends who use al cohol monthly were coded as . These items were included because literature suggests that indi viduals who have peers who use alcohol (Herrenkohl et al., 2007; Kunt sche, Gossrau-Breen, & Gmel, 2009; Leech, Day, Richardson, & Goldschmidt, 2003) are mo re likely to engage in violent behavior. Peer Marijuana Use. Respondents were asked, Of your three best friends, how many use marijuana at least once a mont h? Respondents who reported having one or more friends who use marijuana monthly were coded as . These items were included because literature suggests that indivi duals who have peers who use marijuana (Herrenkohl et al., 2007; Leech, Day, Richar dson, & Goldschmidt, 2003) are more likely to engage in violent behavior. Depression. This mental health status variable was measured with one item, How often in the past week have you felt depressed?. Values for this variable were dichotomized so that 1 = One or more times and 0 = in stances of depression in the past week. Depression was included as a covariate because higher levels of depression have been associated with violence (Elbogen & Johnson, 2009; Senn, Carey, & Vanable, 2010; Thurnherr, Bechtold Michaud, Akre, & Suris, 2008) and other risk behaviors (Latzman & Swisher, 2005; Senn, Carey, & Vanable, 2010). Academic Achievement Academic performance was measured using the variable, On a scale of 1 to 5, where 1 is low and 5 is high, how likely is it that you will go to college?. This item was included as a covariate because academic achievement and IQ have been associated with increased risk of violence (Herrenkohl, McMorris, Catalano, Abbott, Hemphill, & Toumbour ou, 2007; Leech, Day, Richardson, & Goldschmidt, 2003).

PAGE 117

117 Alcohol Use Lifetime alcohol use was evaluat ed using the item, Have you had a drink of beer, wine, or liquornot just a si p or a taste of someone elses drinkmore than 2 or 3 times in your life?. Those who responded affirmatively to this item were categorized as Alcohol Users. Marijuana and other drug use Marijuana use was measured using the item, During your life, how many times hav e you used marijuana? Responses were categorized into users and non-users. Other drug use was created using the selfreported number of times the re spondent used cocaine, inhalants or other drugs in their lifetime. If any of these drugs were used, respondents were categorized as users. These items were included because evidence suggests that the use of marijuana and other drugs (Boles & Miotto, 2003; Dhungana, 2009; Herrenkohl et al., 2007) increases the risk of violent behavior. Desire to leave home This variable was measur ed using the following item: How much do you feel that you want to leave home?. Respondents who reported very much or quite a bit were categorized as , others were categorized as . This variable was included because some ev idence suggests that a negative home environment increases the likelihood of viol ent delinquency (Ou & Reynolds, 2010). Spanish at Home. For Hispanic males and females only, the variable Speaking Spanish at home was be added as an individual -level measure of acculturation. Specifically, all adolescents were asked, What language is usually spoken at home?. Hispanic adolescents who r eported Spanish were coded as Speaking Spanish at Home. Those who reported English were the referent for this category. Those reporting any other language spoken were excluded (n=23) from the analysis.

PAGE 118

118 Generational Status. Generational status has been identified as an important risk factor for multiple health behaviors am ong Hispanics, as the more acculturated adolescents become, the more risky behav iors they tend to engage in (MaldonadoMolina, Reingle, Jennings, & Prado, 2011; Maldonado-Molina, Reingle, Tobler, Jennings, & Komro, 2010; Jennings, Maldonado -Molina, Piquero, & Canino, 2010). In this study, generational status was derived from two variables. Both adolescents and their parents were asked, Wer e you born in the Un ited States?. If neither a parent nor the child reported being born in the United States, the adolesce nt was considered a First Generation Immigrant . If the parent report ed being foreign-born and the adolescent reported being born in the United States, the adolescent was considered Second Generation US-Born. If one or more parents were born in the United States, and the adolescent reported being born in the United States, the adolescent was considered Third Generation US-Born and Beyond. Analytical Methods Group-Based Trajectory Modeling Trajectory groups were fitted to the dat a using group-based trajectory modeling (Nagin & Land, 1993; Nagin, 2005). This method of analysis grouped individuals together based upon common attributes (e.g., le vels of violence over time). This approach is appropriate in this situation bec aus e violence varies ov er time (Farrington, 1986), and individuals with different levels of violence may be substantially different from each other. Grouping participants with heterogeneous levels of violent behavior together and then attempting to predict violence may dilute the effect of risk or protective factors (Nagin, 2005).

PAGE 119

119 Group-based trajectory models are finite mixture models, which use singleand multiple-group models structures (Nagin, 2005) Finite mixture models (also known as latent class models) represent the heter ogeneity in a finite number on unmeasured (latent) classes. The trajectory groups that are created usi ng these analyses are derived from maximum likelihood estimation. In this case, violence data follow a Poisson distribution with a large number of non-violent events (zer o violent events). Therefore, a zero-inflated poisson (ZIP) di stribution was be spec ified in the model (Jones, Nagin, & Roeder, 2001). Models were tested until the most parsi monious number of trajectory groups maximizes the Bayesian Informati on Criterion (BIC). The BIC refers to: BIC = log(L) 0.5 klog(N), where log likelihood at the maximum likelihood estimate is subtracted from half the number of parameters mult iplied by the log of the sample size. The trajectories were descriptive in nature, and quadratic, cubic and linear models was be tested to correctly depict the slopes represented in the data. SAS PROC TRAJ was used to estimate the trajectories ( SAS Institute, Cary, NC; Jones, Nagin, & Roeder, 2001). Individuals were classified into mutually exclusive trajectory groups using the maximum probability procedure (Nagin, 2005, 2001). In other words, participants were assigned to groups in which they have the greatest probab ility of membership (e.g., greater than .80). The modeling st rategy estimated poste rior probabilities of assignment to each group, and individuals were assigned to the group with the highest probability. This does not guar antee that all individuals will have a probability of or membership in a latent group, but the m ean assignment probability for each group is

PAGE 120

120 expected to be high (>.80; Nagin, 2005; 2001) These high assignment probabilities increases confidence in the validity of latent groups. This modeling approach has a number of strengths and weaknesses. Latent group-based modeling allows an estimation of patterns of violent behavior over the lifecourse. These summaries of violence over time provide more information than a traditional dichotomous, violent or non-violent outcome vari able. However, there is a possibility that groups that ar e not meaningful could emerge from the data (e.g., a latent class with 2% or less of observa tions categorized in that group) To avoid this situation, the BIC as well as a judgment of parsimony was used in t he modeling procedure. In addition, the shape of each trajec tory and the number of latent groups is sensitive to the size of the dataset, characteri stics of the sample, or lengt h of the follow-up (Eggleston, Laub, & Sampson, 2004). Specifically, Eggles ton et al. (2004) fou nd that the shape of each trajectory group remained relatively constant despite changes in the length of follow-up; a doubling of the follow-up time changed their conclusion in identifying the time of peak criminality and estimating in dividual-level group membership. These limitations must be considered when interpreting the results of these analyses. Latent group-based trajectory modeling has been used in several studies in estimating trajectories of violence among adolescents and young adults. Piquero (2008) reviewed 80 studies on trajectories of delinquency over the life-course. This study found evidence of the age-crime curve, as criminal behavior decreased over time. The majority of studies found betwe en three and five classes of delinquency, regardless of methodology and the sample (Piquero, 2008; Maldonado-Molina et al.,

PAGE 121

121 2009). Overall, these trajecto ry models have been used in pr evious studies to examine delinquency and violence over time. Three variables were used to construct tr ajectories of violence across three Waves (II-IV). Due to the severi ty of two violent behaviors, having shot or stabbed someone, and having pulled a knife and gun on someone were weighted as (if they have not reported such behavior) or (if they reported shooting or stabbing, or pulling a knife or gun). The third variable (hurting someone badly enough to require care from a doctor or nurse) was coded as if no instances of this behavior were reported, if one, two, or thee instances were reported, or if four or more we re reported. Responses ranged from 0 (non-violent in all cat egories) to 12 (Frequent violence in all three categories). All baseline covariates in the model was recoded as dichotomous (with the exception of parental involvement, poverty, racial dispersion, and age). The distribution of the parental involvement measure is normal, with a relatively high internal consistency ( = 0.74). Trajectory estimation was conducted usi ng the recoded violence measures from Waves II-IV for each racial/ethnic and gender subgroup. Due to sample size limitations, Native Americans and Asians were not be stratified by gender. Instead, risk and protective factors were assessed for those groups as a whole. Other Races were excluded from the analysis. Multinomial Logistic Regression Once trajectory groups have been specified, bivariate and multivariate multinomial logistic regr ession were used to estimate odds-ratios for risk and protective factors on membership in each trajectory. This model is an extension of multiple logistic

PAGE 122

122 regressions; however, the model is more appropriate in this situation because trajectory group membership is a nominal variable, and this procedure compares membership in each trajectory group to a reference category (e.g., low-level violence). Under this model, each ( g 1) odds-ratios were generated (Hedeker, 2003). Multinomial logistic regression proc edures were adapted to account for the multilevel nature of the data (Hedecker, 2003). The Add Health sampling design selected schools as the primary sampli ng unit, and individuals are nested within schools. Clustered robust standard errors were estimated to pro duce error estimates that take into account the autocorrelation due to the sampling design. Failure to account for the sampling deign would result in an inflated Type 1 error rate, artificially increasing the precision of the effects (Twisk, 2006). The adapted multinomial model does not assume that observations are inde pendent; therefore, it is appropriate for longitudinal and clustered designs (Hede cker, 2003). STATA 11 software (College Station, TX) was used to conduct all multinomial logistic regression analyses. The first stage of model selection involved a bivariate test of the association of each predictor variable with the trajectory groups. All variabl es that are not marginally predictive of any dependent variable (traje ctory group) in bivariate analyses were removed from the multivariate model. After this initial model selection, distal variables (e.g., community-level) were added to the model first, followed by parental and peerlevel variables, and then individual-level characteristics. The final model assessed the influence of all risk and protective fact ors, accounting for baseline violence. Mediation Analysis Mediation analyses were conduc ted with any variables that are significantly associated with both the contextual variables and the outcome (violence). Covariance

PAGE 123

123 matrices were generated and regression es timates were standardized to obtain an overall mediated effect of each individual-lev el variable for each c ontextual variable. To test the significance of the mediator, the Sobel test was used to generate a z statistic and standard error (Baron & Kenny, 1986; Sobel, 1982; MacKinnon, Warsi, & Dwyer, 1995). The percent mediation for each mediator was calculated using the formula: ab/a1b1axbx +c In this formula, a represents the effect of the contextual variable on the hypothesized mediator, and b represents the effect of the mediator on the outcome variable (in this case, violence). C represents the direct effect of the contextual variable on the outcome. All of these standardized estimates (including all other variables in the model) were used to calculate the proportion of the variance in each contextual variable that is mediat ed by each proximal variable. These percentages were summed by contextual vari able to estimate the proportion of the contextual variables effect on violence that is medi ated through more proximal variables. Finally, when baseline violence was a signifi cant predictor of violent trajectory membership in the bivariate analyses, pos t hoc analysis was conducted to understand the risk factors associated with baseline violence. For this analysis, weighted proportions and means are calculated for each category of baseline violence (violent or non-violent) to understand the differences in risk factors for violence at age 15. Results Trajectories of Violence To determine the number and shape of trajecto ries of vi olence, latent group-based trajectory modeling was used. Three distinct classes were identified for each subgroup. The non-aggressive group had consistently reported no (or very low) participation in

PAGE 124

124 violence behavior across all waves. Desist ors participated in violence at earlier waves (II and III), but this violent behavior declined by Wave IV. Escalators had lower levels of violence at waves II and III, but their violent behavior increas ed drastically at Wave IV. For White males, 76.3% were non-violen t, 12.6% were desistors, and 11.1% were identified as escalators. Among White fema les, 82.6% were non-violent, 6.3% were identified as desistors, and 11.1% were considered escalators. Among AfricanAmerican males, 55.6% were non-violent, 25.8% were desistors, and 18.6% were escalators. The majority of African-Am erican females were non-violent (73.4%), followed by escalators (14.1%), and desisto rs (12.5%). Among Hispanic males, 59.6% were non-violent, 17.5% were desistors, and 21. 4% were escalators. Hispanic females were less violent overall, with 78.8% c onsidered non-violent, 10.3% desistors, and 10.9% escalators. Among Asians, 74.6% were non-violent, followed by 13.5% escalators, and 11.8% desistors. Finally, among Native Americans, 63.2% were nonviolent, 18.1% were desistors, and 18.6% were escalators. A graphical comparison of the proportion of violent versus non-viol ent adolescents by race/ethnic and gender group is displayed in Figure 4-1. For each race/gender subgroup, a three-group trajectory model showed the lowest AIC and BIC compared to a higher-class model. These AICs and BICs are detailed in Table 4-2. The mean posterior probabilitie s for all groups ranged from (0.75-0.92). Figure 4-2 displays the trajectories of violence from ages 16-26 for each subgroup. The average violence values for eac h trajectory group at each wa ve are detailed in Table 43. Mean violence for those in the non-vi olent trajectory group was zero for all subgroups across all waves.

PAGE 125

125 There were clear differences by race/ethni city in level of violence over time. Consistently among Whites, African-Americans, and Hispanics, males were overrepresented in each violent trajecto ry group compared to females. Effects of Risk and Protect ive Factors at Age 15 on Tr ajectories of Ph ysical Aggression White males Table 4-4 shows the bivariate relationshi p bet ween each risk or protective factor and violence trajectories. For desistors, peer alcohol use (OR = 2.30; 95% CI 1.723.07), peer marijuana use (OR = 1.94; 95% CI 1.41-2.67) indi vidual-level alcohol use (2.46; 95% CI 1.79-3.38), ma rijuana use (OR = 2.35; 95% CI 1.75-3.16), other drug use (OR = 2.56; 95% CI 1.70-3. 87), group fighting (OR = 2. 22; 95% CI 1.79-2.76), and baseline violence (OR = 4.11; 95% CI 3.02-5.60) were identifi ed as risk factors for being in the desistor trajectory group compared to the non-aggressive group. Protective factors for desistors included higher academic achievement (OR = 0.82; 95% CI 0.720.92), and age (OR = 0.93; 95% CI 0.86-1.01; p <0.10). For escalators racial dispersion (OR = 3.62; 95% CI 1.31-3.10) peer alcohol use (OR = 1. 64; 95% CI 1.10-2.44), peer marijuana use (OR = 1.62; 95% CI 1.01-2.59), individual-lev el alcohol use (OR = 1.48; 95% CI 1.01-2.15), marijuana use (OR = 1.70; 95% CI 1.072.68), other drug use (OR = 01.82; 95% CI 0.94-3.53; p <0.10), group fighting (OR = 1. 60; 95% CI 1.08-2.38), and baseline violence (OR = 2.36; 95% CI 1.51-3. 70) were identified as risk factors. Academic achievement (OR = 0.86; 95% CI 0.72-1.03; p <0.10) was identified as a marginally significant protective factor for membership in the escalator group compared to the non-violent group. Because poverty urban neighborhood, desire to leave home,

PAGE 126

126 and depression were not significant in predicting violence for neither desistors nor escalators in bivariate analyses, they were dropped from further mu ltivariate analyses. Table 4-5 shows the multivariate effect of community-, parentand peer-level variables on the desistor and escalator trajectory groups. Racial dispersion was a significant risk factor for e scalators only (OR = 3.38; 95% CI 1.21-5.49). Peer alcohol use was a risk factor for desistors (OR = 2.29; 95% CI 1.56-3.36); however, this effect was only marginally significant for esca lators (OR = 1.53; 95% CI 0.99-2.33; p <0.10). Peer marijuana use remained a risk factor fo r desistors only (OR = 1.58; 95% CI 1.082.31). When individual-level variables were added to the multivariate model (Table 4-6), racial dispersion among escalators was the only contextual variable that remained significant (OR = 3.11; 95% CI 1.19-8.19). Peer alcohol use remained a significant risk factor among desistors only (OR = 1.53; 95 % CI 1.03-2.28). A number of individuallevel variables were significant for desistors only. Specifically, alcohol use (OR = 1.65; 95% CI 1.08-2.52), other drug use (OR = 1.35; 95% CI 1.04-1. 76), and group fighting (OR = 1.82; 95% CI 1.41-2.34) were identified as risk fact ors for membership in the desistors trajectory group. No individ ual-level variables significantly predicted membership in the escalator trajectory group. The final model, adjusted for baseline violence in addition to all other risk and protective factors, is presented in Table 4-7. For escalators racial dispersion remained significant (OR = 3.07; 95% CI 1.17-8.05), a nd baseline violence was also identified as a risk factor (OR = 1.83; 95% CI 1.18-2.84). Peer alcoho l use remained marginally significant (OR = 1.47; 95% CI 0.98-2.21; p <0.10) among desistors. Also among

PAGE 127

127 desistors, alcohol use (OR = 1.55; 95% CI 1.02-2.35), other drug use (OR = 1.35; 95% CI 1.04-1.76), group fighting (OR = 1.50; 95% CI 1.16-1. 93) and baseline violence (OR = 2.66; 95% CI 1.88-3.75) pr edicted membership in the des istor group, compared to the non-aggressive trajectory group. The test of the mediation pathway (based on bivariate analysis) is presented in Figure 4-3. The significance of pathway a indicated that all variabl es met the criteria for inclusion in the mediation analysis (e.g., contextual variables must be significantly associated with at least one mediator (Tabl e 4-8). The indirect effects of each contextual variable by each in dividual-level variable are deta iled in Table 4-9. For the effect of racial dispersion on violence, 64.1% of the effe ct on violence was mediated through individual-level variables (such as alcohol use, marijuana use, other drug use, academic achievement, group fighting, and baseline violence). More than half of the effect of peer alcohol (52.4%), and 83. 5% of peer marijuana use on violence was mediated through proximal variables. Because baseline violence was a significant predictor of both violent trajectory groups, a post-hoc analysis was conducted to understand the characteristics associated with baseline violence (Table 4-10). Higher levels of racial heterogeneity in the neighborhood, peer alcohol use, peer mariju ana use, individual-level alcohol and marijuana use, other drug use, lower academic achievement, greater desire to leave home, depression, and group fighting were all significantly higher among those who were violent at baseline. White females Table 4-11 shows the bivariate relationshi p b etween each risk or protective factor and violence trajectories for White females. For desistors, peer alcohol use (OR = 1.88;

PAGE 128

128 95% CI 1.07-4.77), peer marijuana use (OR = 3.14; 95% CI 1.99-6.44) individual-level alcohol use (3.59; 95% CI 1.99-6.44), mari juana use (OR = 2.35; 95% CI 1.61-3.42), other drug use (OR = 3. 24; 95% CI 1.99-5.28), desire to leave ho me (OR = 1.88; 95% CI 1.41-2.52), depression (OR = 1.92; 95% CI 1.16-3.18), group fighting (OR = 3.64; 95% CI 2.70-4.91), and baseline physical aggression (OR = 7.50; 95% CI 4.37-12.86) were identified as risk factors for being in t he desistor trajectory group compared to the non-aggressive group. Protective factor s for desistors included higher parental involvement (OR = 0.93; 95% CI 0.86-0.99) and academic achievement (OR = 0.82; 95% CI 0.70-0.95). For escalators, none of t he risk and protective factors tested in the model were significant predictors of group membership. Because racial dispersion, poverty, urban neighborhood, and parental alcohol use were not significant in predicting physical aggression trajectories for either group at the bivariate level, they were dropped from further mult ivariate analyses. When considering the multivaria te effect of parentand pee r-level variables only in predicting violent trajectory membership (results not shown), peer marijuana use remained a risk factor for desistors only (O R = 3.04; 95% CI 1. 70-5.41). Parental involvement remained protective for desistors (OR = 0.91; 95% CI 0.85-0.98). When individual-level variables were added to the multivariate model (Table 4-12), parental involvement among desistors was the only contextual variable that remained significant (OR = 0.93; 95% CI 0.87-0.99). A number of individual-level variables were significant for desistors only. Specifically, alcohol use (OR = 2.27; 95% CI 1.16-4.89), desire to leave home (OR = 1.41; 95% CI 1.01-1.98) and group fighting (OR = 2.96; 95% CI 2.12-4.13) were identified as risk factors for membership in the desistors trajectory

PAGE 129

129 group. No risk or protective factors signifi cantly predicted membership in the escalator trajectory group. The final model, adjusted for baseline vi olence and all other risk and protective factors, is presented in Table 4-13. For desistors, parental involvement remained significant (OR = 0.93; 95% CI 0.86-0.99) and peer alcohol use was marginally significant (OR = 1.73; 95% CI 0.90-3.31; p <0.10) among desistors. Also among desistors, alcohol use (OR = 2.13; 95% CI 1.07-4.21), group fight ing (OR = 2.56; 95% CI 1.79-3.66) and baseline violence (OR = 3.82; 95% CI 2.09-6.99) predicted membership in the desistor group, compared to the non-violent trajectory group. None of the risk and protective factor s tested were significant in predicting membership in the escalator trajectory group. The test of the mediation pathway (based on bivariate analysis) is presented in Figure 4-4. The significance of pathway a indicated that all variabl es met the criteria for inclusion in the mediation analysis (e.g., contextual variables must be significantly associated with at least one mediator) (Table 4-14). The indirect effects of each contextual variable by each in dividual-level variable are deta iled in Table 4-15. For the effect of parental involvement on violen ce, 54.3% of the effect on violence was mediated through individual-level variables (such as alcohol use, marijuana use, other drug use, academic achievement, depression, desire to leave home, group fighting, and baseline violence). The majority of the effect of peer alcohol (82.9%) on violence, and 65.1% the effect of peer marijuana use on violence was mediated through proximal variables.

PAGE 130

130 Because baseline violence was a significant predictor of at l east one trajectory group (desistors), a post-hoc analysis was conducted to understand the characteristics associated with baseline violence (Table 4-16). Higher levels of racial heterogeneity in the neighborhood (p<0.10), peer alcohol use, peer marij uana use, individual-level alcohol and marijuana use, ot her drug use, lower academ ic achievement (p<0.10), greater desire to leave home, depression, and gr oup fighting were all significantly higher among those who were violent at baseline. African-American males Table 4-17 shows the bivariate relationshi p b etween each risk or protective factor and violence trajectories. For desistors, peer marijuana use (OR = 2.43; 95% CI 1.175.03), individual-level alcohol use (2.03; 95% CI 1.14-3.63), mari juana use (OR = 3.42; 95% CI 1.89-6.22), desire to leave home (OR = 1.38; 95% CI 0.98-1.94; p <0.10), group fighting (OR = 2.04; 95% CI 1.30-3.18), and baseline physical aggression (OR = 3.57; 95% CI 1.61-7.92) were identified as risk fa ctors for being in the desistor trajectory group compared to the non-aggressive group. For escalators, parental alcohol use (OR = 2.27; 95% CI 1.17-4.41), individual-level marijuana use (OR = 2.08; 95% CI 1.034.18), and group fighting (O R = 1.85; 95% CI 1.22-2.83) were identified as risk factors. Baseline violence was not a significant predict or of escalation (OR = 1.47; 95% CI 0.643.40). Because racial dispersion, poverty, urban neighborhood, parental involvement, peer alcohol use, other drug use, academic achievement, and depression were not significant in predicting physi cal aggression for either group in bivariate analyses, they were dropped from further multivariate analyses. When considering the multivariate effect of parentand peer-level variables only (results not shown), parental alcohol use was a significant risk factor for escalators only

PAGE 131

131 (OR = 2.23; 95% CI 1.12-4.42). Peer marijuana use was a ri sk factor for desistors (OR = 2.77; 95% CI 1.24-6.16). When indivi dual-level variables were added to the multivariate model (Table 4-18), parental al cohol use among escalators was the only contextual variable that re mained significant (OR = 2.16; 95% CI 1.09-4.28). Marijuana use was a significant risk factor for both e scalators (OR = 3.11; 95% CI 1.23-7.88) and desistors (OR = 2.51; 95% CI 1.02-6.16). Gr oup fighting was also a significant predictor of both escalation (OR = 1.97; 95% CI 1. 22-1.36) and desistance (OR = 1.90; 95% CI 1.17-3.08). The final model, adjusted for baseline violence in addition to all other risk and protective factors, is pres ented in Table 4-19. For escala tors, parental alcohol use (OR = 2.22; 95% CI 1.15-4.29) remained a signifi cant risk factor. Ma rijuana use was a risk factor for both desistors (OR = 2.78; 95% CI 1.12-6.89) and escala tors (OR = 2.59; 95% CI 1.08-6.23). Group fighting was marginally predictive of desistance (OR = 1.66; 95% CI 0.96-2.87; p <0.10); however, group fight ing was a significant predictor of escalation (OR = 2.05; 95% CI 1.23-3.42) after account ing for baseline violence. Baseline violence (OR = 2.56; 95% CI 1.16-5.63) predicted me mbership in the desistor group only, compared to the non-violen t trajectory group. The test of the mediation pathway (based on bivariate analysis) is presented in Figure 4-5. The significance of pathway a indicated that all vari ables except parental alcohol use met the criteria for inclusion in the mediation analysis (e.g., contextual variables must be significantly associated with at least one mediator) (Table 4-20). The indirect effects of each peer marijuana use by each individual-level variable are detailed in Table 21. For example, 74.3% of the ef fect of peer marijuana use on violence was

PAGE 132

132 mediated through the individual-le vel variables. The greatest portion of this effect was mediated through individual mar ijuana use (24.9%). Group fight ing (16.9%), desire to leave home (11.0%), alcohol use (10.8%), and baseline ph ysical aggression (10.7%) also significantly mediated the effect of peer marijuana use on violence trajectories. Because baseline violence was a significant predictor of at least one violent trajectory group (desistors), a post-hoc analysis was conducted to understand the characteristics associated with baseline violenc e (Table 4-22). Lower levels of poverty in the neighborhood, peer alcohol use, peer ma rijuana use, individual-level alcohol and marijuana use, lower academic achievement (p<0.10), greater des ire to leave home, depression (p<0.10), group fight ing, and age were all significantly higher among those who were violent at baseline. African-American females Table 4-23 shows the bivariate relationshi p b etween each risk or protective factor and violence trajectories among African-Amer ican females. For desistors, peer marijuana use (OR = 2.44; 95% CI 1.40-4.27) individual-level alcohol use (1.97; 95% CI 1.60-3.34), marijuana use (OR = 2.68; 95% CI 1.66-4.31), group fighting (OR = 2.59; 95% CI 1.83-3.64), and baseline violence (OR = 5.53; 95% CI 3.52-8.68) were identified as risk factors for being in the desistor trajectory group compared to the non-violent group. Parental involvement wa s protective from membership in the desistor group (OR = 0.77; 95% CI 0.64-0.93). For escalators, none of the risk and protective factors tested significantly predicted escalation. Because racial dispersion, poverty, urban neighborhood, parental involvem ent, parental alcohol use, peer alcohol use, other drug use, desire to leave home, and depression were not significant in predicting violence in

PAGE 133

133 bivariate analyses for either group, they were dropped from further multivariate analyses. Table 4-24 shows the multivariate effect of peerand individual-level variables on the desistor and escalator trajectory groups Group fighting was also a significant predictor of desistance (OR = 2.18; 95% CI 1.46-3.25) only, and academic achievement was marginally protective among desis tors (OR = 0.82; 95% CI 0.65-1.02; p <0.10). No risk and protective factors were significant am ong escalators. As displayed in Table 425, the fully adjusted model shows that group fighting (OR = 1.79; 95% CI 1.16-2.77) and baseline violence (OR = 3.14; 95% CI 1. 79-5.62) were the only risk factors that were significantly predictive of desistance. The test of the mediation pathway (based on bivariate analysis) is presented in Figure 4-6. The significance of pathway a indicated that all variabl es met the criteria for inclusion in the mediation analysis (e.g., peer marijuana use must be significantly associated with at least one mediator) (Table 4-26). The indirect effects of peer marijuana use by each individual-level variable are detailed in Table 4-27. For example, 73.5% of the effect of peer marijuana use was mediated through the individual-level variables. The greatest portion of this ef fect was mediated th rough baseline violence (20.8%). Individual-level marijuana use (19. 1%), group fighting (16.9%), alcohol use (13.9%), and academic achievement (2.7%) also significantly mediated the effect of peer marijuana use on violence trajectories. Because baseline violence was a significant predictor of at least one violent trajectory group (desistors), a post-hoc analysis was conducted to understand the characteristics associated with baseline viol ence (Table 4-28). Residing in an urban

PAGE 134

134 neighborhood (p<0.10), parental alcohol use, peer alcohol use, peer marijuana use, individual-level alcohol and marijuana us e, other drug use, lower academic achievement, greater desire to leave hom e, depression and group fighting were all significantly higher among those who were violent at baseline. Hispanic males Table 4-29 shows the bivariate relationshi p b etween each risk or protective factor and violence trajectories among Hispanic males. For desistors, peer alcohol use (OR = 2.00; 95% CI 1.06-3. 76), peer marijuana use (OR = 2. 36; 95% CI 1.234.51), alcohol use (3.09; 95% CI 1.72-5.57) marijuana use (OR = 3.49; 95% CI 1.89-6.46) other drug use (OR = 4.78; 95% CI 2.638.70), group fighting (OR = 2.38; 95% CI 1.61-3.50), and baseline violence (OR = 7.13; 95% CI 3.27-15. 56) were identified as risk factors for being in the desistor trajectory group compared to the non-violent group. For escalators, marijuana use (OR = 2.49; 95% CI 0.99-6.21; p <0.10), depression (OR = 2.39; 95% CI 1.19-9.26), gr oup fighting (OR = 2.23; 95% CI 1.25-3.96), and baseline violence (OR = 3.61; 95% CI 1. 40-9.26) were identified as ri sk factors. Parental alcohol use was identified as a marginally significant protective factor (OR = 0.90; 95% CI 0.281.05; p <0.10) for escalators. Because racial di spersion, poverty, urban neighborhood, parental involvement, academic achievement desire to leave home, and speaking Spanish at home were not significant in predi cting violence for either group in bivariate analyses, they were dropped from fu rther multivariate analyses. When considering the multivariate effect of parentand peer-level variables on the desistor and escalator trajectory groups alo ne (results not shown), parental alcohol use was a marginally significant protective fact or for escalators only (OR = 0.45; 95% CI 0.20-1.03; p <0.10). When individual-level variabl es were added to the multivariate

PAGE 135

135 model (Table 4-30), parental alcohol use am ong escalators was the only contextual variable that remained marginally signi ficant (OR = 0.48; 95% CI 0.21-1.12; p <0.10). Marijuana use was a significant risk factor fo r both escalators (OR = 2.49; 95% CI 1.036.01) and desistors (OR = 2.45; 95% CI 0.97-6.17; p <0.10). Group fighting (OR = 1.97; 95% CI 1.10-3.57) and depression (OR = 2. 21; 95% CI 1.10-4.44) were significant predictors of escalation only. The final model, adjusted for baseline violence in addition to all other risk and protective factors, is presented in T able 4-31. Marijuana use was a marginally significant risk factor for both desis tors (OR = 2.07; 95% CI 0.87-4.93; p <0.10) and escalators (OR = 2.13; 95% CI 0.87-5.22; p <0.10). Depression was a risk factor for escalators only (OR = 2.39; 95% CI 1.24-4.62). Group fighting was not a significant predictor of membership in either traj ectory group once baseline violence was added to the model. Baseline violence pr edicted membership in both high-risk trajectory groups (OR = 5.64; 95% CI 2.18-14.55 for desisto rs, OR = 3.56; 95% CI 1.30-9.62 for escalators). The test of the mediation pathway (based on bivariate analysis) is presented in Figure 4-7. The significance of pathway a indicated that all vari ables except parental alcohol use met the criteria for inclusion in the mediation analysis (e.g., contextual variables must be significantly associated with at least one mediator) (Table 4-32). The indirect effects of each peer-l evel variable by each individual-level variable are detailed in Table 4-33. For example, 86.9% of the effect of peer alcohol use on violence, and 79.0% of the effect of peer marijuana use on violence was mediated through the

PAGE 136

136 individual-level variables, including alcohol use, marijuana use, other drug use, depression, group fighting, and baseline violence). Because baseline violence was a significant predictor of at least one violent trajectory group (desistors), a post-hoc analysis was conducted to understand the characteristics associated with baseline viol ence (Table 4-34). Peer alcohol use, peer marijuana use, individual-level alcohol and marijuana use, other drug use, lower academic achievement (p<0.10), greater desire to leave home (p<0.10), speaking English in the home, group fighting, and being born in the United States were all significantly more likely among those who were violent at baseline. Hispanic females Table 4-35 shows the bivariate relationshi p b etween each risk or protective factor and violence trajectories among Hispanic female s. For desistors, peer alcohol use (OR = 3.37; 95% CI 1.63-6.98), peer marijuana use (OR = 2.52; 95% CI 1.16-5.47), alcohol use (OR = 2.96; 95% CI 1.346.54), group fighting (OR = 1.63; 95% CI 1.03-2.56), and baseline violence (OR = 3.17; 95% CI 1.33-7.53) were identifi ed as risk factors for being in the desistor trajectory group compared to the non-aggressive group. Age (OR = 0.77; 95% CI 0.61-0.97) was a protective factor for des istors. Among escalators, speaking Spanish at Home (O R = 1.72; 95% CI 0.94-3.15; p <0.10) was the only significant risk factor, while residing in an urban neighborhood (OR = 0.31; 95% CI 0.120.80), and being 2nd Generation US-Born (OR = 0.28; 95% CI 0.11-0.69) were identified as protective factors. Because racial di spersion, poverty, parental involvement, parental alcohol use, marijuana use, other drug use, academic ac hievement, desire to leave home, and depression were not significant in predicting violence for either group in bivariate analyses, they were dropped from further multivariate analyses.

PAGE 137

137 When considering the multivariate effect of neighborhoodand peer-level variables only on the desistor and escalator trajectory groups (results not shown), residence in an urban neighborhood was a significant protective factor for escalators only (OR = 0.29; 95% CI 0.13-0.66). Peer alcohol use was a risk factor for both desistors (OR = 3.34; 95% CI 1.14-9.72) and escalators (OR = 2.52; 95% CI 0.96-6. 56). When individuallevel variables were added to the m odel (Table 4-36), residence in an urban neighborhood in the escalator group was the only contextual variable that remained significant (OR = 0.29; 95% CI 0.13-0.68). Al cohol use marginally predicted desistance (OR = 2.40; 95% CI 0.94-6.14; p <0.10), and speaking Spanish at home was a significant risk factor for escalators (OR = 7.04; 95% CI 1.50-32.98) only. Group fighting did not significantly predict membership in ei ther high-risk trajectory group. The results from the model adjusted for baseline in addition to all other risk and protective factors are displayed in Table 4-37. These results did not differ substant ially from the model unadjusted for baseline violence. The test of the mediation pathway (based on bivariate analysis) is presented in Figure 4-8. The significance of pathway a indicated that all variabl es met the criteria for inclusion in the mediation analysis (e.g., contextual variables must be significantly associated with at least one mediator) (Table 4-38). The indirect effects of each contextual-level variable by each individual-level variable are detailed in Table 4-39. Specifically, 59.7% of the effect of re siding in an urban nei ghborhood on violence, 52.3% of peer alcohol use on violence, and 63. 0% of peer marijuana use on violence is mediated through individual-level variables, in cluding alcohol use, group fighting, and baseline violence.

PAGE 138

138 Asians Table 4-40 shows the bivariate relationshi p b etween each risk or protective factor and violence trajectories among Asians. For desistors, other drug use (OR = 3.63; 95% CI 1.25-10.54), desire to leave hom e (OR = 1.85, 95% CI 0.91-3.74; p <0.10), group fighting (OR = 4.45; 95% CI 1.98-9.99), and baseline violence (OR = 3.70; 95% CI 1.0313.25) were identified as risk factors for being in the desistor trajectory group compared to the non-violent group. Protective factors for desistors included living in an urban neighborhood (OR = 0.26; 95% CI 0.08-0.89), high academic achievement (OR = 0.59; 95% CI 0.430.80), and lower age (OR = 0.72; 95% CI 0.51-1.03; p <0.10). For escalators, no risk or protective variables te sted explained the variabi lity in trajectory group membership. Because racial dispersion, poverty, parental involvement, parental alcohol use, peer alcohol use, peer marij uana use, alcohol use, marijuana use, and depression were not significant in predicting violence fo r either group in bivariate analyses, they were dropped from fu rther multivariate analyses. Table 4-41 shows the multivariate effe ct of the remain ing contextualand individual-level variables on trajectories of violence. Desire to leave home (OR = 1.57; 95% CI 0.96-2.59; p <0.10) and group fighting (OR = 3. 08; 95% CI 1.67-5.67) were significant predictors of membership in t he desistor group only. The model adjusted for baseline violence in addition to all other risk and protective factors is presented in Table 4-42. This fully adjusted m odel does not substantially di ffer from the unadjusted model, as baseline violence did not predict trajectory membership among escalators or desistors. Instead, group fighting (OR = 3.09; 95% CI 1.71-5.61) and desire to leave home (OR = 1.57; 95 % CI 0.95-2.60; p <0.10) were identified as risk factors for desistors.

PAGE 139

139 The test of the mediation pathway (based on bivariate analysis) is presented in Figure 4-9. The significance of pathway a indicated that only group fighting met the criteria for inclusion in t he mediation analysis (e.g., c ontextual variables must be significantly associated with at least one mediat or) (Table 4-43). The indirect effect of urban neighborhood on violent traj ectory membership is detailed in Table 4-44. This table shows that 29.4% of t he effect of residing in an ur ban neighborhood on violence is mediated through group fighting. Because baseline violence was a significant predictor of at least one violent trajectory group (desistors), a post-hoc analysis was conducted to understand the characteristics associated with baseline vi olence (Table 4-45). Peer alcohol use (p<0.10), peer marijuana use, individual-level alcohol and marijuana use, other drug use, lower academic achievement, and group fight ing were all significantly more likely among those who were violent at baseline. Native Americans Table 4-46 shows the bivariate relationshi p b etween each risk or protective factor and violence trajectories among Native Americ ans. For desistors, racial dispersion (OR = 5.89; 95% CI 1.76-19.92), peer marij uana use (OR = 2.57; 95% CI 1.00-6.61; p <0.10), alcohol use (OR = 0.87 ; 95% CI 0.87-8.13; p <0.10), marijuana use (OR = 3.19; 95% CI 1.30-7.79), other drug use (OR = 3.04; 95% CI 1.16-7.94), desire to leave home (OR = 1.70, 95% CI 1.092.63), depression (OR = 1.70; 95% CI 1.15-4.55), group fighting (OR = 1.92; 95% CI 1.17-3.14), and baseline violence (OR = 4.67; 95% CI 1.65-13.27) were identified as risk factors for being in the desistor trajectory group compared to the nonaggressive group. Parental involvement was identified as a protective factor for desistors (OR = 0.89; 95% CI 0.78-1.01; p <0.10). Among escalators, peer marijuana

PAGE 140

140 use (OR = 3.25; 95% CI 1.258.44), alcohol use (OR = 2.84; 95% CI 1.35-5.98), marijuana use (OR = 2.54; 95% CI 0.88-7.34; p <0.10), other drug us e (OR = 4.68; 95% CI 1.48-14.73), depression (OR = 2.79; 95% CI 1.05-7.44), and baseline violence (OR = 2.69; 95% CI 0.91-7.97; p <0.10). Because urban neighbo rhood, poverty, parental alcohol use, peer alcohol use, peer alcohol use, and academic achievement were not significant in predicting violence for eit her group in bivariate analyses, they were dropped from further mult ivariate analyses. The multivariate effect of contextual vari ables alone in predicting violent trajectory membership showed that racial dispersi on (OR = 4.21; 95% CI 1.05-16.83) and peer marijuana use (OR = 2.70; 95% CI 1.08-6.73) were significant risk factors for desistors (results not shown). Peer marijuana use was significantly predictive of escalation (OR = 4.63; 95% CI 1.68-12.74). When individual-level variables were added to the multivariate model (Table 4-47), peer mariju ana use was the only contextual variable that retained significance among escalators (OR = 3.48; 95% CI 1.14-10.16). Other drug use was the only significant individual-l evel variable predicting escalation (OR = 1.82; 95% CI 1.07-3.11). None of the variables tested significantly predicted desistance. The final model, adjusted for baseline violence in addition to all other risk and protective factors, is presented in Table 4-48. After adj usting for baseline violence, racial dispersion marginally predicted desistance (OR = 3.88 ; 95% CI 0.91-16.61; p <0.10); however, no other variables predicted membership in the desistor group. Among escalators, alcohol use (OR = 1.89; 95% CI 0.13-3.23; p <0.10), other drug use (OR = 1.92; 95% CI 1.17-3.14) and baseline violence (OR = 3.14; 95% CI 0.92-10.71;

PAGE 141

141 p <0.10) were identified as risk factors fo r membership in the escalator category compared to the non-violent group. The test of the mediation pathway (based on bivariate analysis) is presented in Figure 4-10. The significance of pathway a indicated that all variables except for baseline violence met the criteria for inclusi on in the mediation anal ysis (e.g., mediators must be significantly associated with at leas t one contextual variable) (Table 4-49). The indirect effects of each cont extual-level variable by each individual-level variable are detailed in Table 4-50. This table shows that 66.3% of the effect of racial dispersion on violence, 68.6% of the effect of parental involvement on violence, and 67.9% of the effect of peer marijuana use on violence is mediated through individual-level variables, such as alcohol use, marijuana use, other drug use, depression, desire to leave home, and group fighting. Because baseline violence was a significant predictor of at least one violent trajectory group (desistors), a post-hoc analysis was conducted to understand the characteristics associated with baseline vi olence (Table 4-51). Peer alcohol use, individual-level alcohol use, lower academic achievement (p<0.10), desire to leave home (p<0.10), depression (p<0.10), and group fighting were all significantly more likely among those who were violent at baseline. Discussion The present study examined the number and shape of trajectories of violent behavior by race/ethnic and gender subgroups, as well as the direct and indirect effects of multiple domains of risk and protective fa ctors for membership in each trajectory group. The latent-group based trajectory models found a three-class model across all

PAGE 142

142 genderand racial/ethnic subgroups: a non-violent group, a group who desists violence, and a group of escalators whose severity of violence increases over time. These results are consistent with previous research on trajectories of violence, and risk and protective factors for violent behav ior among adolescents. Three trajectory groups were extracted from the data in this study, and this is consistent with the extant literature that suggests t here are between three and five unique groups of violent adolescents (Piquero, 2008; Maldonado-Molina, Reingle, Tobler, Jennings, & Komro, 2010; Maldonado-Molina, Piquero, Jennings, Bird & Canino, 2009). There were differences in the prevalence of member ship in each trajectory group between gender and racial/ethnic subgroups. The highest proporti on of those in the late-onset escalator group was African-American males and Native Americans, followed by Hispanic males. African-American males and Hispanic males were more likely to be desistors, and White females and Hispanic females were most likel y to be non-violent. Males were more likely than females to be violent for each racial/ethnic group. The findings from this study are unique in that a late-onset group of violent adolescents was identified across racial/e thnic and gender subgroups. Although some studies have found support for the existence of this group (DUnger et al., 1998; Zara & Farrington, 2009), the majority of the literature on trajectories of delinquency supports the age-crime curve, in which adolescent s age out of delinquent behaviors before age 20 (Piquero, 2008; Farrington, 1986). The shar p increase found among the late-onset escalators in this study may be due to the weighting of more severe forms of violence used in this study (e.g., using a knife or gun in a fighting, shooting someone). If this is

PAGE 143

143 the case, participants assigned to the escalato r group initiate late in life, and with more severe forms of violent behavior. This study identified a variety of risk and protective factors that significantly predicted violent trajectory group membership by race/e thnicity and gender. These findings are consistent with prior literature on the disparities by race/ethnicity and gender on community-, familyand peer-level risk factors for violence. First, consistent with Williams, Van Dorn, Ayer s, Bright, Abbott, & Hawkins (2007), results from this study found evidence of gender differences in violence by level of parental involvement, as lower levels of parental involvement was pr edictive of violence for girls but not boys. Second, this study found additional support fo r Broidy et al., (2003) who reported that early, chronic aggression and violence (e.g., elementary school) increases the risk of continual violence and other forms of delinquency throughout adolescence among males only. This study adds to this finding in that prior violence appears to strongly predict future violence among males, however, baseline aggression also predicts future violence among African-American female des istors. These findings provide evidence that there are important differences by ra ce/ethnicity and gender in the predictors of violence. Gender Differences This study found support for the literature suggesting that males have a higher rate of offending compared to females, as males were less likely to be in the non-violent trajectory group than females across all ra cial/ethnic groups (McNulty & Bellair, 2003; Williams, Van Dorn, Ayers, Bright, A bbott, & Hawk ins, 2007; Farrington & Loeber, 2000). Results of this study also suggest that females are participating in more serious violence; specifically, aggravated assault. Pr evious literature has found that since 1980,

PAGE 144

144 aggravated and simple assaults among gi rls have increased by 113% and 257%, respectively (Snyder, 2003). The current fi nding may be due to the weighting of the more severe aggression variables, as fightin g is not weighted as heavily as using a knife or gun in a fight, or havi ng shot or stabbed someone. There were a small number of consistencie s in the predictors of violence within gender groups. Hispanic and African-American males were at increased risk for violence if they used marijuana. Group figh ting was a risk factor for violence among all African-American and White males. Additi onally, alcohol use was a predictor of desistance among Hispanic and White females. These findings are consistent with prior literature on racial/ethnic differences in substance use, as African-Americans and Hispanics are more likely to use marijuana co mpared to Whites, while Whites are more likely to use alcohol (Lee, Mun, White & Simon, 2010; Johnston et al., 2010) compared to African-Americans and Hispanics. Considering these similarities, there were a number of gender differences within racial/ethnic group. First, parental involvem ent was a significant predictor of desistance among White females only. Group fighting and marijuana use predicted violence for African-American males but not African-American females. Baseline violence was not predictive of violence among Hispanic fema les and White females; however, this relationship did exist among African-Amer ican females. Among Hispanics, urban residence and alcohol use predicted violence fo r females only, while marijuana use, depression, and baseline violence predicted vi olence among males only. Overall, the predictors of violence across racial/ethnic ge nder groups were more different than they were similar.

PAGE 145

145 These findings support the hypothesis by Mo ffitt et al. (2001), who suggested that boys are exposed to more risk fa ctors than girls, and the elev ated number of risk factors may play a role in the higher prevalence of violent offending among males. Consistent with tests of this hypothesis (Daigle et al ., 2007), males were gene rally exposed to a greater number of risk factor s compared to females across racial/ethnic group. Among African-American females specifically, vari ables other than previous violence did not predict future violence. This suggests that either the predictors for delinquency among females are substantially different for females compared to males (and those risk factors were not included in this study), or that previous violence was a very strong risk factor for future violence, emphasizing the importance of early risk factors. Racial/Ethnic Differences This study found a number of differences in the predictors of membership in trajectory group by race/ethnicity. Among Whites, peer alcohol use had both a direct and indirect effect on violence This finding was not supported for any other group. Also consis tent among Whites, individual-lev el alcohol use was predictive of violence (desistance only). This direct relationshi p between alcohol and violence was also found among Hispanic females and Native Amer icans. Among African-American and Hispanic males only, marijuana use was a consis tent predictor of violence. Desire to leave home was a significant risk factor am ong Asians only, and Native Americans and White males were at risk for violence if they have reported using illi cit drugs other than marijuana. There were also a number of similarities across racial/ethnic groups. For example, peer marijuana use had a consistent, indirect ef fect on violence in all subgroups except Asians. Group fighting predicted desistanc e among Asians, White males and females,

PAGE 146

146 and African-American males. Baseline violence significantly predicted violence among White males, African-American males, Af rican-American females, Hispanic males and Native Americans. This study was unable to i dentify predictors of escalati on across racial/ethnic and gender subgroups, a high-risk late-onset tr ajectory group. Marijuana use was a significant predictor of escalation among African-American and Hispanic males; and baseline violence predicted e scalation among White males, Hispanic males, and Native Americans. These two risk factors were the only consistent predictors of this high-risk group across race/ethnicity and gender. The lack of significant predictors may be due to the time-varying nature of violence in th is group and the importanc e of early risks. Specifically, risk factors at mo re time points more proximal to the violent behavior (e.g., late adolescence) may be more predictive of me mbership in this trajectory group. Since violence is relatively low at baseline among e scalators, this variable may not be a potent predictor of late-onset violence. This findi ng highlights the need for future research on this group of escalators, as unique or time -varying risk factors may be present. In one study that identified this late-onset esca lator group (Zara & Farri ngton, 2009), a variety of psychological predictors were identifi ed, including high anxiety, low IQ, delinquent friends, having few friends early in life, and la te onset of sexual intercourse. These results indicate that childhood ri sk factors may predict this late-onset group of violent young adults, and more research on this unique group is necessary to further understand the etiology of la te-onset escalation. The number of risk factor s present at baseline highlights the need for early violence prevention programming. Specifically, across all racial/ethnic and gender

PAGE 147

147 subgroups, peer alcohol and peer marijuana use at baseline was associated with violent behavior at age 15. Individual-level alcoho l use was also associated with baseline violence among all subgroups, and marijuana use was more prevalent among those violent as baseline compared to those who we re not violent at baseline in all groups except Native Americans. Lower academic achievement was also more prevalent in those who were violent at baseline compared to those who were non-violent at baseline. These findings highlight the ear ly risk factors that are pres ent prior to age 15 that may serve as targets for large-scale violence prevention programming. This study had several limitations. Firs t, this study was unable to account for some of the variables that are important in predicting violence, such as peer delinquency, cognitive development, and psychological disorders. Second, latent-group based trajectory modeling provides an estima tion of the type and number of groups in the data, and this process is exploratory in nature. Despite the exploratory nature of trajectory estimation, the results of this study were consistent with the expected number and shape of trajectory groups from other studies (Piquero 2008; Zara & Farrington, 2009). Finally, risk factors were analyzed at mu ltiple levels; however, hierarchical linear modeling (HLM) was not used due to the small sample sizes available in some of the trajectory groups. In accordance with t he sampling design, all analyses accounted for the nesting of adolescents within schools, which may account for a portion of the variability in Census tract measures. Us e of HLM to account for the nesting within Census blocks is a direction for future research. Despite these weaknesses, the current study had a number of strengths. First, data were derived from a longi tudinal, nationally representative sample of adolescents

PAGE 148

148 followed into young adulthood. This samplin g design allows generalization to a national sample of adolescents across the United States. Second, the large sample size in the Add Health provides adequate power to stratify by racial/ethnic and gender groups to understand the differential risk factors by subgr oup. Third, although many studies have analyzed the multilevel risk and protective factors for violent behavior, few have assessed the degree to which the effects of contextual variables are mediated through more proximal variables at the individual-level. The me diation analyses allowed this study to quantify the effects of contextual variables as impor tant predictors of violence even though their direct and indirect effe cts, which are other wise mitigated using multivariate regression models. Finally, the trajectories estimated in this study are especially appropriate for studi es of delinquency and violence, as patterns tend to vary significantly over time (Fa rrington, 1986; Piquero, 2008). In conclusion, the findings from this st udy indicate that t here are substantial differences in the risk factors for violen ce by race/ethnicity and gender subgroups. These differential predictors may justify prevention strategies for large-scale prevention programming to include all of these dimens ions to maximize the preventive effect across demographic groups. These findings have a number of implications for prevention programming. First, social infl uences, such as exposure to peers who use alcohol or marijuana, and comm unity-level exposure to al cohol influence adolescents risk for violent behavior. These risk factors t hat were consistent across race/ethnicity may be targeted in a variety of populations to reduce participation in violence. Second, violent behavior begins even before age 15 in t he general population, indicating that the current prevention programming occurs too late. Prevention programming should

PAGE 149

149 begin early in elementary school settings to prev ent initiation of violent behavior. Third, there were substantial differences in the predictors of violence between racial/ethnic groups. Therefore, the compos ition of the intervention population (e.g., characteristics of the social structure, community, family, etc.) should be considered prior to program administration, as different subgroups may be exposed to different risk and protective factors. This differential exposure may equat e to increased or diminished propensity for violence. Disclosure This study was suppor ted by Awar d Numbers K01 AA017480 (PI: Mildred Maldonado-Molina), R01 AA013458 (PI: Kelli A. Komro) from the Na tional Institute on Alcohol Abuse and Alcoholism and the National In stitute for Minority Health and Health Disparities, and the Institute for Child Health Policy at the Un iversity of Florida. The content is solely the respons ibility of the author s and does not necessarily represent the official views of the National Institute on Alcohol Abuse and Alcoholism of the National Institute of Health. This research uses data from Add Health a program project directed by Kathleen Mullan Harris and designed by J. Richard Udry Peter S. Bearman, and Kathleen Mullan Harris at the University of North Caro lina at Chapel Hill, and funded by grant P01HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative f unding from 23 other federal agencies and foundations. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Info rmation on how to obtain the Add Health data files is available on the Add Health w ebsite (http://www.cpc.unc.edu/addhealth). No direct support was received from gr ant P01-HD31921 for this analysis.

PAGE 150

150 White males White females Figure 4-1. Trajectories of violenc e by racial/ethnic and gender subgroups. # Violent EventsMean Age # Violent EventsMean Age

PAGE 151

151 African-American males African-American females Figure 4-1. Continued. # Violent EventsMean Age # Violent EventsMean Age

PAGE 152

152 Hispanic males Hispanic females Figure 4-1. Continued. Asians # Violent EventsMean Age # Violent EventsMean Age

PAGE 153

153 Native Americans Figure 4-1. Continued. # Violent EventsMean Age # Violent EventsMean Age

PAGE 154

154 Figure 4-2. Differences in the prevalence of violence between racial/ethnic and gender subgroups.

PAGE 155

155 Table 4-1. Description of sample, Add Health. N=9421. Variable % Trajectory Groups Violence, Wave II a 0.52(0.04) Violence, Wave III a 0.28(0.03) Violence, Wave IV a 0.91(0.04) Community-level Racial Dispersion a 0.31(0.27) % Poverty a 0.13(0.03) Urban area 61.8 Parental and Peer Influences Parental Involvement a 5.81(3.4) Parental alcohol use 65.7 One or more peers use alcohol 57.6 One or more peers use alcohol 36.4 Individual-level Risk Factors Ever use of alcohol 58.2 Ever use of marijuana 30.5 Lifetime other drug use 17.7 Depression b 42.3 Intend to go to college 72.3 Desire to leave home 38.2 Speaking Spanish at home 6.14 Violence Group fighting in past year 22.1 Baseline violence 22.0 Demographics Gender (Male) 42.8 Age a 15.4(1.60) White 64.5 African-American or Black 23.6 Hispanic or Latino 14.8 Asian or Pacific Islander 5.8 Native American 4.1 Other Race 1.1 a Mean(SE) are reported. b Depression was measured as feeling sad or depressed one or more times in the past month.

PAGE 156

156 Table 4-2. Model fit statistics by racial/ethnic and gender subgroup, 3-group model. Subgroup AIC BIC -Log(L) White males 6267 6291 6259 White females 3602 3627 3594 African-American males 2702 2720 2694 African-American females 2047 2067 2039 Hispanic males 1962 1982 1953 Hispanic females 1039 1018 1009 Native Americans 869 886 860 Asians 1073 1091 1065

PAGE 157

157 Table 4-3. Racial/ethnic and gender group differences in mean violence by trajectory group. Trajectory Group Non-Violent N=2016 Desistors N=514 Escalators N=285 White males Mean Violence (Age 16) 0 2.62(2.83)*** 0.73(1.98)** Mean Violence (Age 21) 0 1.54(2.06)*** 0.52(1.48)*** Mean Violence (Age 26) 0 0.27(0.78)*** 6.75(2.32)*** N=2780 N=175 N=339 White females Mean Violence (Age 16) 0 2.85(2.63)*** 0.10(0.61)* Mean Violence (Age 21) 0 0.86(1.66)*** 0.07(0.50) Mean Violence (Age 26) 0 0.11(0.53) 7.55(1.46)*** N=513 N=191 N=117 African-American males Mean Violence (Age 16) 0 3.86(3.77)*** 0.55(1.54)** Mean Violence (Age 21) 0 1.82(2.61)*** 0.95(2.19)** Mean Violence (Age 26) 0 0.09(0.44) 6.38(2.30)*** N=919 N=127 N=157 African-American females Mean Violence (Age 16) 0 3.34(3.28)*** 0.11(0.51)* Mean Violence (Age 21) 0 1.61(2.59)*** 0.11(0.45)+ Mean Violence (Age 26) 0 0.24(0.97)** 7.27(1.66)*** N=442 N=156 N=388 Hispanic males Mean Violence (Age 16) 0 3.64(3.66)*** 1.87(3.08)** Mean Violence (Age 21) 0 1.65(2.45)*** 0.27(0.79)* Mean Violence (Age 26) 0 0.01(0.14) 6.56(2.44) N=618 N=67 N=77 Hispanic females Mean Violence (Age 16) 0 2.81(3.01)*** 0.45(1.45) Mean Violence (Age 21) 0 1.26(2.35)** 0.06(0.34) Mean Violence (Age 26) 0 0 7.50(1.45)*** N=226 N=58 N=54 Native Americans Mean Violence (Age 16) 0 3.14(2.93)*** 2.16(2.69)*** Mean Violence (Age 21) 0 0.83(1.83)* 0.64(1.24)** Mean Violence (Age 26) 0 0 5.82(2.47)*** N=530 N=60 N=81 Asians Mean Violence (Age 16) 0 2.72(3.03)** 0.05(0.32) Mean Violence (Age 21) 0 1.68(2.50)*** 0.12(0.48) Mean Violence (Age 26) 0 0.80(1.70) 7.28(1.76)*** Note: For all mean difference tests, the violent trajectory group means are compared to the non-violent group. *p<0.05 **p<0.01 ***p<0.001

PAGE 158

158 Table 4-4. Bivariate effects between risk/protec tive factors and trajectories of violence, White males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.50 0.72-3.10 3.62* 1.31-3.10 Poverty 26.8 0.24-3021. 15 25.9 0.09-7481.62 Urban Area 0.99 0.711.37 0.93 0.64-1.35 Parental and Peer Influences Parental Involvement 1. 00 0.95-1.05 0.98 0.94-1.03 Parental alcohol use 0. 85 0.62-1.18 1.09 0.70-1.70 Peer alcohol use 2.30*** 1.72-3.07 1.64*** 1.10-2.44 Peer marijuana use 1.94*** 1.41-2.67 1.62* 1.01-2.59 Individual-level Risk Factors Alcohol use 2.46*** 1. 79-3.38 1.48* 1.01-2.15 Marijuana use 2.35*** 1. 75-3.16 1.70* 1.07-2.68 Other drug use 2.56*** 1.70-3.87 1.82 0.94-3.53 Academic achievement 0.82 ** 0.72-0.92 0.86 0.72-1.03 Desire to leave home 1. 17 0.96-1.43 1.19 0.90-1.57 Depression 1.18 0.821.70 1.28 0.80-2.06 Violence Group fighting 2.22*** 1. 79-2.76 1.60* 1.08-2.38 Baseline violence 4.11*** 3.02-5.60 2.36*** 1.51-3.70 Demographics Age at Baseline 0.93 0.86-1.01 0.94 0.85-1.05 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 **p<0.01 ***p<0.001

PAGE 159

159 Table 4-5. Community, family, and peer effects on trajectories of viol ence, White males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.33 0.61-2.93 3.38* 1.21-5.49 Parental and Peer Influences Peer alcohol use 2.29 *** 1.56-3.36 1.53 0.99-2.33 Peer marijuana use 1.58* 1.08-2.31 1.28 0.78-2.10 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age. *p<0.05 ***p<0.001

PAGE 160

160 Table 4-6. Effects of multiple domains of ri sk factors on trajectories of violence, White males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.08 0.49-2.41 3.11* 1.19-8.19 Parental and Peer Influences Peer alcohol use 1.53* 1.03-2.28 1.29 0.78-2.11 Peer marijuana use 1.03 0.67-1.58 1.01 0.59-1.73 Individual-level Risk Factors Alcohol use 1.65* 1. 08-2.52 1.18 0.76-1.83 Marijuana use 1.30 0.861.99 1.22 0.68-2.17 Other drug use 1.35* 1. 04-1.76 1.20 0.80-1.79 Academic achievement 0.89 0.79-1.02 0.88 0.75-1.05 Desire to leave home Violence Group fighting 1.82*** 1.41-2.34 1.27 0.85-1.89 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age. *p<0.05 ***p<0.001

PAGE 161

161 Table 4-7. Effects of multiple domains of risk factors on traj ectories of violence, adjusted for baseline, White males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.05 0.46-2.393.07* 1.17-8.05 Parental and Peer Influences Peer alcohol use 1.47 0.98-2.211.25 0.76-2.06 Peer marijuana use 1.02 0.66-1.581.00 0.58-1.71 Individual-level Risk Factors Alcohol use 1.55* 1. 02-2.351.14 0.74-1.77 Marijuana use 1.16 0. 77-1.761.12 0.64-1.97 Other drug use 1.35* 1.04-1.761.21 0.81-1.79 Academic achievement 0. 93 0.82-1.060.91 0.77-1.07 Violence Group fighting 1.50** 1. 16-1.931.12 0.73-1.70 Baseline violence 2.66*** 1.88-3.751.83** 1.18-2.84 Note: The Non-Violent trajec tory group serves as the reference category. All analyses are controlling for age. *p<0.05 **p<0.01 ***p<0.001

PAGE 162

162 Figure 4-3. Test of the mediation pathway (based on biva riate analysis), White males. Social/Contextual Risks Racial dispersion Parental Involvement Peer alcohol and marijuana use Violent Trajectory Membership a b c Alcohol, Marijuana, and other drug use Academic Achievement Group fighting Baseline Violence Individual-Level Risks

PAGE 163

163 Table 4-8. Regression models testing t he association between community-level and peer level variables on individual-level risk and protective factors, White males. Mediators Alcohol Use Marijuana Use Other Drug Use Academic Achievement Group Fighting Baseline Violence OR OR OR OR OR OR Community-level variables Racial Dispersion 1.30 2.63* 1.34 0.52 2.23* 1.97* Parental and Peer Influences Peer alcohol use 8.03*** 7.01*** 6.78***0.72* 3.65*** 2.75*** Peer marijuana use 3.98*** 14.33*** 13.52***0.71* 2.14*** 2.34*** Note: All analyses are controlling for age. *p<0.05 ***p<0.001

PAGE 164

164 Table 4-9. Mediated effect of parentand pee r-level variables on violence trajectories, White males. Mediator Indirect Effect ( ab) z SE Percent Mediated Racial Dispersion Alcohol Use 0.075 2.72** 0.28 11.5 Marijuana Use 0.158 2.47* 0.06 22.6 Other Drug Use 0.045 2.26* 0.02 7.0 Academic Achievement 0.009 2.34* 0.01 1.5 Group Fighting 0.126 2.71** 0.05 19.2 Baseline Violence 0.015 3.04** 0.05 2.3 Total 64.1 Peer alcohol use Alcohol Use 0.326 5.03*** 0.06 13.6 Marijuana Use 0.327 4.83*** 0.07 13.6 Other Drug Use 0.257 3.28** 0.07 10.7 Academic Achievement 0.024 4.96*** 0.01 1.0 Group Fighting 0.278 4.14*** 0.07 11.6 Baseline Violence 0.045 5.02*** 0.06 1.9 Total 52.4 Peer marijuana use Alcohol Use 0.294 4.07*** 0.07 16.6 Marijuana Use 0.385 4.30*** 0.08 21.6 Other Drug Use 0.277 3.89*** 0.07 15.6 Academic Achievement 0.026 4.82*** 0.01 1.5 Group Fighting 0.208 4.59*** 0.05 11.7 Baseline Violence 0.293 4.85*** 0.06 16.5 Total 83.5 Notes: All models are adjusted for age. (a): These mediated effects were generated in acco rdance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhsh, & Veblen-Mortenson (2001). The perce nt mediation was generated using the formula: [(a*b/(a*b + c)] (MacKinnon, 2008). (b): Indirect effects are not direct ly comparable across variables. Percent mediation is comparable across variables and groups of variables. *p<0.05 **p<0.01 ***p<0.001

PAGE 165

165 Table 4-10. Post-hoc description (means and proportions) of adolescents who are violent at Wave I, White males. Violence at Baseline Violent NonViolent p Community-level Racial Dispersion a 0.23* 0.19 0.043 Poverty a 0.13 0.13 0.201 Urban Area 0.46 0.43 0.409 Parental and Peer Influences Parental Involvement a 5.74 5.81 0.797 Parental alcohol use 0.59 0.62 0.340 Peer alcohol use 0.73*** 0.52 <0.001 Peer marijuana use 0.47*** 0.27 <0.001 Individual-level Risk Factors Alcohol use 0.76*** 0.53 <0.001 Marijuana use 0.22*** 0.49 <0.001 Other drug use 0.23*** 0.11 <0.001 Academic achievement 0.55*** 0.71 <0.001 Desire to leave home 0.40** 0.31 0.005 Depression 0.41*** 0.27 <0.001 Violence Group fighting 0.47*** 0.13 <0.001 Demographics Age at Baseline a 15.26 15.17 0.373 Notes: Participants were considered violen t at baseline if they reported any of the violence items that were used to estima te violence trajectories: shot or stabbed someone, used knife or gun in a fight, or hurt someone badly enough to need care from a doctor or nurse. Wave IV weighti ng variable used for these analyses. a Mean is reported. *p<0.05 **p<0.01 ***p<0.001

PAGE 166

166 Table 4-11. Bivariate effects between risk/protec tive factors and trajectories of violence, White females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 2.00 0.84-4.77 1.78 0.76-4.12 Poverty 1.71 0.01-238.3369.72 0.28-123.8 Urban Area 1.02 0.641.64 0.76 0.54-1.07 Parental and Peer Influences Parental Involvement 0.93* 0.86-0.99 0.99 0.95-1.04 Parental alcohol use 0. 71 0.44-1.16 0.84 1.19-2.95 Peer alcohol use 1.88** 1.07-4.77 1.33 0.83-1.67 Peer marijuana use 3.14*** 1.99-6.44 1.18 0.72-1.31 Individual-level Risk Factors Alcohol use 3.59***1. 99-6.44 0.97 0.71-1.31 Marijuana use 2.35***1. 61-3.42 1.13 0.80-1.61 Other drug use 3.24***1. 99-5.28 0.70 0.41-1.18 Academic achievement 0.82* 0.70-0.95 0.99 0.86-1.14 Desire to leave home 1.88 ***1.41-2.52 1.16 0.97-1.39 Depression 1.92* 1.16-3.18 0.92 0.65-1.30 Violence Group fighting 3.64***2. 70-4.91 0.91 0.62-1.33 Baseline violence 7.50***4. 37-12.86 1.01 0.61-1.67 Demographics Age at Baseline 0.89 0.78-1.03 1.08 0.96-1.22 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 **p<0.01 ***p<0.001

PAGE 167

167 Table 4-12. Effects of multiple domains of risk factors on trajectories of violence, White females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Parental and Peer Influences Parental involvement 0.93* 0.87-0.99 1.00 0.95-1.06 Peer alcohol use 0.78 0.41-1.43 1.25 0.78-2.01 Peer marijuana use 1.70 0.87-3.33 1.17 0.71-1.91 Individual-level Risk Factors Alcohol use 2.27* 1. 16-4.89 0.79 0.55-1.14 Marijuana use 1.23 0.67-2.23 1.16 0.68-1.98 Other drug use 1.00 0. 79-1.21 0.60 0.33-1.13 Depression 1.18 0.71-1.96 0.86 0.58-1.27 Academic achievement 0.98 0.79-1.21 0.98 0.83-1.15 Desire to leave home 1.41* 1.01-1.98 1.15 0.93-1.41 Violence Group fighting 2.96*** 2.12-4.13 0.95 0.66-1.35 Note: The Non-Violent trajec tory group serves as the reference category. All analyses are controlling for age. *p<0.05 ***p<0.001

PAGE 168

168 Table 4-13. Effects of multip le domains of risk factors on trajectories of violence, adjusted for baseline, White females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Parental and Peer Influences Parental Involvement 0.93* 0.86-0.99 1.00 0.94-1.06 Peer alcohol use 0.70 0.37-1.32 1.25 0.78-2.01 Peer marijuana use 1.73 0.90-3.31 1.17 0.72-1.91 Individual-level Risk Factors Alcohol use 2.13* 1.074.21 0.79 0.55-1.13 Marijuana use 1.13 0.65-1.99 1.16 0.68-1.97 Other drug use 0.79 0.401.57 0.60 0.32-1.12 Academic achievement 0.96 0.77-1.21 0.91 0.83-1.15 Desire to leave home 1.35 0.93-1.94 1.14 0.93-1.40 Depression 1.14 0.69-1.90 0.86 0.58-1.27 Violence Group fighting 2.56** 1.79-3.66 1.12 0.66-1.32 Baseline violence 3.82*** 2.09-6.99 1.08 0.67-1.89 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age. *p<0.05 **p<0.01 ***p<0.001

PAGE 169

169 Figure 4-4. Test of the mediation pathway (based on bivariate analysis), White females. Social/Contextual Risks Peer marijuana use Parental alcohol use Violent Trajectory Membership a b c Alcohol and marijuana use Academic achievement Group fighting Baseline Violence Individual-Level Risks

PAGE 170

170 Table 4-14. Regression models testing t he association between parent and peer level variables and individual-level risk and protective factors, White females. Mediators Alcohol Use MJ Use Other Drug Use Depression Desire to Leave Home Academic Achievement Group Fighting Baseline Violence OR OR OR OR OR OR OR OR Parental and Peer Influences Parental Involvement 0.96* 0.92 0.93** 0.94*** 0.90*** 1.17*** 0.98 0.97 Peer alcohol use 7.02*** 5.35*** 4.66*** 1.75*** 2.23*** 0.67* 3.03*** 2.51*** Peer marijuana use 6.36*** 9.85*** 6.67*** 2.06*** 2.24*** 0.58*** 2.82*** 2.64*** Note: All analyses are controlling for age. *p<0.05 ***p<0.001

PAGE 171

171 Table 4-15. Mediated effect of parentand pee r-level variables on violence trajectories, White females. Mediator Indirect Effect ( ab) z SE Percent Mediated Parental Involvement Alcohol Use 0.164 7.99*** 0.02 8.6 Marijuana Use 0.157 7.68*** 0.02 8.3 Other Drug Use 0.103 6.75*** 0.02 5.4 Academic Achievement 0.020 5.94*** 0.01 1.1 Depression 0.136 5.72*** 0.02 7.1 Desire to Leave Home 0.239 10.09*** 0.02 12.5 Group Fighting 0.017 32.78*** 0.001 0.9 Baseline Violence 0.199 5.07*** 0.04 10.4 Total 54.3 Peer alcohol use Alcohol Use 0.260 5.21*** 0.05 14.0 Marijuana Use 0.242 4.69*** 0.05 13.0 Other Drug Use 0.174 3.39*** 0.05 9.4 Academic Achievement 0.032 4.68*** 0.006 1.7 Depression 0.121 5.01*** 0.02 6.5 Desire to Leave Home 0.236 6.32*** 0.04 12.7 Group Fighting 0.271 3.93*** 0.07 14.6 Baseline Violence 0.202 3.78*** 0.05 10.9 Total 82.9 Peer marijuana use Alcohol Use 0.259 5.12*** 0.05 13.5 Marijuana Use 0.247 4.72*** 0.05 12.9 Other Drug Use 0.133 4.38*** 0.04 6.9 Academic Achievement 0.028 4.74*** 0.005 1.4 Depression 0.132 4.88*** 0.02 6.8 Desire to Leave Home 0.246 7.28*** 0.03 12.9 Group Fighting 0.248 4.15*** 0.06 13.0 Baseline Violence 0.214 3.78*** 0.06 11.2 Total 65.1 Notes: All models are adjusted for age. (a): These mediated effects were generated in acco rdance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhsh, & Veblen-Mortenson (2001). The perce nt mediation was generated using the formula: [(a*b/(a*b + c)] (MacKinnon, 2008). (b): Indirect effects are not direct ly comparable across variables. Percent mediation is comparable across variables and groups of variables. ***p<0.001

PAGE 172

172 Table 4-16. Post-hoc description (means and proportions) of adolescents who were violent at Wave I, White females. Violence at Baseline Violent NonViolent p Community-level Racial Dispersion a 0.24 0.21 0.056 Poverty a 0.13 0.13 0.900 Urban Area 0.50 0.48 0.576 Parental and Peer Influences Parental Involvement a 5.96 6.21 0.301 Parental alcohol use 0.67 0.62 0.176 Peer alcohol use 0.73*** 0.55 <0.001 Peer marijuana use 0.55*** 0.33 <0.001 Individual-level Risk Factors Alcohol use 0.76*** 0.57 <0.001 Marijuana use 0.48*** 0.27 <0.001 Other drug use 0.39*** 0.12 <0.001 Academic achievement 0.70 0.77 0.070 Desire to leave home 0.52*** 0.34 <0.001 Depression 0.57** 0.46 0.003 Violence Group fighting 0.48*** 0.14 <0.001 Demographics Age a 14.92 15.04 0.291 Note: Participants were considered violent at baseline if they reported any of the violence items that were used to estima te violence trajectories: shot or stabbed someone, used knife or gun in a fight, or hurt someone badly enough to need care from a doctor or nurse. Wave IV weighti ng variable used for these analyses. a Mean is reported. **p<0.01 ***p<0.001

PAGE 173

173 Table 4-17. Bivariate effects between risk/protec tive factors and trajectories of violence, African-American males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.90 0.45-8.09 1.27 0.30-5.38 Poverty 0.02 0.0-193. 90 0.84 0.0-189.12 Urban Area 1.18 0.602.30 0.86 0.47-1.59 Parental and Peer Influences Parental Involvement 1. 02 0.95-1.10 1.02 0.92-1.13 Parental alcohol use 1. 50 0.78-2.86 2.27* 1.17-4.41 Peer alcohol use 1.30 0.61-2.76 1.38 0.71-2.65 Peer marijuana use 2.43* 1.17-5.03 1.42 0.69-2.93 Individual-level Risk Factors Alcohol use 2.03* 1. 14-3.63 1.58 0.76-3.28 Marijuana use 3.42*** 1. 89-6.22 2.08* 1.03-4.18 Other drug use 1.49 0. 49-4.55 2.04 0.61-6.82 Academic achievement 0. 93 0.74-1.17 0.89 0.68-1.15 Desire to leave home 1. 38 0.98-1.94 1.39 0.93-2.06 Depression 1.06 0.631.79 0.93 1.30-3.18 Violence Group fighting 2.04** 1. 30-3.18 1.85** 1.22-2.83 Baseline violence 3.57** 1.61-7.92 1.47 0.64-3.40 Demographics Age at Baseline 1.01 0.85-1.19 1.04 0.83-1.32 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 **p<0.01 ***p<0.001

PAGE 174

174 Table 4-18. Effects of multip le domains of risk factors on trajectories of violence, African-American males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Parental and Peer Influences Parental Alcohol use 1. 48 0.76-2.892.16* 1.09-4.28 Peer marijuana use 1.23 0.42-3.640.68 0.31-1.49 Individual-level Risk Factors Alcohol use 1.31 0.572.990.92 0.40-2.14 Marijuana use 3.11* 1. 23-7.882.51* 1.02-6.16 Desire to leave home 1. 21 0.81-1.811.12 0.68-1.83 Violence Group fighting 1.90* 1. 17-3.081.97** 1.22-1.36 Note: The Non-Violent trajec tory group serves as the reference category. All analyses are controlling for age. *p<0.05 **p<0.01

PAGE 175

175 Table 4-19. Effects of multip le domains of risk factors on trajectories of violence adjusted for baseline, African-American males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Parental and Peer Influences Parental Alcohol use 1.33 0.67-2.63 2.22* 1.15-4.29 Peer marijuana use 1.33 0. 42-4.19 0.67 0.31-1.45 Individual-level Risk Factors Alcohol use 1.16 0.532.55 0.94 0.40-2.17 Marijuana use 2.78* 1.126.89 2.59* 1.08-6.23 Desire to leave home 1.16 0.78-1.75 1.13 0.68-1.86 Violence Group fighting 1.66 0.96-2.87 2.05** 1.23-3.41 Baseline violence 2.56* 1. 16-5.63 0.86 0.37-1.99 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age. *p<0.05 **p<0.01

PAGE 176

176 Figure 4-5. Test of the mediation pathway (based on bivariate analysis), AfricanAmerican males. Social/Contextual Risks Parental involvement Peer alcohol and marijuana use Violent Trajectory Membership a b c Alcohol, marijuana, and other drug use Academic achievement Desire to leave home Depression Group fighting Baseline Violence Individual-Level Risks

PAGE 177

177 Table 4-20. Regression models testing t he association between parent and peer level variables and individual-level risk and protective factors, African-American males. Mediators Alcohol Use Marijuana Use Desire to leave home Group Fighting Baseline Violence OR OR OR OR OR Parental and Peer Influences Parental Alcohol Use 0.96 1.39 1.59 1.15 1.55 Peer marijuana use 2.52** 12.09*** 2.00** 4.02** 1.69+ Note: All analyses are controlling for age. +p<0.10 **p<0.01 ***p<0.001

PAGE 178

178 Table 4-21. Mediated effect of peer marij uana use on violence trajectories, AfricanAmerican males. Mediator Indirect Effect (ab )zSE Percent Mediated Peer marijuana use Alcohol Use 0.21 2.68** 0.08 10.8 Marijuana Use 0.47 1.99* 0.23 24.9 Desire to leave home 0.21 3.53***0.06 11.0 Group Fighting 0.32 2.05* 0.16 16.9 Baseline Violence 0.20 2.40* 0.08 10.7 Total 74.3 Notes: All models are adjusted for age. (a): These mediated effects were generated in accordance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhs h, & Veblen-Mortenson (2001). The percent mediation was generated using the formula: [(a*b/(a*b + c) ] (MacKinnon, 2008). (b): Indirect effects are not directly comparable across variables. Percent mediation is comparable across variables and groups of variables. *p<0.05 **p<0.01 ***p<0.001

PAGE 179

179 Table 4-22. Post-hoc description (means and proportions) of adolescents who were violent at Wave I, African-American males. Violence at Baseline Violent NonViolent p Community-level Racial Dispersion a 0.43 0.42 0.702 Poverty a 0.13* 0.14 0.032 Urban Area 0.59 0.54 0.377 Parental and Peer Influences Parental Involvement a 4.72 5.06 0.490 Parental alcohol use 0.54 0.45 0.205 Peer alcohol use 0.62** 0.45 0.006 Peer marijuana use 0.53* 0.37 0.018 Individual-level Risk Factors Alcohol use 0.66***0.39 <0.001 Marijuana use 0.51***0.27 <0.001 Other drug use 0.07 0.05 0.485 Academic achievement 0.62 0.72 0.081 Desire to leave home 0.51 0.37 0.082 Depression 0.35 0.36 0.806 Violence Group fighting 0.47** 0.25 0.006 Demographics Age a 15.86** 15.31 0.007 Note: Participants were considered violent at baseline if they reported any of the violence items that were used to estima te violence trajectories: shot or stabbed someone, used knife or gun in a fight, or hurt someone badly enough to need care from a doctor or nurse. Wave IV weight ing variable used for these analyses. a Mean is reported. *p<0.05 **p<0.01 ***p<0.001

PAGE 180

180 Table 4-23. Bivariate effects between risk/protec tive factors and trajectories of violence, African-American females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.62 0.46-5.700.83 0.27-2.60 Poverty 0.12 0.386. 48 0.15 0-988.41 Urban Area 1.25 0.632.481.11 0.62-1.98 Parental and Peer Influences Parental Involvement 1. 04 0.97-1.120.99 0.89-1.09 Parental alcohol use 0. 84 0.42-1.681.19 0.70-2.01 Peer alcohol use 1.42 0.83-2.421.01 0.66-1.54 Peer marijuana use 2.44 ** 1.40-4.270.95 0.58-1.53 Individual-level Risk Factors Alcohol use 1.97* 1. 60-3.340.94 0.58-1.52 Marijuana use 2.68*** 1. 66-4.310.87 0.45-1.66 Other drug use 2.01 0. 58-6.910.69 0.23-2.11 Academic achievement 0.77 ** 0.64-0.930.99 0.79-1.27 Desire to leave home 1. 21 0.94-1.570.94 0.70-1.28 Depression 1.25 0.732.131.13 0.65-1.97 Violence Group fighting 2.59*** 1. 83-3.640.83 0.46-1.49 Baseline violence 5.53*** 3.52-8.681.26 0.65-2.43 Demographics Age at Baseline 0.98 0.81-1.181.03 0.87-1.21 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 **p<0.01 ***p<0.001

PAGE 181

181 Table 4-24. Effects of multip le domains of risk factors on trajectories of violence, African-American females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Peer Influences Peer marijuana use 1.43 0.75-2.731.07 0.59-1.92 Individual-level Risk Factors Alcohol use 1.35 0.742.450.89 0.53-1.48 Marijuana use 1.32 0.712.450.86 0.43-1.71 Academic achievement 0.82 0.65-1.020.95 0.77-1.18 Violence Group fighting 2.18*** 1. 46-3.250.83 0.45-1.52 Note: The Non-Violent trajec tory group serves as the reference category. All analyses are controlling for age. ***p<0.001

PAGE 182

182 Table 4-25. Effects of multiple domains of risk factors on vi olence, adjusted for baseline, African-American females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Peer Influences Peer marijuana use 1.40 0.74-2.661.06 0.59-1.91 Individual-level Risk Factors Alcohol use 1.16 0.63-2.150.87 0.51-1.46 Marijuana use 1.11 0.54-2.270.82 0.39-1.70 Academic achievement 0.84 0.66-1.060.96 0.77-1.19 Violence Group Fighting 1.79** 1.16-2.770.78 0.42-1.45 Baseline violence 3.14*** 1.79-5.621.42 0.71-2.86 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age. **p<0.01 ***p<0.001

PAGE 183

183 Figure 4-6. Test of the mediation pathway (based on bivariate analysis), AfricanAmerican females. Social/Contextual Risks Peer marijuana use Violent Trajectory Membership a b c Alcohol and marijuana use Academic achievement Group fighting Baseline Violence Individual-Level Risks

PAGE 184

184 Table 4-26. Regression models testing the association between peer level variables on individual-level risk and protective fa ctors, African-American females. Mediators Alcohol Use Marijuana Use Academic Achievement Group Fighting Baseline Violence OR OR OR OR OR Peer Influences Peer marijuana use 3.80*** 17.62*** 0.83* 3.47*** 2.93*** Note: All analyses are controlling for age. *p<0.05 ***p<0.001

PAGE 185

185 Table 4-27. Mediated effect of peer marij uana use on violence trajectories, AfricanAmerican females. Mediator Indirect Effect (ab ) zSE Percent Mediated Peer marijuana use Alcohol Use 0.195 3.67*** 0.05 13.9 Marijuana Use 0.269 2.99** 0.09 19.1 Academic Achievement 0.038 3.15** 0.01 2.7 Group Fighting 0.238 3.12** 0.08 16.9 Baseline Violence0.292 2.99** 0.10 20.8 Total 73.5 Notes: All models are adjusted for age. (a): These mediated effects were generated in accordance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhs h, & Veblen-Mortenson (2001). The percent mediation was generated using the formula: [(a*b/(a*b + c) ] (MacKinnon, 2008). (b): Indirect effects are not directly comparable across variables. Percent mediation is comparable across variables and groups of variables. **p<0.01 ***p<0.001

PAGE 186

186 Table 4-28. Post-hoc description (means and proportions) of adolescents who were violent at Wave I, African-American females. Violence at Baseline Violent NonViolent p Community-level Racial Dispersion a 0.41 0.44 0.890 Poverty a 0.13 0.14 0.170 Urban Area 0.64 0.55 0.087 Parental and Peer Influences Parental Involvement a 5.49 4.85 0.105 Parental alcohol use 0.57* 0.44 0.029 Peer alcohol use 0.72*** 0.49 <0.001 Peer marijuana use 0.56*** 0.32 <0.001 Individual-level Risk Factors Alcohol use 0.70*** 0.47 <0.001 Marijuana use 0.49*** 0.19 <0.001 Other drug use 0.11** 0.03 0.001 Academic achievement 0.64* 0.75 0.041 Desire to leave home 0.58* 0.41 0.013 Depression 0.68*** 0.47 <0.001 Violence Group fighting 0.55*** 0.16 <0.001 Demographics Age at Baseline 15.06* 15.39 0.048 Note: Participants were considered violent at baseline if they reported any of the violence items that were used to estima te violence trajectories: shot or stabbed someone, used knife or gun in a fight, or hurt someone badly enough to need care from a doctor or nurse. Wave IV weighti ng variable used for these analyses. a Mean is reported. *p<0.05 **p<0.01 ***p<0.001

PAGE 187

187 Table 4-29. Bivariate effects between risk/protec tive factors and trajectories of violence, Hispanic males. Trajectory Group Desistors Escalators OR OR Community-level Racial Dispersion 1.62 0.52-5.02 1.99 0.46-8.54 Urban Area 1.43 0.683.00 1.57 0.69-3.59 Parental and Peer Influences Parental Involvement 1. 03 0.94-1.11 0.90 0.79-1.02 Parental alcohol use 1. 14 0.60-2.17 0.55+ 0.28-1.05 Peer alcohol use 2.00* 1.06-3.76 1.01 0.41-2.45 Peer marijuana use 2.36* 1.23-4.51 1.82 0.72-4.57 Individual-level Risk Factors Alcohol use 3.09*** 1. 72-5.57 1.31 0.60-2.86 Marijuana use 3.49*** 1. 89-6.46 2.49+ 0.99-6.21 Other drug use 4.78*** 2. 63-8.70 1.00 0.24-4.18 Academic achievement 0.97 0.78-1.21 0.81 0.58-1.12 Desire to leave home 1. 19 0.83-1.71 1.02 0.52-1.99 Depression 1.51 0.83-2.77 2.39* 1.19-4.78 Acculturation 0.71 0. 34-1.48 0.67 0.26-1.75 Violence Group fighting 2.38*** 1.61-3.50 2.23** 1.25-3.96 Baseline violence 7.13*** 3.27-15.56 3.61** 1.40-9.26 Demographics Age at Baseline 1.07 0.92-1.23 1.14 0.84-1.54 1s t Generation Immigrant --2n d Generation US-Born 1.68 0.78-3.62 0.68 0.23-1.98 3r d Generation US-Born and Beyond 1.86 0.82-4.23 1.26 0.43-3.64 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 ***p<0.001

PAGE 188

188 Table 4-30. Effects of multip le domains of risk factors on violence, Hispanic males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Parental and Peer Influences Parental alcohol use 1. 05 0.52-2.12 0.48 0.21-1.12 Peer alcohol use 0.84 0.34-2.04 0.55 0.18-1.64 Peer marijuana use 0.79 0.31-2.04 1.25 0.45-3.47 Individual-level Risk Factors Alcohol use 1.54 0.504.69 0.81 0.34-1.92 Marijuana use 2.45 0.976.17 2.49* 1.03-6.01 Other drug use 1.49 0. 86-2.55 0.48 0.16-1.45 Depression 0.95 0.46-1.91 2.21* 1.10-4.44 Violence Group fighting 1.75 1.10-2.80 1.97* 1.10-3.57 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age and generation. *p<0.05

PAGE 189

189 Table 4-31. Effects of multip le domains of risk factors on trajectories of violence, adjusted for baseline violence, Hispanic males. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Parental and Peer Influences Parental alcohol use 1. 14 0.59-2.21 0.51 0.23-1.14 Peer alcohol use 0.77 0.30-1.97 0.50 0.17-1.51 Peer marijuana use 0.88 0.35-2.18 1.32 0.44-3.95 Individual-level Risk Factors Alcohol use 1.49 0.53 -4.24 0.82 0.32-2.10 Marijuana use 2.07 0.87 -4.93 2.13 0.87-5.22 Other drug use 1.35 0. 82-2.24 0.43 0.14-1.33 Depression 1.01 0.48-2.12 2.39* 1.24-4.62 Violence Group fighting 1.33 0.88-2.00 1.59 0.88-2.90 Baseline violence 5.64** 2. 18-14.55 3.56* 1.3-9.62 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age and generation. *p<0.05 **p<0.01

PAGE 190

190 Figure 4-7. Test of the mediation pathway (based on bivariate analysis), Hispanic males. Social/Contextual Risks Parental alcohol use Peer alcohol and marijuana use Violent Trajectory Membership a b c Alcohol, marijuana, and other drug use Depression Group fighting Baseline Violence Individual-Level Risks

PAGE 191

191 Table 4-32. Regression models testing t he association between parent and peer level variables on individual-level risk and pr otective factors, Hispanic males. Mediators Alcohol Use MJ Use Other Drug Use Depression Group Fighting Baseline Violence OR OR OR OR OR OR Parental and Peer Influences Parental alcohol use 0.97 0. 95 1.36 1.21 0.71 0.81 Peer alcohol use 7.74*** 5.51 *** 7.74** 1.81+ 7.79*** 2.66** Peer marijuana use 3.08** 11.19*** 6.12*** 1.37 5.00*** 2.55** Note: All analyses are controlling for age and generation. +p<0.10 **p<0.01 ***p<0.001

PAGE 192

192 Table 4-33. Mediated effect of peer-level variables on violence trajectories, Hispanic males. Mediator Indirect Effect (ab ) zSE Percent Mediated Peer alcohol use Alcohol Use 0.376 2.17* 0.18 14.5 Marijuana Use 0.486 1.96+ 0.25 18.7 Other Drug Use 0.427 1.52 0.28 16.4 Depression 0.063 2.32* 0.08 2.4 Group Fighting 0.443 2.20* 0.20 17.1 Baseline Violence0.464 2.17* 0.21 17.8 Total 86.9 Peer marijuana use Alcohol Use 0.219 2.26* 0.10 10.0 Marijuana Use 0.482 2.12* 0.23 21.9 Other Drug Use 0.329 1.96+ 0.17 14.9 Depression 0.124 2.49* 0.05 5.6 Group Fighting 0.176 0.82 0.21 8.0 Baseline Violence0.404 2.21* 0.18 18.4 Total 79.0 Notes: All models are adjusted for age and generation. (a): These mediated effects were generated in accordance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhs h, & Veblen-Mortenson (2001). The percent mediation was generated using the formula: [(a*b/(a*b + c) ] (MacKinnon, 2008). (b): Indirect effects are not directly comparable across variables. Percent mediation is comparable across variables and groups of variables. +p<0.10 *p<0.05

PAGE 193

193 Table 4-34. Post-hoc description (means and proportions) of adolescents who were violent at Wave I, Hispanic males. Violence at Baseline Violent NonViolent p Community-level Racial Dispersion a 0.54 0.49 0.994 Poverty a 0.13 0.14 0.214 Urban Area 0.83 0.79 0.452 Parental and Peer Influences Parental Involvement a 4.81 5.32 0.344 Parental alcohol use 0.49 0.51 0.892 Peer alcohol use 0.73** 0.49 0.002 Peer marijuana use 0.52** 0.28 0.002 Individual-level Risk Factors Alcohol use 0.76*** 0.55 <0.001 Marijuana use 0.59*** 0.25 <0.001 Other drug use 0.29*** 0.07 <0.001 Academic achievement 0.49 0.62 0.056 Desire to leave home 0.38 0.23 0.076 Depression 0.44 0.36 0.339 Acculturation 0.27* 0.40 0.032 Violence Group fighting 0.58*** 0.19 <0.001 Demographics Age at Baseline a 15.62 15.3 0.313 1s t Generation Immigrant 0.20* 0.36 0.033 2n d Generation US-Born 0.13 0.14 3r d Generation US-Born and Beyond 0.66 0.50 Note: Participants were considered violent at baseline if they reported any of the violence items that were used to estima te violence trajectories: shot or stabbed someone, used knife or gun in a fight, or hurt someone badly enough to need care from a doctor or nurse. Wave IV weighti ng variable used for these analyses. a Mean is reported. *p<0.05 **p<0.01 ***p<0.001

PAGE 194

194 Table 4-35. Bivariate effects between risk/protec tive factors and trajectories of violence, Hispanic females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.59 0.46-5.46 4.48 0.59-34.25 Urban Area 0.80 0.36-1. 80 0.31* 0.12-0.80 Parental and Peer Influences Parental Involvement 0.91 0.80-1.03 0.96 0.83-1.12 Parental alcohol use 0.70 0.32-1.53 0.90 0.35-2.31 Peer alcohol use 3.37** 1.63-6.98 2.06 0.87-4.89 Peer marijuana use 2.52* 1.16-5.47 1.31 0.51-3.37 Individual-level Risk Factors Alcohol use 2.96** 1.346.54 1.84 0.74-4.57 Marijuana use 1.76 0.823.76 0.85 0.37-1.96 Other drug use 0.78 0.321.93 0.82 0.25-2.66 Academic achievement 1.05 0.80-1.36 1.00 0.76-1.33 Desire to leave home 1.41 0.92-2.16 1.19 0.54-2.63 Depression 1.43 0.64-3.19 0.90 0.43-1.88 Acculturation 0.94 0.442.01 1.72 0.94-3.15 Violence Group fighting 1.63* 1.032.56 1.10 0.60-2.04 Baseline violence 3.17** 1. 33-7.53 2.34 0.55-9.99 Demographics Age at Baseline 0.77* 0. 61-0.97 1.13 0.88-1.46 1s t Generation Immigrant --2n d Generation US-Born 0.62 0.23-1.72 0.28** 0.11-0.69 3r d Generation US-Born and Beyond 0.93 0.39-2.20 1.00 0.35-2.81 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 **p<0.01

PAGE 195

195 Table 4-36. Effects of multip le domains of risk factors on trajectories of violence, Hispanic females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community and Peer Influences Urban neighborhood 1.25 0.493.19 0.29** 0.13-0.68 Peer alcohol use 2.31 0.82-6.55 2.14 0.72-6.35 Peer marijuana use 1.57 0.62-3.96 0.73 0.31-1.68 Individual-level Risk Factors Alcohol use 2.40 0.946.14 1.58 0.76-3.29 Acculturation 1.40 0.513.88 7.04* 1.50-32.98 Violence Group fighting 1.26 0.792.00 0.92 0.55-1.53 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age and generation. *p<0.05 **p<0.01

PAGE 196

196 Table 4-37. Effects of multip le domains of risk factors on trajectories of violence, adjusted for baseline violence, Hispanic females. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community and Peer Influences Urban neighborhood 1.33 0.523.450.30** 0.13-0.67 Peer alcohol use 2.27 0.83-6.262.13 0.72-6.35 Peer marijuana use 1.48 0.59-3.760.67 0.29-1.55 Individual-level Risk Factors Alcohol use 2.30 0.876.111.51 0.76-3.02 Acculturation 1.33 0.493.586.51* 1.42-29.98 Violence Group fighting 1.11 0.661.880.82 0.51-1.32 Baseline Violence 1.97 0. 67-5.771.79 0.60-5.36 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age and generation. *p<0.05 **p<0.01

PAGE 197

197 Figure 4-8. Test of the mediation pathway (based on bivariate analysis), Hispanic females. Social/Contextual Risks Urban neighborhood Peer alcohol and marijuana use Violent Trajectory Membership a b c Alcohol use Acculturation Group fighting Baseline Violence Individual-Level Risks

PAGE 198

198 Table 4-38. Regression models testing t he association between communityand peer level variables on individual-level risk and protective factors, Hispanic females. Mediators Alcohol Use Acculturation Group Fighting Baseline Violence OR OR OR OR Community and Peer Influences Urban neighborhood 0.56 1.32 0.65 0.51 Peer alcohol use 6.61*** 0.86 3.64** 3.65** Peer marijuana use 6.15*** 0.72 2.90** 3.62*** Note: All analyses are controlling for age and generation. **p<0.01 ***p<0.001

PAGE 199

199 Table 4-39. Mediated effect of parentand pee r-level variables on violence trajectories, Hispanic females. Mediator Indirect Effect (ab ) zSE Percent Mediate d Urban neighborhood Alcohol Use 0.668 2.46* 0.03 24.9 Group Fighting 0.043 2.64** 0.02 15.9 Baseline Violence 0.051 1.63 0.03 18.9 Total 59.7 Peer Alcohol Use Alcohol Use 0.316 2.65** 0.12 23.6 Group Fighting 0.153 2.35* 0.07 11.5 Baseline Violence 0.231 1.86+ 0.12 17.3 Total 52.3 Peer marijuana use Alcoho l Use 0.356 2.38* 0.15 28.1 Group Fighting 0.168 2.45* 0.07 13.3 Baseline Violence 0.275 1.94+ 0.14 21.7 Total 63.0 Notes: All models are adjusted for age and generation. (a): These mediated effects were generated in accordance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhs h, & Veblen-Mortenson (2001). The percent mediation was generated using the formula: [(a*b/(a*b + c) ] (MacKinnon, 2008). (b): Indirect effects are not directly comparable across variables. Percent mediation is comparable across variables and groups of variables. +p<0.10 *p<0.05 **p<0.01

PAGE 200

200 Table 4-40. Bivariate effects between risk/protec tive factors and trajectories of violence, Asians. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 1.65 0.20-13.38 1.11 0.14-8.53 Poverty 0.93 0.0-9.27 6.53 0-99.27 Urban Area 0.26* 0. 08-0.89 0.68 0.24-1.91 Parental and Peer Influences Parental Involvement 1. 06 0.91-1.24 1.05 0.91-1.21 Parental alcohol use 0. 84 0.25-2.78 1.22 0.37-4.09 Peer alcohol use 1.71 0.46-6.32 1.86 0.87-3.96 Peer marijuana use 1.22 0.36-4.08 1.31 0.52-3.29 Individual-level Risk Factors Alcohol use 2.65 0. 78-9.05 0.87 0.44-1.70 Marijuana use 1.96 0. 65-5.85 1.55 0.52-4.61 Other drug use 3.63* 1.25-10.54 0.81 0.21-3.11 Academic achievement 0.59 ** 0.43-0.80 0.90 0.61-1.33 Desire to leave home 1. 85 0.91-3.74 0.84 0.51-1.39 Depression 0.80 0.282.29 0.97 0.45-2.11 Violence Group fighting 4.45*** 1. 98-9.99 0.87 0.43-1.77 Baseline violence 3.70* 1.03-13.25 1.08 0.24-4.78 Demographics Age at Baseline 0.72 0.51-1.03 0.91 0.73-1.14 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 **p<0.01 ***p<0.001

PAGE 201

201 Table 4-41. Effects of multiple domains of risk factors on trajec tories of violence, Asians. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-Level Urban Neighborhood 0.45 0. 14-1.42 0.75 0.29-1.95 Individual-level Risk Factors Other drug use 1.39 0. 84-2.30 0.86 0.37-1.96 Academic achievement 0. 74 0.50-1.12 0.90 0.58-1.40 Desire to leave home 1. 57 0.96-2.59 0.86 0.53-1.41 Violence Group fighting 3.08*** 1. 67-5.67 0.84 0.38-1.87 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age and gender. *p<0.05 **p<0.01 ***p<0.001

PAGE 202

202 Table 4-42. Effects of multip le domains of risk factors on trajectories of violence, adjusted for baseline, Asians. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-Level Urban Neighborhood 0.45 0. 14-1.42 0.75 0.29-1.94 Individual-level Risk Factors Other drug use 1.39 0. 82-2.37 0.86 0.36-2.04 Academic achievement 0.74 0.50-1.11 0.90 0.59-1.39 Desire to leave home 1. 57 0.95-2.60 0.86 0.53-1.41 Violence Group fighting 3.09***1. 71-5.61 0.84 0.35-2.01 Baseline violence 0.98 0. 31-3.09 1.00 0.23-4.40 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age. ***p<0.001

PAGE 203

203 Figure 4-9. Test of the mediation pathway (based on bivariate analysis), Asians. Social/Contextual Risks Urban neighborhood Violent Trajectory Membership a b c Illegal drug use Academic achievement Desire to leave home Group Fighting Baseline violence Individual-Level Risks

PAGE 204

204 Table 4-43. Regression models testing t he association between parent and peer level variables and individual-level ri sk and protective factors, Asians. Mediators Other Drug Use Desire to Leave Home Academic Achievement Group Fighting Baseline Violence OR OR OR OR OR Contextual Influences Urban Neighborhood 0.47 0.77 1.46 0.54+ 0.75 Note: All analyses are controlling for age and gender. +p<0.10

PAGE 205

205 Table 4-44. Mediated effect of contextual variables on violence trajectories, Asians. Mediator Indirect Effect (ab ) zSE Percent Mediated Urban Neighborhood Group Fighting 0.135 1.97* 0.03 29.4 Total 29.4 Notes: All models are adjusted for age. (a): These mediated effects were generated in accordance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhs h, & Veblen-Mortenson (2001). The percent mediation was generated using the formula: [(a*b/(a*b + c) ] (MacKinnon, 2008). (b): Indirect effects are not directly comparable across variables. Percent mediation is comparable across variables and groups of variables. *p<0.05

PAGE 206

206 Table 4-45. Post-hoc description (means and proportions) of adolescents who were violent at Wave I, Asians. Violence at Baseline Violent NonViolent p Community-level Racial Dispersion a 0.59 0.02 0.126 Poverty a 0.12 0.006 0.420 Urban Area 0.65 0.69 0.529 Parental and Peer Influences Parental Involvement a 6.46 5.96 0.582 Parental alcohol use 0.41 0.36 0.679 Peer alcohol use 0.61 0.43 0.071 Peer marijuana use 0.57** 0.25 0.001 Individual-level Risk Factors Alcohol use 0.65* 0.46 0.048 Marijuana use 0.52*** 0.21 <0.001 Other drug use 0.17** 0.06 0.004 Academic achievement 0.62** 0.83 0.002 Desire to leave home 0.40 0.25 0.795 Depression 0.58 0.47 0.355 Violence Group fighting 0.52*** 0.16 <0.001 Demographics Age a 15.36 15.37 0.966 Note: Participants were considered violent at baseline if they reported any of the violence items that were used to estima te violence trajectories: shot or stabbed someone, used knife or gun in a fight, or hurt someone badly enough to need care from a doctor or nurse. Wave IV weight ing variable used for these analyses. a Mean is reported. *p<0.05 **p<0.01 ***p<0.001

PAGE 207

207 Table 4-46. Bivariate effects between risk/protec tive factors and trajectories of violence, Native Americans. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 5.89** 1.74-19.92 2.18 0.37-12.57 Poverty 1.42 0-3.68 4.28 0.0-9.69 Urban Area 0.86 0.362.04 0.48 0.17-1.36 Parental and Peer Influences Parental Involvement 0. 89 0.78-1.01 0.93 0.80-1.09 Parental alcohol use 1. 10 0.53-2.26 0.47 0.14-1.51 Peer alcohol use 2.08 0.69-6.21 2.23 0.81-6.12 Peer marijuana use 2.57* 1.00-6.61 3.25* 1.25-8.44 Individual-level Risk Factors Alcohol use 2.65 0.87-8.13 2.84** 1.35-5.98 Marijuana use 3.19* 1.30-7.79 2.54 0.88-7.34 Other drug use 3.04* 1. 16-7.94 4.68** 1.48-14.73 Academic achievement 0.90 0.65-1.24 0.84 0.61-1.17 Desire to leave home 1.70* 1.09-2.63 1.34 0.64-2.81 Depression 2.29* 1.15-4.55 2.79* 1.05-7.44 Violence Group fighting 1.92* 1.17-3.14 1.26 0.68-2.32 Baseline violence 4.67** 1.65-13.27 2.69 0.91-7.97 Demographics Age at Baseline 0.93 0.75-1.17 0.77 0.54-1.11 Note: The Non-Violent trajectory gr oup serves as the reference category. *p<0.05 **p<0.01

PAGE 208

208 Table 4-47. Effects of multiple domains of risk factors on trajectories of violence, Native Americans. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 3.45 0.77-15.402.40 0.58-9.86 Parental and Peer Influences Parental Involvement 0. 95 0.83-1.09 1.00 0.86-1.16 Peer marijuana use 1.94 0.72-5.27 3.48* 1.14-10.16 Individual-level Risk Factors Alcohol use 1.15 0.353.75 1.27 0.45-3.51 Marijuana use 1.20 0.423.37 0.74 0.24-2.24 Other drug use 0.88 0.332.06 1.82* 1.07-3.11 Depression 1.56 0.643.76 2.05 0.84-5.00 Desire to leave home 1. 59 0.91-2.80 1.31 0.66-2.60 Violence Group fighting 1.52 0.623.69 0.88 0.45-1.72 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age and gender. *p<0.05

PAGE 209

209 Table 4-48. Effects of multip le domains of risk factors on trajectories of violence, adjusted for baseline, Native Americans. Trajectory Group Desistors Escalators OR 95% CI OR 95% CI Community-level Racial Dispersion 3.88 0.91-16.612.74 0.67-11.22 Parental and Peer Influences Parental Involvement 0. 93 0.80-1.08 0.98 0.84-1.14 Peer marijuana use 1.91 0.70-5.27 2.54 1.08-11.59 Individual-level Risk Factors Alcohol use 1.02 0.313.34 1.89* 1.37-3.23 Marijuana use 1.19 0.403.56 0.76 0.25-2.29 Other drug use 0.83 0. 32-2.12 1.92* 1.17-3.14 Depression 1.42 0.593.40 1.89 0.74-4.80 Desire to leave home 1. 57 0.90-2.75 1.24 0.61-2.53 Violence Group fighting 1.21 0.403.61 0.65 0.32-1.32 Baseline Violence 2.96 0. 70-12.493.14 0.92-10.71 Note: The Non-Violent trajectory group serv es as the reference category. All analyses are controlling for age and gender. *p<0.05

PAGE 210

210 Figure 4-10. Test of the m ediation pathway (based on bi variate analysis), Native Americans. Social/Contextual Risks Racial dispersion Parental involvement Peer marijuana use Violent Trajectory Membership a b c Alcohol, marijuana, and other drug use Desire to leave home Depression Group Fighting Baseline violence Individual-Level Risks

PAGE 211

211 Table 4-49. Regression models testing t he association between community-level and parent and peer level variables on indivi dual-level risk and protective factors, Native Americans. Mediators Alcohol Use MJ Use Other Drug Use Depression Desire to Leave Home Group Fighting Baseline Violence OR OR OR OR OR OR OR Contextual Influences Racial Dispersion 1.53 7.25* 3.27 1.17 0.99 1.89 1.58 Parental and Peer Influences Parental Involvement 0.93 0.88* 0.96 0.97 0.89* 1.01 1.04 Peer marijuana use 5.83*** 15.55*** 8.46*** 2.58** 1.38 2.15* 1.87 Note: All analyses are controlling for age and gender. *p<0.05 **p<0.01 ***p<0.001

PAGE 212

212 Table 4-50. Mediated effect of contextual, parentand peer-level variables on violence trajectories, Native Americans. Mediator Indirect Effect (ab ) z SE Percent Mediated Racial Dispersion Alcohol Use 0.858 1.28 0.07 5.4 Marijuana Use 0.421 0.95 0.44 26.5 Other Drug Use 0.273 1.15 0.24 17.2 Depression 0.105 1.42 0.07 6.6 Desire to leave home 0.076 1.44 0.05 4.8 Group Fighting 0.093 1.65+ 0.06 5.8 Total 66.3 Parental Involvement Alcohol Use 0.284 2.56* 0.11 10.4 Marijuana Use 0.346 2.48* 0.14 12.6 Other Drug Use 0.367 5.24*** 0.07 13.4 Depression 0.355 3.12** 0.11 12.9 Desire to leave home 0.288 4.13*** 0.07 10.5 Group Fighting 0.237 2.54* 0.09 8.7 Total 68.6 Peer Marijuana Use Alcohol Use 0.238 1.89+ 0.13 10.6 Marijuana Use 0.331 1.41 0.24 14.7 Other Drug Use 0.413 1.75+ 0.24 18.4 Depression 0.251 2.08* 0.12 11.2 Desire to leave home 0.154 2.17* 0.07 6.8 Group Fighting 0.137 1.85+ 0.07 8.1 Total 67.9 Notes: All models are adjusted for age and gender. (a): These mediated effects were generated in acco rdance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhsh, & Veblen-Mortenson (2001). The perce nt mediation was generated using the formula: [(a*b/(a*b + c)] (MacKinnon, 2008). (b): Indirect effects are not direct ly comparable across variables. Percent mediation is comparable across variables and groups of variables. +p<0.10 *p<0.05 **p<0.01 ***p<0.001

PAGE 213

213 Table 4-51. Post-hoc description (means and proportions) of adolescents who were violent at Wave I, Native Americans. Violence at Baseline Violent NonViolent p Community-level Racial Dispersion a 0.43 0.40 0.584 Poverty a 0.13 0.13 0.493 Urban Area 0.53 0.58 0.641 Parental and Peer Influences Parental Involvement a 5.52 5.28 0.710 Parental alcohol use 0.43 0.54 0.231 Peer alcohol use 0.81* 0.65 0.036 Peer marijuana use 0.61 0.47 0.216 Individual-level Risk Factors Alcohol use 0.80* 0.62 0.020 Marijuana use 0.54 0.35 0.131 Other drug use 0.27 0.18 0.359 Academic achievement 0.44 0.60 0.069 Desire to leave home 0.44 0.29+ 0.089 Depression 0.54 0.36 0.095 Violence Group fighting 0.71*** 0.22 <0.001 Demographics Age at Baseline a 15.21 14.83 0.195 Note: Participants were considered violent at baseline if they reported any of the violence items that were used to estima te violence trajectories: shot or stabbed someone, used knife or gun in a fight, or hurt someone badly enough to need care from a doctor or nurse. Wave IV weighti ng variable used for these analyses. a Mean is reported. *p<0.05 ***p<0.001

PAGE 214

214 CHAPTER 5 RISK FACTORS ASSOCIATED WITH RACIAL/ETHNIC DIFFERENCES IN TRAJECTORIES OF AGGRESSION IN A L O NGITUDINAL SAMPLE OF HIGH-RISK, URBAN YOUTH Racial/ethnic disparities are present in delinquency, and evidence suggests that certain groups (e.g., African-Americans and Hispanics) are at greater risk for delinquency and aggression compared to Whites The purpose of this study is to estimate trajectories of aggression using a longitudinal sample of urban adolescents by racial/ethnic subgroups, testing multiple dom ains of risk factors to evaluate which predictors differentiate profiles of aggre ssion. Methods. Participants included a multiethnic, urban sample of 4,188 adolescents follow ed from ages 11 to 14. Trajectories of aggression were estimated for AfricanAmericans and Hispanics separately, and participants were assigned to each trajectory group using latent traj ectory modeling. Multinomial logistic regression procedures were used to evaluate the effect of multiple domains of risk and protective factors in st ages (e.g., community -level, parentand peer-level, and individual-level) to understand the predictors of me mbership in most aggressive trajectory groups. Mediation analyses were conducted to further evaluate the direct and indirect e ffect of community-, parental and peer-level variables on aggression trajectories. Results. Four groups of aggression trajectories were identified for all subgroups. Among Hispanics, the four groups included: 1) low-aggression, 2) desistors, 3) escalators, and 4) consistent ly aggressive. Among African-Americans, the four groups included: 1) lo w-Aggression, 2) escalators 2) moderate-consistent aggression, and 4) consistent aggression. Group fighting significantly predicted aggression above and beyond baseline aggr ession for both Hispanics and AfricanAmericans who were consistently aggressive. A number of differences in the multiple

PAGE 215

215 domains of risk and protective factors emerged between groups. Conclusion. Group fighting is a significant predictor of vi olence beyond baseline violence for the most aggressive groups of aggressive adolesce nts across racial/ethnic group, beyond baseline aggression. Cont extual variables (peer al cohol use, adult alcohol consumption, and home access to alcohol) appear to increase risk for aggression differentially by subgroup. Background Each year, nearly 700, 000 adolescents and y oung adults (10-24) are treated in the emergency room for injuries re lated to violent activity (CDC, 2009). Evidence suggests that adolescents who engage in delinquent behavior are more likely to engage in other high-risk activities (e.g., alcohol and ot her drug use, dropping out of school, gun ownership, gang membership, risky sexual activity and familial independence) (Thornberry, Huizinga, & Loeber, 1995; CDC, 2009; 2010) and increase their risk of health-related consequences (including serious injury and death) (Conseur, Rivara, & Emanuel, 1997; Farrington & Loeber, 2000). The evidence is clear that individualand family-level characteristics increase the risk for violent and aggressive behavior. Fo r example, neurologic al deficiencies and cognitive impairments (Moffitt et al., 2001), lo w IQ, hyperactivity, difficulty concentrating at school, beliefs and attitudes favorable to violence, antisocial behavior, and impulsivity have been consistently associated with violent behavior at the individual-level (Hawkins et al., 2000; Howell, 2009). At the family level, parental crimi nal behavior, child maltreatment, low levels of parental invo lvement, parental attitudes favorable to violence and drug/alcohol use, and separ ation of the parent and child have been identified as risk factors in a recent meta-ana lysis of longitudinal studies of risk factors

PAGE 216

216 for violence (Hawkins et al ., 2000). Each of these factors has been consistently associated with increases in violent activity. Despite the strong evidence in support of some risk factors for violence, other behavioral risk factors within the family and p eer group are less studied. Hawkins et al. (2000) found that delinquent peers and gang membership have been predictive of violent behavior; however, the effect of peer and parental substance use is unclear. Academic failure and dropout has also been a ssociated with violence, but less drastic measures of academic success have not been ev aluated in the empirical literature on violence (Hawkins et al., 2000) Community-level influences such as availability of firearms, exposure to viol ence, and exposure to raci sm in the neighborhood have consistently been linked to violent behavio r (Kaufman, 2005; Reingle, Jennings, Maldonado-Molina, & Canino, 2010). Finally, although many studies have analyzed the multiple domains of risk and protective fa ctors for violent behavior, few have assessed the degree to which contextual variables are m ediated by more proximal variables at the individual-level. Racial and ethnic differences in the preval ence of violence have been identified. For example, Williams et al. (2007) found that self-reported violence initiation rates were higher for African-Americans compared to Whites for major delinquency, violence, and juvenile justice system involvement. Additi onally, Williams et al. (2007) reported African-Americans higher rates of major del inquent and violent acts when compared to Whites. Additionally, McNulty & Bellair (2003) found that AfricanAmericans, Hispanics, and Native Americans have higher involvemen t in serious physical violence compared to White adolescents at ages 15 to 16.

PAGE 217

217 Evidence suggest that these racial differ ences may be partially attributable to differences in the presence of community-l evel risk factors, such as neighborhood poverty and exposure to guns, delinquent peer s, and violence (Sampson, Morenoff, & Roaudenbush, 2005; Farrington & Loeber, 2000). Fa mily and individual-level variables also contribute to violent behavior, incl uding acculturation level among Hispanics (having immigrated to the Unit ed States more recently is a protective factor from violence), parental marital status (liv ing with parents who are not married puts adolescents at risk for violence), lower verbal reading and writing ability, shorter length of residence in the neighborhood, and lower income, as low socioeconomic position has been shown to increase the risk of violence in itiation (Sampson et al ., 2005; Williams et al., 2007). These disparities may be a function of socioeconomic differences between racial/ethnic groups, including income, neighborhood structural characteristics, oppression, cumulative disadvantage within t he family, employment, and education; as well as racial discrimination and bias (Centerwall, 1995; Williams, 1999; Peterson & Krivo, 2005). These findings support t he hypothesis that the racial and ethnic differences in violence may be at least par tially attributable to neighborhood-level and individual-level socioeconomic variables. Although the risk factors above have expl ained some of the racial/ethnic differences in violent behavior, further resear ch is necessary to more clearly delineate racial and ethnic differences (which hav e been accepted as a proxy for larger socioeconomic differences and discriminat ion) in delinquency and violence between these groups. The research on race and ethnic differences on violence has not consistently attributed violent behavior to di fferences in race/ethnicity or socioeconomic

PAGE 218

218 position. For example, Blum and colleagues (2000) found that gender, race, ethnicity, income, and family structure explained no more than ten percent of the variance in violent behavior. This finding indicates that the differential risk factors for violence by racial/ethnic group is more complex than race or socioeconomic position. Therefore, multiple domains of risk factors that differ by race/ethnicity must be further studied to better understand how these factors c ontribute to violent behavior. Criminological theory provides insight as to the rationale behind racial and ethnic differences in the prevalence of violent offendi ng. First, social learning theory posits that individuals learn to engage in criminal behavior by observing those around them (Akers, 1973). This theoretical framework is compris ed of four central components: excess of definitions favorable to crim inality, association with devi ant peers, reinforcement of criminal behavior, and imitation. When these elements are combined, an individual is more likely to engage in deviant or criminal behavior (Akers, 1973). Using this theory, violent behavior would be more prevalent among racial/ethnic minoriti es if they have more violent peers who teach them to participat e in violence, and this violent behavior is reinforced. Research supports this hypothesis, as African-Americans and Hispanics are more likely to be involved with gangs (Mc Nulty & Bellair, 2003) and have delinquent peers (Stewart, Simons, & Conger, 2002) compared to Whites. Shaw and McKay maintain that the char acteristics of the community facilitate crime, rather than t he individuals within that communi ty. For example, transient individuals are unlikely to watch over their neighbors property, or even become acquainted with their nei ghbors. Therefore, a neighbor could never know if someone is stealing a car out of the driveway, as they do not know who resides in that particular

PAGE 219

219 home. Suburban areas have these types of neighborly ties, and they would recognize when someone does not belong. Using the framework provided by Social Disorganization theory, the la ck of neighborly recognition is one of the reasons why crime is higher in transient, disorganized communities. Sampson, Raudenbush, and Earls (1997) furt her elaborate on this aspect of their theory in defining the collective efficacy of communities. Collective efficacy is the neighborhoods ability to come together to reduce criminal or disorderly behavior. Cohesive, suburban neighborhoods are very lik ely to do this: When there is a problem within the community, member s of the neighborhood organize themselves and remedy the issue. Urban, disadv antaged, disorganized neighborhoods are not likely to have such a mechanism in place to filter out cr iminal behavior among residents. This lack of social control in a community allows crime to multiply. This theory has substantial implications for racial/ethnic differences in violence, as minorities are more likely to reside in these urban, at-risk comm unities (Sampson & Wilson, 2005). To further explore the factors driving ra cial/ethnic differences in delinquency over the life-course (and the risk factors for viol ence over time), this study examined the number and shape of trajectories of aggression among AfricanAmerican and Hispanic urban adolescents, as well as t he direct and indirect effects of multiple domains of risk and protective factors for membership in each trajectory group. This will contribute substantially to the literat ure on disparities and etiol ogy of aggression among high-risk youth. Specifically, I hypot hesize that the trajectories of aggression differ between African-Americans and Hispanic s, and African-Americans are more likely to be involved in high levels of aggression. Finally, I hypothe size that these differences in aggression

PAGE 220

220 are attributable to differences in multiple domains of risk factors, including group fighting. Methods Design Project Northland Chic ago (PNC) was a group-randomized controlled alcohol prevention intervention implemented during 6th through 8th grades in selected Chicago, Illinois schools (Komro et al., 2008; 2004). A ll schools in Chicago were eligible for inclusion if students were enrolled in 5th through 8th grade, had a mobility rate of less than 25%, and had thirty or more students per grade. All m agnet schools were ineligible, as they were less likely to in clude students from a sp ecific neighborhood, and the intervention included a community intervention component. A total of sixty-six schools agreed to participate in the study, and these schools were combined into study units with approx imately two hundred student s per unit. Units were grouped in correspondence with census tr acts, and matched on ethnicity, mobility rate, reading and math scores, and poverty levels within each unit. After matching, units were randomized to treatment condition (30 schools to the intervention, 36 schools served as controls). Five schools withdrew from the study prior to the intervention, leaving 29 intervention and 32 control schools. Baseline surveys were administered in cla ss during the Fall semester of 2002, and three follow-up surveys were conducted duri ng the intervention period (Spring 2003, Spring 2004, Spring 2005). All students who were enrolled in the school during the intervention year were eligible to complete the surveys. During the Fall of 2002, 91% (n=4,259) of eligible students participated in the baseline survey; 94% (n = 4,240) participated in the first follow-up (Spring, 2003); 93% (n = 3,778) completed the second

PAGE 221

221 follow-up (Spring, 2004); and 95% (n = 3,802) of students completed the final middleschool survey (Spring, 2005). Of those pres ent at baseline, the cohort follow-up rate was 61% from Fall 2002 through Spring, 2005. At baseline, parents were surveyed usi ng a sample obtained from the Chicago Public Schools (CPS) address list. When surv eys were administered to students, each student was asked to deliver a modified surv ey to their parent, and parents were given a $25 incentive for returning the survey. Student s were given a $5 gift card to deliver the survey to their parents. Two weeks later, another survey was sent home, and teachers reminded students to prompt thei r parents to return the surv ey. At the end of baseline data collection, a total of 3,250 parents (70% of eligible parents) responded. Participants The sample used in this study includes participants who were present at baseline (6th Grade, Fall) and completed at leas t one additional survey throughout 6th through 8th grades. This cohort consisted 4,188 adolescent s, who were 50.3% male, 12.9% White, 38.8% African-American, and 34.6% Hispanic. The average age in 6th grade was 11.38 (sd = 0.24). Nearly thirty percent of the sa mple reported participati ng in a group fight in 6th grade (28.3%). Due to sample size limitations in the esti mation of trajectory groups, only African-Americans and Hispanics were included in these analyses. Measures The current study used baseline measures as covariates to examine trajectories of violence in the cohort of students who participated in the baseline survey and at least one follow-up conducted during PNC data collection.

PAGE 222

222 Group fighting Participants were asked, During the last month, how many times have you taken part in a fight where a group of your friends were against another group?. Response options included, never, -3 times, and or more times. Aggression. All participants were asked three item s to evaluate past month levels of aggression. These three items included, how many times have you pushed, shoved, pulled someones hair, or gr abbed someone?; how many times have you kicked, hit, or beat up another person?; and how many times have you told someone you were going to hit or beat them up?. All items included the following response options, Never, -3 times, or more times. These items were coded as , , and , and this scale was used to create t he trajectories of aggression. Risk Factors for Violence Area Deprivation. Community deprivation was created using the following Census 2000 indicators of disadvantage (S ingh, 2003): 1) educational distribution (the proportion of residents with less than 9 years and more than 12 years of education); 2) the unemployment rate; 3) median household income; 4) income disparity; 5) occupational composition; 6) median house va lue; 7) median gross rent; 8) median mortgage; 9) home ownership ra te; 10) family poverty ra te; 11) population below 150% of poverty level; 12) single-parent househol d rate; 13) proportion of homes without a motor vehicle, telephone, or complete plumbing; and 16) household crowding. In accordance with Singh (2003), factor scores were used to weight the indicators, and the scale was standardized ( = 0.87). A higher score on this scale indicated greater deprivation. These items were incorporated in to the analysis in accordance with Shaw and McKays (1942) theory of Social Disorganization.

PAGE 223

223 Alcohol outlet density. The number of off-premise alcohol outlets per 1,000 population was obtained by divi ding the raw number of alcoho l outlets in the community by the total population for each neighborhood. This item was incorporated into the analysis as a measure of Social Disorganization (Shaw & McKay, 1942) in the neighborhood. Adults in neighborhood drink alcohol. As a part of the student survey, all participants were asked, How many adults in your neighborhood dr ink alcohol?. Response options included, None, A few, Some, Many, and Almost All. Those who reported that many or almost all adults in their neighborhood drink alcohol were considered to have high neigh borhood exposure to alcohol. This variable was included in the analysis as a measure of social lear ning theory (Akers, 1973). Home access to alcohol. To identify the source of th eir last alcoholic beverage, students were asked, If you hav e ever had an alcoholic drink, think back to the last time you drank. How did you obtain the alcohol?. If the st udent responded either, Your parent or guardian gave it to you, or You t ook it from home, yout h were considered to have home access to alcohol. All other s ources of alcohol (e.g., a commercial source, friends home, friends parent, etc.), as we ll as non-drinkers, comprised the reference group. This measure serves as a measure of family risk, as alcohol access in the home has been associated with increa sed use of alcohol (Komro Maldonado-Molina, Tobler, Bonds, & Muller, 2007), other substanc e use, and delinquency (Swahn & Hammig, 2000). Peer alcohol use Students were asked, How many of your friends drink alcohol?. Response options ranged from None to Almost All. These were recoded

PAGE 224

224 as None, a few, or more than a few. These items were included because literature suggests that individuals who have peers who use alcohol (Guo, Elder, Cai, & Hamilton, 2009; Herrenkohl et al., 2007; Kuntsche, Go ssrau-Breen, & Gmel, 2009; Leech, Day, Richardson, & Goldschmidt, 2003) are more likely to engage in violent behavior. Depression. Students were asked, During the la st month, how often have you felt sad or depressed?. Responses were coded as, Never, or One or more times. Depression was included as a covariate because higher levels of depression have been associated with violence (E lbogen & Johnson, 2009; Senn, Carey, & Vanable, 2010; Thurnherr, Bechtold, Michaud, Akre, & Suri s, 2008) and other risk behaviors (Latzman & Swisher, 2005; Senn, Carey, & Vanable, 2010). Parental Involvement The parental involvement scale included ten items measuring parental communication and involv ement. These items included frequency of parental praise and general talking, aski ng about school and where the adolescent was going, discussing problems at school, talk ing about movies and television marketing alcohol to adolescents, problems with alcohol alcohol rules, and alcohol consequences, dining habits, and music restrictions. Responses included Never, Hardly Ever, Sometimes, A lot, and All the time. Values for each it em ranged from 1 to 5, with higher scores indicating greater parental involvement. The standardized Cronbach coefficient alpha for this scale was 0.74. This scale was included as a covariate because evidence suggests that parenting variables (e.g., m onitoring, involvement) are related to violence (Park, Morash, & Stevens, 2010). Low academic achievement. As a measure of academic success, students were asked, During the last month, how often have you done poorly on a test or important

PAGE 225

225 school project?. Participants were categor ized into those who had done poorly on a test or project one or more times, compar ed to those who have not. This item was included as a covariate because academic achievement and IQ have been associated with increased risk of violence (Herrenkohl McMorris, Catalano, Abbott, Hemphill, & Toumbourou, 2007; Leech, Day, Ric hardson, & Goldschmidt, 2003). Baseline alcohol use Students who have used alcohol in the past year were measured using the item, Durin g the last 12 months, on how many occasions, or times, have you had alcoholic beverages to drink?. Options included, times, -2 occasions, -5 occasions, -9 occasi ons, -19 occasions, 20-39 occasions, and or more occasions. Responses were dichotomized into drinkers and nondrinkers. This item is included as a co variate because alcohol use has been associated with increased levels of violent behavior (Maldonado-Molina, Reingle, & Jennings, 2010). Marijuana use. Marijuana use in the past year was captured using the item, During the last 12 months, on how many o ccasions, or times, have you used marijuana (other names for marijuana are: pot, grass, weed, reefer, blunt, or hashish)?. Response options ranged from occasions to or more occasions. Responses were dichotomized to include marijuana user s and marijuana non-users. These items were included because evidence suggests that the use of marijuana (Boles & Miotto, 2003; Dhungana, 2009; Herrenkohl et al., 2007) increases the risk of violent behavior. Unsupervised time Students were asked, About how many hours a day do you usually spend without an adult around?. Pa rticipants responded, None, less than one hour, -2 hours, -4 hours, and or more hours. This item was included as a

PAGE 226

226 covariate because evidence sug gests that adolescents are mo re likely to be delinquent when they have more leisure time unsuper vised by a parent or guardian (MaldonadoMolina, Jennings, & Komro, 2009). Race/Ethnicity Participants race and ethnicity was measured at baseline using the item, How do you describe y ourself? Mark all that describe you. If you are not sure, mark other and write in how you describe your self. Response options included, Asian American or Asian Indian, Afr ican-American or Black, Lat ino, Hispanic, or MexicanAmerican, Native American or American Indian, White Caucasian, or European American, and Other. Partic ipants were coded as Hispani c if they identified as Hispanic, regardless of the other options selected. Natural Parent Household To evaluate students living arrangements, participants were asked, Who do you live with most of the time? Ma rk all that apply. Response options included, Mother and father together, and other combinations of parents and grandparents. Students were c oded as living with both parents (e.g., natural parent household), or other. For a review of the evidence suggesting that single-parent households are a risk factor for violence, see Anderson (2010). Free or Reduced Lunch An indicator of family socioeconomic position, participants were asked, Do you receive fr ee or reduced price lunches at school? Free or reduced-price lunch means t hat lunch at school is provided for free or you pay less for it. Responses were coded as Yes or N o. Participants who did not know if they received free or reduced lunch were coded as No. Spanish at home. Students were asked, What is the language most often spoken in your home? If more than one language is spoken in your home, mark the one

PAGE 227

227 you speak most often. Responses were categorized into Spanish or English groups for Hispanics only. Hispanics who reported speaking other languages at home were excluded from the analysis. This variable was included as a covariate because speaking Spanish in the home is an indicator of lower acculturation, which has shown a protective effect on violent behavior (Ma ldonado-Molina, Reingle, Tobler, Jennings, & Komro, 2010). Analytical Methods Trajectory groups were fitted to the dat a using group-based trajectory modeling (Nagin & Land, 1993; Nagin, 2005). This method of analysis grouped individuals together based upon common attributes (e.g., le vels of aggression over time). This approach is appropriate in this situation bec ause violence varies over time (Farrington, 1986), and individuals with different levels of violence may be substantially different from each other. Groupi ng participants with heterogeneous levels of aggressive behaviors together and then attempting to predict aggression may dilute the effect of risk or protective factors. Group-Based Trajectory Modeling Group-based trajectory models are finite mixture models, which use singleand multiple-gr oup models structures (Nagin, 2005) Finite mixture models (also known as latent class models) represent the heter ogeneity in a finite number on unmeasured (latent) classes. The trajectory groups that are created usi ng these analyses are derived from maximum likelihood estimation. In this case, aggression data follows a Poisson distribution with a large number of non-violent events (zer o violent events). Therefore, a zero-inflated poisson (ZIP) dist ribution was specified in the model (Jones, Nagin, & Roeder, 2001).

PAGE 228

228 Models were tested until the most parsi monious number of trajectory groups maximized the Bayesian Informati on Criterion (BIC). The BIC refers to: BIC = log(L) 0.5 klog(N), where log likelihood at the maximum likelihood estimate is subtracted from half the number of parameters mult iplied by the log of the sample size. The trajectories were descriptive in nature, and quadratic, cubic and linear models were tested to correctly depict the slopes represented in the data. SAS PROC TRAJ was used to estimate the trajectories (SAS Institute, Cary, NC; Jones, Nagin, & Roeder, 2001). Individuals were categorized into tr ajectory groups using the maximum probability procedure (Nagin, 20 05, 2001). In other words, participants were assigned to groups in which they have the greatest pr obability of membership (e.g., greater than .80). The modeling strategy estimated posterior probabilities of assignment to each group, and individuals were assigned to t he group with the highest probability. This does not guarantee that all individuals will have a perfect probability of membership in a latent group, but the mean assignment probability for each group is expected to be high (>.80; Nagin, 2005; 2001). These high assign ment probabilities increase confidence in the validity of latent groups. This modeling approach has a number of strengths and weaknesses. Latent group-based modeling allows an estimation of patterns of aggressive behavior over the life-course. These summaries of violence over time provide more information than a traditional dichotomous, aggres sive or non-aggressive outcome variable. However, there is a possibility that groups that are not meaningful could emerge from the data (e.g., a latent class with 2% or less of obs ervations categorized in that group). To avoid this situation, the BIC as well as a judgment of parsimo ny was used in the modeling

PAGE 229

229 procedure. In addition, the shape of each trajec tory and the number of latent groups is sensitive to the size of the dataset, characteri stics of the sample, or length of the followup (Eggleston, Laub, & Sampson, 2004). Specific ally, Eggleston et al. (2004) found that the shape of each trajectory group remained relatively constant despite changes in the length of follow-up; a doubling of the fo llow-up time changed their conclusion in identifying the time of peak criminality and estimating individual-level group membership. These limitations must be c onsidered when interpreting the results of these analyses. Latent group-based trajectory modeling has been used in several studies in estimating trajectories of violenc e and aggression among adolescents and young adults. Piquero (2008) reviewed 80 studies on trajectories of delinquency over the lifecourse. This study found evidence of the age-crime curve, as criminal behavior decreased over time. The majority of st udies found between three and five classes of delinquency, regardless of methodology and the sample (Piquero, 2008; MaldonadoMolina et al., 2009). Overall, these trajectory models have been used in previous studies to examine delinquency a nd violence over time. Multinomial Logistic Regression Once trajectory groups have been specified, bivariate and multivariate multinomial logistic regr ession were used to estimate odds-ratios for risk and protective factors on membership in each trajectory for Hispani cs and African-Americans separately. This model is an extension of mu ltiple logistic regressions; however, the model is more appropriate in this situation because trajecto ry group membership is a nominal variable, and this procedure compares membership in each trajectory group to a reference

PAGE 230

230 category (e.g., low-level violence). Under this model, each ( g 1) odds-ratios were generated (Hedeker, 2003). Multinomial logistic regression proc edures were adapted to account for the multilevel nature of the data (Hedecker, 2003). The PNC sampling design features students nested within schools. Clustered robust standard errors were estimated to produce error estimates that ta ke into account the autocorrelation due to the sampling design. Failure to account for the sampli ng deign would result in an inflated Type 1 error rate, artificially increas ing the precision of the effe cts (Twisk, 2006). The adapted multinomial model does not assume that observations are independent; therefore, it is appropriate for longitudinal and cluster ed designs (Hedecker, 2003). STATA 11 software (College Station, TX) was used to conduct all multinomial logistic regression analyses. Mediation Analyses Mediation analyses were conduc ted to evaluat e the indirect effect of community-, parentand peer-level variables on aggressive trajectory member ship. Trajectory groups were dichotomized into aggressive tr ajectory group member (if classified as a desistor or an escalator) and non-aggressi ve trajectory gr oup member (if nonaggressive) groups for logistic regressi on modeling (MacKinnon, 2008). For each mediator and contextual variab le, four logistic regression models were examined: 1) the effect of the contextual va riable on the mediator (slope a) ; 2) the effect of the mediator on the outcome (aggressive tr ajectory membership, slope b ); 3) the direct effect of the contextual variable on the outcome (slope c); and 4) the adjusted effect of both the contextual variable and the mediator on t he outcome (estimating parameters for both slope b and slope c). All regression models were adjusted for other risk factors,

PAGE 231

231 demographics, and sampling design. These effect sizes were standardized and adjusted using the covariance matrix and variance for each variable in the model (MacKinnon, 2008). To test the significance of the mediator, the Sobel test was used to generate a z statistic and standard error (Baron & Kenny, 1986; Sobel, 1982; MacKinnon, Warsi, & Dwyer, 1995). The percent mediation for each mediator was calculated using the formula: ab/a1b1axbx +c In this formula, a represents the effect of the contextual variable on the hypothesized mediator, and b represents the effect of the mediator on the outcome variable (in this case, violence). C represents the direct effect of the contextual variable on the outcome. All of these standardized estimates (including all other variables in the model) were used to calculate the proportion of the variance in each contextual variable that is mediat ed by each proximal variable. These percentages were summed by contextual vari able to estimate the proportion of the contextual variable that is mediated th rough more proximal variables. These percentages were summed withi n categories of each contextual variable (e.g., each contextual variable can be no more than 100% medi ated by individual-level variables) to estimate the proportion of the contextual variable that is mediated by more proximal variables. Results Ethnic Differences in Traj ectories of Aggression To determine the number of trajectori es of aggression, latent group-based trajectory modeling was used. Four distinct classes we re identified for each subgroup. Among African-Americans, 7.9% were in the low-aggression group, 19.5% were escalators, 8.4% had moderate-consistent aggression patterns, and 64.0% were in the

PAGE 232

232 consistently aggressive group. Among Hisp anics, 17.1% were in the low-aggression group, 18.1% were desistors, 21.6% were escalators, and 43.3% were consistently aggressive. For each subgroup, a four-group trajectory model showed the lo west AIC and BIC compared to a higher-class model. For Afri can-Americans, the AIC and BIC for the 4class model were -10,437 and -10467, respec tively. For Hispanics, the AIC and BIC were -8,333 and 8,362, respectively, for the 4-class model. The mean posterior probabilities ranged from (0.79-0. 91) for African-Americans and Hispanics. Figure 5-1 displays the trajectories of aggression from grades 6-8. The mean aggression value by trajectory subgroup is det ailed in Table 5-2. Effects of Risk and Pr otective Factors at 6th Grade on Trajectories of Aggression Bivariate results, African-Americans Table 5-3 shows the bivariate relationship bet ween each risk or protective factor and aggression trajectories for African-Amer icans. For desistors, perception of the number of adults who consume alcohol in the neighborhood (OR = 1. 70; 95% CI 1.132.56), peer alcohol use (OR = 2.32; 95% CI 1.36-3.98), indi vidual-level alcohol use (OR = 3.12; 95% CI 1.05-15.52), marijuana us e (OR = 12.18; 95% CI 2.20-67.31), depression (OR = 1.67; 95% CI 1.05-2.66), group fighting (OR = 2.38; 95% CI 1.394.07), and baseline aggression (O R = 3.33; 95% CI 2.25-4.93) were identified as risk factors for being in the desistor traj ectory group compared to the non-aggressive group. For escalators, alcohol outlet density (O R = 3.32; 95% CI 1.21-9.08), number of adults in the neighborhood who consume alcohol (OR = 1.73; 95% CI 1.11-2.68), peer alcohol use (OR = 2.00; 95% CI 1.12-3.56), depre ssion (OR = 1.73; 95% CI 1.06-2.82), group fighting (OR = 1.88; 95% CI 1.08-2. 26), and baseline aggression (OR = 2.23;

PAGE 233

233 95% CI 1.46-3.42) were identified as risk fa ctors. Among the highe st risk group (those who were consistently aggre ssive), the number of adults in the neighborhood who drank alcohol (OR = 2.96; 95% CI 2.04-4.30), peer alcohol use (OR = 5.09; 95% CI 3.008.66), individual-level alcohol use (OR = 3. 18; 95% CI 1.93-24.51), marijuana use (OR = 4.58; 95% CI 0.86-24.21), lo w academic achievement (OR = 1.60; 95% CI 1.03-2.50), having more unsupervised time (OR = 1.51; 95% CI 1.06-2.16), depression (OR = 2.53; 95% CI 1.73-3.71), group fight ing (OR = 7.14; 95% CI 4.31-11.79), baseline aggression (OR = 10.60; 95% CI 7.45-15.06) and receiving free or reduced price lunch (OR = 1.44; 95% CI 1.01-2.04) increased the likelih ood that youth would be members of the consistently aggressive trajectory group co mpared to the non-aggr essive group. Protective factors from membership in the consistently aggressive group included parental involvement (OR = 0. 96; 95% CI 0.92-1.01) and re siding with both parents at home (OR = 0.61; 95% CI 0. 43-0.86). Because area deprivation and home access to alcohol were not significant predictors of membership in any aggressive trajectory group, they were dropped from fu rther multivariate analyses. Bivariate results, Hispanics Table 5-10 shows the bivariate relationshi p between eac h risk or protective factor and aggressive trajectory membership for His panic youth. For desistors, peer alcohol (OR = 3.39; 95% CI 2.01-5.71), individual-level alcohol use (OR = 2.17; 95% CI 1.084.36), low academic achievement (OR = 2.62; 95% CI 1.714.01), having more unsupervised time (OR = 1.81; 95% CI 1. 31-2.48), depression (OR = 1.88; 95% CI 1.23-2.87), group fighting (OR = 2.49; 95% CI 1.34-4.65) and baseline aggression (OR = 4.39; 95% CI 3.14-6.16) we re identified as risk factor s for being in the desistor trajectory group compared to the non-aggressive group.

PAGE 234

234 Protective factors for desistors included higher parental invo lvement (OR = 0.96; 95% CI 0.92-1.01; p <0.10) and speaking Spanish in t he home (OR = 0.55; 95% CI 0.350.86). For escalators, area deprivation (OR = 0.99; 95% CI 0.980.99), perceived higher number of adults in the nei ghborhood who consume alcohol (OR = 1.47; 95% CI 0.972.23), having peers who use alco hol (OR = 2.08; 95% CI 1.10 -3.93), using alcohol (OR = 2.48; 95% CI 1.12-5.50), having low academ ic achievement (OR = 2.11; 95% CI 1.403.17), depression (OR = 1.50; 95% CI 1.062.13), and baseline aggr ession (OR = 2.83; 95% CI 1.94-4.13) increased the odds of me mbership in the escalator group compared to the non-aggressive group. Protective factor s from membership in the escalator group included speaking Spanish in the home (OR = 0.64; 95% CI 0.43-0.97) and residing with both parents at home (OR = 0.60; 95% CI 0. 41-0.89). Among the highest risk group, the perceiv ed number of adults who use alcohol was identified as a risk factor (OR = 3.42; 95% CI 2.41-4.84), as well as having home access to alcohol (OR = 2.48; 95% CI 1.80-3.42), peers who use alcohol (OR = 5.49; 95% CI 3.35-8.97), alcohol use (OR = 5. 51; 95% CI 2.47-12.28), marijuana use (OR = 17.52; 95% CI 2.08-147.16), low academic achi evement (OR = 3.48; 95% CI 2.47-4.91), having more unsupervised time (OR = 2.05; 95% CI 1.45-2.91), depression (OR = 1.62; 95% CI 1.88-3.65), group fight ing (OR = 5.36; 95% CI 3.12-9.23), baseline aggression (OR = 14.30; 95% CI 10.90-18. 76), and older age (OR = 1.51; 95% CI 1.08-2.09). Protective factors from membership in the consistently aggressive trajectory group included speaking Spanish in the home (OR = 0.50; 95% CI 0.32-0.78) and living with both parents in the home (OR = 0.64; 95% CI 0.58-1.49; p <0.10).

PAGE 235

235 Because alcohol outlet density was not significant in predicting aggression trajectories for any of the three groups this variable was dropped from further multivariate analyses. Marijuana use was also dropped from the analyses due to collinearity with aggression (e .g., n=1 reported marijuana use among non-aggressive youth). Multivariate community-, parent-, peer-, and individual-level results, AfricanAmericans Table 5-4 shows the multivariate effect of community-, parentand peer-level variables on three trajectory groups of African-American youth. Among desisto rs, a higher number of adults in the neighborhood w ho use alcohol remained a marginally significant risk factor (OR = 1.45; 95% CI 0.94-2.25). Ha ving peers who used alcohol doubled the odds of membership in the desistor trajectory group (OR = 2.02; 95% CI 1.16-3.53). A greater density of alcohol outlets in the neighborhood was a marginally significant risk factor among escalators (OR = 2.34; 95% CI 0.88-6.24). Among consistently aggressive youth, the perceptio n that more adults in the neighborhood use alcohol (OR = 2.11; 95% CI 1. 43-3.10) and having peers who use alcohol (3.72; 95% CI 2.12-6.52) were identif ied as risk factors. When individual-level variables were added to the model (Table 5-5), the number of adults who use alcohol in the neighbor hood among consistently aggressive youth was the only contextual variable that rema ined significant (OR = 1.83; 95% CI 1.242.70). Peer alcohol use remained a signifi cant risk factor among desistors (OR = 1.72; 95% CI 0.95-3.11) and the consistently aggr essive group (OR = 2.39; 95% CI 1.274.48). The only individual-l evel variable that significantly predicted consistent aggression was depression (OR = 1.97; 95% CI 1.25-3.09). No individual-level

PAGE 236

236 variables besides group fighting predicted de sistance or escalation. Group fighting significantly increased the odds of membersh ip in the desistor (OR = 1.88; 95% CI 1.053.39) and consistently aggressi ve (OR = 4.33; 95% CI 2. 47-7.58) trajectory group compared to the non-aggressive group. Multivariate parent-, peer-, and individual-level results, Hispanics Table 5-11 shows the multivariate effect of parentand peer-level variables on the three aggressive trajectory groups of Hispanic youth. Nei ghborhood deprivation remained a significant protective factor from membership in the escalator group (OR = 0.99; 95% CI 0.98-0.99). Perception of more adults in the neighborhood who use alcohol remained a risk factor for the consis tently aggressive group (OR = 2.18; 95% CI 1.55-3.05). Among desistors and escalators, peer alcohol use remained a risk factor (OR = 3.91; 95% CI 2.32-6.58 for desisto rs, OR = 2.28; 95% CI 1.28-4.05 for escalators). Home access to alcohol (O R = 1.73; 95% CI 1.292.32) and peer alcohol use (OR = 4.38; 95% CI 2.73-7.05) significantly increased the risk of membership in the consistently violent group, while parental involvement (OR = 0.95; 95% CI 0.91-0.99) was protective from mem bership in this group. When individual-level variables ar e added to the model (Table 5-12), neighborhood deprivation remain s a protective factor am ong escalators (OR = 0.99; 95% CI 0.98-0.99), while the number of adults who consume alcohol remains a risk factor for the consistently aggressive group (OR = 1.92; 95% CI 1.30-2.82). Among desistors, having peers who use alcohol (OR = 2.94; 95% CI 1.72 -5.02), low academic achievement (OR = 2.09; 95% CI 1.25-2.43) and more unsupervised time (OR = 1.72; 95% CI 1.22-2.43) increased risk for membersh ip in this group. Having access to alcohol in the home (OR = 0.72; 95% CI 0.52-1.02) and speaking Spanish in the home

PAGE 237

237 (OR = 0.53; 95% CI 0.33-0.83) was protective from membership in the desistor group compared to the non-aggressive group. Among escalators, having peers who use alcohol (OR = 1.94; 95% CI 1.06-3.54) and low academic achievement (OR = 1.77; 95% CI 1.08-2.90) were ident ified as risk factors, while speaking Spanish in the home was protective (OR = 0.69; 95% CI 0.46-1.04; p <0.10) from membership in this group. Among those who were consistently violent, having home access to alcohol (OR = 1.45; 95% CI 1.11-1.91), having peers who use al cohol (OR = 2.78; 95% CI 1.68-4.61), low academic achievement (OR = 2.09; 95% CI 1.39-3.15), depression (OR = 1.61; 95% CI 1.07-2.41), and group fighting (O R = 3.30; 95% CI 1.84-5.91) were identified as risk factors for membership in t he consistently aggressive group. Protective factors included higher parental involvement (O R = 0.96; 95% CI 0.91-0.99) and speaking Spanish in the home (OR = 0.56; 95% CI 0.35-0.91). Multivariate Results Adjusted for Baseline Aggressi on, African-Americans The model adjusted for base line aggression is presented in Table 5-6. For the consistently violent group, t he perceived number of adults in the neighborhood who use alcohol remained a significant risk factor (OR = 1.75; 95% CI 1.16-2.64), and peer alcohol use remained a marginally significant risk factor for consistently aggressive youth (OR = 1.86; 95% CI 0.95-3.63; p <0.10). Depression remained a risk factor among the high-aggression group (OR = 1.63; 95% CI 1.61-5.39) even after controlling for several risk/protective fa ctors, including baseline aggr ession. Group fighting was only significant among consistently aggressive youth (OR = 2.95; 95% CI 1.61-5.39) and baseline aggression was a significant risk fa ctor for membership in each trajectory group (OR = 2.64; 95% CI 1.793.91 for desistors, OR = 1.84; 95% CI 1.21-2.81 for escalators, OR = 5.67; 95% CI 4.018.02 for consistently violent).

PAGE 238

238 Multivariate Results Adjusted for Baseline Aggression, Hispanics The model adjusted for baseline aggression for Hispanics is presented in Table 513. Neighborhood deprivation remained a signific ant protective factor from membership in the escalator trajectory group (OR = 0.99; 95% CI 0.98-0.99). Among des istors, having peers who use alcohol (OR = 2. 57; 95% CI 1.46-4.55), low academic achievement (OR = 1.82; 95% CI 1.08-3.06), more unsupervi sed time (OR = 1.69; 95% CI 1.20-2.38), and parti cipation in aggression at baseline (OR = 3.28; 95% CI 2.09-5.13) increased risk for membership in this group. Having access to alcohol in the home (OR = 0.65; 95% CI 0.46-0.92) and speaking Spanish in the home (OR = 0.54; 95% CI 0.350.83) were protective from membership in the desistor group compared to the nonaggressive group. Among escalators, havi ng peers who use alcohol (OR = 1.76; 95% CI 0.95-3.24; p <0.10), low academic achievement (OR = 1.60; 95% CI 0.95-2.69; p <0.10), and baseline aggression (OR = 2.38; 95% CI 1.49-3. 80) were identified as risk factors for membership in this group. Among those who were consistently violent, having home access to alcohol (OR = 1.26; 95% CI 0.98-1.64; p <0.10), having peers who use alcohol (OR = 2.28; 95% CI 1. 35-3.88), low academic achievement (OR = 1.59; 95% CI 1.02-2.47), gr oup fighting (OR = 2.00; 95% CI 1.13-3.55), and baseline aggression (OR = 8.23; 95% CI 5.67-11.94) were identified as risk factors for membership in the consistently aggressive group. Protective factors included higher parental involvement (OR = 0. 95; 95% CI 0.91-0.99) and s peaking Spanish in the home (OR = 0.56; 95% CI 0.34-0.90). Mediation, African-Americans The test of the mediation pathway ( based on bivariate analysis) for AfricanAmericans is presented in Figure 52. The significance of pathway a indicated that all

PAGE 239

239 variables except alcohol outlet density met the criteria for inclusion in the mediation analysis (e.g., contextual variables must be significantly associated with at least one mediator) (Table 5-7). The indirect effect s of each contextual variable by each individual-level variable are detailed in Tabl e 5-8. For parental involvement, 65.2% of the effect on aggression was mediated through individual-level variables. For parental alcohol use 15.9% of the effect was mediat ed, and for peer alcohol use 69.4% of the effect on aggression was mediated through individual-level variables. Parental involvement was equally mediated thr ough group fighting and baseline aggression (15.7%), as was parental alc ohol use (15.8 and 15.9%, respec tively). Effects of peer alcohol use on aggression were mediat ed primarily through baseline aggression (21.1%), followed by group fighting (17.2%). Mediation, Hispanics The test of the mediation pathway (based on bivariate analysis) is presented in Figure 5-3. The significance of pathway a indicated that all variables except for neighborhood deprivation met the cr iteria for inclusion in th e mediation analysis (e.g., contextual variables must be significantly associated with at least one mediator) (Table 5-14). The indirect effects of each contextual variable by each individual-level variable are detailed in Table 5-15. For parental involvement, 65.9% of the effect was mediated through individual-level variabl es. The majority of the effects of parental alcohol use (71.7%), home access to alcohol (77.9%), a nd peer alcohol use (66.7%) were mediated through proximal variables. The variable accounting for the largest percentage of the indirect effect of these contextual variables was through baseline aggression. The second largest mediator for peer alcohol us e (12.5%) was group fight ing, while alcohol

PAGE 240

240 use was a significant mediator for effects of home access to alcohol (16.7%), parental alcohol use (11.9%), and par ental involvement (12.5%). Characteristics of sample at base line (po st-hoc), African-Americans Because baseline aggression was a signifi cant predictor of all aggressive trajectory groups among African-Americans a post-hoc analysis was conducted to understand the characteristics associated with baseline aggression (Table 5-9). Parental involvement, perce ived number of adults who consume alcohol in the neighborhood, home access to alcohol, peer alcohol use, individual-level alcohol use, marijuana use, unsupervised time, depression, group fighting, residing in a singleparent home, and receiving free or reduced pr ice lunch were all significantly higher among those who were aggressive at baseline. Those who were aggressive at baseline were also more likely to have poor academic achievement and be younger in age compared to those who were not aggressive at baseline. Characteristics of sample at baseline (post-hoc), Hispanics Among His panics, a post-hoc analysis wa s also conducted to understand the characteristics associated with baseline aggre ssion (Table 5-16). A greater portion of adolescents who were aggressive at baseline perceived there to be more adults in the neighborhood who use alcohol, have more ac cess to alcohol in the home, have lower parental involvement, have peers who use alcohol, use alcohol themselves, use marijuana, have a record of poor academic achievement, have more unsupervised time, report depression, have engaged in group fighti ng, and were older in age compared to those who were not aggressive at baseline.

PAGE 241

241 Discussion The present study examined the number and shape of trajectories of aggression among African-American and His panic urban adolescents, as well as the direct and indirect effects of multiple dom ains of risk and protective fa ctors fo r membership in each trajectory group. The latent-group based traj ectory models found a four-class model for each racial/ethnic subgroup. Among African-Americans, the four groups included a lowaggression group, escalators, moderate-consistent aggression, and consistent aggression. Among Hispanics, the groups included a low-aggression group, desistors, escalators, and consistently aggressive. There were differences between African-Am ericans and Hispanics in the level of violence over time. As expected, African-Am ericans were more likely to be consistently aggressive (64.0%) compared to Hispanics (43.3%), and le ss likely to be in the nonaggressive group (7.9% were non-aggressive among African-Americans, 17.1% among Hispanics). These differences in trajectory membership indicate that African-Americans are more likely to consistently participate in aggression, and are less likely to be nonaggressive, compared to Hispanics. These results are consistent with previous research on trajectories of delinquency. Four trajectory groups were ex tracted from the data in this study, and this is consistent with the extant literature that suggests there are between three and five unique patterns of aggressive behaviors among adolescents (Piquero, 2008; Maldonado-Molina, Reingle, Tobler, Jennings, & Komro, 2010; Maldonado-Molina, Piquero, Jennings, Bird & Canino, 2009). These trajectory groups ar e also consistent with the other studies which have investigated the patterns of delinquency specifically among Hispanic

PAGE 242

242 adolescents (Maldonado-Molina et al., 2010; Jennings, Maldonado-Molina, Piquero, Odgers, Bird, & Canino, 2010). There were a number of similarities in the risk and protective factors across African-Americans and Hispanic s. For example, baseline aggression was a significant predictor of all trajectory gr oups. Peer alcohol use also had a direct and indirect effect for both groups. For Hispanics, peer alc ohol use predicted membership in all three aggressive trajectory groups, while this effect was only significant among the consistently aggressive gr oup of African-Americans. This study also found a substantial number of differences in the predictors of membership in each trajectory group bet ween African-Americans and Hispanics. Among Hispanics, neighborhood depriv ation was significantly protective from membership in the escalators and consis tently aggressive trajectory groups. Among the highest risk group of Hispanics (consistently aggressive), the perception of more adults in the neighborhood consuming alcohol, having access to alcohol in the home, and having more peers who use alcohol a ll had both direct and indirect effects increasing the risk of mem bership in the high-risk group, compared to the nonaggressive group. Also among Hispani cs, having low academic achievement (consistently aggressive and desistors), more unsupervised time (desistors), and group fighting (consistently aggressive) were ri sk factors for membership in an at-risk trajectory group. Speaking Spanish at home was protective for consistent aggression and desistance, and parental involvement was protective for consistent aggression. Among African-Americans only, the number of adults who consumed alcohol and number of peers who use alcohol had a direct effect on memb ership in the consistent

PAGE 243

243 aggression group, and an indirect effect on aggression overall. Among AfricanAmericans only, depression was a risk factor for membership in the consistently aggressive group. This study identified a variety of risk and protective factors that significantly predicted aggressive trajectory group member ship. These findings are consistent with prior literature on the disparities by race/ethnicity on community-, familyand peer-level risk factors for violence. Specifically, the fi nding that the perceiv ed number of adults who use alcohol in the neighborhood and peer alcohol use both increase the risk for consistent aggression across race/ethnicity supports the peer and parental effects of social learning on individual aggression (Akers 1985). The effect of social learning through contextual alcohol use appears to operate differently between AfricanAmericans and Hispanics. Specifically, am ong Hispanics, the effect of adults and peers who use alcohol is mediated by individual-level alcohol use. This is not true for AfricanAmericans. This suggests that the influenc e of contextual su bstance use operates differently by race/ethnic group. At the individual-level, gr oup fighting was significant in predicting membership in the consistently aggressive gr oup of Hispanics only. This finding is supported in the literature on racial/ethnic differences in aggression (McNulty & Bellair, 2003), as gang membership was a predictor of serious violence among Hispanics only. The higher prevalence of gang membership among consistently aggressive Hispanics in this sample may be driving the relationshi p between group fighting and consistent aggression. This finding also indicates that the construct of group fighting operates differently from aggression, as both construc ts independently predicted trajectories of

PAGE 244

244 consistent aggression. Therefore, comb ining group fighting with other violence and aggression variables may be an inappropriate practice. Among Hispanics, speaking Spanish in the home was a significant protective factor from desistance and consis tent aggression. This is consistent with prior literature on a variety of outcomes, as lower levels of acculturati on have been strongly associated with adverse health outcomes, including driving under the influence of alcohol (Maldonado-Molina et al., 2011; Caetano, Ramisetty-Mikler, & Rodriguez, 2008), intimate partner violence, alcohol us e (Caetano, 2006; Caetano, Schafer, Clark, Cunradi, & Raspberry, 2000), and aggression and violence (Maldonado-Molina, Reingle, Tobler, Jennings, & Komro, 2010; Maldonado-Molina, Piquero, Jennings, Bird, & Canino, 2009). It has been proposed that the stress associated with intergenerational conflict generated from internaliz ation of American culture and values may increase a variety of problem behaviors among more acculturated Hispanic youth (Perez et al., 2008). Among African-Americans, this study identif ied few direct effects of contextual variables on aggression among escalators and moderate-consistently aggressive adolescents. The lack of significant predictors may be due to the time-varying nature of aggression in this group. Specifically, risk fact ors at more time poin ts more proximal to the aggressive behavior (e.g., late adolescence) may be more predictive of membership in this trajectory group. Since aggression is relatively low at baseline among escalators, this variable may not be a pot ent predictor of late-onset aggression. This finding highlights the need for future research on this group of escalators, as unique or timevarying risk factors may be present.

PAGE 245

245 The presence of multiple risk factor s for aggression at baseline highlights important targets for preventi on programming. These risk factor s were largely consistent among African-Americans and Hispanics. First, those who were aggressive as baseline had greater neighborhood, school, and family risk (e.g., perceived that there were more adults in the neighborhood who used alcohol, had access to alcohol in the home, and had a record of poor academic achievement) than those who were not aggressive at baseline. Peer alcohol consum ption and individual-level alcohol and marijuana use were more prevalent among those who reported aggression at base line. African-Americans were more likely to be aggressive at baseli ne if they were low-income and lived in a single parent home, while Hispanics who we re aggressive at baseline reported more depression and unsupervised time. These differ ences at baseline highlight the need for early aggression prevention programming, as thes e risk factors were present prior to 6th grade. This study had several limitations. Firs t, this study was unable to account for some of the variables that are important in predicting aggression, such as peer delinquency, cognitive development, and psychological disorders. Second, latent-group based trajectory modeling provides an estima tion of the type and number of groups in the data, and this process is exploratory in nature. Despite the exploratory nature of trajectory estimation, the results of this study were consistent with the expected number and shape of trajectory groups from other studies (Piquero 2008; Zara & Farrington, 2009). Finally, risk factors were analyzed at mu ltiple levels; however, hierarchical linear modeling (HLM) was not used due to the small sample sizes available in some of the trajectory groups. In accordance with the sampling design, all analyses accounted for

PAGE 246

246 the nesting of adolescents within schools, which may account for a portion of the variability in Census tract measures. Us e of HLM to account for the nesting within Census blocks is a direction for future research. Despite these weaknesses, the current study had a number of strengths. First, data were derived from a unique sample of high-risk, urban adolescents who were followed longitudinally for th ree years. Second, the large sample size provided adequate power to stratify groups of Af rican-Americans and Hispanics to understand the differential risk factors by subgroup. Third, although many studies have analyzed the multiple domains of risk and protective factors for aggressive behavior, few have assessed the degree to which contextual va riables are mediated by more proximal variables at the individual-lev el. The mediated effects allows this study to acknowledge that contextual variables are important in predicti ng aggression even though their effects are mitigated using mult ivariate regression models. Finally, the trajectories estimated in this study are especially appropriate for studie s of delinquency and aggression, as patterns tend to vary significant ly over time (Farrington, 1986; Piquero, 2008). In conclusion, the findings from this study i ndicate that there are similarities in the risk factors for aggression between AfricanAmericans and Hispanics. These predictors have significant implication for large-scale prevention programming. First, interventions should target multiple risk and protective fa ctors to maximize the preventive effect across demographic groups. Social influences, such as exposure to peers who use alcohol or marijuana, and comm unity-level exposure to al cohol influence adolescents risk for violent behavior. These risk factors t hat were consistent across race/ethnicity

PAGE 247

247 may be targeted in a variety of populations to reduce participation in violence. Second, aggressive behavior begins even before 6th grade in high-risk settings, indicating that the current prevention programming occurs too late. Prevention programming should begin early in elementary school settings to pr event initiation of aggressive behavior. Third, there were some differences in the predictors of aggression between racial/ethnic groups. Therefore, the composit ion of the intervention populat ion (e.g., characteristics of the social structure, comm unity, family, etc.) should be considered prior to program administration. This differential exposure may equate to increased or diminished propensity for violence. Disclosure This study was supported by Awar d Numbers K01 AA017480 (PI: Mildred Maldonado-Molina), R01 AA013458 (PI: Kelli A. Komro) from the Na tional Institute on Alcohol Abuse and Alcoholism and the National In stitute for Minority Health and Health Disparities, and from the Institute for Child Heal th Policy at the University of Florida. The content is solely the responsibilit y of the authors and does not necessarily represent the official views of the National Institute on Alcohol Abuse and Alcoholism of the National Institute of Health.

PAGE 248

248 African-Americans Hispanics Figure 5-1. Trajectories of aggression over time, PNC. # Violent EventsGrade # Violent EventsGrade

PAGE 249

249 Table 5-1. Description of sample, PNC. N=4,188. Variable % Violence 6 th Grade (ST03) a 5.25(4.21) 7 th Grade (ST04) a 5.82(4.13) 8 th Grade (ST05) a 6.05(4.91) Community-level Neighborhood exposure to alcohol 0.25(0.18) Area deprivation a -29.9(26.4) Adults in neighborhood drink g 42.6 Parental and Peer Influences Parental Involvement a 35.8(7.64) Home access to alcohol b 29.3% Peer alcohol use f 11.4% Parental connection a 20.6(4.44) Individual-level Risk Factors Alcohol use in the past year 9.9% Marijuana use in the past year 2.8% Depression c 72.7% Unsupervised time d 50.2% Natural parent household 55.2% Free or Reduced Price Lunch 72.8% Spanish at Home 23.7% Low academic achievement e 66.8% Violence Group fighting in prior month 28.3% Demographics Male 50.3% Age at Baseline a 11.38(0.24) White 12.9% Black or African-American 38.8% Hispanic or Latino 34.6% Other Race 13.8% a Mean(SD) are reported. b Home access to alcohol was measured as last obtaining alcohol from either the home or the adolescents parent. c Depression was measured as feeling sad or depressed one or more times in the past month. d Unsupervised time was measured as having one or more hours each day without being supervised by an adult. eLow academic achievement was defined as having reported poor performance on a test or project in the past month. f Some, many, or almost all peers use alcohol. g Many or almost all parents in the neighborhood use alcohol.

PAGE 250

250 Table 5-2. Mean (SD) aggression in each trajectory gr oup for African-Americans and Hispanics. Trajectory Group Subgroup LowAggression N=138 Escalators N=339 Desistors N=146 Consistently Aggressive N=1109 AfricanAmericans Mean Aggression (6 th Grade) 0.09(0.43) 0.62(1.01)*** 3. 90(2.28)*** 8.86(2.88)*** Mean Aggression (7 th Grade) 0.31(0.73) 3.80(2.82)*** 3. 94(2.88)*** 9.27(2.73)*** Mean Aggression (8 th Grade) 0.82(1.18) 8.36(2.65)*** 3. 04(2.43)*** 9.23(2.97)*** NonAggressive N=223 Desistors N=236 Escalators N=282 Consistently Aggressive N=565 Hispanics Mean Aggression (6 th Grade) 0.21(0.61) 4.30(2.69)*** 0. 94(1.07) *** 7.92(2.86) *** Mean Aggression (7 th Grade) 0.30(0.71) 4.09(2.87) *** 4. 15(3.06) *** 8.37(3.00) *** Mean Aggression (8 th Grade) 0.29(0.71) 0.86(1.17) *** 6. 05(2.79) *** 8.23(2.93) *** ***p<0.001

PAGE 251

251 Table 5-3. Bivariate effects between risk/ protective factors and trajectories of aggression, African-Americans. Trajectory Group Desistors Escalators Consistent Aggression OR 95% CI OR 95% CI OR 95% CI Community-level Alcohol outlet density 1.11 0.284.47 3.32* 1.21-9.08 2.02 0.71-5.75 Area Deprivation 0.99 0.99-1.01 0.99 0.99-1.00 1.00 0.99-1.01 Adults in Neighborhood Drink 1.70* 1. 13-2.56 1.73* 1.11-2.68 2.96*** 2.04-4.30 Parental and Peer Influences Parental Involvement 0.96 0.931.01 1.01 0.95-1.07 0.96 0.92-1.01 Home Access to Alcohol 0.77 0. 43-1.37 0.59 0.27-1.25 1.16 0.70-1.93 Peer alcohol use 2.32** 1.36-3.98 2.00* 1.12-3.56 5.09*** 3.00-8.66 Individual-level Risk Factors Alcohol use 3.12** 1.05-15.521.25 0.92-15.84 3.18** 1.93-24.51 Marijuana use 12.18** 2.20-67.312.85 0.61-13.23 4.58 0.86-24.21 Low Academic Achievement 1.34 0.832.16 1.05 0.65-1.70 1.60* 1.03-2.50 Unsupervised time 1.08 0.71-1. 62 1.32 0.86-2.02 1.51* 1.06-2.16 Depression 1.67* 1.05-2.66 1. 73* 1.06-2.82 2.53*** 1.73-3.71 Aggression Group fighting 2.38** 1.39-4.07 1. 88* 1.08-3.26 7.14*** 4.31-11.79 Baseline aggression 3.33*** 2.25-4.93 2.23***1.46-3.42 10.60*** 7.45-15.06 1.15 0.79-1.67 0.80 0.53-1.22 0.89 0.57-1.40 Demographics Age at Baseline 1.15 0.79-1. 67 0.80 0.53-1.22 0.89 0.57-1.40 Family Composition 0.78 0.52-1.19 0.87 0.55-1.37 0.61** 0.43-0.86 Reduced Lunch 1.16 0.71-1.90 1.28 0.89-1.74 1.44* 1.01-2.04 Note: The Non-Aggression trajectory group serves as the reference category. All analyses are controlling for treatment. Clustered robust standard e rrors were calculated to account for the clustered sampling design. *p<0.05 **p<0.01 ***p<0.001

PAGE 252

252 Table 5-4. Community, family and peer effect s on trajectories of aggression, AfricanAmericans. Trajectory Group Desistors Escalators Consistent Aggression OR 95% CI OR 95% CI OR 95% CI Community-level Alcohol outlet density 0.83 0.213.20 2.34 0.88-6.24 1.29 0.45-3.66 Adults in Neighborhood Drink 1.45 0.94-2.25 1.49 0. 91-2.45 2.11*** 1.43-3.10 Parental and Peer Influences Parental Involvement 0.97 0.931.02 1.01 0.96-1.07 0.98 0.94-1.02 Peer alcohol use 2.02* 1.16-3.53 1.65 0.89-3.06 3.72*** 2.12-6.52 Note: The Non-Aggression trajectory group serves as the reference category. All analyses are controlling for demographics and treatment. Clustered robu st standard errors were calculated to account for the clustered sampling design. *p<0.05 ***p<0.001

PAGE 253

253 Table 5-5. Effects of multip le domains of risk factors on trajectories of aggression, African-Americans. Trajectory Group Desistors Escalators Consistent Aggression OR 95% CI OR 95% CI OR 95% CI Community-level Alcohol outlet density 0.81 0.22-3. 022.21 0.81-5.99 1.37 0.49-3.90 Adults in Neighborhood Drink 1.39 0. 89-2.131.41 0.86-2.31 1.83** 1.24-2.70 Parental and Peer Influences Parental Involvement 0.98 0.94-1. 021.03 0.97-1.08 0.99 0.95-1.03 Peer alcohol use 1.72 0.95-3.11 1.36 0.68-2.74 2.39** 1.27-4.48 Individual-level Risk Factors Alcohol use 1.82 0.48-6.902.39 0.52-11.03 1.53 0.43-5.43 Marijuana use 0.45 0.11-1.820.79 0.21-2.89 1.32 0.28-6.15 Low Academic Achievement 1.18 0.72-1.950.94 0.58-1.54 1.23 0.78-1.94 Unsupervised time 0.96 0.64-1.45 1.23 0.79-1.92 1.26 0.86-1.85 Depression 1.40 0.83-2.331.47 0.85-2.56 1.97** 1.25-3.09 Aggression Group fighting 1.88* 1.05-3.391.51 0.82-2.79 4.33*** 2.47-7.58 Note: The Non-Aggression trajectory group serves as the reference category. All analyses are controlling for demographics and treatment. Clustered robu st standard errors were calculated to account for the clustered sampling design. *p<0.05 **p<0.01 ***p<0.001

PAGE 254

254 Table 5-6. Effects of multip le domains of risk factors on trajectories of aggression, adjusted for baseline, African-Americans. Trajectory Group Desistors Escalators Consistent aggression OR 95% CI OR 95% CI OR 95% CI Community-level Alcohol outlet density 0.82 0.23-2. 962.03 0.80-5.181.49 0.53-4.16 Adults in Neighborhood Drink 1.35 0. 87-2.091.36 0.82-2.241.75** 1.16-2.64 Parental and Peer Influences Parental Involvement 0.99 0.95-1. 031.03 0.97-1.090.99 0.96-1.04 Peer alcohol use 1.44 0.78-2.63 1.22 0.57-2.581.86 0.95-3.63 Individual-level Risk Factors Alcohol use 1.62 0.42-6.242. 09 0.44-9.851.34 0.38-4.76 Marijuana use 0.40 0.09-1.700.37 0.18-0.761.16 0.23-5.81 Low Academic Achievement 1.04 0.641.710.86 0.52-1.411.04 0.23-5.81 Unsupervised time 0.94 0.62-1. 421.19 0.76-1.841.20 0.80-1.82 Depression 1.23 0.73-2.051. 38 0.80-2.381.63* 1.61-5.39 Aggression Group fighting 1.42 0.73-2.751. 23 0.63-2.392.94** 1.61-5.39 Baseline aggression 2.64*** 1.79-3.91 1.84** 1.21-2.815.67*** 4.01-8.02 Note: The Non-Aggression trajectory group serves as the reference category. All analyses are controlling for demographics and treatment. Clustered robu st standard errors were calculated to account for the clustered sampling design. *p<0.05 **p<0.01 ***p<0.001

PAGE 255

255 Figure 5-2. Test of the mediation pathway (based on bivariate analysis), AfricanAmericans. Social/Contextual Risks Alcohol outlet density Parental involvement Peer alcohol use Adult alcohol use Aggressive Trajectory Membership a b c Alcohol and marijuana use Unsupervised time Depression Academic achievement Group Fighting Baseline aggression Individual-Level Risks

PAGE 256

256 Table 5-7. Regression models testing t he association between parent and peer level variables and individual-level risk and pr otective factors, African-Americans. Mediators Alcohol Use MJ Use Unsupervised time Depress Academic Achievement Group Fighting Baseline aggression OR OR OR OR OR OR OR Contextual Variables Alcohol outlet density 0.66 0.67 1.06 0.78 0.81 1.05 0.77 Parental Involvement 0.96 0.98 0.94*** 0.98 0.98 0.97* 0.96* Parental Alcohol Use 2.81*** 2.63 ** 1.44*** 1.48**1.41** 2.13*** 2.13*** Peer alcohol use 8.02*** 41.5 6*** 1.46** 1.48**1.28* 3.79*** 4.98*** Note: All analyses are controlling for demographics and treatment. Clustered robust standard errors were calculated to account for the clustered sampling design. *p<0.05 **p<0.01 ***p<0.001

PAGE 257

257Table 5-8. Mediated effect of peer marijuana use on aggression trajectories, African-Americans. Mediator Indirect Effect (ab ) z SE Percent Mediated Parental Involvement Alcohol Use 0.255 1.60 0.16 9.7 Marijuana Use 0.176 1.19 0.15 6.7 Unsupervised time 0.138 5.80*** 0.02 5.2 Depression 0.183 5.06*** 0.04 6.9 Academic Achievement 0.139 4.61*** 0.03 5.3 Group Fighting 0.416 3.65*** 0.11 15.7 Baseline Aggression 0.414 6.05*** 0.07 15.7 Total 65.2 Adults in neighborhood use alcohol Alcohol Use 0.398 1.55 0.26 18.2 Marijuana Use 0.195 1.19 0.16 8.9 Unsupervised time 0.109 5.03*** 0.02 4.9 Depression 0.141 4.49*** 0.03 6.4 Academic Achievement 0.101 4.17*** 0.02 4.6 Group Fighting 0.345 3.58*** 0.09 15.8 Baseline Aggression 0.347 4.32*** 0.08 15.9 Total 74.9 Peer Alcohol Use Alcohol Use 0.283 1.39 0.20 12.1 Marijuana Use 0.146 0.94 0.15 6.3 Unsupervised time 0.088 4.41*** 0.02 3.7 Depression 0.118 4.14*** 0.03 5.1 Academic Achievement 0.079 4.13*** 0.02 3.4 Group Fighting 0.411 3.31*** 0.12 17.6 Baseline Aggression 0.492 3.18** 0.15 21.1 Total 69.4 Notes: All models are adjusted for demographics and treatment. Clustered robust standard errors were calculated to account for the clustered sampling design. (a): These mediated effects were generated in acco rdance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhsh, & Veblen-Mortenson (2001). The perce nt mediation was generated using the formula: [(a*b/(a*b + c)] (MacKinnon, 2008). (b): Indirect effects are not direct ly comparable across variables. Percent mediation is comparable across variables and groups of variables. **p<0.01 ***p<0.001

PAGE 258

258 Table 5-9. Post-hoc description (means and proportions) of adolescents who were aggressive at Wave I, African-Americans. Aggression at Baseline AggressiveNonAggressive p Community-level Alcohol outlet density a 0.24 0.24 0.787 Area Deprivation a --27.99 -29.64 0.486 Parental and Peer Influences Parental Involvement a 22.17** 21.36 0.003 Adults in Neighborhood Drink 0.56*** 0.36 <0.001 Home Access to Alcohol 0.25* 0.19 0.019 Peer alcohol use 0.39*** 0.11 <0.001 Individual-level Risk Factors Alcohol use 0.21*** 0.04 <0.001 Marijuana use 0.07** 0.00 0.005 Low Academic Achievement 0.71*** 0.53 <0.001 Unsupervised time 0.53*** 0.42 <0.001 Depression 0.78*** 0.58 <0.001 Aggression Group fighting 0.40*** 0.06 <0.001 Demographics Age at Baseline a 11.77** 11.90 0.002 Family Composition 0.32*** 0.46 <0.001 Reduced Lunch 0.74* 0.66 0.041 Note: Participants were considered physical aggressive at baseline if they reported any of the aggression items that were used to estimate trajectories. a Mean is reported. *p<0.05 **p<0.01 ***p<0.001

PAGE 259

259 Table 5-10. Bivariate effects between risk/ protective factors and trajectories of aggression, Hispanics. Trajectory Group Desistors Escalators Consistently Aggressive OR 95% CI OR 95% CI OR 95% CI Community-level Alcohol outlet density 0.88 0.13-5. 93 2.74 0.78-9.70 2.08 0.21-20.09 Area Deprivation 0.99 0.98-1.00 0.99* 0.98-0.99 0.98 0.97-1.00 Adults in Neighborhood Drink 1.43 0.93-2.20 1.47 0. 97-2.23 3.42*** 2.41-4.84 Parental and Peer Influences Parental Involvement 0.96 0.921.01 0.98 0.95-1.03 0.91 0.87-0.95 Home Access to Alcohol 1.07 0.78-1.45 1.17 0. 79-1.72 2.48*** 1.80-3.42 Peer alcohol use 3.39*** 2.01-5. 71 2.08* 1.10-3.93 5.49*** 3.35-8.97 Individual-level Risk Factors Alcohol use 2.17* 1.08-4.36 2.48* 1.12-5.50 5.51*** 2.47-12.28 Marijuana use 5.08 0.53-47.933.29 0.34-31.75 17.52** 2.08-147.16 Low Academic Achievement 2.62*** 1.71-4.01 2.11** 1. 40-3.17 3.48*** 2.47-4.91 Unsupervised time 1.81*** 1.31-2.48 1.20 0.87-1.63 2.05*** 1.45-2.91 Depression 1.88** 1.23-2.87 1. 50* 1.06-2.13 1.62*** 1.88-3.65 Spanish at Home 0.55** 0.35-0. 86 0.64* 0.43-0.97 0.50** 0.32-0.78 Aggression Group fighting 2.49** 1.34-4.65 1. 42 0.86-2.35 5.36*** 3.12-9.23 Baseline aggression 4.39*** 3.14-6.16 2.83*** 1.94-4.13 14.30*** 10.90-18.76 Demographics Age at Baseline 1.15 0.80-1. 64 1.24 0.96-1.77 1.51* 1.08-2.09 Family Composition 0.86 0.60-1.26 0.60* 0.41-0.89 0.64** 0.47-0.85 Reduced Lunch 1.21 0.72-2.03 1.14 0.75-1.73 0.93 0.58-1.49 Note: The Non-Aggressive trajectory group serves as the reference category. All analyses are controlling for treatment. Clustered robust standard e rrors were calculated to account for the clustered sampling design. Marijuana use was collinear with aggression, and was removed from further models. *p<0.05 **p<0.01 ***p<0.001

PAGE 260

260 Table 5-11. Community, family and peer effects on trajectories of aggression, Hispanics. Trajectory Group Desistors Escalators Consistently Aggressive OR 95% CI OR 95% CI OR 95% CI Community-level Neighborhood Deprivation 0.99 0.981.000.99* 0.98-0.990.99 0.98-1.00 Adults in Neighborhood Drink 1.05 0. 69-1.611.23 0.81-1.852.18*** 1.55-3.05 Parental and Peer Influences Parental Involvement 0.97 0.931.021.00 0.97-1.030.95** 0.91-0.99 Home Access to Alcohol 0.83 0.611.121.02 0.70-1.491.73*** 1.29-2.32 Peer alcohol use 3.91*** 2.32-6. 582.28** 1.28-4.054.38*** 2.73-7.05 Note: The Non-aggressive trajectory group serves as the reference category. All analyses are controlling for demographics and treatment. Clustered robu st standard errors were calculated to account for the clustered sampling design. *p<0.05 **p<0.01 ***p<0.001

PAGE 261

261 Table 5-12. Effects of multiple domains of risk factors on trajectories of aggression, Hispanics. Trajectory Group Desistors Escalators Consistently Aggressive OR 95% CI OR 95% CI OR 95% CI Community-level Neighborhood Deprivation 0.99 0.98-1. 00 0.99* 0.98-0.99 0.99 0.98-1.00 Adults in Neighborhood Drink 0.95 0.59-1.52 1.14 0.731.78 1.92** 1.30-2.82 Parental and Peer Influences Parental Involvement 0.98 0.94-1. 03 0.99 0.96-1.04 0.96* 0.91-0.99 Home Access to Alcohol 0.72* 0.521.02 0.88 0.60-1.29 1.45** 1.11-1.91 Peer alcohol use 2.94*** 1.72-5.02 1.94* 1.06-3.54 2.78*** 1.68-4.61 Individual-level Risk Factors Alcohol use 1.63 0.55-6.26 2.12 0.66-6.81 2.22 0.66-7.52 Low Academic Achievement 2.09** 1.25-2. 43 1.77* 1.08-2.90 2.09** 1.39-3.15 Unsupervised time 1.72** 1.22-2.43 1.05 0.75-1.46 1.46 0.92-2.35 Depression 1.30 0.80-2.12 1.21 0.76-1.91 1.61* 1.07-2.41 Spanish at Home 0.53** 0.33-0.83 0.69 0.46-1.04 0.56* 0.35-0.91 Aggression Group fighting 1.71 0.90-3.25 1.05 0.61-1.80 3.30*** 1.84-5.91 Note: The Non-aggressive trajectory group serves as the reference category. All analyses are controlling for demographics and treatment. Clustered robu st standard errors were calculated to account for the clustered sampling design. Marijuana use was collinear with aggression, and was therefore not included in these analyses. *p<0.05 **p<0.01 ***p<0.001

PAGE 262

262 Table 5-13. Effects of multiple domains of risk factors on trajectories of aggression, adjusted for baseline, Hispanics. Trajectory Group Desistors Escalators Consistently Aggressive OR 95% CI OR 95% CI OR 95% CI Community-level Neighborhood Deprivation 0.99 0.991.000.99* 0.98-0.99 0.99 0.98-1.00 Adults in Neighborhood Drink 0.81 0. 49-1.351.03 0.67-1.58 1.55* 1.03-2.32 Parental and Peer Influences Parental Involvement 0.98 0.94-1. 020.99 0.96-1.03 0.95* 0.91-0.99 Home Access to Alcohol 0.65* 0.460.920.81 0.55-1.20 1.26 0.98-1.64 Peer alcohol use 2.57** 1.46-4.55 1.76 0.95-3.24 2.28** 1.35-3.88 Individual-level Risk Factors Alcohol use 1.31 0.38-4.531. 60 0.50-5.13 1.40 0.41-4.73 Low Academic Achievement 1.82* 1.083.061.60 0.95-2.69 1.59* 1.02-2.47 Unsupervised time 1.69** 1.20-2. 381.04 0.74-1.44 1.39 0.86-2.25 Depression 1.10 0.70-1.731. 07 0.69-1.65 1.26 0.85-1.89 Spanish at Home 0.54** 0.35-0.83 0.71 0.47-1.08 0.56* 0.34-0.90 Aggression Group fighting 1.20 0.60-2.400. 79 0.45-1.39 2.00* 1.13-3.55 Baseline aggression 3.28*** 2.09-5.132.38** 1.49-3.80 8.23*** 5.67-11.94 Note: The Non-aggressive trajectory group serves as the reference category. All analyses are controlling for demographics and treatment. Clustered robu st standard errors were calculated to account for the clustered sampling design. *p<0.05 **p<0.01 ***p<0.001

PAGE 263

263 Figure 5-3. Test of the mediation pathway (based on bivariate analysis), Hispanics. Social/Contextual Risks Neighborhood deprivation Parental involvement Peer alcohol use Adult alcohol use Home access to alcohol Aggressive Trajectory Membership a b c Alcohol use Unsupervised time Depression Academic achievement Acculturation Group Fighting Baseline aggression Individual-Level Risks

PAGE 264

264 Table 5-14. Regression models testing t he association between parent and peer level variables and individual-level risk and protective factors, Hispanics. Mediators Alcohol Use MJ Use Unsuperv ised time Depres sion Academi c Achieve ment Accultu ration Group Fighting Baseline Aggression OR OR OR OR OR OR OR OR Contextual Variables Neighborhood Deprivation 0.99 1.00 1.00 1.00 1.00 0.99 0.99 0.99 Parental Involvement 0.92** 0.87*** 0.93*** 0.99 0.96* 1.02 0.93*** 0.94** Parental Alcohol Use 2.46*** 2.67** 1.76*** 1.71***1.56** 0.76* 1.65*** 2.70*** Home Access to Alcohol 4.41*** 4.03*** 1.74** 1.20 1.79*** 0.91 3.86*** 3.96*** Peer alcohol use 8.82*** 15.52*** 1.41* 2.46***2.00*** 0.71* 1.79*** 2.19*** Note: All analyses are controlling for demographics and treatment. Clustered robust standard errors were calculated to account for the clustered sampling design. *p<0.05 **p<0.01 ***p<0.001

PAGE 265

265 Table 5-15. Mediated effect of multiple domains of risk fa ctors on aggression trajectories, Hispanics. Mediator Indirect Effect (ab ) zSE Percent Mediated Parental Involvement Alcohol Use 0.337 1.69 0.20 12.5 Marijuana Use1 Unsupervised time 0.153 6.31*** 0.02 5.6 Depression 0.187 6.48*** 0.03 6.9 Academic Achievement 0.239 5.94*** 0.04 8.9 Acculturation 0.059 4.95*** 0.01 2.2 Group Fighting 0.274 3.96*** 0.07 10.1 Baseline Aggression 0.053 6.80*** 0.08 19.8 Total 65.9 Adults in neighborhood use alcohol Alcohol Use 0.219 2.39* 0.09 11.9 Marijuana Use1 Unsupervised time 0.143 5.03*** 0.03 7.8 Depression 0.070 2.76** 0.03 3.8 Academic Achievement 0.191 5.00*** 0.04 10.3 Acculturation 0.026 4.51*** 0.006 1.5 Group Fighting 0.209 3.68*** 0.06 11.4 Baseline Aggression 0.458 5.22*** 0.09 24.9 Total 71.7 Home Access to Alcohol Alcohol Use 0.311 2.32* 0.13 16.7 Marijuana Use1

PAGE 266

266 Table 5-15. Continued. Mediator Indirect Effect (ab ) zSE Percent Mediated Home Access to Alcohol Unsupervised time 0.140 5.02*** 0.03 7.6 Depression 0.123 4.29*** 0.03 6.6 Academic Achievement 0.217 5.05*** 0.04 11.7 Acculturation 0.030 4.06*** 0.007 1.6 Group Fighting 0.227 3.50*** 0.06 12.2 Baseline Aggression 0.397 4.84*** 0.08 21.4 Total 77.9 Peer Alcohol Use Alcohol Use 0.209 2.12* 0.09 9.8 Marijuana Use1 Unsupervised time 0.099 4.57*** 0.02 4.6 Depression 0.160 5.03*** 0.03 7.5 Academic Achievement 0.187 4.49*** 0.04 8.8 Acculturation 0.021 4.24*** 0.005 0.9 Group Fighting 0.266 3.45*** 0.08 12.5 Baseline Aggression 0.479 4.57*** 0.10 22.5 Total 66.7 Notes: All models are adjusted for demographics and treatment. Clustered robust standard errors were calculated to acc ount for the clustered sampling design. 1Mediation analyses were not conducted with marijuana use due to collinearity. (a): These mediated effects were generated in accordance with MacKinnon (2008) and Komro, Perry, Williams, Stigler, Farbakhs h, & Veblen-Mortenson (2001). The percent mediation was generated using the formula: [(a*b/(a*b + c) ] (MacKinnon, 2008). (b): Indirect effects are not directly com parable across variables. Percent mediation is comparable across variables and groups of variables. *p<0.05 **p<0.01 ***p<0.001

PAGE 267

267 Table 5-16. Post-hoc description (means and proportions) of adolescents who were aggressive at 6th grade (baseline), Hispanics. Aggression at Baseline AggressiveNonaggressive p Community-level Alcohol outlet density a 0.21 0.21 0.848 Area Deprivation a -38.83 -35.46 0.332 Parental and Peer Influences Parental Involvement a 20.36** 21.57 0.001 Adults in Neighborhood Drink 0.45*** 0.23 <0.001 Home Access to Alcohol 0.36*** 0.21 <0.001 Peer alcohol use 0.44*** 0.16 <0.001 Individual-level Risk Factors Alcohol use 0.28*** 0.08 <0.001 Marijuana use 0.05*** 0.01 <0.001 Low Academic Achievement 0.77*** 0.53 <0.001 Unsupervised time 0.51*** 0.39 <0.001 Depression 0.78*** 0.59 <0.001 Aggression Group fighting 0.33*** 0.08 <0.001 Demographics Age at Baseline a 11.81 11.77 0.087 Family Composition 0.66 0.71 0.151 Reduced Lunch 0.75 0.74 0.862 Note: Participants were considered violent at baseline if they reported any of the aggression items that were used to estimate aggression trajectories. a Mean is reported. **p<0.01 ***p<0.001

PAGE 268

268 CHAPTER 6 DISCUSSION This study used two distinct samples of adolescents: 1) a natio nally representative sample of adolescents (beginning at age 11) followed longitudinally through age 32; and 2) a high-risk sample of urban adolescent s in Chicago, IL, followed from 6th through 8th grade (ages 11-14). Between thes e two contexts, there was c onsiderable variability in the patterns and predictors of violence and aggr ession over time. Within the high-risk sample from Chicago, four distinct pattern s of aggressive behaviors among adolescents were found for both African-Americans and His panics. In the nationally representative sample, three groups were found for the overall sample as well as each subgroup (White males, White females, African-Am erican males, African-American females, Hispanic males, Hispanic females, Asians, and Native Americans). Despite the differences in the number of type of trajectory groups acro ss the samples, the number and type of trajectory groups are consist ent with the prior lit erature seeking to understand trajectories of delinquency among adolescents (Piquero, 2008). The predictors of membership in the hi gh-risk trajectory groups varied between samples and within samples by race/et hnicity and gender. Study 1 (Chapter 3) estimated trajectories of serious violence using a longitudinal sample of adolescents, considering group fighting (independent of baseline violence) as a risk factor in differentiating profiles of violent behavio rs. Participants included a nationally representative sample of 9421 adolescents followed from ages 15 through 26. Three groups of violence trajectories were identif ied: 1) Non-Violent; 2) Escalators; and 3) Desistors. Group fighting significantly predicted violence above and beyond baseline violent behavior for desistors only. Raci al dispersion at the neighborhood level

PAGE 269

269 predicted both escalation and desistance before individual-level ch aracteristics were entered into the model. Peer alcohol use predicted both escalation and desistance, and these effects were mediated through individ ual-level variables. Aside from baseline violence, the models did not explain membersh ip in the escalator trajectory group. Group fighting was a significant predictor of violence beyond baseline violence, but only for those who desisted from violent activi ty. Contextual vari ables (such as peer substance use and racial composition of the neighborhood) appear to increase risk for violence. Study 2 (Chapter 4) used the same longitu dinal, nationally representative sample of adolescents to estimate traj ectories of serious violence stratified by race/ethnic and gender subgroup. Specifically, this study s ought to test where the number and pattern of trajectories vary by race/ethnici ty and gender group, and w hether predictors of membership in the violent trajectory groups differ by race/ethnicity and gender. Similar to Study 1, three groups of violence trajectories were id entified for all subgroups: 1) Non-Violent; 2) Escalators; and 3) Desistor s. Although the number of groups was the same across gender and race/ethnicity, the pr oportion of the sample in the violent groups differed substantially by subgroup. Group fighting significantly predicted violence above and beyond baseline violent behavior for Asian desistors, White male and female desistors, African-American ma le and female desistors, and AfricanAmerican male escalators. Contextual (urban neighborhood residence, racial dispersion, peer alcohol and marijuana us e, and parental alcohol use) appear to increase risk for violence differentially by subgroup.

PAGE 270

270 Finally, the purpose of Study 3 was to esti mate trajectories of aggression using a longitudinal sample of urban adolescents by racial/ethnic subgroups, considering group fighting (independent of baseline aggression) as a risk factor in differentiating profiles of aggression. Participants incl uded a multi-ethnic, urban sa mple of 4,188 adolescents followed from ages 11 to 14. Unlike Studies 1 and 2, four groups of aggression trajectories were identified for African-Americans and Hispanics. Among Hispanics, the four groups included: 1) low-aggr ession, 2) desistors, 3) escalators, and 4) consistently aggressive adolescents. Among African-Americans, the four groups included: 1)lowaggression, 2) escalators, 2) moderate-c onsistent aggression, and 4) consistent aggression. Group fighting significant ly predicted aggression above and beyond baseline aggression for both Hispanics and African-Americans who were consistently aggressive. A number of differences in the multiple domains of risk and protective factors emerged between groups. There were a number of similarities in the contextual-level risk and protective factors found across the three studies. In Study 1, group fighting significantly predicted membership in the desistor group independent of baseline violence, but group fighting was not significant in predicting escalation. This finding was supported in Study 2; as group fighting predicted desistance among As ians, White males, White females, and African-American males. Only among African-American Males did group fighting predict escalation. In Study 1, raci al dispersion (i.e., racial heterogeneity) at the census tract level had a marginal, direct effect on escalation, independent of demographics and baseline violence. This finding was supported am ong White male escalators in Study 2. Adult modeling of alcohol use was a risk fa ctor for violence among African-American

PAGE 271

271 male escalators (Study 2) and African-American consistently violence adolescents (Study 3). A number of individual-level predictors were also consistent across the three studies. Among the full sample (Study 1), effects of peer alcohol and marijuana use were mediated through individual-level alcoho l and marijuana use. Subgroup analyses (Study 2) revealed that peer alcohol us e had a direct effect on desistance among Whites, and an indirect effect on violence among Whites and Hispanics. Marijuana use had an indirect effect on violence across all subgroups except for Asians. In Study 1, the effect of parental involvement was mediat ed by multiple individual-level variables, including the adolescents desire to leave ho me, as well as their drug and alcohol use. Study 2 supported the indirect effect of par ental involvement among White females and Hispanics. Study 3 found an indirect effect for parental involvement among AfricanAmericans and a direct effect among Hispanics who are consistently violent. Finally, baseline violence was significantly associ ated with desistance and escalation in Study 1, and among all subgroups in St udy 3. Study 2 reveals a more complex relationship, as baseline violence directly affected viol ent trajectory membership among White males, African-American desistors, Hispani cs, and Native American escalators. In light of these similarities, there ar e a number of differences between Studies 1 and 2, which use the same sample of adole scents. Specifically, depression appears to be an important predictor of violence among Hispanic males, and speaking Spanish at home is protective from violence among Hispani c females. Racial di spersion is a risk factor for Native Americans and White ma les, but not for other populations. Additionally, peer alcohol use had a direct effect on desistance among Whites, but this

PAGE 272

272 was not true of any other group. If s ubgroup analyses were not conducted, these risk and protective factors woul d not be identified. As expected, distal influences had both di rect and indirect effects on aggressive and violent behaviors. Specifically, effects of distal variables such as peer substance use were mediated through individual-level va riables (individual-level substance use). Within the high-risk sample in Chicago, however, peer alcohol use had a direct effect on consistent aggression among African-Americans, and all le vels of aggression among Hispanics. The racial composition of the neighborhood was indirectly related to violence among White males, and poverty had an indirect effect on risk for violence among Hispanic females. The most potent predi ctors of violent trajectory membership were group fighting and baseline violence. Due to the strength of the association between past violence and future violent behav ior, the substantia l influence of these previous violent behaviors mask the effect that contextualand peer-level variables may have on violence over time or prior to the onset of aggressive and violent behaviors. These findings are consistent with the theoretical models guiding this research. All three studies find trajectory groups of adolescence-limited offenders (e.g., desistors) and life-course persistent offenders (e.g., escalators and consistently violent) (Moffitt, 2005; Piquero, 2008). The direct effect of peer alcohol use on violence among those in urban Chicago and adolescents nationwide highli ghts the importance of social learning on violent behavior (Akers, 1973). Also in su pport of this theory, modeling of alcohol consumption by parents increases the risk for violence across both samples. These findings highlight the role that parents and peers play in risk behavior. The importance of community-level variables, such as perceived adult alcohol consumption in the

PAGE 273

273 community, racial dispersion, and urbanici ty, provide support for Shaw and McKays (1942) Theory of Social Disorganization. Finally, direct effects for academic performance (Study 3) and desire to leave hom e (Study 2) provide support for Hirschis (1969) Social Bond Theory. Adolescents who seek to leave their neighborhoods, friends, and families, and perform poorly in academics may be less bonded to conventional society and therefore, mo re likely to engage in violent behavior. Collectively, these findings have significant implications for violence and delinquency prevention programming. First, a ll three studies consis tently found that violent and aggressive behavior begins early, even before age 11 in both the general population and high-risk settings. This i ndicates that the current prevention programming occurs too late, as violence has already begun as early as elementary school. Second, these findings indicate that social influences, such as exposure to peers who use alcohol or marijuana, and community-level exposure to alcohol influence adolescents risk for violent behavior. These risk factors that were consistent across race/ethnicity and gender may be targeted in a variety of populations to reduce participation in violence. These findings also suggest that there are substantial differences in the variable that influence violent behavior between subgroups of adolescents. Therefore, the demographic co mposition of the in tervention population should be considered prior to program adminis tration, as different subgroups may be exposed to different risk and pr otective factors. This differential exposure may equate to increased or diminished propensity for violence. Findings from this study should be consider ed in light of several limitations. First, this study was unable to acc ount for some of the variables that are important in

PAGE 274

274 predicting violence, such as peer violence cognitive development, and psychological disorders (Moffitt, 1993; Zara & Farringt on, 2009). Second, latent-group based trajectory modeling provides an estimation of the type and num ber of groups in the data, and this process is exploratory in nature. Despite the expl oratory nature of trajectory estimation, the results of this study we re consistent with the expected number and shape of trajectory groups from other studies (Piquero, 2008; Zara & Farrington, 2009). Finally, risk factors were analyzed at multiple levels; however, hierarchical linear modeling (HLM) was not used due to the small sample sizes available in some of the trajectory groups. In accordance with the sampling design, all analyses accounted for the nesting of adolescents within schools, which may account for a portion of the variability in Census tract measures. Us e of HLM to account for the nesting within Census blocks is a direction for future research. Despite these weaknesses, the current study had a number of strengths. First, data were derived two longitudinal studies, a llowing an evaluation of intra-individual change in violent and aggressive behavior over ti me. The trajectory models estimated in this study are especially appropriate for studies of delinquency and aggression, as patterns tend to change over time (Farringt on, 1986; Piquero, 2008). Second, the nationally representative sample used in Studies 1 and 2 allows generalization to a national sample of adolescents across the United States. The high-risk, urban sample in Study 3 provides additional support fo r the findings that there are substantial differences across race/ethnicity and gender, as well as across geographical location and context. Third, although many studies have analyzed the multiple domains of risk and protective factors for violent behavio r, few have assessed the degree to which

PAGE 275

275 contextual variables are mediated by more pr oximal variables at t he individual-level. Consideration of the mediat ed effects acknowledge that contextual variables are important in predicting violence even t hough their effects are mitigated using multivariate regr ession models. In summary, this study contributes to the literature on violent behavior among adolescents. Findings from Studies 1 and 2 pr ovide additional evidence for a late-onset group of escalators (Zara & Farrington, 2009), as well as racial/ethnic and gender differences within this group of escalators. Because of the large sample size used in Studies 1 and 2, and the considerable racial/et hnic diversity in Study 3, this study was able to identify risk and protective factor s for aggression and violence broken down by subgroup. In addition, this study was able to identify the key variables that serve as mediators for contextual-, peerand family-l evel variables, an important component of violence prevention programming. These find ings lead future research to examine the disparities in violence by subgroup, and to further understand the risk factors for lateonset escalation of violent behavior.

PAGE 276

276 LIST OF REFERENCES Adams, M. (2009). A delin quent discipline: The rise and fall of criminology. A cademic Questions, 22 491-503. Agnew, R. (1994). The techniques of neutralization and violence. Criminology, 32(4), 555-580. Akers, R. (1973). Deviant behavior: A social learning approach. Belmont, CA: Wadsworth. Anderson, K. L. (2010). Conflict, power, and violence in families. Journal of Marriage and Family, 72, 726-742. Ball, R. A., & Curry, G. D. ( 1995). The logic of definition in criminology: Purposes and method for defining gangs. Criminology, 33, 225-245. Bandura, A. (1971). Social learning t heory: New York: G eneral Learning. Barnes, G. M., Welte, J. W ., & Hoffman, J. H. (2002). Rela tionship of alcohol use to delinquency and illicit drug use in adolescents: Gender, age, and racial/ethnic differences. Journal of Drug Issues, 22 ,153-178. Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. Belknap, J., & Holsinger, K. (2006). The gendered nature of risk fa ctors for delinquency. Feminist Criminology, 1 48-71. Bellair, P. E., & McNulty, T. L. (2005). Beyond the bell curve: Community disadvantage and the explanation of Afric an-American-white differences in adolescent violence. Criminology, 43 1135-1168. Bendixen, M., Endresen, I. M., & Olweus, D. (2006). Joining and leaving gangs: Selection and facilitation effects on se lf-reported antisocial behavior in early adolescence. European Journal of Criminology, 3 85-114. Bjerregaard, B., & Smith, C. (1993). Gender differenc es in gang participation, delinquency, and substance use. Journal of Quantitative Criminology, 4 329-355. Blum, R. W., Beuhring, T., Shew, M. L., Beari nger, L. H., Sieving, R. E., & Resnick, M. D. (2000). The effects of race/ethnicity, income, and family structure on adolescent risk behaviors. American Journal of Public Health, 90 1879-1884. Blum, J., Ireland, M., & Blum, R. W. (2003). Gender difference s in juvenile violence: A report from Add Health. Journal of Adolescent Health, 32 234-240.

PAGE 277

277 Booth, J. A., Farrell, A., & Varano, S. P. ( 2008). Social control, serious delinquency, and risky behavior: A gendered analysis. Crime & Delinquency, 54 423-456. Boles, S. M., & Miotto, K. ( 2003). Substance use and violence: A review of the literature. Aggression and Violence Behavior, 8 155-174. Brady, S. S., Tschann, J. M., Pasch, L. A., Flores, E., & Ozer, E. J. (2008). Violence involvement, substance use, and sexual activity among Mexican-American and European-American adolescents. Jour nal of Adolescent Health, 43, 285-295. Broidy, L. M., Tremblay, R. E., Brame, B., Fergusson, D ., Horwood, J. et al. (2003). Developmental trajectories of childhood disruptive behaviors and adolescent delinquency: A six-site, cross-national study. Developmental Psychology, 39 222245. Caetano, R. (2006). Acculturation and drinking patterns among us hispanics. Addiction, 82, 789-799. Caetano, R., Schafer, J., Clark, C. L., Cunradi C. B., & Raspberry, K. (2000). Intimate partner violence, acculturation, and alc ohol consumption among hispanic couples in the united states. Journal of Interpersonal Violence, 15, 30-45. Caetano, R., Ramisetty-Mikler, S., & Rodri guez, L. A. (2008). The hispanic americans baseline alcohol survey (hablas): DUI ra tes, birthplace, and acculturation across hispanic national groups. Journal of Studies on Alcohol and Drugs, 69, 259-265. Centers for Disease Control and Prevention (2009). Youth risk behavioral surveillance United States, 2009. MMWR, 2010 59. Centerwall, B. S. (1995). Race, socioec onomic status, and domestic homicide. Journal of the American Medical Association 273 1755-1758. Chantala, K., Tabor, J. (1999) Strategies to Perform a Design-Based Analysis Using the Add Health Data. National Longitudinal Study of Adolescent Health. Chesney-Lind, M. (1983). Girls and violence An exploration of the gender gap in serious delinquent behavior. Rockville, MD: National Institute of Justice. Chesney-Lind, M., & Shelden, R. G. (1998). Girls, delinquen cy, and juvenile justice (2nd Edition). Belmont, CA: Wadsworth. Conseur, A., Rivara, F. P., & Emanuel, I. (1997). Juvenile delinquency and adolescent trauma: How strong is the connection? Pediatrics, 99 : e5. Curry, G. D., & Spergel, I. A. (1988). Gang homicide, delinquency, and community. Criminology, 26 381-406.

PAGE 278

278 DAmico, D. J., Edelen, M. O., Miles, J. N., & Morral, A. R. (2008). The longitudinal association between substance use and delinquency among high-risk youth. Drug and Alcohol Dependence 93, 85-92. Daigle, L. E., Cullen, F. T., & Wright, J. P. (2007). Gender differences in the predictors of juvenile delinquency: Assessing the Generality-S pecificity debate. Youth Violence and Juvenile Justice, 5, 254-285. Dembo, R., Williams, L., Getr eu, A., Genung, L., Schmeidler J., Berry, E., LaVoie, L. (1991). A longitudinal study of the relati onships among marijuana/hashish use, cocaine use, and delinquency in a cohort of high risk youths. Journal of Drug Issues 21, 271-312. Dhungana, K. J. (2009). Risk factors of gang membersh ip: A study of community, school, family, peer and individual level predictors among th ree south florida counties. Unpublished doctoral dissertat ion, Florida State University. DUnger, A. V., Land, K. L., & McCall, P. L. (2002). Sex differences in age patterns of delinquent/criminal careers: Results from poisson latent class analyses of the Philadelphia cohort study. Journal of Quantitative Criminology, 18 349-375. Eggleston, E. P., Laub, J. H., & Sampson, R. J. (2004). Methodologi cal sensitivities to latent class analysis of longterm criminal trajectories. Journal of Quantitative Criminology, 20, 1-26. Elbogen, E. B., & Johnson, S. C. (2009). The intricate link bet ween violence and mental disorder: Results from the national epidemiologic survey of alcohol and related conditions. Archives of General Psychicatry, 66 152-161. Ellickson, P. L., Tucker, J. S. & Klein, D. J. ( 2003). Ten-year prospective study of public health problems associated with drinking. Peditrics, 111 949-955. Elliott, D. (1984). National Youth Survey, Wave VI (Version 1) [Data file]. Ann Arbor, MI: Inter-university Consortium for Political and Social Research. Esbensen, F. (2000, September). Pr eventing adolescent gang involvement. Juvenile Justice Bulletin, 1-6 Esbensen, F., & Deschenes, E. P. (1998). A multisite exami nation of youth gang membership: Does gender matter? Criminology, 36 799-828. Esbensen, F., & Weerman, F. M. (2005). Y outh gangs and troublesome youth groups in the United States and the Netherlands : A cross-national comparison. European Journal of Criminology, 2 5-37. Esbensen, F., Winfree, L. T., He, N., & Taylor T. J. (2001). Youth gangs and definition issues: When is a gang a gang, and why does it matter? Crim e & Delinquency, 47 105-130.

PAGE 279

279 Esbensen, F., Peterson, D., Taylor, T. J., & Freng, A. (2010). Youth violence: Sex and race differences in offending, victimiz ation, and gang membership. Philadelphia, PA: Temple. Farrington, D. P. ( 1986). Age and crime. Crime & Justice, 7 189-250. Farrington, D. P. & Loeber, R. (2000). Epidemiology of juvenile violence. Juvenile Violence, 9, 733-748. Farrington ,D. P., Loeber, R. & Stouthamer-Loeber, M. (2003). How can the relationship between race and violence be explained? In Hawkins, D. F. (Ed.), Violent crime: Assessing race and ethnic differences (213-237). New York: Cambridge. Goldstein, P. J. (1985). Drugs and violent behavior. Journal of Drug Issues, 15 493506. Guo, G. Elder, G. H., Cai, T., & Hamilton, N. (2009). Gene-environment interactions: Peers alcohol use moderates genetic contribution to adolescent drinking behavior. Social Science Research, 38 213-224. Harris, K.M., C.T. Halpern, E. Whitsel, J. Hussey, J. Tabor P. Entzel, and J.R. Udry. (2009). The National Longitudinal Study of Adolescent Health: Research Design Retrieved January 5, 2011, from the Univer sity of North Carolina Chapel Hill Web Site: http://www.cpc.unc.edu/projects/addhealth/design. Hawkins, J. D., Herrenkohl, T. I., Farrington, D. P., Brewer, D., Catalano, R. F. et al. (2000, April). Predictors of youth violence. Juvenile Justice Bulletin, 1-10. Hedeker, D. (2003). A mixed-effects mu ltinomial logistic regression model. Statistics in Medicine, 22 1433-1446. Herrenkohl, T. I., McMorris, B. J., Catalano, R. F., Abbott, R. D., Hemphill, S. A., & Toumbourou, J. W. (2007). Risk factors for violence a nd relational aggression in adolescence. Journal of Interpersonal Violence, 22 386-405. Hirschi, T. (1969). Causes of delinquency. Berkeley, CA: University of California. Hirschi, T., & Gottfredson, M. (1983). Age and the explanation of crime. The American Journal of Sociology, 89 552-584. Howell, J. C. (1998, August) Youth gangs: An overview. Juvenile Justice Bulletin ,1-20. Huizinga, D., Loeber, R., & Thornberry, T. P. (1995). Recent findings from the program of research on the causes and correlate s of delinquency. Rockville, MD: National Institutes of Justice.

PAGE 280

280 Jennings, W. G., Maldonado-Molina, M. M., Pi quero, A. R., Odgers, C. L., Bird, H., & Canino, G. (2010). Sex differences in traj ectories of offending among Puerto Rican youth. Crime & Delinquency, 56, 327-357. Jessor, R., Donovan, J. E., & Costa, F. M. (1991). Beyond adolescence: Problem behavior and young adulthood. New York: Cambridge. Johnston, L. D., OMalley, P. M., Bachm an, J. G., & Schulen berg, J. E. (2010). Monitoring the Future: National Results on Adolescent Drug UseOverview of Key Findings, 2009. Bethesda, MD: National Institutes of Drug Abuse. Johnston, L. D., OMalley, P. M., & Ev eland, L. (1978). Drugs and delinquency: A search for causal connections In D. B. Kandel (Ed.), Longitudinal research on drug use: Empirical findi ngs and methodological issues (pp. 137-156). Washington, DC: Hemisphere. Jones, B. L., Nagin, D. S., & Roeder, K. (2001). A SAS procedure based on mixed models for estimating devel opmental trajectories. Sociological Methods and Research, 29 374-393. Kaufman, J. M. (2005). Ex plaining the race/ethnici ty-violence relationship: Neighborhood context and social psychological processes. Justice Quarterly, 22, 224-251. Kawachi, I., Kennedy, B. P., & Wilkinson, R. G. (1999). Crime: Soci al disorganization and relative deprivation. Social Science & Medicine, 48, 719-731. Klein, M. W. (1971) Street gangs and street worker s. Englewood Cliffs, NJ: Prentice Hall. Klein, M. W., Weerman, F. M., & Thornberry, T. P. (200 6). Street gang violence in Europe. European Journal of Criminology, 3 413-437. Komro, K. A., Maldonado-Molina, M., Tobler, A. L., Bonds, J. R., & Miller, K. (2007). Effects of home access and availability of alcohol on young adolescents alcohol use. Addiction, 102, 1597-1608. Komro, K. A., Perry, C. L., V eblen-Mortenson, S., Farbakhsh, K., Toomey, T. L., Stigler, M. H., Jones-Webb, R., Kugler, K. C., Pa sch, K. E., & William s, C. L. (2008). Outcomes from a randomized controlled tr ial of a multi-component alcohol use preventive intervention for urban y outh: Project Northland Chicago. Addiction, 103 606-618. Komro, K. A., Perry, C. L., V eblen-Mortenson, S., Bosma, L. Dudovitz, B. S., Williams, C. L., Jones-Webb, R., & Toomey, T. (2004). Brief report: The adaptation of project northland for urban youth. Journal of Pediatric Psychology, 29 457-466.

PAGE 281

281 Kuntsche, E., Gossrau-Breen, D., & Gmel, G. (2009). The role of older siblings and drunken peers in the alcohol-violence nexus. European Journal of Public Health, 1-6. Latzman, R. D., & Swisher, R. R. (2005). The interactive relationship among adolescent violence, street violence, and depression. Journal of Community Psychology, 33 355-371. Leech, S. L., Day, N. L., Richardson, G. A., Goldschmidt, L. (2003). Predictors of selfreported delinquent behavior in a sample of young adolescents. The Journal of Early Adolescence, 23 78-106. Lytton, H., & Romney, D. M. (1991). Parents differential so cialization of boys and girls: A meta-analysis. Psychological Bulletin, 109, 267-296. MacKinnon, D, P. (2008). Introduction to statistical mediation analysis New York: Taylor. MacKinnon, D. P., Warsi, G., & Dwyer, J. H. (1995). A simulation study of mediated effect measures. Multivariate Behavioral Research, 30, 41-62. Maldonado-Molina, M., Piquero, A. R., Jennings, W. G., Bi rd, H., & Canino, G. (2009). Trajectories of delinquency among puerto rican children and adolescents at two sites. Journal of Research in Crime and Delinquency, 46, 144-181. Maldonado-Molina, M., Reingle, J., & Jennings W. (2010). Does alcohol predict violent behaviors? The relationship between alcohol and violence in a nationally representative long itudinal sample. Youth Violence and Juvenile Justice, 9, 99111 Maldonado-Molina, M., Jennings, W. G., & Ko mro, K. A. (2009). Effects of alcohol on trajectories of physical aggression among urban youth: An application of latent trajectory modeling. Journal of Youth and Adolescence, 39 1012-1026. Mattila, V. M., Parkkari, J. P., & Rimpela, A. H. (2006). Risk factors for violence and violence-related injuries among 14to 18-year-old Finns. Journal of Adolescent Health, 38 617-620. McNulty, T. L., & Bellair, P. E. (2003). Explaining racial and ethnic differences in serious adolescent violent behavior. Criminology, 41 709-747. Menard, S. & Elliott, D. S. ( 1993). Data set comparability and short-term trends in crime and delinquency. Journal of Criminal Justice, 21 433-445. Miller, J. (1998). Gender and victimizat ion risk among young women in gangs. Journal of Research in Crime and Delinquency, 35 429-453.

PAGE 282

282 Moffitt, T. E. (1993). The neuropsychology of conduct disorder. Development and Psychopathology, 5 135-151. Moffitt, T. E. (2005). Pathways in the Life Cour se to Crime. In Cullen, F. C., & Agnew, R. (Eds.), Criminological Theory (pp. 502-521). Los Ange les, CA: Roxbury. Moffitt, T. E., Caspi, A., Rutte r, M., & Silva, P. A. (2001). Sex differences in antisocial behavior: Conduct disorder, delinquency, and violence in the Dunedin Longitudinal Study Cambridge, UK: Cambridge. Nagin, D. S, & Tremblay, R. E. (2001). Analyz ing developmental trajec tories of distinct but related behaviors: A group-based method. Psychological Methods, 6, 18-34. Nagin, D. S. (2005). Groupbased modeling of development. Cambridge, MA: Harvard. Nagin, D. S., & Land, K. C. (1993). Age, criminal car eers, and population heterogeneity: Specification and estimation of a nonparametric, mixed poisson model. Criminology, 31 327-362. Ou, S., & Reynolds, A. J. (2010). Childhoo d predictors of young adult male crime. Child and youth services review, 32 1097-1107. Park, S., Morash, M., & Stevens, T. (2010) Gender differences in predictors of assaultive behavior in late adolescence. Youth Violence and Juvenile Justice, 8 314-331. Peterson, R. D., & Krivo, L. J. (2005). Ma crostructural analyses of race, ethnicity, and violent crime: Recent lessons and new directions for research. Annual Review of Sociology, 31, 331-356. Piquero, A. R. (2008). Taking stock of developmental trajec tories of criminal activity over the life course. In Lieberman, A. M. (Ed.), The long view of crime: A synthesis of longitudinal research (pp. 23-78). Washi ngton, DC: Springer. Piquero, N. L., Gover, A., MacDonald, J. M., & Piquero, A. R. (2005). The influence of delinquent peers on delinquency: Does gender matter? Youth and Society, 36, 251-275. Prado, G. J., Schwartz, S. J., Maldonado-Molina, M., Huang, S., Pant in, H., Lopez, B., & Szapocznik, J. (2009). Ecodevelopmental X Interpersonal: Substance use and sexual behavior in Hispanic adolescents. Health Education & Behavior, 36 45-61. Reingle, J., Jennings, W. G., & Maldonado-Molina, M. (in press). Group fighting and Violence Among Hispanic Adolescents and Young Adults: Results from a Nationally Representativ e, Longitudinal Study. Race & Justice.

PAGE 283

283 Resnick, M. D., Ireland, M., & Borowsky, I. (2004). Youth violen ce perpetration: What protects? What predicts? Findings from the national longitudinal study of adolescent health. Journal of Adolescent Health, 35, e1-e10. Sampson, R. J., Morenoff, J. D., & Raudenbush, S. (2005). Social anatomy of racial and ethnic disparities in violence. Public Health Matters, 95 224-232. Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Collective efficacy and crime. Science, 227 ,918-924. Sampson, R. J., & Wilson, W. J. (1995). Toward a theory of ra ce, crime, and urban inequality. In Hagan, J., & Peterson, R. D. (Eds.), Crime and Inequality (pp. 3754). Stanford, CA: Stanford University Press. Senn, T. E., Carey, M. P., & Vanable, P. A. (2010). The intersection of violence, substance use, depression, and STDs: Testing of a syndemic patterns among patients attending an urban STD clinic. Journal of the National Medical Association, 102 614-620. Shaw, C. R., & McKay, H. D. (1942). Ju venile Delinquency and Ur ban Areas. Chicago, IL: University. Snyder, H. N. (2003, Decem ber). Juvenile arrests 2001. Juvenile Justice Bulletin 1-12. Snyder, H. N., & Sickmund, M. (2006). Juvenile offenders and victims: 2006 National Report. Washington, DC: OJJDP. Sobel, M. E. (1982). A symptotic confidence intervals for direct effects in structural equation models. In Leinhardt, S. (Ed.), Sociological Methodology 1982 (pp. 290312). Washington, DC: American Sociological Association. Spergel, I. A. (1993). Gang suppression & intervention: An assessment. Office of Juvenile Justice and Delinquency Prevention (DOJ Report No. 87-JS-CX-K001). Spergel, I. A., Chance, R., Ehr ensaft, K., Regulus, T., Kane, C., Laseter, R. Alexander, A., & Oh, S. (1994). Gang Suppression and Intervention: Community models, Research summary Retrieved on January 10, 2011, from http://www.ncjrs.gov/pdffiles/gangcorr.pdf. Spergel, I. A., Wa, K. M., & Sosa, R. V. (2005). Evaluation of the san antonio comprehensive communitywide approach to gang prevention, intervention and suppression program (DOJ Report No. 209189). StataCorp. (2009). Stata st atistical software: Releas e 11. College Station, TX: StataCorp LP.

PAGE 284

284 Steffensmeier, D., Zhong, H., Ackerman, J. Schwartz, J., & Agha, S. (2006). Gender gap trends for violent crime, 1980 to 2003: A UCR-NCVS comparison. Feminist Criminology, 1 72-98. Stewart, E. A., Simons, R. L., & Conger R. D. (2002). Assessing neighborhood and social psychological influences on ch ildhood violence in an African-American sample. Criminology, 40, 801-830. Sutherland, E. H. (1939). Pr inciples of Criminology, 3rd ed. Philadelphia: J. B. Lippincott. Swahn, M. H., & Hammig, B. J. (2000). Prevalence of youth access to alcohol, guns, illegal drugs, or cigarette s in the home and association with health-risk behaviors. Annals of Epidemiology, 10, 457. Thornberry, T. (1998). Membership in y outh gangs and involvement in serious and violent offending. In Loeber, R., & Farrington, D. P. (Eds.), From songerious and violent juvenile offenders: Risk factors and successful interventions (pp. 147-166). Thousand Oaks, CA: Sage. Thornberry, T. P., Huizinga, D., & Loeber, R. (1995). In Howell, J. C., Krisberg, B. J., Hawkins, J. D., & W ilson, J. J. (Eds.), Sourcebook on Serious, Violent, and Chronic Juvenile Offenders (pp. 61-141). Thousand Oaks: Sage. Thrasher, F. M. (1963). The gang. Chicago, IL: University. Thurnherr, J., Berchtold, A., Michaud, P ., Akre, C., & Suris, J. (2008). Violent adolescents and their educational envir onment: A multilevel analysis. Journal of Developmental and Behavioral Pediatrics, 29, 351-359. Twisk, J. W. R. (2006). App lied multilevel analysis: A prac tical guide. Cambridge, UK: Cambridge. Uniform Crime Reports (2009) [Data file]. Washington, DC: United States Department of Justice. White, H. R., Brick, J., & H ansell, S. (1993). A l ongitudinal investigation of alcohol use and aggression in adolescence. Journal of Studies on Alcohol, 11, 62-77. White, H. R., & Hansell, S. (1996). The moderating effect s of gender and hostility on the alcohol-aggression relationship. Journal of Research in Crime and Delinquency, 33, 450-470. White, H. R., Loeber, R., Stouthamer-Loeber M., & Farrington, D. P. (1999). Developmental associations between subs tance use and violence. Developmental Psychopathology, 11, 785-803. White, R. (2002, September). Sc hool strategies to deal with gangs. Australian Institute of Criminology Trends & Issues, 240, 1-6.

PAGE 285

285 Williams, D. R. (1999). Race, socioeconomic status, and health: The added effects of racism and discrimination. Annals of the New York Academy of Sciences, 896 173-188. Williams, J. H., Van Dorn, R. A., Ayers, C. D ., Bright, C. L., Abbott, R. D. & Hawkins, J. D. (2007). Understanding ra ce and gender differences in delinquent acts and alcohol and marijuana use: A devel opmental analysis of initiation. Social Work Research, 31 71-81. Windle, M. (1990). A longitudinal study of an tisocial behaviors in early adolescence as predictors of late adolescent substance use: Gender and ethnic group differences. Journal of Abnormal Psychology, 99, 86-91. Winfree, L. T., Backstorm, T. V., & Mays, G. L. (1994). Social learning theory, selfreported delinquency, and youth gangs: A new twist on a general theory of crime and delinquency. Youth & Society, 26 147-177. Xue, Y., Zimmerman, M. A. & Cunningham, R. (2009). Re lationship between alcohol use and violent behavior among urban African American youths from adolescence to emerging adulthood: A l ongitudinal study. American J ournal of Public Health, 99, 2041-2048. Youth Risk Factor Surveillance System (2010) [Data file] Atlanta, GA: Centers for Disease Control and Prevention. Zara, G., & Farrington, D. P. (2009). Childho od and adolescent predictors of late onset criminal careers. Journal of Youth and Adolescence, 38, 287-300.

PAGE 286

286 BIOGRAPHICAL SKETCH Jennifer Marie Reingle was born in Aber deen, New Jersey. She has one younger sister, Tricia Reingle. She attended Ma tawan Regional High School in Aberdeen, graduating in 2002. She completed her under graduate education at t he University of North Carolina Wilmington in 2002, with a B.A. in criminal justice (with Univ ersity Honors), and a B.A. in sociology. She ear ned the Honors Medallion for completion of the University Honors progr am at the university. After graduating in December 2002, Jennife r briefly entered the workforce, serving as a Lectures Intern at CAMPUSPEAK, Inc., in Denver, CO. Eight months later, she began her masters degree in criminal justice at the University of Cincinnati. One year later (2007), Jennifer earned her M.S. in crim inal justice, and enrolled as a PhD student in the Department of Health Education and Behavior at the University of Florida. Between 2007 and 2008, she taught two courses and served as a research assistant for Dr. Dennis Thombs. Concurrently, she served as a research assistant on a large metaanalysis in the Department of Epidemio logy and Health Policy Research. In 2009, Jennifer enrolled as a doctoral student in epidemiology at the University of Florida. During her time as a doctor al student, Jennifer co-authored thirteen peerreviewed manuscripts and three book chapt ers. She receiv ed the William L. Simon/Anderson Publishing Outstanding Pa per Award (2011) from the Academy of Criminal Justice Sciences, an Institute for Child Health Policy Pre-Doctoral Fellowship (2009-2011) and the Health Solutions scholarship in 2008. She is a member of the Society for Prevention Research (SPR), and a graduate student mem ber of the National Hispanic Science Network (NHSN). Jennifer seeks to pursue a faculty position where

PAGE 287

287 she can continue her research. She currently lives in Gainesville, Florida, with Bryon and their two labradors, Floyd and Jax.