EFFECTS OF COUNTY AND STATE ECONOMIC, SOCIAL, AND POLITICAL CONTEXTS ON RACIAL/ETHNIC AND GENDER DIFFERENCES IN YOUTH’S PENETRATION INTO THE JUSTICE SYSTEM By Tia Stevens A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Criminal Justice - Doctor of Philosophy 2013 ABSTRACT EFFECTS OF COUNTY AND STATE ECONOMIC, SOCIAL, AND POLITICAL CONTEXTS ON RACIAL/ETHNIC AND GENDER DIFFERENCES IN YOUTH’S PENETRATION INTO THE JUSTICE SYSTEM By Tia Stevens The current study is designed to extend the empirical and theoretical research on disproportionate youth contact with the justice system. Missing from the considerable body of work examining the effects of extralegal factors on police behavior and justice system processing is an examination of the social, political, and economic contextual factors that may influence disparities in justice system contact. The current study addresses this gap by identifying contextual factors associated with severity of justice system response to youth and by identifying the macro-structural environments that disproportionately affect young women and youth of color. Specifically, it examines the direct effects of county and state characteristics on youth risk of arrest and probabilities of charge, a court appearance, conviction, and placement and how the effects of individual characteristics and county and state characteristics interact to disproportionately impact certain groups of youth in certain environments. The main dataset for this study was constructed from the National Longitudinal Survey of Youth (NLSY97). Using the confidential NLSY97 Geocode File, the NLSY97 was appended with county- and statespecific data from various publically available sources indicating structural disadvantage, population composition, political conservatism, prosecutor’s office characteristics, delinquency petition and crime rates, gender inequity, child health and well-being, and juvenile justice policy punitiveness. To take advantage of the longitudinal nature of the NLSY97 data, a combination of multilevel modeling techniques, event history analysis, and generalized linear modeling was employed to examine the effects of individual characteristics and contextual conditions on youths’ risk of arrest and probabilities of charge, a court appearance, conviction, and placement. The findings suggest that the effects of gender and racial/ethnic group on youth penetration into the justice system are more pronounced at some decision-making levels and depend on contextual environment. The results of the analyses by race, gender, and ethnicity suggest three major findings. First, racial disparities are present in youth risk of arrest, which are magnified in predominately non-Black communities. However, this study also found evidence of a compensatory effect whereby Black youth receive more favorable court dispositions than their non-Black counterparts. Second, the gender gap in youth justice system processing depends on state climates of women and children’s health and wellbeing. Specifically, as women and children’s health and wellbeing decrease, the gender gap in processing narrows and, in the case of court appearance, reverses. Third and finally, Hispanic youth are treated disproportionately more harshly in states with poor climates of children’s health and wellbeing and in states with less punitive juvenile justice systems. Overall, the findings indicate that the reduction of gender and racial/ethnic disparities is unlikely without commitment to the structural reform of inequalities. Intervention efforts to reduce disparities should be multifaceted and include community-based youth-serving organizations and human services agencies, in addition to criminal and juvenile justice agencies. I lovingly dedicate this dissertation to my dear husband, Seth, and our precious children, Isaac and Amelia. iv ACKNOWLEDGEMENTS I would like to thank Dr. Merry Morash, my graduate advisor and committee chairperson, for her invaluable guidance and support. For this, I am truly thankful. Since my first day of the doctoral program, you taught me how to become a good scholar by emphasizing the importance of hard work, attention to detail, and persistence. I have learned so much from you, and I appreciate every opportunity you have given me. I hope I can make the same kind of lasting impact on my future students’ lives and academic careers as you have on mine. I would also like to thank Drs. Steven Chermak, Soma Chaudhuri, and John Hudzik for serving on my dissertation committee. Thank you for taking the time to provide me with insightful feedback and knowledgeable guidance. I also would like to thank the agencies that supported this research. I would like to thank the staff at the Bureau of Labor Statistics (BLS) for allowing access to the National Longitudinal Survey of Youth 1997 (NLSY97) Confidential Geocode Data. I am also pleased to acknowledge and thank the National Institute of Justice for their financial support of the project. This project was supported by Award No. 2012-IJ-CX-0018, awarded by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice. The opinions, findings, and conclusions or recommendations expressed in this research are those of the author(s) and do not necessarily reflect those of the Department of Justice. I would also like to thank my graduate school friends, such as Miriam Northcutt Bohmert, Deitra Suter, Julie Yingling, and Rebecca Stone. Graduate school was so much easier and more enjoyable because of you. I am especially thankful for the kindness of Miriam and v Ben, who “adopted” me, Isaac, and Amelia while Seth was out-of-town and who always went out of their way to help our family in whatever way they could. I would like to express my sincerest gratitude to Sara Fedewa, Pat and Lynda Andersen, Shirley Andersen, and Sandra Stevens for caring for Isaac and Amelia as I finished my dissertation. Additionally, I owe a special debt of gratitude to Dr. Jennifer Dierickx. You not only encouraged me to apply to graduate school, but you went out of your way to guide me through the program and academic profession. Thank you for your boundless advice and encouragement and for being one of my biggest supporters and closest friends. I owe my greatest thanks to my family. My parents, Sandra and Robert, and parents-inlaw, Steve and Shirley, have been constant sources of love and support. Mom, thank you for being a phenomenal woman and my biggest cheerleader. To my brother, Brandon, thank you for your constant encouragement and enthusiasm. I also thank my sister-in-law, Alysson, for her friendship and support. Most of all, I would like to thank my dear husband, Seth, and our precious children, Isaac and Amelia, who have been my greatest supporters. Thank you for your inspiration, sacrifice, encouragement, steadfast support, and unconditional love. vi TABLE OF CONTENTS LIST OF TABLES ....................................................................................................................... ix LIST OF FIGURES ..................................................................................................................... xi CHAPTER 1 INTRODUCTION .........................................................................................................................1 Statement of the Problem ................................................................................................1 Significance of the Current Study ..................................................................................2 CHAPTER 2 REVIEW OF THE LITERATURE ..............................................................................................6 Conceptual Framework ...................................................................................................6 Predictors of Youth Contact with the Justice System ..................................................8 CHAPTER 3 RESEARCH METHODOLOGY ................................................................................................12 Data Sources and Sample ..............................................................................................12 Research Variables ........................................................................................................14 Outcome variables .............................................................................................14 Independent variables .......................................................................................16 Statistical Methods .........................................................................................................22 Analysis of arrest: Event history modeling .....................................................23 Identifying the inception of risk .............................................................24 Specifying a metric for time ....................................................................27 Analysis procedures ................................................................................27 Analysis of charge, court appearance, conviction, and placement: Generalized linear modeling for dichotomous and ordered categorical outcome variables...............................................................................................29 Steps in the Data Analysis .............................................................................................32 Imputation of Missing Data ..........................................................................................33 CHAPTER 4 RESULTS ....................................................................................................................................34 Multivariate Analyses of Youth Hazard of Arrest......................................................54 Multivariate Analyses of Youth Probability of Charge .............................................58 Multivariate Analyses of Youth Probability of Court Appearance ..........................61 Multivariate Analyses of Youth Probability of Conviction .......................................64 Multivariate Analyses of Youth Probability of Placement ........................................67 vii CHAPTER 5 DISCUSSION AND CONCLUSION .........................................................................................71 Summary and Discussion ..............................................................................................71 Limitations and Recommendations ..............................................................................83 Conclusion and Policy Implications .............................................................................84 REFERENCES ............................................................................................................................87 viii LIST OF TABLES Table 1 NLSY97 Sample Sizes and Retention Rates ....................................................... 13 Table 2 Juvenile Justice System Punitiveness Indicators ................................................. 22 Table 3 Youth Demographic Characteristics .................................................................... 35 Table 4 Youth Self-Reported Number of Delinquent Acts Committed (N = 2821) ......... 36 Table 5 Youth Contact with the Justice System, Ages 12-18 (N = 2821) ........................ 37 Table 6 Life-Table Estimates of Survival and Hazard Functions for Youth First Arrest, Time Measured in Months Beginning with Youths’ th 12 Birthdays (N = 2821) ................................................................................... 40 Table 7 Youth Self-Reported Number of Delinquent Acts Committed, by Sex (N = 2821) ................................................................................................ 42 Table 8 Youth Self-Reported Number of Delinquent Acts Committed, by Race (N = 2821) .............................................................................................. 43 Table 9 Youth Self-Reported Number of Delinquent Acts Committed, by Ethnicity (N = 2821) ....................................................................................... 44 Table 10 Youth Contact with the Justice System, by Sex (N = 2821)................................ 51 Table 11 Youth Contact with the Justice System, by Race (N = 2821) .............................. 52 Table 12 Youth Contact with the Justice System, by Ethnicity (N = 2821) ....................... 53 Table 13 Comparison of Alternative Representations for the Main Effect of Time, Ranked from Lowest to Highest AIC .................................................................. 55 Table 14 Estimating the Hazard of Arrest: Three-Level Discrete-Time Hazard Model Estimates for the Regression of Arrest on Individual And Contextual Characteristics, Random Intercepts for County and State (N = 2821) ............................................................................................ 57 Table 15 Estimating the Probability of Charge: Two-Level Logit Model Estimates for the Regression of Charge on Individual and Contextual Characteristics, Random Intercept for State (n = 490) .................................................................. 60 Table 16 Estimating the Probability of Court Appearance: Two-Level Ordered ix Logit Model Estimates for the Regression of Court Appearance on Individual and Contextual Characteristics, Random Intercept for State (n = 261) ...................................................................................................... 63 Table 17 Estimating the Probability of Conviction: Two-Level Logit Model Estimates for the Regression of Conviction on Individual and Contextual Characteristics, Random Intercept for State (n = 191) ........................................ 66 Table 18 Estimating the Probability of Placement in a Correctional Institution: Two-Level Logit Model Estimates for the Regression of Placement on Individual and Contextual Characteristics, Random Intercept for State (n = 141) ...................................................................................................... 69 Table 19 Summary of the Results by Race: Main Effects and Interactions with Variables Indicating Contextual Environment..................................................... 76 Table 20 Summary of the Results by Gender: Main Effects and Interactions with Variables Indicating Contextual Environment..................................................... 78 Table 21 Summary of the Results by Ethnicity: Main Effects and Interactions with Variables Indicating Contextual Environment..................................................... 80 x LIST OF FIGURES Figure 1 Conceptual Framework for Contextual Influences on Youth Contact with the Justice System ..........................................................................................................7 Figure 2 Schematic of Event History Data for First Arrest ..................................................26 Figure 3 Fitted Survival Function for Age at First Arrest, All Youth and by Sex, Race, and Ethnicity (N = 2821) ....................................................................47 xi CHAPTER 1 INTRODUCTION Statement of the Problem Although racial disparities in juvenile justice system processing have long been of concern to scholars and policy makers, the problem of disproportionate minority confinement was brought to national attention in the late 1980s by the National Coalition of State Juvenile Justice Advisory Groups (now the Coalition for Juvenile Justice) (Hsia, 2008). Since the identification of DMC, much of the research and policy attention has been the result of federal funding connected to three amendments to the Juvenile Justice and Delinquency Prevention (JJDP) Act of 1974 (Kempf-Leonard, 2007). The 1988 amendment required that states address disproportionate minority contact and institute efforts to reduce the proportion of minority youth confined in state institutions if their numbers exceed the proportion of such groups in the general population. A 1992 amendment to the JJDP Act elevated the requirement to a “core requirement” and tied funding eligibility to state compliance. In 2002, Congress broadened the previous amendment to require that states address the disproportionate number of minority youth who come in contact with the juvenile justice system. To be eligible for participation in the Formula Grants Program, states are now required to identify the extent to which disproportionate minority contact (DMC) exists, and if it exists, assess the reasons for DMC, develop and implement interventions to address these reasons, evaluate intervention strategies, monitor DMC trends, and adjust interventions if necessary (Hsia, 2008; Kempf-Leonard, 2007; Office of Juvenile Justice and Delinquency Prevention (OJJDP), 2008). 1 The past three decades have seen an accumulation of empirical and theoretical research documenting the effects of extralegal factors, especially race, on police behavior and juvenile justice processing (for reviews, see Bishop, 2005; Engen, Steen, & Bridges, 2002; Leiber, 2002; Paternoster & Iovanni, 1989; Pope & Feyerherm, 1990a, 1990b; Pope, Lovell, & Hsia, 2002). For example, race has been found to significantly predict the arrest decision (Tapia, 2010), intake (Leiber & Fox, 2005; Leiber, Johnson, Fox, & Lacks, 2007; Leiber & Johnson, 2008; Leiber & Mack, 2003), pre-trial release (Rodriguez, 2007, 2010), diversion (Leiber & Johnson, 2008; Leiber & Mack, 2003; Rodriguez, 2010), petition (Leiber & Mack, 2003; Rodriguez, 2010), adjudication (Leiber et al., 2007), and disposition (Rodriguez, 2010). This recent research clearly supports the conclusion that race plays an important role in the processing of youth. Significance of the Current Study Researchers examining disproportionate minority contact in the juvenile justice system generally embrace one of two opposing theoretical explanations: the differential involvement hypothesis , which attributes the disparities to legitimate and legally permissible factors, such as more serious offending (see, for example, Tracy, 2002), or the differential selection thesis, which attributes the disparities to racial stereotyping and bias within the justice system (see, for example, Leiber et al., 2007). In addition to favoring individual-level explanations for disparities in youth penetration into the justice system, the research literature is predominated by studies that rely on official records and focus on particular court jurisdictions, thereby having limited generalizability, the inability to control for variations delinquent behavior, and the inability to examine the effects of race/ethnicity on likelihood of arrest. Moreover, the overwhelming majority of studies include only Black and White youth, and do not disentangle the effects of race and ethnicity with gender. The few studies that do 2 include Hispanic youth have shown mixed findings. Some research suggests that Hispanic youth are treated more severely than Black youth (Dannefer & Schutt, 1982; DeJong & Jackson, 1998; Maupin & Bond-Maupin, 1999), while other research indicates Hispanic youth are treated less severely than Black youth (Kempf-Leonard & Sontheimer, 1995). Additionally, the research literature examining young women’s penetration into the justice system provides evidence that young women may be disproportionately impacted by changes in police policies and practices that began during the Reagan era (Chesney-Lind & Irwin, 2008; Stevens, Morash, & ChesneyLind, 2011). Furthermore, there is evidence of an interaction effect between youth race/ethnicity and gender, such that the effects of race and ethnicity on youths’ likelihood of court involvement depend on gender (Bishop, Leiber, & Johnson, 2010; Stevens et al., 2011). Even fewer studies of youth contact with the justice system have included social, political, and economic contextual measures in their empirical work (for exceptions, see Armstrong & Rodriguez, 2005; DeJong & Jackson, 1998; Feld, 1991, 1995; Rodriguez, 2007, 2010; Sampson & Laub, 1993). Of these studies, only Rodriguez (2007) examined the indirect effects of community characteristics. Her findings suggest that racial/ethnic effects are moderated by community characteristics. Given that police jurisdictions and juvenile courts are organized at the local level, analysis that ignores structural and contextual variations across communities may be misleading (Feld, 1991). Failing to account methodologically for contextual variations may lead to omitted variable bias and model misspecification. Ignoring potential contextual effects in disparate treatment also masks larger issues of social welfare and inequality (Johnson, 2006) and may obscure effective strategies for addressing disproportionate contact with the justice system. 3 The current study extends recent efforts to examine the effects of social context on youth justice system processing and addresses the lack of a focus on the macro-structural environments that may partially explain disproportionate levels of youth penetration into the justice system. Specifically, it examines how structural and environmental context influences youth risk of arrest and probabilities of charge, a court appearance, conviction, and placement. Using a nationally representative survey, it tests for statistical effects of economic, political, and social factors hypothesized to influence the differential treatment of youth in the justice system, controlling for self-reported levels of delinquent behavior. Furthermore, this research explores the intersection of contextual environments with race and ethnicity and gender, and examines whether some climates disproportionately affect young women. The analysis of a nationally representative survey of youth and county- and state- level data that indicate contextual climate provides a unique opportunity to contribute to the existing research on disproportionate youth contact with the justice system and the increased proportion of young women on court caseloads. It may suggest reasons for difficulty in reducing disproportionate contact, and strategies that may be differentially effective depending on the setting. The current project extends previous research on disproportionate contact with the justice system in a number of ways. First, the overwhelming majority of studies of juvenile contact with the justice system relies on official data and centers on judicial decisions in particular jurisdictions. The project’s use of a nationally-representative sample in a dataset that includes detailed information on youth delinquency and contact with the justice system allows the examination of individual and contextual effects at each level of court processing, net of reported delinquent behavior; the results of which can then be broadly generalized. Second, unlike most other studies, the research focuses on economic, political, and social contexts as potentially 4 important predictors of youth contact with the justice system. Specifically, this project delineates contextual factors that either increase or reduce the severity of justice system response to youth. Lastly, no study, to date, has examined how race/ethnicity and gender intersect with contextual environments to produce disproportionately high rates of justice system contact among minority youth and young women. This project also has potential implications for criminal justice policy and practice. An identification of the community characteristics that significantly predict disproportionate youth contact with the justice system may provide direction to local police jurisdictions, courts, and community agencies about how best to respond. In addition, the identification of economic, political, and social influences on juvenile justice outcomes can support monitoring activities and enable policymakers to choose and establish priorities for local interventions approaches. Additionally, the project has the potential to suggest whether DMC-reduction strategies should be multi-faceted and incorporate coalition-building strategies among community youth-serving agencies, health and human services agencies, and criminal and juvenile justice agencies. 5 CHAPTER 2 REVIEW OF THE LITERATURE Conceptual Framework The current study contributes to the existing body of literature on racial/ethnic and gender disparities in youth penetration into the justice system by examining the contextual environments hypothesized to increase youth risk of arrest and probabilities of charge, an appearance in court, conviction, and placement, controlling for their self-reported delinquency. The contextual effects examined include county population composition and levels of structural disadvantage, political conservatism, county court characteristics, and delinquency petition and crime rates, and state climates of gender inequality, children’s health and wellbeing, and juvenile justice system policies and practices. See Figure 1 for the proposed conceptual model. 6 Figure 1. Conceptual Framework for Contextual Influences on Youth Contact with the Justice System County-level factors Structural disadvantage Black population composition Political conservatism Court characteristics Delinquency petition rates Crime State climate Gender inequality Children’s health and wellbeing Juvenile justice policies and practices Interaction Socio-demographic factors Gender, Race/Ethnicity Penetration into the justice system Delinquent behavior 7 Predictors of Youth Contact with the Justice System The first, and most obvious, factors that are expected to predict youth contact with the justice system are racial/ethnic group and gender. Research supports the notion that, net of delinquent behavior, race affects juvenile justice outcomes (for reviews, see Pope & Feyerherm, 1990a, 1990b; Pope et al., 2002). In particular, research demonstrates that relative to White youth, racial minority youth have a higher likelihood of arrest (Brownfield, Sorenson, & Thompson, 2001; Tapia, 2010), are more likely to be detained before adjudication (Leiber & Fox, 2005; Leiber et al., 2007; Leiber & Johnson, 2008; Leiber & Mack, 2003; Rodriguez, 2007, 2010), are less likely to be informally processed (i.e., diverted) (Leiber & Johnson, 2008; Leiber & Mack, 2003; Rodriguez, 2007), and are more likely to be placed out of the home (Rodriguez, 2010). Previous research also demonstrates that gender impacts juvenile court processing decisions (Armstrong & Rodriguez, 2005; Bishop et al., 2010; Stevens et al., 2011). For example, Bishop and colleagues (2010) found evidence for gender-based decision-making at the intake decision but no gender effects at the petition, adjudication, or disposition stages of processing. Furthermore, the effects of gender depended on racial/ethnic group. In general, males were more likely than females to be formally processed; however, this effect was reversed for young women of color, who were more likely than their male counterparts to be formally processed. Therefore, the first hypothesis of the current research is: Net of self-reported levels of delinquent behavior, racial/ethnic group status and gender affect youth penetration into the justice system. It is expected that racial or ethnic minority youth will have a greater risk of arrest and probabilities of charge, a court appearance, conviction, and placement than White youth, and male youth will have a greater risk of arrest and higher probabilities of charge, a court appearance, conviction, and placement than female youth. Furthermore, it is expected that the 8 effects of racial/ethnic group status and gender will interact in their effect on youth penetration into the justice system, such that young women of color will experience more severe justice system outcomes than their White counterparts. Research supports the notion that community factors contribute to youth involvement in the justice system. Contextual factors relevant to youth contact with the justice system may include county composition (e.g., structural disadvantage, racial composition) (Bridges, 1993; Parker, Stults, & Rice, 2005; Sampson & Laub, 1993), political conservatism, county court characteristics (Johnson, 2006) and petition rates, crimes rate (Frazier & Lee, 1992; McCarthy, 1990; Secret & Johnson, 1997), gender inequity, children’s wellbeing (Short & Sharp, 2005), and the policies and procedures of the juvenile justice system (Poe-Yamagata & Jones, 2000; Short & Sharp, 2005). Although no researchers, to date, have examined the contextual effects of gender inequality on disparities in justice system contact, there is evidence that levels of gender inequity may affect responses to young women’s delinquency. Girls’ and women’s offending is often a response to discrimination, economic marginality, and victimization, and some justice system responses result in a criminalization of their survival strategies (Chesney-Lind, 1989; Goodkind, Wallace, Shook, Bachman, & O'Malley, 2009; Richie, 1995; Steffensmeier & Allan, 1996). Thus, it is reasonable to suspect a relationship between structural levels of gender inequality and justice system involvement and, more broadly, that contextual risk factors contribute to disproportionate levels of youth contact with the justice system. The study’s second hypothesis is then: Controlling for self-reported delinquent behavior, youth penetration into the justice system depends on economic, political, and social contexts. In particular, it is expected that youth at a greater risk of arrest and higher probabilities of charge, a court appearance, conviction and placement will reside in areas with greater levels of structural 9 disadvantage, higher proportions of Black residents, higher levels of political conservatism, lower levels of court resources for handling juvenile cases, higher delinquency petition and crime rates, higher levels of gender inequity, lower levels of children’s health and well-being, and more punitive and less progressive juvenile justice policies. Finally, although there is a dearth of research examining the interactive effects of individual characteristics with contextual variables on youth disproportionate contact with the justice system, there is some evidence that the effects of race/ethnicity depend on community context. Rodriguez’s (2007) study of juvenile court processing data from Maricopa County suggests that the effect of youth ethnicity on penetration into the justice system is moderated by the characteristics of youth’s residential communities. The study’s third hypothesis is then: Controlling for self-reported delinquent behavior, the effects of individual characteristics on youth penetration into the justice system depend on economic, political, and social context. It is expected that the effects of racial/ethnic group and gender on youth risk of arrest and probabilities of charge, a court appearance, conviction, and placement depend on economic, political, and social context. Specifically, it is expected that, controlling for delinquent behavior, minority youth and young women penetrate deeper into the justice system than their counterparts in areas of greater structural disadvantage, higher proportions of Black residents, high levels of political conservatism, low levels of court resources for processing juvenile cases, higher delinquency petition and crime rates, greater gender inequity, and lower child well-being. Furthermore, it is expected that racial/ethnic and gender differences in penetration into the justice system will depend on juvenile justice policy context, such that in some settings, the effect of juvenile justice system punitiveness is magnified for minority youth and young women, and in 10 other settings, the effect of juvenile justice system progressiveness is attenuated for minority youth and young women. 11 CHAPTER 3 RESEARCH METHODOLOGY Data Sources and Sample The main dataset for this study was constructed from the National Longitudinal Survey of Youth 1997 (NLSY97). The NLSY97, conducted by the Bureau of Labor Statistics, uses a multistage stratified random sampling design. After weighting to account for the biases created by over-sampling, probability of selection into the original sample, and non-random attrition, the NLSY97 is representative of the U.S. population born between 1980 and 1984. The NLSY97 was launched to enable researchers to examine the life-course experiences of representative samples of men and women born in the United States. Although the main purpose of the survey is to examine youth transitions from school into the labor market, the NLSY97 also collects extensive information on youths’ personal characteristics, family and community background, educational experiences, and risky and delinquent behaviors. The initial wave of NLSY97 respondents included 8,984 youth ages 12 to 16 on December 31, 1996. NLSY97 sample sizes and retention rates for each round are included in Table 1. Given the focus on juvenile contact with the justice system, including during early adolescence, youth ages 14 and older during the first round of data collection were excluded from analysis, and the subsample was limited to youth ages 12 and 13. Fifty-seven additional youth were excluded from the analyses, which included: a) arrested youth who had missing data for the month and year of the arrest (n = 14), b) arrested youth who reported that their first arrest occurred during a month and year preceding their birth (n = 2), and c) arrested youth who reported that their first arrest occurred prior to the observation period and who therefore had 12 time-dependent covariates corresponding with their date of arrest (n = 41). In other words, because they were arrested prior to the observation period, they were missing valid data on their participation in delinquent behavior and their county and state of residence at the time of their arrest. The study’s observation period extended until youths’ eighteenth birthdays and was then linked to the corresponding data from the NLSY97 rounds 1 through 5. The sample size for analysis is 2,821 youth. Table 1. NLSY97 Sample Sizes and Retention Rates Round Fielding Sample Size Period Retention Rate 1 1997-1998 8984 --- 2 1998-1999 8386 93.3 3 1999-2000 8209 91.4 4 2000-2001 8081 89.9 5 2001-2002 7883 87.7 County- and state-specific data from 1997 through 2001 from various publically available sources were appended to the NLSY97 data using the confidential NLSY97 Geocode File, which provides detailed information on respondent migration patterns. County-level scales reflecting structural disadvantage and population composition were created from the 2000 Decennial Census of Population and Housing (US Census Bureau, 2004). A scale reflecting political conservatism was created from the 2000 presidential election results (Lublin & Voss, 2001). 13 County court characteristics and petition rates were compiled respectively from the National Prosecutors Survey (Bureau of Justice Statistics, 2002) and OJJDP’s Easy Access to State and County Juvenile Court Case Counts (Stahl & Kang, 2005; Stahl, Kang, & Wan, 2002; Stahl, Kang, & Wilt, 2003; Stahl, McGlynn, & Wan, 2000; Stahl & Wan, 2001). County-level crime rates were calculated from 1997-2001 Uniform Crime Reports (UCR) (Federal Bureau of Investigation, 2001a, 2001b, 2006a, 2006b, 2006c) . State-level composite indicators related to gender equity were obtained from the Institute for Women’s Policy Research (IWPR) (2000), and state-level indicators related to children’s health and wellbeing were gathered from the annual Annie E. Casey Foundation Kids Count data books (2000, 2001, 2002, 2003, 2004) . Information about state-level juvenile justice system policies and legislation were gathered from state statutes and Office of Juvenile Justice and Delinquency Prevention (OJJDP) publications (Bilchik, 1999; Griffin, Torbet, & Szymanski, 1998; Logan, 1998; Redding, 2002; Samuels, Dwyer, Halberstadt, & Lachman, 2011; Sickmund, 2003; Snyder & Sickmund, 2006). Research Variables Outcome variables. The outcomes of interest in the current research are youths’ risk of arrest and probabilities of charge, a court appearance, conviction, and placement in a correctional facility. A dichotomous variable indicates whether or not youths were arrested as juveniles (1=arrested, 0=not arrested). Longitudinal data that provides information on the month and year the arrest occurred allows measurement of the timing of arrest (conceptualized as youths’ age in months at the time of their first juvenile arrest). 14 Previous research documents the importance of examining multiple stages of justice system processing when examining disproportionate penetration into the justice system. For example, racial/ethnic group and gender differences may be more pronounced at some decisionmaking levels, and decisions made in earlier decision points may affect subsequent outcomes (Bishop, 2005; Bishop et al., 2010; Dannefer & Schutt, 1982). Small differentials may also accumulate over each decision level, “transforming a more or less heterogeneous racial arrest population into a homogeneous institutional black population” (Liska & Tausig, 1979, p. 197). Some research (Dannefer & Schutt, 1982; Rodriguez, 2007, 2010) also provides evidence of a compensatory effect in the courts where judges may attempt to counteract differential treatment by the police. Dannefer and Schutt (1982), for example, demonstrated the presence of courtlevel corrections processes in urban areas, where Black youth received more favorable court dispositions than White youth, indicating that the effects of race on youth penetration into the justice system depend on the processing stage and social context. Four additional variables are used to capture youth penetration into the justice system as a result of their first arrest. Charge is coded as 1 for youth who reported being charged with an offense as a result of their first arrest. Court is a trichotomous variable coded as 1 for youth who reported appearing in adult court as a result of their first arrest, 2 for youth who reported appearing in juvenile court, and 3 for youth who reported that they did not appear in court. Convicted is coded as 1 for youth who reported pleading guilty or being convicted/adjudicated delinquent as a result their first arrest. Finally, placement is coded as 1 for youth who reported that they were sentenced to spend time in a correctional institution or a youth institution, like a training or reform school. 15 Independent variables. The independent variables used in this study consist of youth-, county-, and state-level variables. Youth-level variables include,  Age: Conceptualized as youths’ age (in days) at the time of first arrest; this variable was included in the analysis of charge, court appearance, conviction, and placement  Sex: A dichotomous variable with Female as the reference group  Black Racial Group: A categorical variable with Non-Black as the reference group  Hispanic Ethnicity: A dichotomous variable with Non-Hispanic as the reference group  Delinquent Behavior: An index variable defined as the number of times youth reported purposively destroying property, stealing something worth $50 or more, attacking someone, and selling drugs in the last year. This is treated as a time-varying covariate in analysis of arrest, with values varying according to annual reports of participation in delinquent behavior. For the analysis of charge, court appearance, conviction, and placement, this variable consists of youth reports of their delinquency during the interview period immediately preceding their first arrest.  Adult Court: A dichotomous variable coded as 1 for youth who reported that they appeared in adult court. This variable was included as a control in the analysis of conviction and placement.  Violent Charge: A dichotomous variable coded as 1 for youth who reported that they were charged with a violent offense. This variable was included as a control in the analysis of conviction and placement. Four county-level scales indicating structural disadvantage employing data from the 2000 decennial census (US Census Bureau, 2004) were created modeled on the work of Sampson and 16 colleagues (Sampson, Raudenbush, & Earls, 1997, 1999). For each scale, items were transformed to z-scores and then averaged.  Concentrated Disadvantage: A county-level scale consisting of the percent of residents who live in poverty, the percent of residents who receive public assistance, the percent of residents who are unemployed, the percent of female-headed families with children, and the percent of residents who are Black or African American (alpha = 0.82).  Concentrated Immigration: A county-level scale consisting of the percent of residents who are Latino and the percent of residents who are foreign born (alpha = 0.79).  Residential Stability: Comprised of the percent of residents ages 5 and older who resided in the same house 5 years earlier and the percent of owner-occupied homes (alpha = 0.68).  Concentrated Affluence: A county-level scale consisting of the percent of adults with annual incomes of $75,000 or more, the percent of adults with a college education, and the percent of the civilian labor force employed in professional or managerial occupations (alpha = 0.86). Black Population Composition, obtained from the 2000 decennial census (US Census Bureau, 2004), consists of the percent of county residents who are Black or African American. 1 1 One possible limitation of the analyses is the reliance on a single measure of racial threat. Two additional measures supported in the research literature (see, for example, Parker et al., 2005) include black immigration (i.e., the percent of Black residents who recently immigrated) and racial inequality (i.e., the ratio of White to Black educational attainment and unemployment). However, given the large proportion of exclusively White counties included in the analyses, I was unable to compute Black immigration or White-to-Black ratio measures. 17 County-level election data were obtained from the Federal Elections Project (Lublin & Voss, 2001) and used to create a scale of Political Conservatism, a county-level measure consisting of the percent of voters who voted for George W. Bush in the 2000 presidential election. County-level prosecutor’s office characteristics were obtained from the 2001 National Prosecutors Survey (Bureau of Justice Statistics, 2002). Variables indicating prosecutor office characteristics and prosecutorial practices include,  Number Waived: A count variable indicating the number of times the county prosecutor’s office pursued criminal charges against juveniles in the preceding year.  Juvenile Unit: A dichotomous variable coded as 1 if the county prosecutor’s office has a special unit for handling juvenile cases.  Juvenile Prosecuting Attorney: A dichotomous variable coded as 1 if the county prosecutor’s office has a special prosecuting attorney for juvenile cases.  Juvenile Case Guidelines: A dichotomous variable coded as 1 if the county prosecutor’s office has written guidelines for handling juvenile cases. County-level juvenile justice system punitiveness was tapped using Delinquency Petition Rates (1997-2001) obtained from Easy Access to State and County Juvenile Court Case Counts (Stahl & Kang, 2005; Stahl et al., 2002; Stahl et al., 2003; Stahl et al., 2000; Stahl & Wan, 2001). Rates of juvenile court petitions for delinquency offenses were calculated per 1,000 residents aged 10 through the upper age limit for the years 1997 through 2001. This is treated as a timevarying covariate in analysis of arrest, with values varying annually. For the analysis of charge, court appearance, conviction, and placement, this variable consists of the delinquency court petition rate during the year of arrest. 18 County-level crime rates were obtained for the years 1997 through 2001 from the Uniform Crime Reports (Federal Bureau of Investigation, 2001a, 2001b, 2006a, 2006b, 2006c).  Violent Crime Rate (1997-2001): The rate of major violent crimes (murder, rape, robbery, and assault) per 1,000 residents. For the analysis of arrest, this is treated as a time-varying covariate. For the analysis of charge, court appearance, conviction, and placement, this variable consists of the county violent crime rate during the year of arrest.  Property Crime Rate (1997-2001): The rate of major property crimes (burglary, larceny, motor vehicle theft, and arson) per 1,000 residents. In the analysis of arrest, this is treated as a time-varying covariate. For the analysis of charge, court appearance, conviction, and placement, this variable consists of the county property crime rate during the year of arrest. A state-level scale indicating Gender Inequity was extracted from the Institute for Women’s Policy Research (IWPR) (2001) composite indicators, which are reported semiannually for each of the 50 states. Each of the following indicators were standardized and averaged, and the composite scale was reverse-coded so that higher values correspond to higher levels of gender inequity (alpha=0.79). For further detail on the construction of women’s status indicators, see Werschkul and Williams (2000).  Women’s Political Participation: Comprised of the proportion of women in elected office, the percent of women in the state who reported registering to vote for the presidential and congressional elections, the percent of women in the state who reported voting in the presidential and congressional elections, and the number of institutional resources available for women in the state, including a state commission 19 for women and/or a legislative caucus for women. Data reflect the years 1992 to 2000.  Women’s Employment and Earnings: Comprised of the median annual earnings for women who are employed full-time and year-round, the women’s to men’s earnings ratio for full-time, year-round employees, the percent of the female population aged 16 or older employed or looking for work, and the percent of the civilian female population aged 16 or older employed in managerial or professional occupations. Data reflect the years 1997 to 2000.  Women’s Social and Economic Autonomy: Comprised of indicators of the percent of women in the state with health insurance, the percent of women with four or more years of college, the percent of firms in the state that are women-owned, and the proportion of women in the state living above the federal poverty level. Data reflect the years 1989 to 2000.  Women’s Reproductive Rights: Compiled from information on state policies and legislation concerning abortion, contraception, gay and lesbian adoption, infertility, sex education, and racial/ethnic disparities in reproductive health. Data reflect the years 1996 to 2000.  Women’s Health and Wellbeing: Comprised of the following indicators: women’s mortality rates from heart disease, lung cancer, breast cancer, and suicide; women’s incidence rates for diabetes, Chlamydia, and AIDS; women’s self reported mean number of days in the past 30 days on which their mental health was not good; and women’s self reported mean number of days in the past 30 days on which their activities were limited due to health reasons. Data reflect the years 1995-1999. 20 State-level data indicating children’s health and wellbeing were obtained from the annual Annie E. Casey Foundation Kids Count data books (2000, 2001, 2002, 2003, 2004). For these analyses, data were taken from the 2000-2004 reports and time-varying scales were constructed reflecting the years 1997-2001. Exploratory factor analysis revealed the presence of two underlying factors, with factor loadings well over .5. The two scales were:  Children’s Health: A state-level scale comprised of percent of low birth weight babies, infant mortality rate (deaths per 1,000 live births), child death rate (deaths per 100,000 children ages 1-14), and rate of teen death by accident, homicide, and suicide (deaths per 100,000 teens ages 15-19). The scale was reverse-coded so that higher values correspond to higher levels of children’s health. The reliability for this scale was consistently found to be in the high range for the years 1997 to 2001 (alpha=0.86 to 0.89).  Youth Disconnectedness: A state-level scale comprised of percent of teens who are high school dropouts (ages 16-19), percent of teens not attending school and not working (ages 16-19), percent of children living with parents who do not have fulltime, year-round employment, percent of children in poverty. Cronbach’s alpha ranged from 0.85 to 0.90 for the years 1997 to 2001. A scale comprised of information regarding state-level policies and legislation was used to indicate Juvenile Justice System Punitiveness. An index variable, modeled after Buck Willison (2010), is comprised of information concerning minimum and maximum ages of juvenile court jurisdiction, life without parole for crimes committed under age 18, youth waiver/transfer to criminal courts, blended sanctions, and the collection of juvenile DNA (See Table 2). Information about state-level juvenile justice system policies and legislation in effect 21 in 2000 were gathered from state statutes and Office of Juvenile Justice and Delinquency Prevention (OJJDP) publications (Bilchik, 1999; Griffin et al., 1998; Logan, 1998; Redding, 2002; Samuels et al., 2011; Sickmund, 2003; Snyder & Sickmund, 2006). Table 2. Juvenile Justice System Punitiveness Indicators State excludes 17-year-olds from juvenile court jurisdiction 1 point State excludes 16-year-olds from juvenile court jurisdiction 1 point State does not specify a minimum age for juvenile court jurisdiction 1 point State mandates life without parole for crimes committed under age 18 1 point State permits life without parole for crimes committed under age 18 1 point State uses automatic (i.e., legislative) exclusion for criminal courts 1 point State uses prosecutor discretion for criminal court transfer 1 point State juvenile courts permitted to “blend” adult-system sanctions 1 point State permits the collection of juvenile DNA 1 point 9 points possible Statistical Methods The study employs a combination of event history, generalized linear modeling for dichotomous data, and multilevel modeling techniques to capture the nature of youth penetration into the justice system. Due to the nested structure of the data (youth nested within counties nested within states), multilevel modeling techniques were used to fit nonlinear random effects models using maximum likelihood estimation. Multilevel modeling techniques are able to 22 account for the correlations between individual-level variables on youth from the same counties and states (i.e., intraclass correlation) and possible interactions between individual-level and contextual variables. Analysis of arrest: Event history modeling. Event history modeling techniques allow the study of “length of time until the occurrence of some event” (Hox, 2010, p. 159). Defined as a “qualitative change in state” (DeMaris, 2004, p. 383), the event modeled in the proposed research is a first arrest. These techniques have been most commonly used in criminological research to examine recidivism (for examples, see Benda, Toombs, & Peacock, 2002; Bushway, Nieuwbeerta, & Blokland, 2011; Grattet, Lin, & Petersilia, 2011; Hepburn & Albonetti, 1994). The use of event history methods has two distinct advantages for the study of youth risk (hazard) of arrest. The first advantage is the ability to deal with right-censoring of the data, which takes place when individuals have not experienced the event (an arrest) by the end of the study period (Singer & Willett, 2003). In these cases, the researchers have no way of determining if or when the individuals will experience the event. Traditional studies of disproportionate youth contact with the justice system rely on official data from youth who have already made contact with the justice system. However, to learn about disproportionate risk of contact with the justice system, one must understand what youth are likely to come in contact with the system and which characteristics predict such contact. To do so, data from both censored and uncensored cases must be simultaneously incorporated in the analysis. Censored cases provide important information, especially about the probability that youth will avoid contact with the justice system. The second advantage of event history methods is its ability to incorporate into the analysis predictor variables whose values vary over time (i.e., time-varying covariates) (Allison, 23 1984). In this study, many individual factors and contextual variables varied considerably over time. For example, youth participation in delinquent behavior varied as youth aged, and youths’ 2 contextual climates varied as they moved and as the characteristics of geographic areas themselves fluctuated (i.e., rates of crime). Identifying the inception of risk. A critical concept in event history analysis is the point at which time begins and everyone in the study population is first exposed to the risk the event, which can be referred to as the “beginning of time” (Singer & Willett, 2003, p. 311) or the “inception of risk” (DeMaris, 2004, p. 384). For these analyses, it is assumed that youth were first exposed to the risk of arrest on their twelfth birthday. Thus, survival time is calculated as youths’ age in months minus their age at the inception of risk (144 months). Youth who were lost through attrition or turned eighteen years of age before experiencing their first arrest are considered to be censored. For these cases, censored survival time is calculated as their age (in months) at last observation minus their age at inception of risk (144 months). Figure 2 provides a schematic of various hypothetical event histories of arrest. The horizontal axis represents survival time. Each horizontal line represents a youth’s event history and represents the time period from inception of risk (12 years of age) until censoring, first arrest, or the end of the observation period (18 years of age). Solid lines represent observed durations of survival, and dotted lines represent survival time that is outside the observation period. Each line moves from left to right, with Xs at the end of the line indicating an arrest and circles at the end of the line indicating right-censoring due to attrition. For example, 2 Youth reported an average of one (0.91) move during the observation period, and more than one-quarter (26.4 percent) of youth moved two or more times. Youths’ average number of moves was also significantly related to likelihood of experiencing an arrest at the bivariate level, with arrested youth reported a greater average number of times moved (t = 6.8, p < .001) 24 Respondents A, C, and F had not experienced their first arrest by their eighteenth birthdays; they are, therefore, right-censored by the ending date of the study. Respondents B and E both experienced their first arrest during the observation period between their twelfth and eighteenth birthdays; the timings of their arrests are symbolized by Xs. Respondent G, however, did not experience an arrest while under observation and dropped out of the study prior to his or her eighteenth birthday. This respondent is consequently right-censored due to attrition. Finally, Respondent D was arrested prior to his or her twelfth birthday and therefore prior to the start of observation. As a result, this case is left-censored. As described above, forty-three cases were left-censored and therefore excluded from analyses. 25 Figure 2. Schematic of Event History Data for First Arrest A B C D E F G t0 th 12 Birthday t1 th 13 Birthday th 14 Birthday th 15 Birthday 26 th 16 Birthday th 17 Birthday th 18 Birthday Specifying a metric for time. The time-to-event-occurrence can be measured continuously or discretely. Although, in practice, time is always measured in discrete units, when these discrete units are very small, it is acceptable to assume that the time of event occurrence is precisely known. Event history models that use a continuous measure of time-toevent-occurrence are known as continuous-time event history methods (Allison, 1984; DeMaris, 2004). However, even if the latent event-generating process is continuous, some data are interval-censored, meaning that events are only known to occur within a given interval of time (DeMaris, 2004; Singer & Willett, 2003). When time-to-event-occurrence is measured discretely, it is more appropriate to use discrete-time event history methods (Allison, 1984; DeMaris, 2004). Because the NLSY97 data provides the month and year of arrest, it is therefore appropriate to apply discrete-time event history methods. Analysis procedures. The first step in analyzing event occurrence data typically involves summarizing the sample distribution of event occurrence in a life table (DeMaris, 2004; Singer & Willett, 2003). Traditional summary statistics (e.g., means, standard deviations, ranges) cannot incorporate censoring and are therefore inappropriate tools for the analysis of event occurrence (Singer & Willett, 2003). The life table technique involves partitioning the data into a series of time intervals and computing the number of individuals who were eligible to experience the event during that interval, experienced the event during the interval, and were censored at the end of the interval (DeMaris, 2004; Singer & Willett, 2003). This information is then used to construct estimates of the probability of conditional failure for each time interval, the survival function, and the hazard function, ignoring the potential effects of explanatory variables. These basic statistical methods of describing event history data incorporate both observed and censored event occurrences. 27 The conditional probability of failure (i.e., the probability of event occurrence among those individuals eligible to experience the event during a given interval) is defined as ˆ k  dk Rk  wk  / 2 , th where dk represents the number of individuals who experienced the event in the k interval, Rk is the number of individuals who were eligible to experience the event at the beginning of the k th th interval, and wk is the number of censored cases in the k interval. The conditional probability for failure is then used to construct estimates of the survival function and the hazard function. The survival probability cumulates the risks of event occurrence associated with each time interval to assess the probability that individuals will not experience the event during a given time interval. The discrete-time hazard probability is the conditional probability an individual will experience the event in a given time interval, given that the individual has not experienced it in any earlier time intervals (Allison, 1984; Singer & th Willett, 2003). The average hazard rate at the midpoint of the k interval is estimated as   ˆ h tj  ˆ 2k ˆ Lk 2     , where Lk is the interval length. The hazard function is the most sensitive tool for describing event timing and occurrence because it displays the unique risk associated with each time 28 interval and allows the examination of variation in hazard rate over time (Singer & Willett, 2003). To investigate the effects of individual characteristics and contextual conditions on youth hazard of arrest, I employ the multilevel extension of the discrete time interval-censored hazard model (DeMaris, 2004; Hox, 2010). The general approach to analyzing discrete time survival data is to model the hazard as a linear function of the covariates and then transform the linear function of the hazard using the appropriate link function. The logit link function is generally used when data are interval-censored (DeMaris, 2004), and the three-level extension can be obtained by introducing normally distributed random effects into the linear regression equation (Hox, 2010). Specifically, for individual i of county j in state k, the model becomes:  P ln  it 1 P it      a(t )  xijkt   k   jk   , where Pit is the conditional probability for failure in the time interval t, a(t) is a function of time representing the baseline hazard for each time interval, x′ijkt β represents a weighted sum of covariates times parameters, υk represents the random effect for state, and µjk represents the random effect for county j nested in state k. Analysis of charge, court appearance, conviction, and placement: Generalized linear modeling for dichotomous and ordered categorical outcome variables. For the analysis of youth probabilities of further contact with the justice system, the sample for analysis was restricted to youth at risk of experiencing the outcome variable (e.g., the analysis of youth probability of a charge was limited to those who reported an arrest). Youth probabilities of 29 charge, conviction, and placement were analyzed using the multilevel extension of the logit model for dichotomous data. Youth probabilities of a court appearance (an outcome variable trichotomized into adult court appearance, juvenile court appearance, and no court appearance) were analyzed using the multilevel extension of the ordered logit model, also known as the proportional odds model. Because the three categories of court appearance represent different degrees of severity of justice system contact, the ordered logit model appropriately models the log odds of “more severe” versus “less severe” justice system contact. 3 Generalized linear models are defined by three components: an outcome variable with a specific error distribution, a linear additive regression equation that predicts the latent predictor of the outcome variable, and a link function that specifies the transformation function for the mean of the latent outcome variable (DeMaris, 2004; Hox, 2010). The multilevel extension of generalized linear models uses the multilevel regression equation for linear predictor of the mean of the unobserved latent variable and the logit link function for the transformed mean (Hox, 2010). Although the structure of the data is such that youth are nested within counties, which are nested within states, the data were limited by sparseness and small cluster sizes at the countylevel (i.e., small numbers of youth per county) once the samples were restricted for analysis of charge, court appearance, conviction, and placement. For example, in the analysis of charge, the average cluster size per county was slightly greater than two (2.3 youth per county), and 3 The proportional odds assumption was tested using the score test, which was non-significant, indicating that the predictors have the same effects on the log odds of more severe court appearance versus less severe court appearance (i.e., adult court appearance versus other; court appearance versus no court appearance). That is, because there was insufficient evidence to reject the proportional odds assumption, regression coefficients are constrained to be the same for each contrast but model intercepts are allowed to vary (DeMaris, 2004). 30 approximately 45 percent (45.16) of counties in the analysis were singleton clusters (i.e., one youth per county). Thus, given county-level data sparseness and the large proportion of singleton county clusters, youth probabilities of further contact with the justice system were modeled using standard two-level modeling techniques, with county-level variables 4 disaggregated to the individual-level. Therefore, the multilevel models for these data include state-level residual variance terms. The regression equations for the probabilities of charge, court appearance, conviction, and placement for an individual i who resides in state k then become: y e 0  xikt    k   1  e 0  xikt    k   . where x′ikt β represents a weighted sum of covariates times parameters and µk represents the random effect for state k. 4 5 One possible limitation of this analysis procedure is that risk of Type I error may be elevated. 5 Average state-level cluster sizes for the analysis of charge, court appearance, conviction, and placement were 11.5, 6.6, 4.9, and 3.7, respectively. Corresponding group sizes (i.e., number of states included in the analysis) were 43, 40, 39, and 38. The results of simulation studies (see, for example, Hox, 2010; Moineddin, Matheson, & Glazier, 2007; Raudenbush, 2008) assessing problems with data sparseness indicate that regression coefficient point estimates, their standard errors , and random variance component point estimates are generally unbiased by small sample sizes. However, the standard errors of random variance components are negatively biased. As a result, the significance of random variance components should be interpreted with caution. 31 Steps in the Data Analysis All analyses were conducted using the software SAS, which has the capacity to generate accurate parameter estimates, standard errors, and tests of significance for complex sample designs. Collinearity diagnostics were preformed prior to analysis. All continuous predictor variables were grand mean centered and standardized to allow the model intercepts to be interpreted as the expected outcome for the average youth. Descriptive summary statistics are presented using the data weighted to be representative of the population. Because of the interest in parameter effects rather than point estimates, multivariate analyses are presented using unweighted data (DuMouchel & Duncan, 1983; Olsen, 2009; Winship & Radbill, 1994). The initial stage of data analysis involved generating descriptive statistics for the sample. The distribution of baseline hazard and survival rates of arrest and probabilities of a charge, court appearance, conviction, and placement were then examined for each gender and racial/ethnic group. Next, the effects of individual and contextual variables on youth contact with the justice system were estimated using multilevel modeling techniques in the SAS GLIMMIX procedure, which has the capacity to fit two- and three-level generalized linear mixed models (GLMM), integrated over random effects (Gharibvand & Liu, 2009; SAS Institute, 2008). The analysis strategy followed Hox’s (2010, pp. 56-59) bottom-up approach, which simplifies and limits the number of parameters in the model by adding each type of parameter one step at a time. The strategy for each outcome variable was to begin with an intercept-only model, which determined the proportion of the variance in youth penetration into the justice system that is explained by the grouping structure of the data. Next, the individual-level explanatory variables were added as fixed effects to determine their contribution to youth penetration into the justice system. The 32 blocks of county-level and state-level explanatory variables were added in succession as fixed effects to examine the direct effects of structural disadvantage, population composition, political conservatism, court characteristics, delinquency petition and crime rates, gender inequity, child health and well-being, and juvenile justice policies. Random coefficient models were then generated to examine if the effects of youth sex and racial/ethnic group varied significantly at the group level (i.e., at the county- or state-level for the analysis of arrest and at the state-level for the analysis of further justice system contact). Finally, cross-level interaction effects models were generated to determine whether significant variance components were explained by variation in county and/or state climate. Imputation of Missing Data Only modest amounts of data are missing in the NLSY97. In such cases, missing values were imputed via multiple imputation following a two step process. First, the SAS PROC MI command was used to generate and impute sets of plausible values that incorporate the uncertainty about the nonresponse model (Rubin, 1996; SAS Institute, 2008b). Each set of plausible values for missing observations created a distinct version of the original dataset. Second, the parameter estimates and standard errors from statistical analyses on each of the imputed datasets were combined using the MIANALYZE procedure (SAS Institute, 2008c). 33 CHAPTER 4 RESULTS This chapter begins by reporting univariate and bivariate descriptive statistics for youth demographic characteristics, participation in delinquent behavior, and justice system involvement. The results from a series of the regression analyses of youth hazard of arrest and probabilities of charge, court appearance, conviction, and placement are then presented and discussed in order to identify and analyze the effects of youth demographic characteristics and contextual environments on contact with the justice system, controlling for self-reported delinquency. Interactions between youth characteristics and variables indicating county and state climate are explored to examine whether certain groups of youth are disproportionately affected by certain contextual environments. Table 3 summarizes the demographic characteristics of youth involved in this study. Slightly more than half of youth were male (51 percent), and approximately 15 percent were Black. The race of the remaining 85 percent was categorized as “other.” Slightly more than one in eight youth (12.7 percent) were of Hispanic ethnicity. These racial/ethnic group proportions are consistent with the US Census population estimates (US Census Bureau, 2004). 34 Table 3. Youth Demographic Characteristics (N = 2821) % Sex Male 51.1 Female 48.9 Racial Group Black 15.3 Non-Black 84.7 Ethnicity Hispanic Non-Hispanic Note: Estimates are weighted. 12.7 87.3 Table 4 presents the means and linearized standard errors for youth self-reported participation in delinquent behavior. The fewest number of delinquent acts were reported when youth were ages twelve and thirteen (1.37 acts). Mean delinquency increased throughout adolescent and was at its highest during youths’ seventeenth year of age (2.47 acts). 35 Table 4. Youth Self-Reported Number of Delinquent Acts Committed (N = 2821) Mean St. Error Delinquency, age 12 1.37 0.16 Delinquency, age 13 1.37 0.16 Delinquency, age 14 1.67 0.19 Delinquency, age 15 1.93 0.19 Delinquency, age 16 2.21 0.25 2.47 0.25 Delinquency, age 17 Note: Estimates are weighted. Table 5 presents the results of descriptive statistics for youth contact with the justice system. Approximately 17 percent of youth were arrested as juveniles, and slightly more than half of youths’ first juvenile arrests resulted in one or more charges. The most common charges were for theft, followed by public order offenses, destruction of property, drug possession, assault, and burglary, respectively. Nearly three-quarters of youth who were charged with an offense appeared in court as a result of those charges, and of those youth, nearly one-third appeared in adult court and the remaining appeared in juvenile court. Of youth whose charges resulted in a court appearance, more than three-quarters received a conviction, and of those who were convicted, approximately 27 percent were placed in an institution. 36 Table 5. Youth Contact with the Justice System, Ages 12-17 (N = 2821) % Arrest Not Arrested 83.3 Arrested 16.7 Charge a Not Charged 45.7 First Juvenile Arrest Resulted in a Charge Charged with Assault Charged with Burglary Charged with Theft 15.6 b 11.9 b 30.6 Charged with Destruction of Property Charged with Drug Possession b 20.5 b Charged with Public Order Offense Court Appearance 54.3 b 19.2 b 23.0 b Did Not Appeared in Court 26.3 Appeared in Court Appeared in Adult Court Conviction 73.7 c 29.1 c Not Convicted 22.1 Convicted 77.9 Placement d Not Placed in an Institution 73.3 Placed in an Correctional Institution 26.7 a. Among Youth Arrested as a Juvenile (n = 494) b. Among Those Whose First Juvenile Arrest Resulted in a Charge (n = 262) c. Among Those Whose First Juvenile Arrest Resulted in a Charge and a Court Appearance (n = 191) d. Among Those Whose First Juvenile Arrest Resulted in a Charge, a Court Appearance, and a Conviction (n = 141) Note: Estimates are weighted. Survey design taken into account using Taylor series method. Percentages of offenses do not add to 100 because youth may have been charged with more than one offense as a result of their arrest. 37 Table 6 presents the results from the life table analysis of youth age at first arrest. Specifically, it collapses the seventy-two months corresponding with the dates in between th th youths’ 12 and 18 birthdays into a series of year-long intervals. The table summarizes the information about the total number of respondents who were at-risk of experiencing their first arrest at the beginning of the year, the number of respondents who experienced their first arrest during the year, the number of respondents who were censured during the year, the conditional probability of being arrested for the first time during each year (hazard function), and the cumulative probability of surviving to the beginning of each year (survival function). Looking at the first row of Table 6, of the 2,821 youth included in the analyses, 62 experienced their first th arrest by the end of their 12 year of age, and 23 were censored by the end of the year, leaving 2,735 youth to be included in the group of youth at risk of experiencing their first arrest at the th beginning of their 13 year of age. During the year corresponding with youths’ 13 th year of age, of the 2,735 youth at risk of experiencing their first arrest, 58 experienced an arrest, and 16 th were censored by the end of the year. During the year corresponding with youths 14 year of age, of the 2,661 youth at risk, 95 were arrested, and 6 were censored. This process continues th until youths’ 17 year of age, when 89 experienced their first arrest, and the remaining 2236 were censored. The sixth column of the life table presented in Table 6 presents the proportion of youth arrested during each year (i.e., hazard function for youth arrest) and indicates that youth hazard of arrest increased with age, and youth were at the greatest risk of experiencing their first arrest at age 17. For instance, youth hazards of arrest during ages 12 and 13 were less than 0.002, whereas, youth hazard of arrest at age 17 was approximately 0.006. The last column of Table 6 presents the cumulative probability of surviving to the beginning of each year without 38 experiencing an arrest (i.e., survival function). For these analyses, it is assumed that youth were th first exposed to risk of arrest on their 12 birthday, and therefore, the probability of surviving to th the beginning of youths’ 12 year of age is 1.0. The cumulative probability of surviving to the th beginning of youths’ 13 year of age without experiencing an arrest is 0.9779. The survival rate continues to decrease in this fashion, until youths 17 th year of age, when the cumulative probability of beginning the year without an arrest declines to 0.8545. 39 Table 6. Life-Table Estimates of Survival and Hazard Functions for Youth First Arrest, Time Measured in Months Beginning with th Youths’ 12 Birthdays (N = 2821) Number Time interval Age (months) No justice system contact at beginning of year Arrested during the year Censored at the end of the year Proportion of youth arrested during the year (Hazard function) Cumulative probability of surviving to the beginning of each year (survival function) 12 [1,13) 2820 62 23 0.0019 1.0000 13 [13,25) 2735 58 16 0.0018 0.9779 14 [25,37) 2661 95 6 0.0030 0.9571 15 [37,49) 2560 99 22 0.0033 0.9229 16 [49,61) 2439 89 25 0.0031 0.8871 17 [61,73) 2325 89 2236 0.0059 0.8545 492 2328 Total 40 Before testing the impact of youth individual characteristics on likelihood of justice system contact, it was of interest to examine whether self-reported delinquent behavior was associated with youth sex, race, or ethnicity. Tables 7, 8, and 9 present the results of the bivariate analyses of youth self-reported count of delinquent acts committed each year from age 12 to 17 by sex, race, and ethnicity, respectively. As shown in Table 7, male youth generally reported significantly higher mean levels of delinquent behavior. The exception was youth selfreported delinquency at age 15; although male youth reported a greater average number of delinquency acts committed, this difference did not reach statistical significance. Turning to Table 8, which presents the results of the analysis by race, differences in self-reported delinquency did not reach statistical significance at ages 12, 13, 14, and 16. At ages 15 and 17, however, Black youth reported committing significantly fewer delinquent acts than non-Black youth (1.4 acts vs. 2.0 acts; 1.5 acts vs. 2.7 acts, respectively). Differences in self-reported delinquency reported by Hispanic and non-Hispanic youth did not reach statistical significance (Table 9). 41 Table 7. Youth Self-Reported Number of Delinquent Acts Committed, by Sex (N = 2821) Male Female Mean St. Error Mean St. Error Test Statistic Delinquency, age 12 2.03 0.28 0.67 0.11 4.67 *** Delinquency, age 13 2.03 0.28 0.67 0.11 4.69 *** Delinquency, age 14 2.19 0.31 1.13 0.22 2.81 ** Delinquency, age 15 2.27 0.26 1.57 0.29 1.80 Delinquency, age 16 3.17 0.43 1.21 0.22 4.18 *** Delinquency, age 17 3.47 0.44 1.41 0.23 3.99 *** * p < .05. ** p < .01. *** p < .001. Note: Estimates are weighted. Comparisons done with independent samples t-test. Survey design taken into account using Taylor series method 42 Table 8. Youth Self-Reported Number of Delinquent Acts Committed, by Race (N = 2821) Black Non-Black Mean St. Error Mean St. Error Test Statistic Delinquency, age 12 1.15 0.17 1.41 0.19 0.98 Delinquency, age 13 1.51 0.17 1.41 0.19 1.01 Delinquency, age 14 1.32 0.20 1.73 0.22 1.48 Delinquency, age 15 1.41 0.21 2.02 0.22 2.03 * Delinquency, age 16 2.39 0.92 2.18 0.24 0.22 Delinquency, age 17 1.46 0.24 2.65 0.28 3.44 *** * p < .05. ** p < .01. *** p < .001. Note: Estimates are weighted. Comparisons done with independent samples t-test. Survey design taken into account using Taylor series method. 43 Table 9. Youth Self-Reported Number of Delinquent Acts Committed, by Ethnicity (N = 2821) Hispanic Non-Hispanic Mean St. Error Mean St. Error Test Statistic Delinquency, age 12 1.11 0.21 1.40 0.18 1.04 Delinquency, age 13 1.13 0.21 1.40 0.18 0.96 Delinquency, age 14 1.56 0.34 1.69 0.22 0.31 Delinquency, age 15 1.68 0.31 1.97 0.22 0.73 Delinquency, age 16 1.64 0.31 2.30 0.28 1.58 Delinquency, age 17 2.85 0.52 2.41 0.27 0.74 * p < .05. ** p < .01. *** p < .001. Note: Estimates are weighted. Comparisons done with independent samples t-test. Survey design taken into account using Taylor series method. 44 Figure 3 illustrates changes in youth hazard of arrest graphically and explores differences in arrest hazard by sex, racial group, and ethnicity. Specifically, Figure 3 plots values of the survivor functions (i.e., the cumulative risk of event non-occurrence) to assess the probability that youth will not experience an arrest during each time period of the observation period (each month from youths’ twelfth birthdays to their eighteenth birthdays) for all youth. It also illustrates how those trajectories vary for male and female youth, as well as those youth who belong to different racial/ethnic groups. Consistent with the life table presented in Table 6, Figure 3-1 indicates that few youth experienced an arrest at either age twelve (months 1-12) or thirteen (months 13-24). Beginning at age fourteen (month 25), as time progresses, the survivor th function declines more steadily as risk of arrest increases. By the end of youths’ 17 year of age, slightly more than 80 percent had never experienced an arrest. Figures 3-2 through 3-4 plot the distribution of baseline survival rates of arrest for each gender and racial/ethnic group. Looking at Figure 3-2, which presents the plotted values for females and males separately, the survival rate for female youth is consistently higher than for males. In addition, the survivor function estimated for male youth decreases more sharply throughout the observation period. By the end of adolescence, approximately 88 percent of young women had survived without experiencing their first arrest, compared to only 78 percent of male youth. As shown in Figure 3-3, which presents the functions separately for Black and non-Black youth, the relative level of the survivor function is consistently higher for non-Black youth, indicating that a greater proportion of non-Black youth than Black youth survived each observation period without experiencing their first arrest. In addition, the relative magnitude of the differential in survival rates between each group is remarkably consistent. During each month from youths’ twelfth through seventeenth birthdays, Black youth were approximately four 45 percent more likely than non-Black youth to have experienced their first arrest. Finally, Figure 3-4 shows a strong similarity between the plotted survivor functions for non-Hispanic and Hispanic youth. In other words, non-Hispanic and Hispanic youth had similar rates of survival during each time period of observation. 46 Figure 3. Fitted Survival Function for Age at First Arrest, All Youth and by Sex, Race, and Ethnicity (N = 2821) 47 Figure 3 (cont’d) 48 Tables 10, 11, and 12 present the results of the bivariate analyses of youth contact with the justice system by gender, race, and ethnicity. The design-based F statistic was used to examine whether youth individual characteristics were significantly related to contact with the justice system at each level: arrest, charge, court appearance, conviction, and placement. As shown in Table 10, a greater proportion of male youth reported an arrest during the observation period (21.3 percent versus 11.8 percent, F-statistic = 45.8, p < .001). Once arrested, male youth and female youth were equally likely to report that they had been charged with an offense. However, male youth were more likely than female youth to appear in court as a result of the charge(s) (80.2 percent versus 60.3 percent, F-statistic = 6.57, p < .05). No significant differences were found in the proportions of male and female youth who appeared in adult court, were convicted of their offense, and were placed in a correctional institution as a result of their conviction. Looking at Table 11, youth race was significantly related to justice system contact at the bivariate level. A greater proportion of Black youth reported an arrest during the observation period (22.4 percent versus 15.6 percent, F-statistic = 10.0, p < .01). Once arrested, Black and non-Black youth were equally likely to report that they had been charged with an offense, appeared in court as a result of the charge(s), and appear in adult court as a result of the charge(s). However, Black youth were less likely than non-Black youth to report that their court appearance resulted in a conviction (66.7 percent versus 80.4 percent, F-statistic = 4.11, p < .05). No significant differences were found in the proportions of Black and non-Black youth who appeared were placed in a correctional institution as a result of their conviction. Finally, Table 12 indicates that youth ethnicity was not significantly related to justice system contact at the bivariate level. Similar proportions of Hispanic and non-Hispanic youth 49 reported an arrest during the observation period and reported that their first arrest resulted in a charge. Once charged, they were equally likely to appear in court as a result of the charge(s). Among those who appeared in court, Hispanic and non-Hispanic youth were equally likely to appear in adult court and equally likely to report that they were convicted of their offense. No significant differences were found at the bivariate level between youth ethnicity and likelihood of being placed in a correctional institution as a result of their conviction. 50 Table 10. Youth Contact with the Justice System, by Sex (N = 2821) Male Female % % Arrest Test Statistic Not Arrested 78.7 88.2 45.8 Arrested 21.3 11.8 44.0 48.9 56.0 51.1 19.8 36.7 80.2 60.3 28.4 31.1 0.11 24.6 15.1 1.91 75.4 84.9 73.5 72.6 26.5 *** 27.4 a Charge Not Charged First Juvenile Arrest Resulted in a Charge 0.93 b Court Appearance Did Not Appeared in Court Appeared in Court Appeared in Adult Court c 6.57 * c Conviction Not Convicted Convicted d Placement Not Placed in an Institution Placed in an Correctional Institution 0.01 * p < .05. ** p < .01. *** p < .001. a. Among Youth Arrested as a Juvenile (n = 494) b. Among Those Whose First Arrest Resulted in a Charge (n = 262) c. Among Those Whose First Arrest Resulted in a Charge and a Court Appearance (n = 191) d. Among Those Whose First Arrest Resulted in a Charge, a Court Appearance, and a Conviction (n = 141) Note: Estimates are weighted. Comparisons done with the design-based F test. Survey design taken into account using Taylor series method. 51 Table 11. Youth Contact with the Justice System, by Race (N = 2821) NonBlack Black % % Test Statistic Arrest Not Arrested 77.6 84.4 Arrested 22.4 15.6 49.9 44.7 50.1 55.3 27.7 26.0 72.3 74.0 27.3 29.5 0.10 Not Convicted 33.3 19.6 4.11 Convicted 66.7 80.4 73.8 73.2 Charge 10.0 ** a Not Charged First Juvenile Arrest Resulted in a Charge Court Appearance b Did Not Appeared in Court Appeared in Court Appeared in Adult Court Conviction Placement 1.04 c 0.07 c * d Not Placed in an Institution 0.01 Placed in an Correctional Institution 26.2 26.8 * p < .05. ** p < .01. *** p < .001. a. Among Youth Arrested as a Juvenile (n = 494) b. Among Those Whose First Arrest Resulted in a Charge (n = 262) c. Among Those Whose First Arrest Resulted in a Charge and a Court Appearance (n = 191) d. Among Those Whose First Arrest Resulted in a Charge, a Court Appearance, and a Conviction (n = 141) Note: Estimates are weighted. Comparisons done with the design-based F test. Survey design taken into account using Taylor series method. 52 Table 12. Youth Contact with the Justice System, by Ethnicity (N = 2821) Hispanic Non% Hispanic % Arrest Test Statistic Not Arrested 83.3 83.3 0.00 Arrested 16.7 16.7 50.3 45.1 49.7 54.9 29.4 25.9 70.6 74.1 24.5 29.7 0.27 28.7 21.3 0.86 71.3 78.7 63.8 74.3 36.2 25.7 a Charge Not Charged First Juvenile Arrest Resulted in a Charge 0.41 b Court Appearance Did Not Appeared in Court Appeared in Court Appeared in Adult Court c 0.22 c Conviction Not Convicted Convicted d Placement Not Placed in an Institution Placed in an Correctional Institution 0.84 * p < .05. ** p < .01. *** p < .001. a. Among Youth Arrested as a Juvenile (n = 494) b. Among Those Whose First Arrest Resulted in a Charge (n = 262) c. Among Those Whose First Arrest Resulted in a Charge and a Court Appearance (n = 191) d. Among Those Whose First Arrest Resulted in a Charge, a Court Appearance, and a Conviction (n = 141) Note: Estimates are weighted. Comparisons done with the design-based F test. Survey design taken into account using Taylor series method. 53 Multivariate Analyses of Youth Hazard of Arrest This section presents the results of fitting discrete-time hazard models for youth risk of arrest. The most basic discrete-time event history model uses a set of dummy variables to identify each period of observation. This specification of time places no constraints on the shape of the baseline hazard. However, unrestricted models of time may yield fitted hazard functions that fluctuate erratically across consecutive time periods due to sampling variation, and they require the inclusion of a large number of dummy predictors when the study involves many discrete time periods (Singer & Willett, 2003, p. 408). In these analyses, the 72 discrete time periods corresponding with each month from youths twelfth to their eighteenth birthdays would necessitate the use of 72 dummy variables (alternatively, 71 dummy variables and an intercept interpretable as the log-odds of arrest in month 72). As a result, although recorded discretely, alternative continuous specifications of time were examined. Table 13 presents the results of fitting the unrestricted model of time and four continuous specifications of time, following the ladder of powers, ranked from lowest to highest Akaike’s Information Criteria (AIC). AIC statistics, which penalize deviance statistics (-2-loglikelihoods) for the presence of additional parameters (Singer & Willett, 2003, p. 416), were used to compare models. As shown in Table 13, the AIC is at its minimum for the quadratic model. However, the linear model also performs well with one fewer parameter than the quadratic model, and the difference in deviance statistics between these two models confirms that the 2 quadratic model fits these data no better than the linear model (χ = 3.6, n.s., 1 df). Therefore, the following analyses constrain the effect of time to be linear on the log-odds of arrest. 54 Table 13. Comparison of Alternative Representations for the Main Effect of Time (Measured in Months), Ranked from Lowest to Highest AIC Specification of Transformation Number of Time AIC Deviance Parameters 2 α0 + α1(time) + α2(time) 3 6806.1 6796.1 Linear α0 + α1(time) 2 6807.7 6799.7 Cubic α0 + α1(time) + α2(time) + α3(time) 4 6808.1 6796.1 α0 + log(time) 2 6810.9 6802.9 Quadratic 2 Logarithmic 3 Unrestricted Series of dummy variables 72 6864.3 6716.3 Note: Models examining fit statistics for time specifications include random intercepts for county and state levels Table 14 shows the parameter estimates for the null model, which includes only the linear specification for the effect of time, and for the models that include youth individual characteristics, contextual characteristics, and cross-level interactions between youth characteristics and contextual environment. To facilitate interpretation, a centering constant of one is subtracted from the time variable so that the intercept is interpretable as the value of the logit hazard during the first observation period (i.e., when the time variable equals zero). The baseline null model, presented in Model One, indicates that youths’ estimated hazard of arrest during the time period corresponding with the month following youths’ twelfth birthdays is approximately 0.002 (e -6.401 ). As time increases, youth hazard of arrest increases. Specifically, 0.013 each month raises the odds of first arrest by approximately 1.3 percent (100[e 55 – 1] ). After the addition of individual-level variables in Model Two, blocks of county- and state-level contextual effects were added in succession to examine the direct effects of county levels of structural disadvantage, black population composition, political conservatism, court characteristics, delinquency petition and crime rates, gender inequity, child health and wellbeing, and juvenile justice policies. The main effects of county- and state-level contextual effects were non-significant in all analyses (results not shown). Next, cross-level interaction effects models were generated to examine whether the effects of individual characteristics depended on contextual climate. Results indicated that the effect of belonging a Black racial group on youth risk of arrest depends on county racial population composition (Model Three). 56 Table 14. Estimating the Hazard of Arrest: Three-Level Discrete-Time Hazard Model Estimates for the Regression of Arrest on Individual and Contextual Characteristics, Random Intercepts for County and State (N = 2821) Model One Model Two Model Three Fixed Individual Characteristics Intercept -6.401 *** -7.009 *** -6.977 *** 0.013 *** 0.014 *** 0.014 *** Male 0.691 *** 0.697 *** Black 0.440 *** 0.502 *** Hispanic 0.168 0.185 0.110 *** 0.108 (Months-1) a Delinquency Fixed County Characteristics a 0.082 Black Composition Cross-Level Interactions Black X Black Composition Random Part *** a -0.299 ** County-Level Intercept 0.021 0.021 0.029 State-Level Intercept 0.024 0.038 <.001 Black Slope, County-Level 0.256 Deviance 6799.7 6692.6 6680.6 AIC 6807.7 6708.6 6700.6 * p < .05. ** p < .01. *** p < .001. a. Variable is grand mean centered and standardized 57 Looking at Model Three in Table 14, the main effects of time (measured in months), gender, racial group, and self-reported delinquency were significantly related to youth hazard of arrest. In addition, the significant interaction between race and county-level black population composition indicates that the magnitude of the effect of race depends on the composition of the population. During each month of observation, the odds of experiencing an arrest are nearly 0.697 twice as high for males as for females (e ). Youth who reported greater levels of self- reported delinquency are more likely to experience their first arrest. In particular, a one standard deviation increase in self-reported delinquency increases the odds of experiencing an arrest by an 0.108 estimated 11.4 percent (100*[e -1]). The significant cross-level interaction between Black racial group and Black Composition indicates that as the proportion of the Black county population decreases, racial disparities in youth hazard of arrest increase. Controlling for delinquent behavior, in counties with a large proportion of Black residents (one standard deviation above the mean), the hazard of arrest for Black youth is approximately 23 percent higher than for their non-Black counterparts 0.502 + (1*-0.299) (100*[e -1]). The main effect of Black racial group indicates that in areas of average Black population composition, the odds of experiencing an arrest is approximately 65 0.502 percent greater for Black youth in comparison to non-Black youth (100*[e -1]). Correspondingly, in counties with a small proportion of Black residents (one standard deviation below the mean), the hazard of arrest for Black youth is approximately 123 percent higher than 0.502 + (-1*-0.299) for their non-Black counterparts (100*[e -1]). Multivariate Analyses of Youth Probability of Charge Table 15 shows the effects of individual characteristics and contextual climate on the likelihood of youths’ first arrest resulting in a charge, with county-level variables disaggregated 58 to the individual level and a random intercept for state. Model One presents the null model with no parameters. Model Two adds the collection of individual characteristics, including youth age at the time of the arrest. Model Three includes significant county characteristics as fixed effects, and Model Four includes the state-level scale of children’s health and as a fixed effect and a cross-level interaction between Hispanic ethnicity and children’s health. In the interest of parsimony, non-significant effects are excluded from the analyses, which did not substantively affect the results. Looking at Model Four, results indicate that, of the individual characteristics, only youth age at the time of the arrest significantly predicts likelihood of the arrest resulting in a criminal charge. A one standard deviation increase in age increases the odds of charge by approximately 0.249 28 percent (100*[e -1]). Characteristics of the county prosecutor’s office also significantly impacts youth likelihood of a criminal charge. Residing in a county with a special prosecuting attorney for handling juvenile cases reduces youths’ likelihood of a criminal charge by 28 -0.329 percent (100*[e -1]). The significant interaction effect between Hispanic ethnicity and children’s health indicates that ethnic disparities in likelihood of charge depend on state climate of children’s health. In particular, Hispanic and non-Hispanic disparities in likelihood of a charge increase as state-level children’s health decreases. In states with poorer levels of children’s health (one standard deviation below the mean), youth odds of a criminal charge are approximately 216 0.177 + (-1*-0.973) percent higher for Hispanic youth, compared to non-Hispanic youth (100*[e - 1]). Conversely, in states with higher levels of children’s health (one standard deviation above the mean), youth odds of a criminal charge are approximately 55 percent lower for Hispanic 0.177 + (1*-0.973) youth, compared to their non-Hispanic counterparts (100*[e 59 -1]). Table 15. Estimating the Probability of Charge: Two-Level Logit Model Estimates for the Regression of Charge on Individual and Contextual Characteristics, Random Intercept for State (n = 490) Model One Model Two Model Three Model Four Fixed Individual Characteristics Intercept 0.171 0.167 0.359 0.255 Male 0.169 0.142 0.169 Black -0.167 -0.200 -0.122 Hispanic -0.274 -0.308 0.177 Age at Arrest (in days) a 0.245 ** 0.245 ** 0.249 ** Fixed County Characteristics Juvenile Prosecuting Attorney -0.430 * -0.329 * Fixed State Characteristics Child Health Interactions a Hispanic X Child Health Random Part 0.144 a State-Level Intercept -0.973 * 0.062 0.078 0.058 Hispanic Slope, State Level 0.011 0.124 Deviance 675.6 666.0 660.8 651.2 AIC 679.6 678.0 674.8 673.2 * p < .05. ** p < .01. *** p < .001. Note: Self-reported delinquency was non-significant and was therefore excluded from the model for the sake of parsimony. This did not substantively affect the results. a. Variable is grand mean centered and standardized. 60 Multivariate Analyses of Youth Probability of Court Appearance Table 16 presents the results for the analysis of court appearance, a trichotomous variable indicating youth adult court appearance, juvenile court appearance, and no court appearance. The two-level ordered logit model includes county-level variables disaggregated to the individual level and a random intercept for state. The model presents the results of the effects of the regressors on the log odds of more severe versus less severe court appearance (i.e., adult court appearance versus other; appearance in any court versus no court appearance). Model One presents the null model with no parameters. Model Two adds the collection of individual characteristics, including youth age at the time of the arrest. Model Three includes the statelevel scales of juvenile justice system punitiveness and gender inequality as fixed effects and cross-level interactions between state-level scales and both gender and Hispanic ethnicity. Again, in the interest of parsimony non-significant effects, including all county-level covariates, are excluded from the analyses; this exclusion did not substantively affect the results. Net of self-reported delinquency, severity of court appearance depends on gender, youth age at the time of the arrest, state levels of gender inequity, and state juvenile justice policy context. In particular, a one standard deviation increase in age increases the odds of a more 0.655 severe court appearance by approximately 93 percent (100*[e -1]). The significant interaction effect between gender and Hispanic ethnicity and state-level covariates indicate that gender and ethnic disparities in likelihood of court appearance depend on state climate of gender inequity and juvenile justice system punitiveness. Looking at the significant interaction effect for gender, severity of court appearance for male and female youth depends on state-levels of gender inequity. In areas of lower gender inequity (one standard deviation below mean gender inequity), youth odds of a more severe court 61 appearance are approximately 289 percent greater for male youth compared to female youth 0.575 + (-1*-0.784) (100*[e -1]). Taking the reciprocal of the odds ratio, it follows that the odds of a severe court appearance for female youth are approximately 26 percent of the odds for male 0.575 + (-1*-0.784) youth (1/[e ]). Conversely, in areas of higher gender inequity (one standard deviation above mean gender inequity), youth odds of a more severe court appearance are 0.575 + (1*- approximately 19 percent lower for male youth compared to female youth (100*[e 0.784) -1]). What this implies is that male youth are treated disproportionately harsher in states with high levels of male and female equality; whereas, female youth are treated disproportionately harsher in states with large inequalities between males and females. Looking at the significant interaction effect for ethnicity, ethnic differences in the odds of a severe court appearance depend on juvenile justice policy context. In states with less punitive juvenile justice policies, Hispanic youth have an approximately 98 percent higher odds of a severe court appearance than their non-Hispanic counterparts (100*[e -0.213 + (-1*-0.895) -1]). The opposite is true in states with more punitive juvenile justice policies, where the odds of a severe court appearance for Hispanic youth are approximately 67 percent lower than for non-Hispanic youth (100*[e -0.213 + (1*-0.895) -1]). 62 Table 16. Estimating the Probability of Court Appearance: Two-Level Ordered Logit Model Estimates for the Regression of Court Appearance on Individual and Contextual Characteristics, Random Intercept for State (n = 261) Model One Model Two Model Three -1.926 *** -1.819 *** Fixed Individual Characteristics Intercept Intercept Male a -1.367 *** b 1.069 *** 0.575 * 0.060 Age at Arrest (measured in days) Fixed State Characteristics c -0.198 -0.234 Hispanic Gender Inequity 0.940 ** 0.666 ** Black 0.755 * -0.213 0.629 *** c 0.454 Juvenile Justice System Punitiveness Interactions Male X Gender Inequity 0.655 *** c 0.400 c -0.784 * Male X Juv. Justice Sys. Punitiveness Hispanic X Gender Inequity c c Hispanic X Juv. Justice Sys. Punitiveness Random Part State-Level Intercept -0.020 0.813 c -0.895 * 0.170 0.185 0.115 Male Slope, State Level <.001 Hispanic Slope, State Level 0.002 Deviance 529.2 498.2 484.6 AIC 535.2 512.2 512.6 * p < .05. ** p < .01. *** p < .001. Note: Self-reported delinquency and a dummy variable indicating a charge for a violent offense were non-significant and were therefore excluded from the model for the sake of parsimony. Their exclusion did not substantively affect the results. a. Model intercept for appearance in adult court versus juvenile court or no court b. Model intercept for appearance in any court versus no court appearance c. Variable is grand mean centered and standardized. 63 Multivariate Analyses of Youth Probability of Conviction The results of the two-level logistic regression of conviction on individual and contextual characteristics are presented in Table 17. County-level covariates are disaggregated to the individual level, and the model intercept is allowed to vary at the state level. Model One presents the baseline intercept-only model, and Model Two adds the collection of individual characteristics. Age at the time of arrest, self-reported delinquency, and two dummy variables indicating charge in adult court and charge for a violent offense were found to be non-significant in all analyses. Their exclusion from the analyses did not substantively affect the results for subsequently tested models. Model Three includes the significant county-level variables, including a dummy variable indicating that the county prosecutor’s office has a specialized unit for handling juvenile cases and standardized rates of violent and property crimes. State levels of gender inequity, children’s health and wellbeing, and juvenile justice system punitiveness did not significantly affect youth likelihood of conviction. No significant interaction effects were found between individual characteristics and either county-level or state-level variables indicating contextual climate. Looking at Model Three, net of self-reported delinquency, Black youth are significantly less likely to be convicted of an offense than their non-Black counterparts. The lack of significant interaction effects indicates that regardless of county or state contextual climate, the odds of conviction are approximately 67 percent lower for Black youth than for non-Black youth -1.001 (100*[e -1]). Turning to the county-level fixed effects, the estimated odds of conviction for youth who reside in a county that has a prosecutor’s office with a specialized unit for handling juvenile cases are approximately 56 percent lower than the odds of conviction for youth whose -0.804 county prosecutor’s office lacks a specialized juvenile unit (100*[e 64 -1]). Furthermore, county levels of violent and property crime rates significantly impact likelihood of conviction but in opposite directions. A one standard deviation increase in the rate of violent crime 1.086 increases youth odds of conviction by approximately 196 percent (100*[e -1]); whereas, a one standard deviation increase in the rate of property crime decreases youth odds of conviction by approximately 41 percent (100*[e -0.531 -1]). 65 Table 17. Estimating the Probability of Conviction: Two-Level Logit Model Estimates for the Regression of Conviction on Individual and Contextual Characteristics, Random Intercept for State (n = 191) Model Model Model One Two Three Fixed Individual Characteristics Intercept 1.037 1.902 2.387 *** Male -0.571 -0.710 Black -0.842 ** -1.001 *** Hispanic -0.713 -0.689 Fixed County Characteristics Juvenile Unit Violent Crime Rate -0.804 * a Property Crime Rate 1.086 * a -0.531 * Random Part State-Level Intercept Deviance <.001 <.001 <.001 219.6 211.5 198.4 AIC 221.6 219.5 212.4 * p < .05. ** p < .01. *** p < .001. Note: The following individual-level variables were non-significant and were therefore excluded from the model for the sake of parsimony: self-reported delinquency, age at arrest, a dummy variable indicating a charge for a violent offense, and a dummy variable indicating an appearance in adult court. Their exclusion did not substantively affect the results. a. Variable is grand mean centered and standardized. 66 Multivariate Analyses of Youth Probability of Placement The regression results for the two-level logit model for placement in a correctional institution are shown in Table 18. Again, county-level effects were disaggregated to the individual level, and the model intercept was allowed to vary at the state level. Model One presents the baseline intercept-only model, and Model Two adds the collection of individual characteristics. Age at the time of arrest, self-reported delinquency, and a dummy variable indicating charge in adult court were non-significant in all analyses and therefore dropped from the models, which did not substantively affect the results. Model Three includes the significant county-level variables, including a standardized count variable indicating the number of times in the previous year the county prosecutor’s office pursued criminal charges against a juvenile and two standardized scales indicating county structural disadvantage: residential stability and concentrated affluence. Model Four adds state-level variables indicating gender inequity and children’s health, and Model Five includes significant cross-level interactions. Model Five shows that neither youth race nor ethnicity significantly impacts likelihood of placement in a correctional institution. However, youth who were convicted of a violent offense are approximately five times more likely than those convicted of a non-violent offense to be 1.671 placed in a correctional institution (e ). At the county level, a one standard deviation increase in the number of youth waived to adult court decreases youth odds of placement by -1.046 approximately 65 percent (100*[e 6 -1]). Furthermore, county-levels of residential stability and affluence reduce youth likelihood of placement. Specifically, a one standard deviation 6 To examine whether the effect of youth waiver was due to variations in jurisdiction sizes or variations in state upper age limits of juvenile court jurisdiction, the models predicting placement in a correctional institution were rerun with the inclusion of county population size and upper age limit as control variables. Their inclusion as covariates did not substantively affect the results. 67 increase in residential stability reduces youth likelihood of placement by approximately 67 -1.118 percent (100*[e -1]), and a one standard deviation increase in concentrated affluence reduces youth likelihood of placement by approximately 68 percent (100*[e -1.135 -1]), net of self- reported delinquent behavior. The significant interaction effects between gender and state-level covariates indicate that gender disparities in likelihood of placement in a correctional institution depend on state climate of gender inequity and children’s health and wellbeing. Taken together, the interactions suggest that in areas of lower levels of women and children’s wellbeing (one standard deviation above mean Gender Inequity and one standard deviation below mean Child Health), youth odds of placement in a correctional institution are approximately 3 percent greater for male youth 0.391 + (1*1.808) + (-1*2.358) compared to female youth (100*[e -1]). On the contrary, in areas of higher levels of women and children’s wellbeing (one standard deviation below mean Gender Inequity and one standard deviation above mean Child Health), youth odds of placement in a correctional institution are approximately 208 percent greater for male youth compared to female 0.391 + (-1*1.808) + (1*2.358) youth (100*[e -1]). Taking the reciprocal of the odds ratio, it follows that the odds of placement for female youth are approximately 32 percent of the odds for male 0.391 + (-1*1.808) + youth in states with high levels of women and children’s wellbeing (1/[e (1*2.358) ]). These findings suggest that male and female youth are nearly equally likely to be placed in a correctional institution in states with negative climates of women’s wellbeing and children’s health. What this implies is that female youth are treated disproportionately harsher in states with negative climates of women’s and children’s wellbeing; whereas, male youth are treated disproportionately harsher in states with positive climates women’s and children’s wellbeing. 68 Table 18. Estimating the Placement in a Correctional Institution: Two-Level Logit Model Estimates for the Regression of Placement on Individual and Contextual Characteristics, Random Intercept for State (n = 141) Model One Model Two Model Three Fixed Individual Characteristics Intercept -0.838 *** -1.229 * -1.717 ** Male 0.155 0.274 Black 0.215 0.571 Hispanic 0.576 0.902 Convicted of a Violent Offense 0.707 1.336 Fixed County Characteristics Number Waived a Residential Stability -0.863 *** a -0.943 * a -0.966 ** Concentrated Affluence Fixed State Characteristics Gender Inequity Child Health Interactions a a Male X Gender Inequity Male X Child Health Random Part a a State-Level Intercept 0.073 0.038 0.176 Deviance 173.4 171.3 156.5 AIC 177.4 181.3 172.5 Male Slope, State Level * p < .05. ** p < .01. *** p < .001. Note: The following individual-level variables were non-significant and were therefore excluded from the model for the sake of parsimony: self-reported delinquency, age at arrest, and a dummy variable indicating an appearance in adult court. Their exclusion did not substantively affect the results. a. Variable is grand mean centered and standardized. 69 Table 18 (cont’d) Model Four Model Five Fixed Individual Characteristics Intercept -1.662 ** -1.958 ** Male 0.215 0.391 Black 0.381 0.508 Hispanic 0.984 1.086 Convicted of a Violent Offense 1.497 * 1.671 * -0.852 ** -1.046 ** -0.983 ** -1.118 ** -1.006 ** -1.135 ** -0.515 -1.879 * Fixed County Characteristics Number Waived a Residential Stability a a Concentrated Affluence Fixed State Characteristics Gender Inequity Child Health Interactions a a 0.669 * Male X Gender Inequity Male X Child Health Random Part a 1.808 a State-Level Intercept -2.314 *** 2.358 ** <.001 Male Slope, State Level <.001 0.314 Deviance 147.3 138.0 AIC 167.3 164.0 * p < .05. ** p < .01. *** p < .001. Note: The following individual-level variables were non-significant and were therefore excluded from the model for the sake of parsimony: self-reported delinquency, age at arrest, and a dummy variable indicating an appearance in adult court. Their exclusion did not substantively affect the results. a. Variable is grand mean centered and standardized. 70 CHAPTER 5 DISCUSSION AND CONCLUSION Despite the large body of research examining the effects of extralegal factors, especially race, on police behavior and justice system processing (for reviews, see Bishop, 2005; Engen et al., 2002; Leiber, 2002; Paternoster & Iovanni, 1989; Pope & Feyerherm, 1990a, 1990b; Pope et al., 2002), researchers have paid little attention to exploring the contextual environments associated with disparities in justice system processing (for exceptions, see Armstrong & Rodriguez, 2005; DeJong & Jackson, 1998; Feld, 1991, 1995; Rodriguez, 2007, 2010; Sampson & Laub, 1993). Even fewer studies have examined the indirect effects of community characteristics and disentangled the effects of race and ethnicity with gender (for exceptions, see Bishop et al., 2010; Rodriguez, 2007; Stevens et al., 2011). In response to this gap, the present study delineates contextual factors that either increase or reduce the severity of justice system response to youth and examines how racial/ethnic group and gender intersect with contextual factors to produce disproportionately high rates of justice system contact. Using a nationally representative and longitudinal sample of youth, this study focuses on how the effects of racial/ethnic group status and gender differentially affect youth penetration into the justice system depending on contextual environment, net of youth participation in delinquent behavior. The concluding chapter begins with a summary of the research findings relevant to the hypotheses and then offers discussion of study limitations and recommendations for future research, followed by a brief conclusion and discussion of implications for policy practice. Summary and Discussion The central hypotheses explored in this study include: 71 1. Racial and ethnic group status and gender affect youth penetration into the justice system 2. Youth penetration into the justice system depends on economic, political, and social contexts 3. The effects of individual characteristics on youth penetration into the justice system depend on economic, political, and social context The conceptual model that frames these hypotheses extends existing models of disproportionate youth contact with the justice system that favor individual-level explanations for disparities in youth penetration into the justice system. The study’s use of a nationally representative sample drawn from the general youth population allowed the incorporation of data from youth who avoided contact with the justice system during adolescence, thus allowing the empirical examination of what characteristics predict such contact, net of reported delinquent behavior, with broadly generalizable results. Further extending previous research on disproportionate contact with the justice system, the study’s use of event history analysis allowed the effects of predictors on youth risk of arrest to vary over time, as youth changed their behavior, moved residences, and as the characteristics of geographic areas changed. Finally, the study’s multilevel and longitudinal design allowed the examination of individual and contextual effects at each level of police contact and court processing. The results of the bivariate analyses show significant differences in justice system contact by gender and race. With the exception of age 15, male youth reported significantly higher levels of participation in delinquent behavior each year from age 12 to age 17. Compared to female youth, male youth had a higher hazard of arrest during each year of observation and a survivor function that decreased more sharply with age. Once arrested, equal proportions of 72 male and female youth reported being charged with an offense, appearing in court as a result of the charge(s), and being convicted, and being placed in a correctional institution. Although bivariate analyses revealed that Black youth reported either similar (at ages 12, 13, 14, and 16) or lower (at ages 15 and 17) levels of participation in delinquent behavior than their non-Black counterparts, they had a higher hazard of arrest during each year from age 12 through age 17. Once arrested, however, similar proportions of Black and non-Black youth reported being charged with an offense and appearing in court as a result of the charge(s), and a smaller proportion of Black youth reported being convicted. Once convicted of an offense, similar proportions of Black and non-Black youth reported being placed in a correctional institution. Unlike the findings for gender and race, no significant differences were found in delinquent behavior and justice system contact between Hispanic and non-Hispanic youth at the bivariate level. Consistent with the hypotheses, the results of the multivariate analyses show the direct effects of gender, racial/ethnic group, and contextual characteristics on penetration into the justice system. Although it was expected that the effects of racial/ethnic group status and gender would interact in their effect on youth justice system contact, no significant interactions were found, indicating that racial disparities in justice system processing do not depend on gender. In other words, racial and ethnic disparities in police and court processing are experienced equally by young men and women of color. Moreover, one of the more notable findings of this dissertation is that the effects of gender and racial/ethnic group are not consistent across police and court processes. Consistent with prior research (see, for example, Bishop, 2005; Bishop et al., 2010; Dannefer & Schutt, 1982; Rodriguez, 2007, 2010), the results of the analyses reveal that the effects of gender and racial/ethnic group are more pronounced at some decision-making 73 levels. Also notable is the identification of the effects of individual characteristics in combination with contextual environment. In particular, the findings indicate that disparities in justice system processing depend on county population composition, state climates of women and children’s wellbeing, and state juvenile justice system punitiveness. The results of the analyses by race reveal that, net of self-reported delinquency, Black youth have a higher hazard of arrest than their non-Black counterparts (See Table 19 for a summary of the results by race). Furthermore, cross-level interaction effects suggest the magnitude of the effect of race depends on county racial composition. In particular, racial disparities in hazard of arrest are magnified as the proportion of county residents who are Black decreases. In other words, racial disparities in risk of arrest are most pronounced among youth who reside in predominately non-Black communities, and Black youth who reside in predominately non-Black communities have the higher risk of arrest. This finding lends support to the benign neglect hypothesis of the conflict perspective, which posits an inverse relationship between Black population percentage and measures of crime control for that population (see, for 7 example, Liska & Chamlin, 1984; Myer & Chamlin, 2011; Parker et al., 2005). Additionally, consistent with prior research documenting variations across decision-making levels in the effects of race on justice system processing, the results indicate that although Black youth have a disproportionately high risk of arrest, once arrested, they are no more likely to be charged with a crime or appear in court, and they are less likely to be convicted of an offense. Once convicted, Black and non-Black youth are equally likely to be placed in a correctional institution. These 7 This relationship is said to result from two mechanisms (Myer & Chamlin, 2011). First, crime in predominately Black communities is likely to be intraracial, and intraracial crime may be viewed by police and victims as a personal problem that does not require official intervention. Second, residents of predominately Black communities may lack the sociopolitical capital necessary to legitimize their problems, resulting in a decreased level of crime control. 74 findings support research documenting a compensatory effect in the courts whereby Black youth receive more favorable court dispositions than their non-Black counterparts (see, for example, Dannefer & Schutt, 1982). 75 Charge Court Appearance Conviction Placement (+) X (-) Juvenile Justice System Punitiveness Youth Disconnectedness Children’s Health Gender Inequity Crime Rate Delinquency Petition Rate Prosecutor’s Office Characteristics /Practices Political Conservatism Population Composition Structural Disadvantage Outcome Arrest Main Effect of Black Racial Group Table 19. Summary of the Results by Race: Main Effects and Interactions with Variables Indicating Contextual Environment County-Level State-Level Interaction Variables Interaction Variables Conclusion Black youth have a higher risk of arrest in all contextual environments, but racial disparities are magnified in non-Black counties Black youth are less likely to be convicted 76 Turning to the results of the analyses by gender, consistent with the hypotheses, male youth have a higher hazard of arrest than female youth, controlling for participation in delinquent behavior (for a summary, see Table 20). Once arrested, however, male and female youth have equal likelihoods of charge. Once charged, the effect of gender on severity of court appearance (i.e., adult court appearance versus other; court appearance versus no court appearance) depends on state levels of women’s wellbeing. In particular, in areas with low levels of political-, economic-, and health-related gender inequality, the gender gap in processing is considerably wide, as male youth are much more likely than their female counterparts to receive a severe court appearance. Conversely, the gender gap in processing narrows and reverses as state levels of gender inequality increase; as a result, in areas with high levels of gender inequality, female youth are treated disproportionately harsher and are more likely than their male counterparts to receive a severe court appearance. Yet once male and female youth appear in court, they are equally likely to be convicted. However, consistent with the findings for court appearance, the results of the analyses suggest that gender disparities in youth likelihood of placement in a correctional institution depend on state climates of women and children’s wellbeing. Again, the gender gap in placement is more pronounced in states with positive climates of women’s and children’s wellbeing (i.e., low levels of gender inequity and high levels of children’s health) and narrows considerably in states with negative climates of women’s wellbeing and children’s health (i.e., high levels of gender inequity and low levels of children’s health). These findings support research that suggests that levels of gender inequity may affect responses to young women’s delinquency (see, for example, Chesney-Lind, 1989; Goodkind et al., 2009; Richie, 1995; Steffensmeier & Allan, 1996). 77 Conviction Placement (+) X X Juvenile Justice System Punitiveness Youth Disconnectedness Children’s Health Gender Inequity Crime Rate Delinquency Petition Rate Prosecutor’s Office Characteristics /Practices Political Conservatism Population Composition Structural Disadvantage Outcome Arrest Charge Court Appearance Main Effect of Male Sex Table 20. Summary of the Results by Gender: Main Effects and Interactions with Variables Indicating Contextual Environment County-Level State-Level Interaction Variables Interaction Variables Conclusion Males have a higher risk of arrest Males are more likely to appear in court in climates of women’s wellbeing. This gender gap closes as women’s wellbeing decreases. X Males are more likely to be placed in a correctional institution in climates of high women’s and children’s wellbeing. This gender gap closes as women’s and children’s wellbeing decreases. 78 Finally, turning to the results of the analyses by ethnicity, although Hispanic ethnicity did not significantly predict youth hazard of arrest, once arrested, cross-level interaction effects indicated that Hispanic youth are disproportionately more likely to be charged with an offense in states with poor climates of children’s health and wellbeing (for a summary, see Table 21). In states with poor levels of children’s health, Hispanic youth are more likely than their nonHispanic counterparts to be charged with an offense once arrested. This ethnic gap in processing narrows considerably and then reverses as state levels of children’s health and wellbeing improve, and Hispanic youth are less likely than their counterparts to be charged with an offense in states with high levels of children’s health and wellbeing. Once charged with an offense, ethnic disparities in severity of court appearance (i.e., adult court appearance versus other; court appearance versus no court appearance) depend on juvenile justice policy context in an unexpected direction. In states with less punitive juvenile justice policies, Hispanic youth receive a considerably more severe court appearance than non-Hispanic youth. Again, the ethnic gap in processing narrows and reverses as state juvenile justice system punitiveness increases, and Hispanic youth receive a considerably less severe court appearance than non-Hispanic youth in states with more punitive juvenile justice system policies. Once youth have appeared in court, however, Hispanic and non-Hispanic youth are equally likely to be convicted of an offense and placed in a correctional institution once convicted. 79 Juvenile Justice System Punitiveness Youth Disconnectedness Children’s Health Gender Inequity Crime Rate Delinquency Petition Rate Prosecutor’s Office Characteristics /Practices Political Conservatism Population Composition Structural Disadvantage Outcome Arrest Charge Main Effect of Hispanic Ethnicity Table 21. Summary of the Results by Ethnicity: Main Effects and Interactions with Variables Indicating Contextual Environment County-Level State-Level Interaction Variables Interaction Variables X Court Appearance X Conviction Placement 80 Conclusion Hispanic youth are more likely to be charged in climates of poor children’s health. This ethnicity gap closes and reverses as children’s health increases. Hispanic youth are more likely to appear in court in climates of low juvenile justice system punitiveness. This ethnicity gap closes and reverses as juvenile justice system punitiveness increases. Recall that the indicator of state juvenile justice system punitiveness was comprised of policies concerning minimum and maximum ages of juvenile court jurisdiction, life without parole for crimes committed under age 18, youth waiver/transfer to criminal courts, blended sanctions, and the collection of juvenile DNA. It is possible that the indicator juvenile justice system punitiveness is a proxy for prosecutorial and judicial discretion in handling juvenile court cases (i.e., level of legislation mandating the handling of juvenile cases). In states with low degrees of policy punitiveness, court officials may have far more discretion in handling juvenile court cases and therefore more discretion in trying juveniles as adults, resulting in ethnic disparities in processing. However, the aggregate nature of the data makes it hard to examine exactly what processes are at play, and future research should examine the direct effects of juvenile policies concerning ages of juvenile court jurisdiction and youth waiver/transfer to criminal courts. It should also be noted that it would be inappropriate to attempt to explain ethnic disparities in processing as the result of the biases of justice system personnel without the appropriate data needed to make such inferences. The results of the multivariate analyses also reveal the direct effects of county characteristics on youth penetration into the justice system. In particular, controlling for rates of violent and property crime, variables indicating prosecutor’s office characteristics and practices and community-level affluence significantly predict youth likelihood of justice system intervention at various stages of processing. Taking the results of the effects of county-level prosecutor’s office characteristics and practices as a whole, the findings indicate that youth who reside in counties with specialized resources within the prosecutor’s office for handling juvenile cases are treated disproportionately less severely than their counterparts. For instance, youth who reside in counties with a dedicated juvenile prosecuting attorney are less likely to be 81 charged with a crime, and youth who reside in counties with a specialized unit for handling juvenile cases are less likely to be convicted of an offense. Furthermore, the number of cases within the county waived to adult court has a significant and negative relationship to likelihood 8 of placement. In other words, youth who reside in counties with higher rates of waiver are less likely to be placed in a correctional institution. If, as Bortner (1986) suggests, organizational practicalities and constraints, including lack bed space and programming for juvenile offenders, account for high rates of juvenile remand to adult court, it would be reasonable to expect that jurisdictions with high rates of youth waiver have lower levels of resources for handling juvenile cases. This would suggest that the relationship between county-level waiver rate and placement is consistent with the findings documenting lower levels of court involvement for youth who reside in counties with greater levels of resources for handling juvenile cases. Finally, youth who resided in counties with higher levels affluence are less likely than their counterparts to be placed in a correctional institution. This is perhaps because affluent communities are more likely to offer a greater number of services to offenders as alternatives to incarceration, including mental health and substance abuse treatment and restorative justice programming (Levrant, Cullen, Fulton, & Wozniak, 1999). 8 It should be noted that the effect of county-level waiver on likelihood of placement was significant controlling for age at arrest, self-reported delinquency, a dummy variable indicating a conviction for a violent offense, and a dummy variable indicating an appearance in adult court. 82 Limitations and Recommendations There are several limitations of this study. Firstly, because the data analyzed are from a nationally representative sample of youth, it is likely that these findings cannot be generalized to more severe offending populations. Although the NLSY97 has an adequately large sample size for the analysis of delinquency and contact with the justice system, which is a relatively rare occurrence, it is likely that youth who have committed severe offenses appear in the sample in very small numbers and are therefore not sufficiently represented in these data. However, evidence indicates that racial disparities in justice system processing may be more pronounced when offences are less serious (Piquero, 2008). Moreover, because the overwhelming majority of research examining disproportionate minority contact relies on official data within particular jurisdictions, the present study complements and extends previous research precisely because of its use of a nationally representative sample and broad generalizability. Secondly, because the current study relies on self-report measures of the occurrence and timing of arrest and levels of subsequent court involvement, the validity of the findings may be affected by respondent recall and/or underreporting (Thornberry & Krohn, 2000). Little is known about the validity of self-report data on the timing of arrests; however, Morris and Slocum (2010) found that retrospective self-report data on the prevalence and frequency of arrest over a three-year period yielded accurate measures. Self-reported timing of arrest, however, was recalled with somewhat less accuracy but yielded accurate measures for the timing of recent arrests. The NLSY97’s collection of information on the occurrence and timing of youth arrest during each annual interview increases the likelihood that the data provides accurate estimates. Furthermore, research generally indicates that self-report data on contacts with the justice system are valid, and it is likely that reporting accuracy increases as youth became more enmeshed in 83 the justice system because court appearances, convictions, and placements are more salient events (for a review of the research on the validity self-report data on contact with the justice system, see Roberts & Wells, 2010). I also considered whether the study’s findings might be biased as a result of gender and racial/ethnic differences in self-reporting of justice system contacts. Comparisons in the validity of self-reported justice system contacts across gender and racial groups yield inconsistent results. For example, several studies report that Black respondents are less likely to report their justice system contacts than White respondents (see, for example, Kirk, 2006). Other research, however, finds no race differences (Thornberry & Krohn, 2000) or suggests that minority respondents may more accurately report their justice system contacts (Farrington, Loeber, Stouthamer-Loeber, van Kammen, & Schmidt, 1996). Despite these limitations, the self-report method remains one of the most valid and reliable methods of measuring delinquency and justice system contacts (Thornberry & Krohn, 2000). Although not the focus of the present study, future research should examine issues of differential validity in self-reported justice system contacts for gender and racial/ethnic groups and develop techniques for eliminating this bias, if it exists (Thornberry & Krohn, 2000). Conclusion and Policy Implications Although there is a considerable body of empirical and theoretical research examining disparities in justice system processing and documenting the effects of extralegal factors, especially race, on police behavior and justice system processing, few studies have included contextual measures in their empirical work, and I could only locate one study (Rodriguez, 2007) examining the indirect effects of contextual characteristics. The research literature is also predominated by studies that rely on official records and focus on particular court jurisdictions, 84 thus having limited generalizability, the inability to control for variations delinquent behavior, and the inability to examine the effects of extralegal variables on likelihood of arrest. The present study is unique in that it uses a dataset constructed from a nationally representative sample of youth merged with county- and state-level data indicating community characteristics and contextual climate. The results, overall, suggest that disparities in justice system processing are partially explained by macro-structural environments. Specifically, this study found racial disparities in youth risk of arrest, which are magnified in predominately non-Black communities. However, this study also found that Black youth receive more favorable court dispositions than their non-Black counterparts, lending support to the hypothesis that judges may compensate for earlier disparities in processing rather than simply reinforce them (see also Dannefer & Schutt, 1982). Consistent with feminist theory (see, for example, Chesney-Lind, 1989), the study’s findings also suggest that the gender gap in youth justice system processing depends on state climates of women and children’s health and wellbeing. Specifically, as women and children’s health and wellbeing decrease, the gender gap in processing narrows and, in the case of severity of court appearance, reverses. The results of the analyses also suggest that Hispanic youth are treated disproportionately more harshly in states with poor climates of children’s health and wellbeing and in states with less punitive juvenile justice systems, perhaps due to higher levels of judicial and prosecutorial discretion in handling juvenile cases in states with low levels of legislation mandating the handling of juvenile cases. Overall, these findings suggest two major directions for policy. First, like most matters of gender, race, and ethnicity, disproportionate contact with the justice system is rooted in the social contexts into which youth are born (Hoytt, Schiraldi, Smith, & Ziedenberg, 2003). Disparities in justice system processing are partially attributable to the social and economic 85 conditions that youth face in the United States, including residential segregation, health and wellbeing, and gender inequality. Without the commitment to the structural reform of inequalities, reducing gender and racial/ethnic disparities is unlikely. Second, intervention efforts to reduce disparities must be sensitive to the context of youth’s lives and recognize the social context of gender, race, and ethnicity. These efforts should be broad, diverse, and multifaceted. It is further recommended that these efforts incorporate broad strategies aimed at reducing structural inequalities and include community-based youth-serving organizations and human services agencies, in addition to criminal and juvenile justice agencies. 86 REFERENCES 87 REFERENCES Allison, P. D. (1984). Event history analysis: Regression for longitudinal event data. Newbury Park, CA: Sage. Annie E. Casey Foundation. (2000). Kids Count Data Book: State Profiles of Child Well-Being. Baltimore: Author. Annie E. Casey Foundation. (2001). Kids Count Data Book: State Profiles of Child Well-Being. Baltimore: Author. Annie E. Casey Foundation. (2002). Kids Count Data Book: State Profiles of Child Well-Being. Baltimore: Author. Annie E. Casey Foundation. (2003). Kids Count Data Book: State Profiles of Child Well-Being. Baltimore: Author. Annie E. Casey Foundation. (2004). Kids Count Data Book: State Profiles of Child Well-Being. Baltimore: Author. Armstrong, G. S., & Rodriguez, N. (2005). Effects of individual and contextual characteristics on preadjudication detention of juvenile delinquents. Justice Quarterly, 22, 521-539. Benda, B. B., Toombs, N. J., & Peacock, M. (2002). Ecological factors in recidivism: Survival analysis of boot camp graduates after three years. Journal of Offender Rehabilitation, 35, 63-85. Bilchik, S. (1999). Juvenile Justice: A Century of Change. Washington, DC: Office of Juvenile Justice and Delinquency Prevention. Bishop, D. M. (2005). The role of race and ethnicity in juvenile justice processing. In D. F. Hawkins & K. Kempf-Leonard (Eds.), Our children, their children: Confronting racial and ethnic differences in American juvenile justice (pp. 23-82). Chicago: University of Chicago Press. Bishop, D. M., Leiber, M. J., & Johnson, J. (2010). Contexts of decision making in the juvenile justice system: An organizational approach to understanding minority overrepresentation. Youth Violence and Juvenile Justice, 8, 213-233. 88 Bortner, M. A. (1986). Traditional Rhetoric, Organizational Realities: Remand of Juveniles to Adult Court. Crime & Delinquency, 32, 53-73. Bridges, G. S. (1993). Racial disproportionality in the juvenile justice system. Olympia, WA: Department of Social and Health Services, State of Washington. Brownfield, D., Sorenson, A. M., & Thompson, K. M. (2001). Gang membership, race, and social class: A test of the group hazard and master status hypotheses. Deviant Behavior: An Interdisciplinary Journal, 22, 73-89. Buck-Willison, J. (2010). Juvenile Justice in the United States: Examining Current Practice. Paper presented at the BJS/JRSA 2010 National Conference, Portland, ME. Bureau of Justice Statistics. (2002). National Prosecutors Survey [Census], 2001. Bushway, S., Nieuwbeerta, P., & Blokland, A. (2011). The predictive value of criminal background checks: Do age and criminal history affect time to redemption? Criminology, 49, 27-60. Chesney-Lind, M. (1989). Girls' crime and woman's place: Toward a feminist model of female delinquency. Crime & Delinquency, 35(5), 5-29. Chesney-Lind, M., & Irwin, K. (2008). Beyond bad girls: Gender, violence and hype. New York: Routledge. Dannefer, D., & Schutt, R. K. (1982). Race and juvenile justice processing in court and police agencies. American Journal of Sociology, 87, 1113-1132. DeJong, C., & Jackson, K. C. (1998). Putting race into context: Race, juvenile justice processing, and urbanization. Justice Quarterly, 15, 487-504. DeMaris, A. (2004). Regression with social data: Modeling continuous and limited response variables. Hoboken, NJ: John Wiley & Sons, Inc. DuMouchel, W. H., & Duncan, G. J. (1983). Using Sample Survey Weights in Multiple Regression Analysis of Stratified Sample. Journal of the American Statistical Association, 78, 535-543. 89 Engen, R. L., Steen, S., & Bridges, G. S. (2002). Racial disparities in the punishment of youth: A theoretical and empirical assessment of the literature. Social Problems, 49, 194-220. Farrington, D. P., Loeber, R., Stouthamer-Loeber, M., van Kammen, W. B., & Schmidt, L. (1996). Self-reported delinquency and a combined delinquency seriousness scale based on boys, mothers, and teachers: Concurrent and predictive validity for African-Americans and Caucasians. Criminology, 34, 501-525. Federal Bureau of Investigation. (2001a). Uniform Crime Reporting Program Data [United States]: County-Level Detailed Arrest and Offense Data, 1997. Federal Bureau of Investigation. (2001b). Uniform Crime Reporting Program Data [United States]: County-Level Detailed Arrest and Offense Data, 1998. Federal Bureau of Investigation. (2006a). Uniform Crime Reporting Program Data [United States]: County-Level Detailed Arrest and Offense Data, 1999. Federal Bureau of Investigation. (2006b). Uniform Crime Reporting Program Data [United States]: County-Level Detailed Arrest and Offense Data, 2000. Federal Bureau of Investigation. (2006c). Uniform Crime Reporting Program Data [United States]: County-Level Detailed Arrest and Offense Data, 2001. Feld, B. C. (1991). Justice by geography: Urban, suburban, and rural variations in juvenile justice administration. The Journal of Criminal Law and Criminology, 82, 156-210. Feld, B. C. (1995). The social context of juvenile administration: Racial disparities in an urban juvenile court. In K. Kempf-Leonard, C. E. Pope & W. Feyerherm (Eds.), Minorities in Juvenile Justice (pp. 66-97). Thousand Oaks, CA: Sage. Frazier, C. E., & Lee, S. R. (1992). Reducing juvenile detention rates or expending the official control nets: An evaluation of a legislative reform effort. Crime & Delinquency, 38, 204218. Gharibvand, L., & Liu, L. (2009). Analysis of survival data with clustered events. Cary, NC: SAS Institute, Inc. 90 Goodkind, S., Wallace, J. M., Shook, J. J., Bachman, J., & O'Malley, P. (2009). Are girls really becoming more delinquent? Testing the gender convergence hypothesis by race and ethnicity, 1976–2005. Children and Youth Services Review, 31, 885-895. Grattet, R., Lin, J., & Petersilia, J. (2011). Supervision regimes, risk, and official reactions to parolee deviance. Criminology, 49, 371-399. Griffin, P., Torbet, P., & Szymanski, L. (1998). Trying Juveniles as Adults in Criminal Court: An Analysis of State Transfer Provisions. Washington, DC: Office of Juvenile Justice and Delinquency Prevention. Hepburn, J. R., & Albonetti, C. A. (1994). Recidivism among drug offenders: A survival analysis of the effect of offender characteristics, type of offense, and two types of intervention. Journal of Quantitative Criminology, 10, 158-179. Hox, J. J. (2010). Multivariate analysis: Techniques and applications (2nd ed.). New York: Routledge. Hoytt, E. H., Schiraldi, V., Smith, B. V., & Ziedenberg, J. (2003). Reducing Racial Disparities in Juvneile Detention: Eight Pathways to Juvenile Detention Reform. Baltimore, MD: Annie E. Casey Foundation. Hsia, H. M. (2008). About DMC: A disproportionate minority contact (DMC) chronology: 1988 to date. Retrieved 20 October, 2008, from http://ojjdp.ncjrs.org/dmc/chronology.html Institute for Women's Policy Research. (2000). The Status of Women in the States. Johnson, B. D. (2006). The multilevel context of criminal sentencing: Integrating judge- and county-level influences. Criminology, 44, 259-298. Kempf-Leonard, K. (2007). Minority youths and juvenile justice: Disproportionate minority contact after nearly 20 years of reform efforts. Youth Violence and Juvenile Justice, 5, 71-87. Kempf-Leonard, K., & Sontheimer, H. (1995). The role of race in juvenile justice in Pennsylvania. In K. Kempf-Leonard, C. E. Pope & W. Feyerherm (Eds.), Minorities in juvenile justice (pp. 98-127). Thousand Oaks, CA: Sage. 91 Kirk, D. S. (2006). Examining the Divergence Across Self-report and Official Data Sources on Inferences About the Adolescent Life-course of Crime. Journal of Quantitative Criminology, 22, 107-129. Leiber, M. J. (2002). Disproportionate minority confinement (DMC) of youth: An analysis of state and federal efforts to address the issue. Crime & Delinquency, 48, 3-45. Leiber, M. J., & Fox, K. C. (2005). Race and the impact of detention of juvenile justice decisionmaking. Crime & Delinquency, 51, 470-497. Leiber, M. J., Johnson, J., Fox, K., & Lacks, R. (2007). Differentiating among racial/ethnic groups and its implications for understanding juvenile justice decision making. Journal of Criminal Justice, 35, 471-484. Leiber, M. J., & Johnson, J. D. (2008). Being young and Black: What are their effects on juvenile justice decision making? Crime & Delinquency, 54, 560-581. Leiber, M. J., & Mack, K. Y. (2003). The individual and joint effects of race, gender, and familystatus on juvenile justice decision-making. Journal of Research in Crime and Delinquency, 40, 34-70. Levrant, S., Cullen, F. T., Fulton, B., & Wozniak, J. F. (1999). Reconsidering Restorative Justice: The Corruption of Benevolence Revisited? Crime & Delinquency, 45, 3-27. Liska, A. E., & Chamlin, M. B. (1984). Social Structure and Crime Control Among Macrosocial Units. American Journal of Sociology, 90, 383-395. Liska, A. E., & Tausig, M. (1979). Theoretical interpretations of social class and race differentials in legal decision-making for juveniles. The Sociological Quarterly, 20, 197207. Logan, W. A. (1998). Proportionality and Punishment: Imposing Life without Parole on Juveniles. Wake Forest Law Review, 681-725. Lublin, D., & Voss, D. S. (2001). Federal Elections Project: American University, Washington, DC and the University of Kentucky, Lexington, KY. 92 Maupin, J. R., & Bond-Maupin, L. J. (1999). Detention decision-making in a predominately Hispanic region: Rural and non-rural differences. Juvenile and Family Court Journal, 50, 11-23. McCarthy, B. R. (1990). A micro-level analysis of social structure and social control: Intrastate use of jail and prison confinement. Justice Quarterly, 7, 325-340. Moineddin, R., Matheson, F. I., & Glazier, R. H. (2007). A simulation study of sample size for multilevel logistic regression. BMC Medical Research Methodology, 7. Morris, N. A., & Slocum, L. A. (2010). The Validity of Self-reported Prevalence, Frequency, and Timing of Arrest: An Evaluation of Data Collected Using a Life Event Calendar. Journal of Research in Crime and Delinquency, 47 210-240. Myer, A. J., & Chamlin, M. B. (2011). Aggregation bias and the benign neglect hypothesis. Journal of Crime and Justice, 34, 124-138. Office of Juvenile Justice and Delinquency Prevention (OJJDP). (2008). About DMC: Overview Retrieved 20 October, 2008, from http://ojjdp.ncjrs.org/dmc/about.html Olsen, R. (2009). Weighting NLS Data. Paper presented at the General NLS Worksop, Columbus, OH. Parker, K. F., Stults, B. J., & Rice, S. K. (2005). Racial threat, concentrated disadvantage and social control: Considering the macro-level sources of variation in arrests. Criminology, 43, 1111-1134. Paternoster, R., & Iovanni, L. (1989). The labeling perspective and delinquency: An elaboration of the theory and an assessment of the evidence. Justice Quarterly, 6, 359-394. Piquero, A. R. (2008). Disproportionate Minority Contact. Future of Children, 18, 59-79. Poe-Yamagata, E., & Jones, M. A. (2000). And justice for some. Davis, CA: National Council on Crime and Delinquency. Pope, C. E., & Feyerherm, W. W. (1990a). Minority status and juvenile justice processing: An assessment of the research literature (Part I). Criminal Justice Abstracts, 22, 327-335. 93 Pope, C. E., & Feyerherm, W. W. (1990b). Minority status and juvenile justice processing: An assessment of the research literature (Part II). Criminal Justice Abstracts, 22, 527-542. Pope, C. E., Lovell, R., & Hsia, H. M. (2002). Disproportionate minority confinement: A review of the research literature from 1989 through 2001. Washington, DC: Office of Juvenile Justice and Delinquency Prevention. Raudenbush, S. W. (2008). Many Small Groups. In J. de Leeuw & E. Meijer (Eds.), Handbook of Multilevel Analysis (pp. 207-236). New York: Springer. Redding, R. E. (2002). Adjudicative Competence in the Modern Juvenile Court. Virginia Journal of Social Policy & the Law, 9, 353-410. Richie, B. E. (1995). Compelled to crime: The gendered entrapment of battered black women. London: Routledge. Roberts, J., & Wells, W. (2010). The validity of criminal justice contacts reported by inmates: A comparison of self-reported data with official prison records. Journal of Criminal Justice, 38, 1031-1037. Rodriguez, N. (2007). Juvenile court context and detention decisions: Reconsidering the role of race, ethnicity, and community characteristics in juvenile court processes. Justice Quarterly, 24, 629-656. Rodriguez, N. (2010). The cumulative effect of race and ethnicity in juvenile court outcomes and why preadjudication detention matters. Journal of Research in Crime and Delinquency, 47, 391-413. Sampson, R. J., & Laub, J. H. (1993). Structural variations in juvenile court processing: Inequality, the underclass, and social control. Law & Society Review, 27, 285-311. Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy. Science, 277, 918-924. Sampson, R. J., Raudenbush, S. W., & Earls, F. (1999). Beyond Social Capital: Spatial Dynamics of Collective Efficacy for Children. American Sociological Review, 64, 633660. 94 Samuels, J. E., Dwyer, A. M., Halberstadt, R., & Lachman, P. (2011). Collecting DNA from Juveniles. Washington, DC: Urban Institute. SAS Institute, I. (2008). The GLIMMIX Procedure: SAS/STAT User's Guide. Cary, NC: SAS Institute, Inc.,. Secret, P. E., & Johnson, J. B. (1997). The effect of race on juvenile, justice decision making in Nebraska: Detention, adjudication, and disposition. Justice Quarterly, 14, 445-478. Short, J., & Sharp, C. (2005). Disproportionate minority contact in the juvenile justice system. Washington, DC: Child Welfare League of America. Sickmund, M. (2003). Juveniles in court. http://www.ncjrs.gov/html/ojjdp/195420/contents.html Singer, J. D., & Willett, J. B. (2003). Applied Longitudinal Data Analysis: Modeling Change and Event Occurence. Oxford: Oxford University Press. Snyder, H. N., & Sickmund, M. (2006). Juvenile Offenders and Victims: 2006 National Report. Washington, DC: Office of Juvenile Justice and Delinquency Prevention. Stahl, A., & Kang, W. (2005). Easy Access to State and County Juvenile Court Case Counts, 2001. Retrieved January 26, 2013, from http://www.ojjdp.gov/ojstatbb/ezaco/ Stahl, A., Kang, W., & Wan, Y. (2002). Easy Access to State and County Juvenile Court Case Counts, 1999. Retrieved January 26, 2013, from http://www.ojjdp.gov/ojstatbb/ezaco/ Stahl, A., Kang, W., & Wilt, D. (2003). Easy Access to State and County Juvenile Court Case Counts, 2000. Retrieved March 26, 2012, from http://www.ojjdp.gov/ojstatbb/ezaco/ Stahl, A., McGlynn, E., & Wan, Y. (2000). Easy Access to State and County Juvenile Court Case Counts, 1997. Retrieved January 26, 2013, from http://www.ojjdp.gov/ojstatbb/ezaco/ Stahl, A., & Wan, Y. (2001). Easy Access to State and County Juvenile Court Case Counts, 1998. Retrieved January 26, 2013, from http://www.ojjdp.gov/ojstatbb/ezaco/ Steffensmeier, D., & Allan, E. (1996). Gender and crime: Toward a gendered paradigm of female offending. Annual Review of Sociology, 22, 459-487. 95 Stevens, T., Morash, M., & Chesney-Lind, M. (2011). Are girls getting tougher, or are we tougher on girls? Probability of arrest and juvenile court oversight in 1980 and 2000. Justice Quarterly, 28, 719-744. Tapia, M. (2010). Untangling race and class effects on juvenile arrests. Journal of Criminal Justice, 38(3), 255-265. doi: 10.1016/j.jcrimjus.2010.03.002 Thornberry, T. P., & Krohn, M. D. (2000). The self-report method for measuring delinquency and crime. In D. Duffee, R. D. Crutchfield, S. Mastrofski, L. Mazerolle, D. McDowall & B. Ostrom (Eds.), CJ 2000: Innovations in measurement and analysis (pp. 33-83). Washington, DC: National Institute of Justice. Tracy, P. E. (2002). Decision making and juvenile justice: An analysis of bias in case processing. Westport, CT: Greenwood Press US Census Bureau. (2004). National estimates by demographic characteristics—Single year of age, sex, race, and Hispanic Origin. Washington, DC: Population Estimates Program, Population Division, US Census Bureau. Winship, C., & Radbill, L. (1994). Sampling Weights and Regression Analysis. Sociological Methods & Research, 23, 230-257. 96